Adaptive recommendations systems

Abstract

An adaptive recommendation system and a mobile adaptive recommendation system are disclosed. The adaptive recommendation system and the mobile adaptive recommendation system include algorithms for monitoring user usage behaviors across a plurality of usage behavior categories associated with a computer-based system, and generating recommendations based on inferences on user preferences and interests. Privacy control functions and compensatory functions related to insincere usage behaviors can be applied. adaptive recommendation delivery can take the form of visual-based or audio-based formats.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

for a given affinity level, affinity_ij, where 0<affinity_ij≦1, for Node i and Node j, and where 1 is the strongest possible relationship, excluding the identity relationship, and 0 implies no direct relationship. “Scaling factor” is a number between 0 and 1 chosen to normalize the degrees of separation for the fuzzy network consistent with the specific definition and distributions of the affinities between nodes in the fuzzy network.

For example, if an affinity of 1.0 is defined as the identity function, then the scaling factor could be set to 0 so that the degree of separation of an affinity of 1.0, the identity degree of separation, is defined as 0. Alternatively, if an affinity of 0 is defined as no relationship whatsoever, then the degree of separation should logically be greater than 1.0, so the scaling factor may be chosen as a number up to and including 1.0.

The scaling factor may be a function of the specific distribution of the intensity level of affinities in a fuzzy network. These intensities may be linear across the range of 0 and 1, or may be nonlinear. If, for example, the mean intensity is defined at 0.5, then the scaling factor for the fractional degree of separation calculation could be set at 0.5.

In summary, for fuzzy networks, the general case of “distance” relationship between two directly linked nodes is a fractional degree of separation. More generally, the degree of separation between any two nodes in a fuzzy network is defined as the minimum of the degrees of separation (which may be calculated on the basis of a specific directional orientation of relationships among the nodes) among all possible paths between the two nodes, where the degrees of separation between any two nodes along the path may be fractional. Where a network has multiple relationships between nodes, multiple potentially fractional degrees of separation may be calculated between any two nodes in the network.

For convenience, the term fractional degrees of separation may be shortened to the acronym “FREES” (FRactional degrEEs of Separation)—as in, say, “Node X is 2.7 FREES from Node Y.” FIG. 22 represents a fuzzy, a-directional network 610 and the associated degrees of separation 622 (using a scaling factor of 0.5) from Node X.

The degree of separation within the fuzzy or non-fuzzy network may be calculated and displayed on demand for any two nodes in the network. All nodes within a specified degree of separation of a specified node may be calculated and displayed. Optionally, the associated fractional degrees of separation between the base node and the nodes within the specified fractional degrees of separation may be displayed.

FIG. 23 depicts a subset 620 of the non-fuzzy a-directional network 600 of FIG. 21, according to the prior art, where the subset 620 is defined as all nodes within two degrees of separation of Node X. FIG. 24 depicts a subset 630 of the fuzzy a-directional network 610 of FIG. 22, according to some embodiments, where the subset 630 is defined as all nodes within 2.5 degrees of separation of Node X.

The degrees of separation among nodes in a fuzzy network may be described by a fractional degrees-of-separation (FREES) matrix. For a network with N nodes, n₁ . . . n_un, the degree-of-separation matrix will have N rows and columns. Each cell of the matrix contains a number that describes the degree of separation between the associated two nodes, i_n and n_o. For non-fuzzy networks, each cell will contain an integer value; for fuzzy networks each cell of the FREES matrix may contain non-integer values. For both fuzzy and non-fuzzy networks, the diagonal of the affinity matrix will be 0's—the identity degree of separation. If a fuzzy network is described by multiple affinity matrices, then the multiple affinity matrices correspond on a one-to-one basis with multiple associated FREES matrices.

The degrees of separation for networks with multiple relationship types, whether for fuzzy or non-fuzzy networks, may be calculated as a function across some or all of the relationship types. For example, such a function could be the minimum of degree of separation from Node X to Node Y of all associated relationship types, or the function could be an average, or any other relevant mathematical function.

According to some embodiments, the adaptive recombinant system 800 of FIG. 18 employs fractional degrees of separation in its syndication and recombination operations, as described in more detail, below.

Fuzzy Network Subsets and Adaptive Operators

The adaptive recombinant system 800 of FIG. 18 includes fuzzy network operators 820. The fuzzy network operators 820 may manipulate one or more fuzzy or non-fuzzy networks. Some of the operators 820 may incorporate usage behavioral inferences associated with the fuzzy networks that the operators act on, and therefore these operators may be termed “adaptive fuzzy network operators.” The fuzzy network operators 820 may apply to any fuzzy network-based system structure, including fuzzy content network system structures, described further below.

FIG. 20 is a block diagram depicting some fuzzy network operators 820, also called functions or algorithms, used by the adaptive recombinant system 800. A selection operator 822, a union operator 824, an intersection operator 826, a difference operator 828, and a complement operator 832 are included, although additional logical operations may be used by the adaptive recombinant system 800. Additionally, the fuzzy network operators 820 include a resolution function 834, which is used in conjunction with one or more of the operators in the fuzzy network operators 820.

A selection operator 822, which selects subsets of networks, may designate the selected network subsets based on degrees of separation. For example, subsets of a fuzzy network may be selected from the neighborhood, designated by a FREES metric, around a given node, say Node X. The selection may take the form of selecting all nodes within the designated network neighborhood, or all the nodes and all the associated links as well within the designated network neighborhood, where the network neighborhood is defined as being within a certain degree of separation from Node X. A non-null fuzzy network subset will therefore contain at least one node, and possibly multiple nodes and relationships.

Two or more fuzzy network subsets may then be operated on by network operations such as union, intersection, difference, and complement, as well as any other Boolean set operators. An example is an operation that outputs the intersection (intersection operator 826) of the network subset defined by the first degree or less of separation from Node X and the network subset defined by the second or less degree of separation from Node Y. The operation would result in the set of nodes and relationships common to these two network subsets, with special auxiliary rules optionally applied to resolve duplicative relationships as will be explained below.

The network operations may apply explicitly to fractional degrees of separation. For example, the union operator 824 may be applied to the network subset defined by half a degree of separation (0.5) or less from Node X and the network subset defined as 2.4 degrees of separation or less from Node Y. The union of the two network subsets results in a unique set of nodes and relationships that are contained in both of these network subsets. Special auxiliary rules may optionally be applied to resolve duplicative relationships. Fuzzy network operations may also be chained together, e.g., a union of two network subsets intersected with a third network subset, etc.

The fuzzy network operators 820 may have special capabilities to resolve the situation in which union 824 and intersection 826 operators define common nodes, but with differing relationships or values of the relationships among the common nodes. The fuzzy network intersection operator 826, Fuzzy_Network_Intersection, may be defined as follows:
Z=Fuzzy_Network_Intersection(X, Y, W)
where X, Y, and Z are network subsets and W is the resolution function 834. The resolution function 834 designates how duplicative relationships among nodes common to fuzzy network subsets X and Y are resolved.

Specifically, the fuzzy network intersection operator 826 first determines the common nodes of network subsets X and Y, to form a set of nodes, network subset Z. The fuzzy network intersection operator 826 then determines the relationships and associated relationship value and indicators uniquely deriving from X among the nodes in Z (that is, relationships that do not also exist in Y), and adds them into Z (attaching them to the associated nodes in Z). The operator then determines the relationships and relationship indicators and associated values uniquely deriving from Y (that is, relationships that do not also exist in X) and applies them to Z (attaching them to the associated nodes in Z).

For relationships that are common to X and Y, the resolution function 834, is applied. The resolution function 834 may be any mathematical function or algorithm that takes the relationship values of X and Y as arguments, and determines a new relationship value and associated relationship indicator.

The resolution function 834, Resolution_Function may be a linear combination of the corresponding relationship value of X and the corresponding relationship value of Y, scaled accordingly. For example:
Resolution_Function (X_RV, Y_RV)=(c₁*X_RV+c₂*Y_RV)/(c₁+c₂)
where X_RV and Y_RV are relationship values of X and Y, respectively, and c₁ and c₂ are coefficients. If c₁=1, and c₂=0, then X_RV completely overrides Y_RV. If c₁=0 and c₂=1, then Y_RV completely overrides X_RV. If c₁=1 and c₂=1, then the derived relationship is a simple average of X_RV and Y_RV. Other values of c₁ and c₂ may be selected to create weighted averages of X_RV and Y_RV. Nonlinear combinations of the associated relationships values, scaled appropriately, may also be employed.

The Fuzzy_Network_Union operator 824 may be derived from the Fuzzy_Network_Intersection operator 826, as follows:
Z=Fuzzy_Network_Union(X, Y, W)
where X, Y and Z are network subsets and W is the resolution function 834. Accordingly,
Z=Fuzzy_Network_Intersection(X, Y, W)+(X−Y)+(Y−X)
That is, fuzzy network unions of two network subsets may be defined as the sum of the differences of the two network subsets (the nodes and relationships that are uniquely in X and Y, respectively) and the fuzzy network intersection of the two network subsets. The resulting network subset of the difference operator contains any unique relationships between nodes uniquely in an originating network subset and the fuzzy network intersection of the two subsets. These relationships are then added to the fuzzy network intersection along with all the unique nodes of each originating network subset, and all the relationships among the unique nodes, to complete the resulting fuzzy network subset.

It should be noted that, unlike the corresponding classic set operators, the fuzzy network intersection 826 and union 824 operators are not necessarily mathematically commutative—that is, the order of the operands may matter. The operators will be commutative if the resolution function or algorithm is commutative.

For the adaptive recombinant system 800, the resolution function 834 that applies to operations that combine multiple networks may incorporate usage behavioral inferences related to one or all of the networks. The resolution function 834 may be instantiated directly by the adaptive recommendations function 240 (FIG. 18), or the resolution function 834 may be a separate function that invokes the adaptive recommendations function. The resulting relationships in the combined network will therefore be those that are inferred by the system to best reflect the collective usage histories and preference inferences of the predecessor networks.

For example, where one of the predecessor networks was used by larger numbers of individuals, or by individuals that members of communities or affinity groups that are inferred to be best informed on the subject of the associated content, then the resolution function 834 may choose to preferentially weight the relationships of that predecessor network higher versus the other predecessor networks. The resolution function 834 may use any or all of the usage behaviors 270, along with associated user segmentations and affinities obtained during usage behavior pre-processing 204 (see FIG. 3C), as illustrated in FIG. 8 and Table 1, and combinations thereof, to determine the appropriate resolution of common relationships and relationship values among two or more networks that are combined into a new network.

Fuzzy Network Metrics

Special metrics may be used to measure the characteristics of fuzzy networks and fuzzy network subsets. For example, these metrics may provide measures associated with the relationship of a network node or object to other parts of the network, and relative to other network nodes or objects. A metric may be provided that indicates the degree to which nodes are connected to the rest of the network. This metric may be calculated as the sum of the affinities of first degree or less separated directionally distinct relationships or links. The metric may be called a first degree connectedness parameter for the specific node.

The first degree connectedness metric may be generalized for zeroth to N^th degrees of connectedness as follows. The zeroth degree of connectedness is, by definition, zero. The N^th degree of connectedness of Node X is the sum of the affinities among all nodes within N degrees of separation of Node X. For fuzzy networks, N may not necessarily be an integer value. The connectedness parameters may be indexed to provide a convenient relative metric among all other nodes in the network.

As an example, in the fuzzy network 630 of FIG. 24, the first degree of connectedness of Node X is determined by summing all relationship values associated with Node X to objects within a fractional degree of separation, defined here as less than 1.5 degrees of separation. Four nodes which have less than 1.5 degrees of separation from Node X are shaded in FIG. 24. By summing the affinities of the four nodes (0.9+0.4+0.3+0.3), a connectedness metric of 1.9 for Node X is obtained.

In networks in which there are multiple types of relationships among nodes, there may be multiple connectedness measures for any specific Node X to the subset of the fuzzy network specified by a degree of separation, N, from X.

In summary, connectedness for a specific Node X may have variations associated with relationship type, the specified directions of the relationships selected for computation, and the degree of separation from the Node X. The general connectedness metric function may be defined as follows:
Connectedness(Node X, T, D, S)
where T is the relationship indicator type, D is the relationship direction, and S is the degree of separation. The Connectedness metric may be normalized to provide a convenient relative measure by indexing the metric across all nodes in a network.

A metric of the popularity of the network nodes or objects, or popularity metric, may also be provided. The fuzzy or non-fuzzy network may be implemented on a computer system, or on a network of computer systems such as the Internet or on an Intranet. The system usage behavioral patterns of users of the fuzzy network may be recorded. The number of accesses of particular nodes or objects of a fuzzy to non-fuzzy network may be recorded. The accesses may be defined as the actual display of the node or object to the user or the accesses may be defined as the display of information associated with the node or object to user, such as access to an associated editorial review. In some of these embodiments, the popularity metric may be based on the number of user accesses of the associated node or object, or associated—information. The popularity metric may be calculated for prescribed time periods. Popularity may be recorded for various user segments, in addition to, or instead of, the usage associated with the entire user community. The usage traffic may be stored so that popularity trends over time may be accessed. In the most general case, popularity for a specific Node X will have variations by user segments and time periods. A general popularity function may therefore be represented as follows:
Popularity(Node X, user segment, time period)
The Popularity metric may be normalized to provide a convenient relative measure by indexing the metric across all nodes in a network.

Metrics may be generated that go beyond the connectedness metrics, to provide information on additional characteristics associated with a node or object within the network relative to other nodes or objects in the network. A metric that combines aspects of connectedness and popularity measures, an influence metric, may be generated. The influence metric may provide a sense of the degree of importance or “influence” a particular node or object has within the fuzzy network.

The influence metric for Node X is calculated by adding the popularity of Node X to a term that is the sum of the popularities of the nodes or objects separated by one degree of separation or less from Node X, weighted by the associated affinities between Node X and each associated related node. The term associated with the weighted average of the popularities of the first degree of separation nodes of Node X is scaled by a coefficient. This coefficient may be defined as the inverse of the first degree connectedness metric of Node X.

For fuzzy networks with directionally distinct relationships and affinities, the influence metric may be calculated based only on the first degree affinities or less for relationships that are oriented in a particular direction. For example, influence may be calculated based on all relationships directed to Node X (as opposed to those directed away from Node X).

A generalized influence metric may also be provided, where the N^th degree of influence of node or object X is defined as the popularity of Node X added to a term that is the weighted average of the popularities of all nodes within N degrees of separation from Node X (where N may be a non-integer, implying a fractional degree of separation). The weights for each node may be a function of the affinities of the shortest path between Node X and the associated node. The generalized influence metric may be a multiplicative function, that is, the affinities along the path from Node X to each node within N degrees separation are multiplied together and then multiplied by the popularity of the associated node. Or, the metric may be a summation function, or any other mathematical function that combines the affinities along the associated network path. The generalized influence metric may be specified as a recursive function, satisfying the following difference equations and “initial condition”:
Nth Degree of Influence(Node X)=(N−1)th Degree of Influence(Node X)+Influence of Nodes of N Degrees of Separation from Node X.   (1)
Zeroth Degree of Influence(Node X)=Popularity(Node X)   (2)

Where there are directionally distinct affinities, the affinities that are multiplied, summed, or otherwise mathematically operated on, between Node X and all other nodes within a directionally distinct degree of separation (where the degree of separation may be fractional), may be of relationships with a selected directional orientation. The relationship direction term (D, in the connectedness metric function, above, may be scaled by the N^th degree of connectedness (of a given directional orientation) of Node X.

The zeroth degree of influence may be defined as just the popularity of Node X. The N^th degree of influence is indexed to enable convenient comparison of influence among nodes or objects in the network. Where there are multiple types of relationships between any two nodes in the network, influence may be calculated for each type of relationship. An influence metric may also be generated that averages (or applies any other mathematical function that combines values) across multiple influence metrics associated with two or more relationship types.

FIG. 25 illustrates an example of influence calculations, using a multiplicative scaling method, in accordance with some embodiments. Fuzzy network 650 depicts Node X having a popularity metric 652 of “10”. The zeroth degree of influence of Node X is therefore just “10.” The first degree of influence of Node X is calculated by multiplying the affinities or relationship indicators associated with relationships from Node X and nodes that are within one degree of separation, by the associated popularities, for example 654, of these nodes. The first degree of influence of Node X is thus the popularity of Node X (10) plus the sum of the popularities of the nodes within one degree of separation, multiplied by their associated relationship values. In FIG. 25, the first degree of influence of Node X is:
10+(45*0.3)+(23*0.9)+(85*0.4)+(42*0.3)=90.8

The second degree of influence of Node X is calculated as the first degree of influence of Node X (already calculated) plus the influence contributed by each node that is two degrees of separation from Node X, and may likewise be calculated, as follows:
90.8+(20*0.4*0.9)+(30*0.8*0.3)+(150*0.2*0.3)+(80*0.6*0.3)+(90*0.9*0.3)+(5*0.4*0.3)+(20*0.5*0.3)+(200*0.8*0.3)=204.5
Table 3 lists the first degree affinities, second degree affinities, popularity, calculated influence, and cumulative influence, relative to Node X, for the fuzzy network 650 of FIG. 25.

TABLE 3
Affinity, popularity, & influence data for fuzzy network 650.
					cum.
Node	1st ° affinities	2nd ° affinities	popularity	influence	influence
0th	1		10	10	10
1st	0.4		85	34
1st	0.9		23	20.7
1st	0.3		42	12.6
1st	0.3		45	13.5	90.8
2nd	0.9	0.4	20	7.2
2nd	0.3	0.8	30	7.2
2nd	0.3	0.2	150	9
2nd	0.3	0.9	90	24.3
2nd	0.3	0.8	200	48
2nd	0.3	0.5	20	3
2nd	0.3	0.4	5	0.6
2nd	0.3	0.6	80	14.4	204.5

In summary, the influence metric for Node X may have variations associated with a specific relationship indicator type, a specific direction of relationships for the relationship indicator type, a degree of separation from Node X, and a scaling coefficient that tunes the desired degradation of weighting for nodes and relationships increasingly distant from Node X. The metric function may therefore be represented as follows:

Influence(Node X, relationship indicator type or types, relationship direction, degree of separation, affinity path function, scaling coefficient). The influence metric may be normalized to provide a convenient relative measure by indexing the metric across all nodes in a network. Metrics associated with nodes of fuzzy networks, such as popularity, connectedness, and influence, may be displayed in textual or graphical forms to users of the fuzzy network-based system. The adaptive recombinant system 800 of FIG. 18 may use connectedness, popularity, and influence metrics in order to syndicate and recombine structural subsets 280 of the adaptive system 100.

Fuzzy Network Syndication and Combination

The adaptive recombinant system 800 of FIG. 18 is able to syndicate and combine structural subsets 280 of the structural aspect 210 (where a structural subset 280 may contain the entire structural aspect 210). The structural subsets 280, which are fuzzy networks, in some embodiments, may be syndicated in whole or in part to other computer networks, physical computing devices, or in a virtual manner on the same computing platform or computing network. Although the adaptive recombinant system 800 is not limited to generating structural subsets which are fuzzy networks, the following figures and descriptions, used to illustrate the concepts of syndication and recombination, feature fuzzy networks. Designers of ordinary skill in the art will recognize that the concepts of syndication and recombination may be generalized to other types of networks.

FIG. 26 illustrates a fuzzy network 500, including a subset 502 of fuzzy network 500. The subset 502 includes three objects 504, 506, and 508, designated as shaded in FIG. 26. The subset 502 also includes associated relationships (arrows) and relationship indicators (values) among the three objects. The separated, or syndicated, subset of the network 502 yields a fuzzy network (subset) 510.

The adaptive system 100 of FIG. 1 may operate in a fuzzy network environment, such as the fuzzy network 500 of FIG. 26. In FIG. 27, an adaptive system 100C includes a structural aspect 210C that is a fuzzy network 500. Thus, adaptive recommendations 250 generated by the adaptive system 100C are also structural subsets that are themselves fuzzy networks.

Similarly, the adaptive recombinant system 800 of FIG. 18 may operate in a fuzzy network environment. In FIG. 28, an adaptive recombinant system 800C includes the adaptive system 100C of FIG. 27. Thus, the adaptive recombinant system 800C may perform syndication and recombination operations, as described above, to generate structural subsets that are fuzzy networks.

The structural aspect 210 of adaptive system 100 may be comprised of multiple structures, comprising network-based structures, non-network-based structures, or combinations of network-based structures and non-network-based structures. In FIG. 29, the structural aspect 210C includes multiple network-based structures and non-network-based structures. The multiple structures of 210c may reside on the same computer system, or the structures may reside on separate computer systems.

FIG. 30 depicts a fuzzy network 520 syndicated to, and combined with, a fuzzy network 530. Fuzzy network 520 contains objects 522 and 532. Fuzzy network 530 contains identical objects 522 and 532, which are depicted by shading.

The adaptive recombinant system 800 may determine objects, such as the objects 522 and 532 of FIG. 30, to be identical through the object evaluation function 830 (see FIG. 18). The object evaluation function 830 may include a global or distributed management of unique identifiers for each distinct object. These identifiers may be managed directly by the adaptive recombinant system 800, or the adaptive recombinant system may rely on an auxiliary system, such as an operating system or another application, to manage object identification. Alternatively, the identity relationship between objects may be determined though comparisons of information associated with the object or through a comparison of the actual object content (information 232) itself. Associated content may be compared using text, graphic, video, or audio matching techniques. A threshold may be set in determining identicalness between two objects that is less than perfect matching to compensate for minor differences, versions, errors, or other non-substantive differences between the two objects, or to increase the speed of object comparisons by sacrificing some level of accuracy in identification of identicalness.

The combination of the fuzzy network 520 and the fuzzy network 530 yields fuzzy network 540. In the fuzzy network 540, relationships that were unique in networks 520 and 530 are maintained. Where relationships or relationship indicators are common in fuzzy networks 520 and 530, the resolution function 834 (FIG. 20) is applied to create the relationship and associated relationship indicators in the newly formed fuzzy network 540.

For example, object 522 and object 532 are part of both fuzzy network 520 and fuzzy network 530. A relationship 521 is depicted between object 522 and object 532 in the fuzzy network 520, while a relationship 531 is depicted between object 522 and object 532 in the fuzzy network 530. Where relationships 521 and 530 are of the same type, the resulting relationship indicators 541 in the newly created fuzzy network 540 is an average of relationship indicators 521 and 531. That is, for determining the relationship between objects 522 and 532 in the fuzzy network 540, the resolution function 834 is a simple average function. In general, the resolution function 834 may be any mathematical function or algorithm that takes as input two numbers between 0 and 1 inclusive, and outputs a number between 0 and 1 inclusive.

The resolution function 834 may be derived from algorithms that apply appropriate usage behavior inferences. As a simple example, if the relationship value and associated indicator of one network has been derived from the usage behaviors of highly informed or expert users, then this may have more weighting than the relationship value and associated indicator of a second network for which the corresponding relationship value was based on inferences associated with the usage behaviors of a relatively sparse set of relatively uniformed users.

New relationships and associated relationship indicators that do not exist in originating fuzzy networks may also be generated by the adaptive recombinant system 800 upon fuzzy network creation. The adaptive recommendations function 240 may be invoked directly to effect such relationship modifications, or it may be invoked in conjunction with fuzzy network maintenance functions.

For example, in FIG. 30, the fuzzy network 540 also contains a new relationship and associated relationship indicators 542 that did not explicitly exist in predecessor fuzzy networks 520 or 530. This is an example of the invocation of the adaptive recommendations function 240 being used by the adaptive recombinant system 800 in conjunction with the fuzzy network operators 820, to automatically or semi-automatically add a new relationship and associated relationship indicators to the newly created fuzzy network.

The determination of a new relationship may be based on fuzzy network structural, usage, or content characteristics, and associated inferencing algorithms. For example, in predecessor network 530, the traffic patterns, combined with the organization of user referenced subsets of 530, as one example, may support adding the relationship 542 in the new network 540 that did not exist in the predecessor networks. The same procedure may be used to delete existing relationships (which may be alternatively viewed as just equivalent to setting a relationship indicator to “0”), as desired. The algorithms for modifying relationships and relationship indicators, including adding and deleting relationships, may incorporate global considerations with regard to optimizing the overall topology of the fuzzy network by creating effective balance of relationships among objects to maximize overall usability of the network.

FIGS. 31A-31D illustrate the general approaches associated with fuzzy network syndication and combination by the adaptive recombinant system 800, according to some embodiments. FIG. 31A illustrates a hypothetical starting condition, and depicts three individuals or organizations, 350, 355, 360. It should be understood that where the term “organization” is used, it may imply a single individual or set of individuals that may or may not be affiliated with any specific organization. A fuzzy network 565 is used solely by, or resides within an organization 550. A fuzzy network 570 is used solely by, or resides within an organization 555. An organization 560 does not have a fuzzy network initially.

In FIG. 31B, a subset of the fuzzy network 565 is selected to form fuzzy network 565a. Fuzzy network 565a is then syndicated to the organization 555, as fuzzy network 565b. The organization 555 then syndicates the fuzzy network 565b to the organization 560, as fuzzy network 565c. Fuzzy network 565a is a subset of fuzzy network 565, fuzzy network 565b is syndicated from fuzzy network 565a, and fuzzy network 565c is syndicated from fuzzy network 565b. Thus, FIG. 31B illustrates how fuzzy networks, or subsets of networks, may be indefinitely syndicated among individuals or organizations by the adaptive recombinant system 800.

In FIG. 31C, the fuzzy network 565b in the organization 555, which was syndicated from fuzzy network 565 (FIG. 31B), may be combined with the fuzzy network 570 already present in organization 555 (FIG. 31A), to form new fuzzy network 575. Fuzzy network 575 is then syndicated to the organization 560 as fuzzy network 575a. Thus, FIG. 31C illustrates how fuzzy networks, or subsets of networks, may be combined to form new fuzzy networks.

In FIG. 31D, the organization 550 includes fuzzy network 565 (FIG. 31A) and fuzzy network 565a, a subset of fuzzy network 565 (FIG. 31B). Fuzzy network 575a, in the organization 560, is syndicated to the organization 550, as fuzzy network 575b, such that organization 550 has three fuzzy networks 565, 565a, and 575b. Fuzzy networks 565 and 575b may be combined, as shown, to form new fuzzy network 580 in the organization 550.

The adaptive recombinant system 800 of FIG. 18 is capable of generating subsets, combining, and syndicating networks, as depicted in FIGS. 31A-31D. The adaptive recombinant system may indefinitely enable sub-setting of fuzzy networks, syndicating them to one or more destination fuzzy networks, and enabling the syndicated fuzzy networks to be combined with one or more fuzzy networks at the destinations. At each combination step, the resolution function 834, through application of the adaptive recommendations function 240 and network maintenance functions, may be invoked to create and update the structural aspect 210, as appropriate.

The adaptive recombinant system 800 may efficiently support multiple adaptive systems 100, without reproducing the components used to support syndication and recombination for each adaptive system. FIG. 32, for example, includes three adaptive systems 100P, 100Q, and 100R. These three adaptive systems share the syndication function 810, the fuzzy network operators 820, and the object evaluation function 830. In addition, it should be remembered that multiple fuzzy networks may exist inside an adaptive system 100, which may in turn form part of the adaptive recombinant system 800.

In addition to the resolution function 834, the adaptive recombinant system 800 may use the object evaluation function 830, to evaluate the “fitness” of the recombined fuzzy networks. The object evaluation function 830 may be completely automated, or it may incorporate explicit human judgment. The networks that are evaluated to be most fit are then recombined among themselves, to create a new generation of fuzzy networks.

The adaptive recombinant system 800 may also create random structural changes to enhance the diversity of the fuzzy networks in the next generation. Or, the adaptive recombinant system 800 may use explicit non-random-based rules to enhance the diversity of the fuzzy networks in the next generation. Preferably, the inheritance characteristics from generation to generation of adaptive recombinant fuzzy networks may be that of acquired traits (Lamarckian). Or, the inheritance characteristics from generation to generation of adaptive recombinant fuzzy networks may be that of non-acquired, or random mutational, traits (Darwinian). For the Lamarckian embodiments, the acquired traits include any structural adaptations that have occurred through system usage, syndications, and combinations.

Through application of these multi-generational approaches, fuzzy networks are able to evolve against the selection criteria that are provided. The fitness selection criteria may be determined through inferences associated with fuzzy network usage behaviors, and may itself co-evolve with the generations of adaptive fuzzy networks.

Fuzzy Content Network

In some embodiments, the structural aspect 210 of the adaptive system 100 and of the adaptive recombinant system 800, as well as the respective structural subsets 280 generated by the adaptive recommendations function 240, are networks of a particular form, a fuzzy content network. A fuzzy content network 700 is depicted in FIG. 33.

The fuzzy content network 700, including content sub-networks 700a, 700b, and 700c. The content network 700 includes “content,” “data,” or “information,” packaged in modules known as objects 710.

The content network 700 employs features commonly associated with “object-oriented” software to manage the objects 710. That is, the content network 700 discretizes information as “objects.” In contrast to typical procedural computer programming structures, objects are defined at a higher level of abstraction. This level of abstraction allows for powerful, yet simple, software architectures.

One benefit to organizing information as objects is known as encapsulation. An object is encapsulated when only essential elements of interaction with other objects are revealed. Details about how the object works internally may be hidden. In FIG. 34A, for example, the object 710 includes meta-information 712 and information 714. The object 710 thus encapsulates information 714.

Another benefit to organizing information as objects is known as inheritance. The encapsulation of FIG. 34A, for example, may form discrete object classes, with particular characteristics ascribed to each object class. A newly defined object class may inherit some of the characteristics of a parent class. Both encapsulation and inheritance enable a rich set of relationships between objects that may be effectively managed as the number of individual objects and associated object classes grows.

In the content network 700, the objects 710 may be either topic objects 710t or content objects 710c, as depicted in FIGS. 34B and 34C, respectively. Topic objects 710t are encapsulations that contain meta-information 712t and relationships to other objects (not shown), but do not contain an embedded pointer to reference associated information. The topic object 710t thus essentially operates as a “label” to a class of information. The topic object 710 therefore just refers to “itself” and the network of relationships it has with other objects 710.

Content objects 710c, as shown in FIG. 34C, are encapsulations that contain meta-information 36c and relationships to other objects 710 (not shown). Additionally, content objects 710c may include either an embedded pointer to information or the information 714 itself (hereinafter, “information 714”).

The referenced information 714 may include files, text, documents, articles, images, audio, video, multi-media, software applications and electronic or magnetic media or signals. Where the content object 714c supplies a pointer to information, the pointer may be a memory address. Where the content network 700 encapsulates information on the Internet, the pointer may be a Uniform Resource Locator (URL).

The meta-information 712 supplies a summary or abstract of the object 710. So, for example, the meta-information 712t for the topic object 710t may include a high-level description of the topic being managed. Examples of meta-information 712t include a title, a sub-title, one or more descriptions of the topic provided at different levels of detail, the publisher of the topic meta-information, the date the topic object 710t was created, and subjective attributes such as the quality, and attributes based on user feedback associated with the referenced information. Meta-information may also include a pointer to referenced information, such as a uniform resource locator (URL), in one embodiment.

The meta-information 712c for the content object 710c may include relevant keywords associated with the information 714, a summary of the information 714, and so on. The meta-information 712c may supply a “first look” at the objects 710c. The meta-information 712c may include a title, a sub-title, a description of the information 714, the author of the information 714, the publisher of the information 714, the publisher of the meta-information 712c, and the date the content object 710c was created, as examples. As with the topic object 710t, meta-information for the content object 710c may also include a pointer.

In FIG. 33, the content sub-network 700a is expanded, such that both content objects 710c and topic objects 710t are visible. The various objects 34 of the content network 700 are interrelated by degrees, using relationships 716 (unidirectional and bidirectional arrows) and relationship indicators 716 (values). (The relationships 716 and relationship indicators 718 are similar to the relationships and relationship indicators depicted in FIG. 13A, above, as well as other figures included herein.) Each object 710 may be related to any other object 710, and may be related by a relationship indicator 718, as shown. Thus, while information 714 is encapsulated in the objects 710, the information 714 is also interrelated to other information 714 by a degree manifested by the relationship indicators 718.

The relationship indicator 718 is a numerical indicator of the relationship between objects 710. Thus, for example, the relationship indicator 718 may be normalized to between 0 and 1, inclusive, where 0 indicates no relationship, and 1 indicates a subset relationship. Or, the relationship indicators 718 may be expressed using subjective descriptors that depict the “quality” of the relationship. For example, subjective descriptors “high,” “medium,” and “low” may indicate a relationship between two objects 710.

The relationship 716 between objects 710 may be bidirectional, as indicated by the double-pointing arrows. Each double-pointing arrow includes two relationship indicators 718, one for each “direction” of the relationships between the objects 710.

As FIG. 33 indicates, the relationships 716 between any two objects 710 need not be symmetrical. That is, topic object 710t 1 has a relationship of “0.3” with content object 710c2, while content object 710c2 has a relationship of “0.5” with topic object 710t1. Furthermore, the relationships 716 need not be bi-directional—they may be in one direction only. This could be designated by a directed arrow, or by simply setting one relationship indicator 718 of a bi-directional arrow to “0,” the null relationship value.

The content networks 700A, 700B, 700C may be related to one another using relationships of multiple types and associated relationship indicators 718. For example, in FIG. 33, content sub-network 700a is related to content sub-network 700b and content sub-network 700c, using relationships of multiple types and associated relationship indicators 718. Likewise, content sub-network 700b is related to content sub-network 700a and content sub-network 700c using relationships of multiple types and associated relationship indicators 718.

Individual content and topic objects 710 within a selected content sub-network 700a may be related to individual content and topic objects 710 in another content sub-network 700b. Further, multiple sets of relationships of multiple types and associated relationship indicators 718 may be defined between two objects 710

For example, a first set of relationships 716 and associated relationship indicators 718 may be used for a first purpose or be available to a first set of users while a second set of relationships 716 and associated relationship indicators 718 may be used for a second purpose or available to a second set of users. For example, in FIG. 33, topic object 710t1 is bi-directionally related to topic object 710t2, not once, but twice, as indicated by the two double arrows. An indefinite number of relationships 716 and associated relationship indicators 718 may therefore exist between any two objects 710 in the fuzzy content network 700. The multiple relationships 716 may correspond to distinct relationship types. For example, a relationship type might be the degree an object 710 supports the thesis of a second object 710, while another relationship type might be the degree an object 710 disconfirms the thesis of a second object 710. The content network 700 may thus be customized for various purposes and accessible to different user groups in distinct ways simultaneously.

The relationships among objects 710 in the content network 700, as well as the relationships between content networks 700a and 700b, may be modeled after fuzzy set theory. Each object 710, for example, may be considered a fuzzy set with respect to all other objects 710, which are also considered fuzzy sets. The relationships among objects 710 are the degrees to which each object 710 belongs to the fuzzy set represented by any other object 710. Although not essential, every object 710 in the content network 700 may conceivably have a relationship with every other object 710.

The topic objects 710t encompass, and are labels for, very broad fuzzy sets of the content network 700. The topic objects 710t thus may be labels for the fuzzy set, and the fuzzy set may include relationships to other topic objects 710t as well as related content objects 710c. Content objects 710c, in contrast, typically refer to a narrower domain of information in the content network 700.

The adaptive system 100 of FIG. 1 may operate in a fuzzy content network environment, such as the one depicted in FIG. 33. In FIG. 35, an adaptive system 100D includes a structural aspect 210D that is a fuzzy content network. Thus, adaptive recommendations 250 generated by the adaptive system 100D are also structural subsets that are themselves fuzzy content networks.

Similarly, the adaptive recombinant system 800 of FIG. 18 may operate in a fuzzy content network environment. In FIG. 36, an adaptive recombinant system 800D includes the adaptive system 100D of FIG. 35. Thus, the adaptive recombinant system 800D may perform syndication and recombination operations, as described above, to generate structural subsets that are fuzzy content networks.

Extended Fuzzy Structures in Fuzzy Networks

The fuzzy network model may be extended to the organizational structure of the meta-information and other affiliated information associated with each network node or object. In a fractional degree of separation system structure, depicted in FIG. 37, meta-information and affiliated information may be structured in distinct tiers or rings around the information, with each tier designated as a fractional degree of separation 750. The chosen parameters for the degrees of separation of the meta-information will depend on the definition of the calculation of the degrees of separation between any two nodes, specifically depending on the choice of the scaling factor on in the formula. This extended fuzzy network structure may be utilized to implement a fuzzy content network system structure, or any other fuzzy network-based structure.

Meta-information 754 associated with information or interactive applications 752 may include, but is not limited to, descriptive information about the object such as title, publishing organization, date published, physical location of a physical object, an associated photo or picture, summary or abstracts, a plurality of reviews, etc. Meta-information 754 may also include dynamic information such as expert and community ratings of the information, feedback from users, and more generally, any relevant set of, or history of, usage behaviors described in Table 1. The meta-information 754 may also include information about relationships to other nodes in the network. For example, the meta-information 754 may include the relationships with other nodes in the networks, including an identification code for each related node, the types of relationships, the direction of the relationships, and the degree of relatedness of each relationship.

The meta-information 754 may be defined within tiers of fractional degree of separation between zero and one. For example, the most tightly bound meta-information might be in a tier at degree of separation of 0.1 and less tightly bound meta-information might be in a tier at degree of separation of 0.8.

Where the degrees of separation calculated between any two nodes in the fuzzy network are between 0 and 1, the meta-information tiers would more appropriately be designated with negative (possibly fractional) degrees of separation. For example, the most tightly bound meta-information 752 may be in a tier at degree of separation of −5 and less tightly bound meta-information may be in a tier at degree of separation of −1.

The meta-information tiers may distinguish between static meta-information such as the original author of the associated information, and dynamic information such as the total number of accesses of the associated information through a computer system.

The fractional degree of separations of less than one may correspond to compound objects 756. For example, a picture object plus a text biography object may constitute a person object. For typical fuzzy content network operations the compound object would generally behave as if it was one object.

The fractional degree of separations of less than one may correspond to a list of objects with which the present object has a specific sequential relation 758. For example, this may include workflow sequences in processes. These sequential relationships imply a tighter “binding” between objects than the relationships associated with other objects in the fuzzy network 770, hence a smaller fractional degree of separation is employed for sequential relationships.

All meta-information may explicitly be content objects that relate to associated information by a fractional degree of separation of less than one, and may relate to other content objects in the network by a fractional degree of separation that may be greater than or equal to one. This can be described by a degree-of-separation matrix. Every object is arrayed in sequence along both the matrix columns and the matrix rows. Each cell of the matrix corresponds to the degree of separation between the two associated objects. The cells in the main diagonal of the degree of separation matrix are all zeroes, indicating the degree of separation between an object and itself is zero. All other cells will contain a non-zero number, indicating the degree of separation between the associated objects, or a designator indicating that the degree of separation is essentially infinite in the case when there is no linked path at all between the associated objects.

Personalized Fuzzy Content Network Subsets

Recall that users 200 of the adaptive system 100 of FIG. 1 may tag or store subsets of the structural aspect 210 for personal use, or to share with others. Likewise, users 200 of the adaptive recombinant system 800 may tag subsets of the fuzzy content network, whether for personal use or to share with others.

FIG. 38 is a screenshot 770 generated by the Epiture software system. A “My World” icon 772 invites the viewer to “create your own knowledge network” by clicking on the icon. The icon 772 further states, “Make your own topics and store relevant resources in them.” The term “store” in the icon 772 may simply imply tagging information—no referenced information need necessarily be physically copied and stored, although physical copying and storing may be implemented.

Users of the Epiture software system may select content objects and tag them for storage in their personal fuzzy network. Optionally, related meta-information and links to other objects in the original fuzzy network may be stored with the content object. Users may also store entire topics in their “My World” personal fuzzy network. Furthermore, users may use fuzzy network operators to create synthetic topics. For example, a user might apply an intersection operator to Topic A and Topic B, to yield Topic C. Topic C could then be stored in the personal fuzzy network. Union, difference and other fuzzy network operators may also be used in creating new fuzzy network subsets to be stored in a private fuzzy content network.

Users of the Epiture software system may directly edit their personal fuzzy networks, including the names or labels associated with content objects and topic objects, as well as other meta-information associated with content and topic objects. The screenshot 770 of FIG. 38 features a “personal topics” icon, allowing the user to explicitly edit the network, thus generating an explicitly requested structural subset 280. Users may also create new links among content and topics in their personal fuzzy network, alter the degree of relationship of existing links, or delete existing links altogether, to name a few features of the Epiture software system.

Users may selectively share their personal fuzzy networks by allowing other users to have access to their personal networks. Convenient security options may be provided to facilitate this feature.

Usage Behavior Information

Users of the Epiture software system may have the ability to review personal, sub-community or community usage behaviors over time. This may include trends related to popularity, connectedness, influence or any other relevant usage metric. FIG. 39 is a screenshot 780 showing trend information display functionality associated with the Epiture software system.

Navigational histories, such as access paths, may be available for review, with capabilities for making queries against the histories though application of selection criteria. FIG. 40 depicts a screenshot 790. The screenshot 790 is an example of navigational usage behavior information display and query functions associated with the “MyPaths” function of the Epiture software system. With appropriate authorizations and permissions, users may be able to access any other usage behaviors, such as online information accesses, traffic patterns and click streams associated with navigating the system structure, including buying and selling behaviors; physical locational cues associated with stationary or mobile use of the system; collaborative behaviors among system users that include written and oral communications, and among and with groups of system users (communities) or system users and people outside of the system; referencing behaviors of system users—for example, the tagging of information for future reference; subscription and other self-profiling behavior of users and associated attributes e.g., subscribing to updates associated with particular aspects of the system or explicitly identifying interests or affiliations, such as job function, profession, organization, etc, and preferences such as representative skill level (for example, novice, business user, advanced etc), preferred method of information receipt or learning style such as visual or audio; and relative interest levels in other communities and direct feedback behaviors, such as the ratings or direct written feedback associated with objects or their attributes such as the objects' author, publisher, etc.

Users may also have access to system usage information that may be captured and organized to retain temporal information associated with usage behaviors, including the duration of behaviors and the timing of the behaviors, where the behaviors may include those associated with reading or writing of written or graphical material, oral communications, including listening and talking, or duration of physical location of a system user, potentially segmented by user communities or affinity groups may be available for review by users.

The above usage behaviors may be available to users in raw form, or in summarized form, potentially after application of statistical or other mathematical functions are applied to facilitate interpretation. This information may be presented in a graphical format.

Adaptive Recommendations in Fuzzy Content Networks

Adaptive recommendations or suggestions may enable users to more effectively navigate through the fuzzy content network. As with other network embodiments described herein, the adaptive recommendations generated from a fuzzy content network may be in the context of a currently accessed content object or historical path of accessed content objects during a specific user session, or the adaptive recommendations may be without context of a currently accessed content object or current session path.

In the most generalized approach, adaptive recommendations in a fuzzy content network combine inferences from user community behaviors and preferences, inferences of sub-community or expert behaviors and preferences, and inferences of personal user behaviors and preferences. Usage behaviors that may be used to make preference inferences include, but are not limited to, those that are described in Table 1. These usage-based inferences may be augmented by automated inferences about the content within individual and sets of content objects using statistical pattern matching of words or phrases within the content. Such statistical pattern matching may include, but not limited to, Bayesian analysis, neural network-based methods, k-nearest neighbor, support vector machine-based techniques, or other statistical analytical techniques.

Community Preference Inferences

Where the structural aspect 210 of the adaptive system 100 or the adaptive recombinant system 800 is a fuzzy content network, user community preferences may be inferred from the popularity of individual content objects and the influence of topic or content objects, as popularity and influence were defined above. The duration of access or interaction with topic or content objects by the user community may be used to infer preferences of the community.

Users may subscribe to selected topics, for the purposes of e-mail updates on these topics. The relative frequency of topics subscribed to by the user community as a whole, or by selected sub-communities, may be used to infer community or sub-community preferences. Users may also create their own personalized fuzzy content networks through selection and saving of content objects and/or topics objects. The relative frequency of content objects and/or topic objects being saved in personal fuzzy content networks by the user community as a whole, or by selected sub-communities, may be used to also infer community and sub-community preferences. These inferences may be derived directly from saved content objects and/or topics, but also from affinities the saved content and/or topic objects have with other content objects or topic objects. Users can directly rate content objects when they are accessed, and in such embodiments, community and sub-community preferences may also be inferred through these ratings of individual content objects.

The ratings may apply against both the information referenced by the content object, as well as meta-information such as an expert review of the information referenced by the content object. Users may have the ability to suggest content objects to other individuals and preferences may be inferred from these human-based suggestions. The inferences may be derived from correlating these human-based suggestions with inferred interests of the receivers if the receivers of the human-based suggestions are users of the fuzzy content object system and have a personal history of content objects viewed and/or a personal fuzzy content network that they may have created.

The physical location and duration of remaining in a location of the community of users, as determined by, for example, a global positioning system or any other positionally aware system or device associated with users or sets of users, may be used to infer preferences of the overall user community.

Sub-Community and Expert Preference Inferences

Community subsets, such as experts, may also be designated. Expert opinions on the relationship between content objects may be encoded as affinities between content objects. Expert views may be directly inferred from these affinities. An expert or set of experts may directly rate individual content items and expert preferences may be directly inferred from these ratings.

The history of access of objects or associated meta-information by sub-communities, such as experts, may be used to infer preferences of the associated sub-community. The duration of access or interaction with objects by sub-communities may be used to infer preferences of the associated sub-community.

Experts or other user sub-communities may have the ability to create their own personalized fuzzy content networks through selection and saving of content objects. The relative frequency of content objects saved in personal fuzzy content networks by experts or communities of experts may be used to also infer expert preferences. These inferences may be derived directly from saved content objects, but also from affinities the saved content objects have with other content objects or topic objects.

The physical location and duration of remaining in a location of sub-community users, as determined by, for example, a global positioning system or any other positionally aware system or device associated with users or sets of users, may be used to infer preferences of the user sub-community.

Personal Preference Inferences

Users may subscribe to selected topics, for the purposes of, for example, e-mail updates on these topics. The topic objects subscribed to by the user may be used to infer personal preferences. Users may also create their own personalized fuzzy content networks through selection and saving of content objects. The relative frequency of content objects saved in personal fuzzy content networks by the user may be used to infer the individual's personal preferences. These inferences may be derived directly from saved content objects, but also from affinities the saved content objects have with other content objects or topic objects. Users may directly rate content objects when they are accessed, and in such embodiments, personal preferences may also be inferred through these ratings of individual content objects.

The ratings may apply against both the information referenced by the content object, as well as any of the associated meta-information, such as an expert review of the information referenced by the content object. A personal history of paths of content objects viewed may be stored. This personal history may be used to infer user preferences, as well as tuning adaptive recommendations and suggestions by avoiding recommending or suggesting content objects that have already been recently viewed by the individual. The duration of access or interaction with topic or content objects by the user may be used to infer preferences of the user.

The physical location and duration of remaining in a location of the user as determined by, for example, a global positioning system or any other positionally aware system or device associated with the user, may be used to infer preferences of the user.

Adaptive Recommendations and Suggestions

Adaptive recommendations in fuzzy content networks combine inferences from user community behaviors and preferences, inferences of sub-community or expert behaviors and preferences, and inferences of personal user behaviors and preferences as discussed above, to present to a fuzzy network user or set of users one or more fuzzy network subsets (one or more objects and associated relationships) that users may find particularly interesting given the user's current navigational context. These sources of information, all of which are external to the referenced information within specific content objects, may be augmented by search algorithms that use text matching or statistical pattern matching or learning algorithms to provide information on the likely themes of the information embedded or pointed to by individual content objects.

The navigational context for a recommendation may be at any stage of navigation of a fuzzy network (e.g., during viewing a particular content object) or may be at a time when the recommendation recipient is not engaged in directly navigating the fuzzy network. In fact, the recommendation recipient need never have explicitly used the fuzzy network associated with the recommendation. As an example, FIG. 41 depicts in-context, displayed adaptive recommendations associated with the Epiture system.

Some inferences will be weighted as more important than other inferences in generating a recommendation, and theses weightings may vary over time, and across recommendation recipients, whether individual recipients or sub-community recipients. For example, characteristics of content and topics explicitly stored by a user in a personal fuzzy network would typically be a particularly strong indication of preference as storing network subsets requires explicit action by a user. In most recommendation algorithms, this information will therefore be more influential in driving adaptive recommendations than, say, general community traffic patterns in the fuzzy network.

The recommendation algorithm may particularly try to avoid recommending to a user content that the user is already familiar with. For example, if the user has already stored a content object in a personal fuzzy network, then the content object might be a very low ranking candidate for recommending to the user. Likewise, if the user has recently already viewed the associated content object (regardless of whether it was saved to his personal fuzzy network), then the content object would typically rank low for inclusion in a set of recommended content objects. This may be further tuned through inferences with regard to the duration that an associated content object was viewed (for example, it may be inferred that a lengthy viewing of a content object is indicative of increased levels of familiarity.

The algorithms for integrating the inferences may be tuned or adjusted by the individual user. The tuning may occur as adaptive recommendations are provided to the user, by allowing the user to explicitly rate the adaptive recommendations. The user may also set explicit recommendation tuning controls to tune the adaptive recommendations to her particular preferences. For example, a user might guide the recommendation function to place more relative weight on inferences of expert or other user communities' preferences versus inferences of the user's own personal preferences. This might be particularly true if the user was relatively inexperienced in the particular domain of knowledge. As the user's experience grew, he might adjust the weighting toward inferences of the user's personal preferences versus inferences of expert preferences.

Fuzzy network usage metrics described above such as popularity, connectedness, and influence may be employed by the recommendation algorithm as convenient summaries of community, sub-community and individual user behavior with regard to the fuzzy network. These metrics may be used individually or collectively by the recommendation algorithm in determining the recommended network subset or subsets to present to the recommendation recipient.

Adaptive recommendations which are fuzzy network subsets may be displayed in variety of ways to the user. They may be displayed as a list of content objects (where the list may be null or a single content object), they may include content topic objects, and they may display a varying degree of meta-information associated with the content objects and/or topic objects. Adaptive recommendations may be delivered through a web browser interface, through e-mail, through instant messaging, through XML-based feeds, RSS, or any other approach in which the user visually or acoustically interprets the adaptive recommendations. The recommended fuzzy network subset may be displayed graphically. The graphical display may provide enhanced information that may include depicting linkages among objects, including the degree of relationship, among the objects of the recommended fuzzy network subset, and may optionally indicate through such means of size of displayed object or color of displayed object, designate usage characteristics such as popularity of influence associated with content objects and topic objects in the recommended network subset. Adaptive recommendations may be delivered for interpretation of users by other than visual senses; for example, the recommendation may be delivered acoustically, typically through oral messaging.

The recommended structural subsets 280, combinations of topic objects, content objects, and associated relationships, may constitute most or even all of the user interface, which may be presented to a system user on a periodic or continuous basis. Such embodiments correspond to embodiment variations of 2130, 2140, 2150 and 2160 of the framework 2000 in FIG. 42, below.

In addition to the recommended fuzzy network subset, the recommendation recipient may be able to access information to help gain an understanding from the system why the particular fuzzy network subset was selected as the recommendation to be presented to the user. The reasoning may be fully presented to the recommendation recipient as desired by the recommendation recipient, or it may be presented through a series of interactive queries and associated answers, as a recommendation recipient desires more detail. The reasoning may be presented through display of the logic of the recommendation algorithm. A natural language (such as English) interface may be employed to enable the reasoning displayed to the user to be as explanatory and human-like as possible.

In addition to adaptive recommendations of fuzzy network subsets, adaptive recommendations of some set of users of the fuzzy network may be determined and displayed to recommendation recipients, typically assuming either implicit or explicit permission is granted by such users that might be recommended to other users. The recommendation algorithm may match preferences of other users of the fuzzy network with the current user. The preference matches may include the characteristics of fuzzy network subsets stored by users or other fuzzy network referencing, their topic subscriptions and self-profiling, their collaborative patterns, their direct feedback patterns, their physical location patterns, their fuzzy network navigational and access patterns, and related temporal cues associated with these usage patterns. Information about the recommended set of users may be displayed to a user. This information may include names, as well as other relevant information such as affiliated organizations and contact information. It may also include fuzzy network usage behavioral information, such as, for example, common topics subscribed to, common physical locations, etc. As in the case of fuzzy network subset adaptive recommendations, the adaptive recommendations of other users may be tuned by an individual user through interactive feedback with the system.

Adaptability/Extensibility Framework

FIG. 42 depicts an adaptability/extensibility framework 2000 used to distinguish the adaptive system 100 and the adaptive recombinant system 800 from the prior art, described herein as an “identified system.” The framework 2000 is a two-dimensional representation comprising a vertical dimension 2002 and a horizontal dimension 2004, each dimension having four categories. The vertical dimension 2002 of the framework 2000 indicates the “degree of adaptiveness” of the identified system. The “degree of adaptiveness” is the degree to which the identified system is adaptive to individual users or to communities of users of the system.

The vertical dimension 2002 includes four categories across a range, the first category being least adaptive and the fourth category being the most adaptive. The categories are: non-adaptive (does not dynamically customize); displays adaptive recommendations 250 (where “displays” includes not only visual delivery of adaptive recommendations, but delivery in other modes, such as audio); provides adaptive recommendations 250 that update structure or content (where the structure and/or content of the system are dynamically updated); and provides a continuous, fully adaptive interface. The adaptive system 100 and the adaptive recombinant system 800 are capable of all degrees of adaptiveness depicted in the framework 2000, including providing a continuous, fully adaptive interface.

The horizontal dimension 2004 of the framework 2000 represents the degree of extensibility of the identified system. The “degree of extensibility” or “degree of portability” denotes the ability to “syndicate” the system 100 or subsets of the system 100, as well as the ability to create combinations of systems. Syndication, as used herein, describes ability to share systems or portions of systems, which may include actual transfer of the system structural and content aspects across computer and communications network hardware, or may describe the virtual transfer of a system on a particular set of computer hardware. Recall that a structural subset 280 is a portion of the structural aspect 210 of a system, including one or more objects 212 and their associated relationships 214, which may be replicated (see FIG. 4). Structural subsets may be syndicated by the adaptive recombinant system 800.

The horizontal dimension 2004 includes four categories across a range, the first category being least extensible and the fourth category being the most extensible. The categories are: no syndication (the identified system has no ability to share content); individual content syndication (individual items of content within the identified system can be shared); structural subset syndication (structural subsets of the identified system can be shared); and recombinant structures syndication (structural subsets of the identified system can be shared and combined to create new systems). The adaptive recombinant system 800 is capable of all degrees of extensibility depicted in the framework 2000, including the most portable feature, recombinant structures syndication.

The framework 2000 is divided into sixteen numbered blocks, arranged according to their relationship to the horizontal dimension 2002 (degree of adaptiveness) and the vertical dimension 2004 (degree of extensibility). The majority of prior art systems are confined to the lower left portion of the framework 2000. For example, most prior art system are non-adaptive and include no syndication capabilities (block 2010). Current computer operating systems (e.g. Microsoft XP™), business productivity applications (e.g., Microsoft Office™), enterprise applications (e.g., SAP), and search utilities (e.g., Google®) are associated with block 2010 of the framework 2000.

Some prior art systems syndicate items of content or sets of content files. These may be based on a central syndication clearinghouse (e.g., Napster), or may be more purely peer-to-peer in operation (e.g., Gnutella). Such systems are associated with block 2020 of the framework 2000.

Other prior art systems provide users with merchandise recommendations based on their buying habits, as well as the buying habits of customers who have purchased common merchandise (e.g., Amazon.com®). However, these systems do not truly deliver adaptive recommendations as defined herein, whether by displaying adaptive recommendations 250 (block 2050), updating structure or content (block 2090) or providing a continuous, fully adaptive interface (block 2130). This is because, among other reasons, the scope of the usage behaviors tracked by such prior art systems is limited to purchasing and associated behaviors.

In contrast, for the adaptive system 100 and the adaptive recombinant system 800, more generalized system usage behaviors 247 are tracked and used to deliver adaptive recommendations 250 to the user 200 and to the adaptive (recombinant) system itself. Thus, prior art systems such as Amazon.com are deemed non-adaptive (block 2010) in the framework 2000. Blocks 2010 and 2020 of the framework 2000 thus represent the extent of prior art system capabilities with regard to system adaptation (vertical dimension 2002) and portability (horizontal dimension 2004).

In contrast, the adaptive recombinant system 800 includes the adaptability and portability associated with the remaining blocks of the framework 2000. For example, the adaptive recombinant system 800 is capable of syndicating non-adaptive structural subsets 280 of the system 800 (block 2030), as well as syndicating non-adaptive recombinant structures (block 2040). Thus, the adaptive recombinant system 800 exhibits a high degree of extensibility, fully covering the horizontal dimension 2004 of the framework 2000.

The vertical dimension 2002 is likewise embodied both by the adaptive system 100 and the adaptive recombinant system 800. While the adaptive system 100 displays adaptive recommendations 250 where no syndication occurs (block 2050), the adaptive recombinant system 800 further displays adaptive recommendations 250 where individual content is syndicated (block 2060), where structural subsets 280 are syndicated (block 2070) and where recombinant structures are syndicated (block 2080).

Moving up the vertical dimension 2002, the adaptive system 100 provides adaptive recommendations 250 that update the structural aspect 210 and/or the content aspect 230 of the system where there is no syndication (block 2090), and the adaptive recombinant system 800 provides adaptive recommendations that update the structural or content aspects where individual content is syndicated (block 2100), where structural subsets 280 are syndicated (block 2110), and where recombinant structures are syndicated (block 2120).

Finally, the adaptive recombinant system 800 provides a continuous, fully adaptive interface for all four categories of syndication (blocks 2130, 2140, 2150, and 2160) while the adaptive system 100 does so where there is no syndication (block 2130). Thus, the adaptive system 100 and the adaptive recombinant system 800 provide various degrees of adaptiveness and extensibility, as represented in the framework 2000.

Sample Recommendations Function and Algorithm

In this example, two types of adaptive recommendations are delivered to the user. The adaptive recommendations are calculated by a set of algorithms based on the systems objects being currently navigated, the relationships of the currently accessed object, the user's navigation path, profile preferences, community membership and level of relevance depending on context and the user's personal library of referenced objects. Recall that a “user” may refer to not only humans, but to another system or adaptive network. In other words, two or more adaptive systems may be “users” of each other.

Two types of adaptive recommendations based on a fuzzy content network structure are described in Table 4. One skilled in the art may apply other variations of adaptive recommendations and associated algorithms.

TABLE 4
Two Recommendations Algorithms
Type	Delivery	characteristics
in-context	when user is accessing or	may be delivered in real-
(suggestions)	interacting, accessing, or	time
	updating content object	available in display pages
		for retrieval/editing
		may be optimized for
		responsiveness and “fast”
		learning of user
		preferences
out-of-context	no explicit access of	inferences may be
(recommendations)	content object by user	updated in real time or
		periodically available in
		display pages for retrieval
		may be optimized for
		accuracy and understand-
		ing of user preferences

The first adaptive recommendations type, in-context recommendations, or suggestions, are delivered to the user when the user is interacting, accessing, or updating a content object. In-context recommendations may be delivered in real time, may be displayed for retrieval and editing, and may be optimized for responsiveness and the “fast” learning of the user's preferences.

The second adaptive recommendations type, out-of-context recommendations, is a “push” recommendation approach. Based on inferences about the user's preferences, the network is aligned to adapt to the preferences. The out-of-context recommendations thus “surprise” the user with recommendations of relevant objects of interest without specific explicit context from the user. Relevant characteristics for out-of-context recommendations include the real-time or periodic updating of inferences and the ability to provide adaptive recommendations in display pages or via other modes of communication for retrieval Further, the out-of-context recommendations algorithm may be optimized for accuracy and understanding of user preferences

Adaptive Recommendations Function Example

FIG. 43 is a flow diagram depicting the operation of an adaptive recommendations function 900 used in the Epiture software system, according to some embodiments. The Epiture software system is one implementation of an adaptive recombinant system, such as the system 800 depicted in FIG. 18. The network described in this example is a fuzzy content network. Recall that the adaptive recommendations function includes algorithms for generating adaptive recommendations to a user in the form of structural subsets 280.

The following data is used by the adaptive recommendations function 900 in generating recommendations:

• • 1) the communities that a user is a member of
  • 2) relationships between those communities and user's preferences (including temporal dimensions that may indicate strengthened or weakened interest in those communities)
  • 3) a user's or other pre-defined system explicit preference of those communities in this context (e,g, business rules for a process, novice vs advanced users),
  • 4) the user's personal topics (where objects of high relevance have been ‘saved’ for future explorations) and the relationships between those topics
  • 5) the content in those topics and their interrelations, or personal highest recommendation objects

The adaptive recommendations function 900 begins by determining personal highest recommendation areas, or PHRAs of the user (block 902). PHRAs are, generated by determining the highest relevance sums of co-topic-community relationships. To illustrate this step, Table 5 includes an abbreviated matrix of topics and communities on one axis versus content objects and topic objects on the other matrix, with numerical relationships between the two axes.

TABLE 5
Relationships between objects in fuzzy content network
	topic A	topic B	topic C	community X
object 1 (article)	5	3	2	0
object 2 (presentation)	1	4	—	5
object 3 (book)	3	3	5	2
topic A	—	3	—	5
total	9	12	7	12
In this limited example, there are three topics, topic A, topic B, and topic C, and one community, community X, that have varying degrees of relationship (rated between 1 and 5) to other objects in the system: object 1 (an article), object 2 (a presentation), object 3 (a book), and topic A. Calculating the highest sum of relationships for the particular context (total row) results in the generation of PHRAs.

In Table 5, topic B and community X have the highest relationship sums thus two PHRAs are found in this example. This method will often generate many PHRAs, which sometimes may be too many to make useful suggestions from. For example, there may be a dozen or more PHRAs with the same value. In this case, the tie breakers are the data that informs on relationships between topics and communities.

For example, in Table 5, topic A has a strong relationship (5) to community X. Topic A itself has a high total score. Thus, the adaptive recommendations function 900 assigns a dynamic weighting to topic A's relevance to community X, to strengthen community X's result. In this case, if it was desirable to have only one PHPA, community X would be chosen. In some embodiments, the top 3-5 PHRAs are selected by the adaptive recommendations function 900.

Building on this procedure, the storing of the dynamic weightings generated in this process can be useful as an additional recommendation mechanism. This approach allows the adaptive recommendations function 900, at the end of processing, to compare which recommendation is actually selected by the user from the top suggestions generated. If there is a discrepancy or convergence, the weightings may be examined and used as a way to strengthen or weaken the relationships between topics, objects and communities for this user's particular context.

The adaptive recommendations function 900 also determines Epiture's highest recommendation area, or EHRA (block 904). Recall that, in the adaptive recombinant system 800, relationships between objects, topics and communities, may be made by experts. There may also be explicit business rules in the system to conform) to, for example in the form of a business process. II addition, the relationship context may be delivered from another fuzzy content network or instance of the adaptive recombinant system, in particular when ‘training’ a new knowledge network or integrating existing networks. The Epiture software system includes these features in determining EHRAs.

A set of Epiture's highest recommendation areas (EHRA) may be generated by selecting related topics or communities with higher relevance values to the current object. The EHRAs are weighted appropriately to the situation, either by system rules, or by user preferences.

The adaptive recommendations function 900 also determines Epiture's highest recommended objects (block 906). Again, this step uses relationships already in existence in the system, either an average across all relationships and quality ratings, or tuned to select a particular set of relationship types or quality ratings. From these data, a set of Epitures highest recommendation objects (EHRO) may be generated by selecting related content objects with higher relevance values (with relevance defined by context of both the object in question and system ‘priorities’) to the current object.

Although steps 902, 904, and 906 are presented in a particular order in FIG. 43, they may be implemented by the adaptive recommendations function 900 in a different order than the one shown. The adaptive recommendations function 900 next combines the PHRA, EHRA and EHRO data to determine what will be recommended to the user (block 908). Initially, if a set of objects score highly in both PHRA and EHRA, then they will be the objects recommended first. Depending on the amount of recommendation results that are prespecified by the adaptive recommendations function, this initial set of recommended objects may be sufficient.

If not, however, the adaptive recommendations function 900 determines whether it can find any objects in EHRO that also exist in the PHRA. If so, those results will be returned and the operation ends even though the selected objects are a second tier of the recommended objects. To ensure that the user realizes this, a relevance weighting may be assigned, and graphically indicated if needed.

A third tier of recommended objects may be found by finding any objects in the EHRO that exist in the EHPA, using quality, relationships types and values and other attributes as guides for making the selection.

If a sufficient set of recommendation objects have been found (the “yes” prong of block 910), the adaptive recommendations function 900 removes duplicated objects in the potential recommendations determined thus far (block 908). This step is particularly relevant where the users of the Epiture software system are human users who have been browsing the system for some time period. Such users generally do not wish to be recommended content they have already read, visited, or used. If the user has already visited some of the selected recommended objects within a predetermined time period, say, in the last 24 hours, or, if some of recommended objects are already in the user's personal topic library, the adaptive recommendations function 900 determines the object to be unnecessary to recommend. Thus, such objects are removed from the recommendation object set.

Where objects removed in this manner cause the available adaptive recommendations to be insufficient or empty (the “no” prong of block 914), or where enough adaptive recommendations were not produced initially (the “no” prong of block 910), the adaptive recommendations function 900 proceeds to determine the most popular jump objects in the path of a community (block 916).

The adaptive recommendations function 900 examines the paths of other users who have browsed the object. Given criteria such as similar community membership to the current user, content quality rating and distribution, overall popularity, and other attributes, it is determined which objects to recommend based on prior usage. This fourth tier of recommendation objects (besides PHRAs, EHRAs, and EHROs) is designated as a second set of Epiture's highest recommended objects or EHRO2.

This step (block 916) may be helpful in the case of integrating two or more networks together. Since the relationship context and attributes of the objects in the network may be ‘carried’ over or ported into th e new network, the objects may ‘look’ for their prior relationships and segment based on usage criteria. In addition, influence and other metrics and attribute patterns may be used to determine similarities between objects. Thus, the adaptive recommendations function 900 may connect objects which have not been in contact before, providing the user a targeted recommendation, and generating a relationship between those objects. That newly formed relationship may cascade to affect other objects in the system such as communities and topics

Finally, the adaptive recommendations function 900 may track usage of adaptive recommendations (block 918). As the embedded algorithms are optimized for speed and real-time performance for in-context recommendations, the ‘understanding’ and true relevance (as inferred from user usage behavior) of the adaptive recommendations may be processed later As such, tracking the selection and usage of adaptive recommendations at this time may be beneficial Criteria such as placement position on a list or other display mechanism, determined (estimated) relevance as predicted by the algorithm versus first selections by the user, and choice of object type (such as article, subject matter expert, multimedia, image etc), are just a few examples of how the adaptive recommendations function may self-monitor its performance. This performance analysis may ultimately generate better quality recommendations for the user, and be used in updating system structure such as EHRA inputs. Or, the system may be self-policing, in effect, making changes as usage data builds up.

It should be noted that the adaptive recommendations function 900 depicted in FIG. 43 is a simplified embodiment of the adaptive recommendations function 240, as one algorithm of possibly many is examined Many complex variations of the recommendations algorithms may be implemented, in accordance with the descriptions of the adaptive system 100 of FIG. 1 and the adaptive recombinant system 800 of FIG. 18, above.

Further Example Embodiment Description

The screenshot 770 also depicts a user personal library function 714, denoted “My Personal Topics,” for a particular user. A screenshot 720 in FIG. 45 illustrates the use of the adaptive recommendations function, as shown in a “Recommended For You” graphic 722, with a list of suggestions. A “My Path” graphic 724 also with a list, represents the path of objects the user has already browsed. The recommendations in 722 adapt as the user browses different objects.

In the screen image 790 of FIG. 40, the ‘MyPath’ function represents the journey a user has made in the network during their session. The user may browse the list of objects that they have visited during a session. There are further options to save an object as part of their My World personal library and also to remove an object from their path. The MyPath function way be useful to users in identifying areas of the network they have browsed before, and users may also elect to share a specific path or all paths with other users of the system.

Path data can be used to strengthen adaptive recommendations on an automatic basis, while also contributing to input of an automatic or semi-automatic recommendation for the setup of a new community or new topical area.

Cumulative usage data may also be of interest to users of the system as illustrated in the screen image 780 of FIG. 39 Table 782 shows an example of usage patterns shown on a temporal bases to reflect amount of interest in certain topical areas. While human users of the system can be easily overwhelmed with the amount of statistical information generated by usage data of many different kinds, the screen image displays the information in a manner so as to express multifaceted data for input into its adaptive recommendation functions.

Automatic Fuzzy Content Network Maintenance

The adaptive recommendations function and related sets of algorithms, in conjunction with the fuzzy network maintenance functions, may be used to automatically or semi-automatically update and enhance the fuzzy content network. These functions may be employed to determine new affinities and the appropriate degree of relationship among fuzzy network objects in the fuzzy network as a whole, within personal fuzzy network subsets, or sub-community-specific fuzzy network subsets. The automatic updating may include potentially setting a relationship between any two objects to zero (effectively deleting a relationship link).

The recommendation function and fuzzy network maintenance functions may operate completely automatically, performing in the background and updating affinities independently of human intervention, or the function may be used by users or special experts who rely on the adaptive recommendations to provide guidance in maintaining the fuzzy network as a whole, or maintaining specific fuzzy network subsets.

In either an autonomous mode of operation, or in conjunction with human expertise, the recommendation function may be used to integrate new content or content objects into the fuzzy content network.

As in the case of adaptive recommendations that are delivered to recipients to enhance their ability to effectively navigate and use the system, adaptive recommendations that function to update the fuzzy content network include algorithms that make inferences from the usage behaviors of system users. These inferences may be at the community level, sub-community level, or individual user level. Usage behaviors that may be included in the inferencing include online information accesses, traffic patterns and click streams associated with navigating the system structure, including buying and selling behaviors; physical locational information associated with stationary or mobile use of the system; collaborative behaviors among system users or systems users and people outside the system, that include written and oral communications; referencing behaviors of system users—for example, the tagging of information for future reference; subscription and other self-profiling behavior of users; and direct feedback behaviors, such as the ratings or direct written feedback associated with objects or their attributes such as the objects' author, publisher, etc. The algorithms may also use information associated with temporal information associated with usage behaviors, including the duration of behaviors and the timing of the behaviors, where the behaviors may include those associated with reading or writing of written or graphical material, oral communications, including listening and talking, or duration of physical location of a system user.

In some embodiments, inferences regarding a plurality of usage behaviors may be used to adjust relationships and associated relationship values and indicators, as explained in the sample embodiment above. These fuzzy network structural modifications may be applied to multiple relationship types. Navigational access information may be used by the algorithms; that is, the relative level of traffic between two objects (each either a content object or a topic object) will influence the degree of relationship between the two objects. However, access information alone is likely to be insufficient for best results as navigation accesses are highly influenced by the current system structure, and therefore current structures would tend to be reinforced, limiting the level of adaptation. Therefore, other or additional behavioral information is preferentially used to overcome this bias. For example, duration of viewing objects typically provides a better indication of value of an object to a user than does just an object access, as does, for example, reference and reference organization cues, collaboration cues, and direct feedback. Therefore, this additional behavioral information may be used to adjust the strengths of relationships among objects.

As an example, where referenced or tagged information can be organized by users, the system may scan the referenced information and how it is organized, and the frequency of the organizational structures among users, to determine a preliminary degree of relationships in the system. This may be augmented by information associated with navigational accesses and the duration of the accesses.

As a simplified example, FIG. 44A depicts a simple fuzzy network 670a before application of the recommendation function and associated fuzzy network maintenance functions. FIG. 44B depicts fuzzy network 670b, resulting from the application of the recommendation function and associated fuzzy network maintenance functions to fuzzy network 670a. (For the sake of simplicity, relationship indicators are not shown.)

The fuzzy network 670a may have a popular access path 672a from Node X to Node Y which in turn has a popular access path 674a to Node Z. Assuming the existing relationships along that path are of similar strength, it might suggest, without any additional information, that these relationships should perhaps be strengthened due to the high popularity of the path. However, more usage behavioral information may suggest a different fuzzy network updating approach. For example, the duration of accesses of Node X and Node Z were generally significantly higher than for Node Y, a better structural update might be to increase, or establish, the relationship between Node X and Node Z, as is shown in the fuzzy network 670b. After application of an algorithm that incorporates the durational usage behavioral cues, a relationship 676b is established between Node X and Node Z. In addition, in this example, the former relationship 672a between Node X and Node Y is deleted (in practice, it might just be weakened in strength).

The structural transformation from fuzzy network 670a to 670b as shown would be even more reinforced if additional usage behavioral information supported reinforced the access durational-based inferences on preferences. For example, if Node X and Node Z were more frequently referenced by users than Node Y, and were organized such as to imply close affinity (for example, stored in the same personal topical area). This would be more confirming information to strengthen the relationship between Node X and Node Z, and to weaken or eliminate the relationship between Node X and Node Y.

The relationship updating algorithm may temper potential relationship updating, including adding new relationships, with global considerations related to optimal connections among network objects. For example, too few relationships, or relationships with insufficient spread of strength values tend to inhibit effective navigation, but on the other hand too many relationships also is not optimal. The algorithm may strive to maintain an optimal richness of relationships while updating the fuzzy content network based on usage characteristics. The algorithm may use preferential distributions based on fuzzy network metrics such as connectedness and influence to optimize the fuzzy network relationship topologies.

The recommendation function or related algorithms, in conjunction with the fuzzy content network maintenance functions, may also be extended to scan, evaluate, and determine fuzzy network subsets that have special characteristics. For example, the recommendation function or related algorithms may suggest that certain of the fuzzy network subsets that have been evaluated are candidates for special designation. This may include being a candidate for becoming a topical area. The recommendation function may suggest to human users or experts the fuzzy network subset that is suggested to become a topical area, along with existing topical areas that are deemed by the recommendation function to be “closest” in relationship to the new suggested topical area. A human user or expert may then be invited to add a topic, along with associated meta-information, and may manually create relationships between the new topic and existing topics. Statistical pattern matching or learning algorithms used to identify such fuzzy network subsets may include, but are not limited to, semantic network techniques, Bayesian analytical techniques, neural network-based techniques, k-nearest neighbor, support vector machine-based techniques, or other statistical analytical techniques.

The algorithms may apply fuzzy network usage behaviors, along with user community segmentations, to determine new topical areas. The algorithms may be augmented with global considerations related to optimal topologies of fuzzy network structures so as to deliver the most effective usability. For example, too many topics, or topics not sufficiently spread across the over domain of information or knowledge addressed by the system, tend to inhibit effective navigation and use. The algorithm may strive to maintain an optimal richness of topical areas. The algorithm may use preferential distributions based on fuzzy network metrics such as connectedness and influence to optimize the fuzzy network relationship topologies. This approach may also be employed in suggesting topical areas for deletion.

Or, the recommendation function or related algorithms, in conjunction with the fuzzy content network maintenance functions, may automatically generate the topic object and associated meta-information, and may automatically generate the relationships and relationship indicators and their values between the newly created topic object and other topic objects in the fuzzy network.

In some embodiments this capability may be extended such that the recommendation function or related algorithms, along with fuzzy network maintenance functions, automatically maintain the fuzzy network and identified fuzzy network subsets. The recommendation function may not only identify new topical areas, generate associated topic objects, associated relationships and relationship indicators among the new topic objects and existing topic objects, and the associated values of the relationships indicators, but also identify topic objects that are candidates for deletion, and in some embodiments may automatically delete the topic object and its associated relationships.

The adaptive recommendations function, in conjunction with the fuzzy network maintenance functions, may likewise identify content objects that are candidates for deletion, and may automatically delete the associated content objects and their associated relationships.

In this way the adaptive recommendations function or related algorithms, along with the fuzzy content network maintenance functions, may automatically adapt the structure of the fuzzy network itself on a periodic or continuous basis to enable the best possible experience for the fuzzy network's users.

As in network embodiments, when a new fuzzy content network is initialized, the adaptive recommendation function may also serve as a training mechanism for the new network. Given a distribution of content, relationships and relationships types, metrics and usage behaviors associated with scope, subject and other experiential data of other fuzzy content networks, a module of the adaptive recommendation function may automatically begin assimilation of content objects into a fuzzy content network, with intervention as required by humans. Clusters of newly assimilated content objects may enable inferences resulting in the suggestion of new topical objects and communities, and associated relationship types and indicators may also be automatically created and updated. This functionality of the adaptive recommendation engine may also be applied when two or more fuzzy content networks are brought together and require integration.

Each of the automatic steps listed above may be interactive with human users and experts as desired.

Social Network Analysis in Fuzzy Content Object Networks

Social network analysis may be conducted with adaptive recombinant system 800 in multiple ways. First, the representation of a person or people may be explicitly through content objects in the fuzzy content network. Special people-type content objects may be available, for example. Such a content object may have relevant meta-information such as an image of the person, and associated biography, affiliated organization, contact information, etc. The content object may be related to other content objects that the person or persons personally contributed to, topics that they have particular interest or expertise in, or any other system objects with which the person or persons have an affinity. Tracking information associated with access to these content objects by specific users, and/or user sub-communities may be determined as described above.

Furthermore, collaborative usage patterns may be used to understand direct communications interactions among persons, in addition to indirect interactions (e.g., interactions related to the content associated with a person). The physical location of people may be tracked, enabling an inference of in-person interactions, in addition to collaborations at a distance.

Second, specific people may be associated with specific content and topic objects—for example, the author of a particular content object. These people may or may not have explicit associated people-type content objects. Metrics related to the popularity, connectedness, and influence of a person's associated content may be calculated to provide measurement and insights associated with the underlying social network. The associations with content objects may be with a group of people rather than a single individual such as an author. For example, the metrics may be calculated for organizations affiliated with content objects. An example is the publisher of the associated content.

In either of the approaches described above, report-based and graphical-based formats may be used to display attributes of the underlying social network. These may include on-line or printed displays that illustrate how communities or sub-communities of users directly access a set of people (through the associated content objects), or indirectly through associated content affiliated with the set of people.

Adaptive Processes and Process Networks

The adaptive system 100 and the adaptive recombinant system 800 enable the effective implementation of computer-based or computer-assisted processes. Processes involve a sequence of activity steps or stages that may be explicitly defined, and such sequences are sometimes termed “workflow.” These processes may involve structures that require, or encourage, a step or stage to be completed before the next step or stage may be conducted. Additional relevant details on process-based applications and implementations of adaptive networks is disclosed in U.S. Provisional Patent Application, No. 60/572,565, entitled “A Method and System for Adaptive Processes,” which is incorporated herein by reference, as if set forth in its entirety.

A set of relationships and associated relationship indicators may be employed to designate process flows among objects in a fuzzy network, or fuzzy content network. The existence of a process relationship between object x and object y implies that x precedes y in a specified process. A process relationship may exist between object x and a plurality of other objects. In these embodiments, a user may have a choice of multiple process step options from an originating process step. The values of a plurality relationship indicators associated with the process relationships between an object and a plurality of objects may be different.

A plurality of process relationship indicators may be designated among the objects in a fuzzy content network, which enables objects to be organized in a plurality of processes.

Display functions enable a user to navigate through a fuzzy network or fuzzy network subset via objects that have process relations between them. At each process step, corresponding to accessing the corresponding object, the user may have the ability to navigate to other related objects, which can be advantageous in providing the user with relevant information to facilitate executing the corresponding process step.

Fuzzy processes may be organized into fuzzy sub-processes through selection of a subset of objects corresponding to a contiguous set of process steps, along with all other objects related to the process step objects, or more generally, as the set of all objects within a specified fractional degrees of separation from each of the process step objects.

New fuzzy processes may be generated by combining fuzzy process sub-networks into new fuzzy process networks using the fuzzy network union, intersection and other operators.

FIG. 45 depicts various hardware topologies that the adaptive system 100 or the adaptive recombinant system 800 may embody. Servers 950, 952, and 954 are shown, perhaps residing a different physical locations, and potentially belonging to different organizations or individuals. A standard PC workstation 956 is connected to the server in a contemporary fashion. In this instance, the systems 100 or 800 may reside on the server 950, but may be accessed by the workstation 956. A terminal or display-only device 958 and a workstation setup 960 are also shown. The PC workstation 956 may be connected to a portable processing device (not shown), such as a mobile telephony device, which may be a mobile phone or a personal digital assistant (PDA). The mobile telephony device or PDA may, in turn, be connected to another wireless device such as a telephone or a GPS receiver.

FIG. 45 also features a network of wireless or other portable devices 962. The adaptive system 100 or the adaptive recombinant system 800 may reside, in part or as a whole, on all of the devices 962, periodically or continuously communicating with the central server 952, as required. A workstation 964 connected in a peer-to-peer fashion with a plurality of other computers is also shown. In this computing topology, the systems 100 or 800, as a whole or in part, may reside on each of the peer computers 964.

Computing system 966 represents a PC or other computing system which connects through a gateway or other host in order to access the server 952 on which the systems 100 or 800 reside. An appliance 968, includes software “hardwired” into a physical device, or may utilize software running on another system that does not itself host the systems 100 or 800. The appliance 968 is able to access a computing system that hosts an instance of the system 100 or 800, such as the server 952, and is able to interact with the instance of the system 100 or 800. P While the present invention has been described with respect to a limited number of embodiments, those skilled in the art will appreciate numerous modifications and variations therefrom. It is intended that the appended claims cover all such modifications and variations as fall within the true spirit and scope of this present invention.

Claims

1. An adaptive recommendation system, comprising:

at least one storage device configured to store a plurality of aspects comprising:

a content aspect comprising information;

a computer-implemented structural aspect comprising the content aspect and associated relationships; and

a usage aspect, comprising captured usage behaviors, wherein the usage behaviors are associated with one or more users of the system; and

at least one processing device configured to execute a plurality of functions comprising:

a function to generate a user tunable adaptive recommendation based, at least in part, on a user's navigational context and on an automatic inference of the user's interests from a plurality of usage behaviors associated with the one or more users corresponding to a plurality of usage behavior categories; and

a function to deliver the adaptive recommendation to the user one or more users.

2. The adaptive recommendation system of claim 1, wherein the information is selected from a group consisting of text, graphics, audio, video, interactive forms of content, applets, tutorials, advertising content, courseware, demonstrations, representations of people, modules, executable code, and computer programs.

3. The adaptive recommendation system of claim 1, wherein the computer-implemented structural aspect further comprises:

one or more objects, each object comprising the information; and

one or more relationships, wherein each relationship is associated with each pair of the one or more objects.

4. The adaptive recommendation system of claim 1, the usage aspect further comprising one or more usage behaviors, wherein each usage behavior is associated with either a user, one or more user communities, or a the user and the one or more user communities simultaneously, wherein the user comprises a single-member subset of the one or more users and a community of the one or more user communities comprises a multiple-member subset of the one or more users.

5. The adaptive recommendation system of claim 1, wherein a user of the one or more users is selected from a group consisting of a computer-based system, a second adaptive system, and a human being.

6. The adaptive recommendation system of claim 1, the plurality of usage behaviors further comprising private behaviors and non-private behaviors.

7. The adaptive recommendation system of claim 1, further comprising a privacy control, the privacy control enabling a user of the one or more users to restrict usage behaviors associated with the user from being deemed non-private behaviors.

8. The adaptive recommendation system of claim 1, wherein a the function to generate a the user tunable adaptive recommendation based, at least in part, on a user's the navigational context of the one or more users and on an the automatic inference of the user's interests of the one or more users from a the plurality of usage behaviors associated with the one or more users corresponding to a the plurality of usage behavior categories further comprises:

usage behavior categories, wherein the usage behavior categories are selected from a group consisting of navigation and access patterns, collaborative patterns, direct feedback patterns, subscription patterns, self-profiling patterns, reference patterns, and physical location patterns.

9. The adaptive recommendation system of claim 1, wherein a the function to generate a the user tunable adaptive recommendation based, at least in part, on a user's the navigational context of the one or more users and on an the automatic inference of the user's interests of the one or more users from a the plurality of usage behaviors associated with the one or more users corresponding to a the plurality of usage behavior categories further comprises:

a function that infers user preferences from the plurality of usage behaviors.

10. The adaptive recommendation system of claim 9, wherein a the function that infers user preferences from the plurality of usage behaviors further comprises:

an algorithm that prioritizes application of usage patterns associated with a the plurality of usage behavior categories.

11. The adaptive recommendation system of claim 9, wherein a the function that infers user preferences the algorithm from the plurality of usage behaviors comprises a statistical learning algorithm, wherein the statistical learning algorithm is selected from a group consisting of Bayesian modeling, neural network modeling, k-nearest neighbor modeling, and support vector machine modeling.

12. The adaptive recommendation system of claim 1, wherein a the function to generate a the user tunable adaptive recommendation based, at least in part, on a user's the navigational context of the one or more users and on an the automatic inference of the user's interests of the one or more users from a the plurality of usage behaviors associated with the one or more users corresponding to a the plurality of usage behavior categories further comprises:

an algorithm that detects apparent insincere system usage behaviors or other inferred “gaming” gaming behaviors by the one or more users.

13. The adaptive recommendation system of claim 1, wherein a the function to generate a the user tunable adaptive recommendation based, at least in part, on a user's the navigational context of the one or more users and on an the automatic inference of the user's interests of the one or more users from a the plurality of usage behaviors associated with the one or more users corresponding to a the plurality of usage behavior categories further comprises:

a compensatory algorithm associated with the detection of apparent insincere system usage behaviors or other inferred “gaming” gaming behaviors by the one or more users.

14. The adaptive recommendation system of claim 1, wherein a the function to generate a the user tunable adaptive recommendation based, at least in part, on a user's the navigational context of the one or more users and on an the automatic inference of the user's interests of the one or more users from a the plurality of usage behaviors associated with the one or more users corresponding to a the plurality of usage behavior categories further comprises:

an algorithm that applies pattern matching of information embodied in the structural aspect and content aspect to produce content interpretation patterns, and associates the content interpretation patterns with usage patterns.

15. The adaptive recommendation system of claim 1, wherein the user tunable adaptive recommendation further comprises:

a structural subset of the structural aspect, the structural subset comprising at least one of the one or more objects and associated relationships of the one or more objects of the structural aspect.

16. The adaptive recommendation system of claim 1, wherein a the function to deliver the adaptive recommendation to the user one or more users comprises:

a recommendation delivery mode, wherein the recommendation delivery mode is selected from a group consisting of visual, audio, and a combination of visual and audio.

17. The adaptive recommendation system of claim 1, wherein a the function to deliver the adaptive recommendation to the user one or more users comprises:

a recommendation delivery means, wherein the recommendation delivery means is selected from a group consisting of user in-context system usage, direct user requests, and out-of-the-context of system usage.

18. A mobile adaptive recommendation system, comprising:

at least one storage device configured to store a plurality of aspects comprising:

a content aspect comprising information;

a computer-implemented structural aspect comprising the content aspect and associated relationships; and

a usage aspect, comprising captured usage behaviors, wherein the usage behaviors are associated with one or more users; and

at least one processing device configured to execute a plurality of functions comprising:

a function to automatically determine the location of a user based on physical location data generated by a location-aware device;

a user-controlled recommendation tuning function;

a function to generate an adaptive recommendation based, at least in part, on the user's recommendation tuning settings and on the automatically determined location of the user and at least one other of the usage behavior behaviors associated with the one or more users corresponding to at least one other usage behavior category; and

a function to deliver the adaptive recommendation to the user.

19. The mobile adaptive recommendation system of claim 18, wherein a the function to generate an the adaptive recommendation based, at least in part, on the user's recommendation tuning settings and on the automatically determined location of the user and at least one other usage behavior associated with the one or more users corresponding to at least one other usage behavior category further comprises:

an algorithm to determine the change in location of a the user as a function of time.

20. An article comprising a physical non-transitory computer-readable medium storing instructions for enabling a processor-based system to:

access a content aspect comprising information;

access a structural aspect comprising the content aspect and associated relationships;

access a usage aspect, comprising captured usage behaviors, wherein the usage behaviors are associated with one or more users;

generate an user tunable adaptive recommendation based, at least in part, on a user's navigational context and on an automatic inference of the user's interests from a plurality of usage behaviors associated with the one or more users corresponding to a plurality of usage behavior categories; and

deliver the adaptive recommendation to the user.

21. An adaptive recommendation system, comprising:

at least one storage device configured to store a plurality of aspects comprising:

a content aspect comprising information;

a structural aspect comprising the content aspect and associated relationships; and

a usage aspect comprising captured usage behaviors associated with users of the adaptive recommendation system and corresponding to a plurality of usage behavior categories; and

at least one processing device configured to execute:

a function to generate user tunable adaptive recommendations based, at least in part, on a navigational context of the users and on an automatic inference of interests of the users from a plurality of the captured usage behaviors associated with the users and corresponding to the plurality of usage behavior categories; and

a function to deliver the user tunable adaptive recommendations to at least one of the users.

22. The adaptive recommendation system of claim 21, wherein the information includes text, graphics, audio, video, interactive forms of content, applets, tutorials, advertising content, courseware, demonstrations, representations of people, modules, executable code, or computer programs.

23. The adaptive recommendation system of claim 21,

wherein the structural aspect further includes objects with at least a portion of the information; and

wherein each of the associated relationships is configured to associate a pair of the objects.

24. The adaptive recommendation system of claim 21,

wherein the captured usage behaviors are associated with either the users or one or more user communities or the users and the one or more user communities;

wherein each of the users comprises a single-member subset of the users; and

wherein a community of the one or more user communities comprises a multiple-member subset of the users.

25. The adaptive recommendation system of claim 21, wherein the users are selected from a group consisting of a computer-based system, a second adaptive system, and a human being.

26. The adaptive recommendation system of claim 21, wherein the captured usage behaviors further comprise private behaviors and non-private behaviors.

27. The adaptive recommendation system of claim 21, further comprising a privacy control configured to enable the users to restrict the captured usage behaviors from being deemed non-private behaviors.

28. The adaptive recommendation system of claim 21, wherein the plurality of usage behavior categories includes navigation and access patterns, collaborative patterns, direct feedback patterns, subscription patterns, self-profiling patterns, reference patterns, or physical location patterns.

29. The adaptive recommendation system of claim 21, wherein the function to generate the user tunable adaptive recommendations includes a function configured to infer preferences of the users from the captured usage behaviors.

30. The adaptive recommendation system of claim 29, wherein the function configured to infer preferences of the users from the captured usage behaviors includes an algorithm configured to prioritize application of usage patterns associated with the plurality of usage behavior categories.

31. The adaptive recommendation system of claim 29, wherein the function configured to infer preferences of the users from the captured usage behaviors includes a statistical learning algorithm selected from a group consisting of Bayesian modeling, neural network modeling, k-nearest neighbor modeling, and support vector machine modeling.

32. The adaptive recommendation system of claim 21, wherein the function to generate user tunable adaptive recommendations includes an algorithm configured to detect apparent insincere system usage behaviors or other inferred gaming behaviors by the users.

33. The adaptive recommendation system of claim 21, wherein the function to generate user tunable adaptive recommendations includes a compensatory algorithm configured to detect apparent insincere system usage behaviors or other inferred gaming behaviors by the users.

34. The adaptive recommendation system of claim 21, wherein the function to generate user tunable adaptive recommendations includes an algorithm configured to apply pattern matching of the information embodied in the structural aspect and the content aspect to produce content interpretation patterns and configured to associate the content interpretation patterns with usage patterns.

35. The adaptive recommendation system of claim 21, wherein a structural subset of the structural aspect includes at least one object and wherein the relationships are associated with the least one object.

36. The adaptive recommendation system of claim 21,

wherein the function to deliver the user tunable adaptive recommendations includes a recommendation delivery mode; and

wherein the recommendation delivery mode is selected from a group consisting of visual, audio, and a combination of visual and audio.

37. The adaptive recommendation system of claim 21, wherein the function to deliver the user tunable adaptive recommendations includes a delivery means comprising in-context system usage by the users, direct requests by the users, or out-of-context system usage.

38. A mobile adaptive recommendation system, comprising:

at least one storage device configured to store a plurality of aspects comprising:

a content aspect comprising information;

a structural aspect comprising the content aspect and associated relationships; and

a usage aspect comprising captured usage behaviors associated with a user; and

at least one processing device configured to execute:

a function to automatically determine a location of the user based on physical location data generated by a location-aware device;

a user-controlled recommendation tuning function;

a function to generate an adaptive recommendation based, at least in part, on recommendation tuning settings, the determined location of the user, and at least one of the captured usage behaviors associated with the user and corresponding to at least one usage behavior category; and

a function to deliver the adaptive recommendation to the user.

39. The mobile adaptive recommendation system of claim 38, wherein the function to generate the adaptive recommendation includes an algorithm to determine a change in the location of the user as a function of time.

40. An article comprising a non-transitory computer-readable medium storing instructions that, in response to execution by a processing device, cause the processing device to perform operations comprising:

access a content aspect comprising information;

access a structural aspect comprising the content aspect and associated relationships;

access a usage aspect comprising captured usage behaviors associated with a user;

generate a user tunable adaptive recommendation based, at least in part, on a navigational context of the user and on an automatic inference of interests of the user from the captured usage behaviors associated with the user and corresponding to usage behavior categories; and

deliver the user tunable adaptive recommendation to the user.