Method and system of classifying, ranking and relating information based on weights of network links

• US 8,156,056 B2
• Filed: 03/25/2008
• Issued: 04/10/2012
• Est. Priority Date: 04/03/2007
• Status: Expired due to Fees

• Current Assignee: Ondore, Inc.
• Original Assignee: Fernando Luege Mateos
• Inventors: Luege Mateos, Fernando
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): CHANG, LI WU
• Application Number:
  US12/076,917
• Publication Number:
  US 20080249966A1
• Time in Patent Office: 1,477 Days
• Field of Search: None
• US Class Current: 706/20
• CPC Class Codes: G06F 16/3347 using vector based model
• CPC Class Codes: G06F 16/337 Profile generation, learning or modification

Abstract

What is disclosed is an automatic method of classification, qualification and relation of information that is based on a network built starting from the semantic, numeric, conceptual, etc. relationships between elements of different sets that are structurally alike. Through the topological properties of the network, we generate a vector of relations for each element that allows us to analyze the similarity among them, improve and facilitate the clustering processes, carry out personalized dynamic ranking, analyze conduct patterns as well as maintain the auto evolving dynamic of the network through the interaction and feedback of its elements, among other characteristics.

Claims

1. A computer-implemented method, comprising:

identifying a first data set and a second data set,

establishing a plurality of relationships, each of the plurality of relationships linking an element of the first data set to an element of the second data set,

identifying a set of categories for categorizing elements of the first and second data sets,

initializing a plurality of first ranking vector, each of the first ranking vectors corresponding to an element of the first data set, each component of the each first ranking vector a representing a degree to which the element is related to one of the categories,

initializing a plurality of second ranking vector, each of the second ranking vectors corresponding to an element of the second data set, each component of the each second ranking vector u representing a degree to which the element is related to one of the categories,

assigning a link weight to each of the plurality of relationships of interconnected elements of the first and second data sets, said link weight being modifiable by interaction between the elements linked by the each relationship,

computing a similarity quotient for the each relationship of the plurality of relationships from a formula;

2. The method of claim 1, further comprising:

ranking the plurality of relationships based on values of the similarity quotients.

3. The method of claim 1, further comprising:

applying a clustering operation to the plurality of relationships to generate clusters of similarity quotients,

wherein the relationships are classified based on the clusters of similarity quotients.

4. The method of claim 3, wherein the plurality of relationships form a network of interconnected elements of the first and second data sets, the network including indirect links between elements of the same data set and between elements of different data sets;

and wherein the clustering operation is applied to a section of the network.

5. The method of claim 1, wherein each interaction between any pair of elements modifies at least one of the ranking vectors corresponding to the pair of interacting elements, the pair of interacting elements being 1) any two elements of the first data set, 2) any two elements of the second data set, or 3) an element of the first data set and an element of the second data set.

6. The method of claim 5, wherein each component a^j of the at least one ranking vector a is incremented by an amount proportional to a difference between components a^j and u^j of the ranking vectors a and u, the amount being computed from κ

*(u^j−

a^j), κ

being an interaction coefficient.

7. The method of claim 6, further comprising:

keeping a record of the at least one ranking vector a after each interaction of the interactions,

generating a trajectory of the at least one ranking vector a as a function of the interactions, and

predicting values of components of the at least one ranking vector a by extrapolating the trajectory to extrapolate the at least one ranking vector a and evaluating the extrapolated at least one ranking vector a.

8. The method of claim 6, wherein the interaction coefficient varies during an interaction.

9. A computer-implemented method, comprising:

identifying a first data set and a second data set,

establishing a plurality of relationships, each of the plurality of relationships linking an element of the first data set to an element of the second data set,

identifying a set of categories for categorizing elements of the first and second data sets,

initializing a plurality of first ranking vectors A, each of the first ranking vectors corresponding to an element of the first data set, each component of the each first ranking vector representing a degree to which the element is related to one of the categories,

initializing a plurality of second ranking vectors U, each of the second ranking vectors corresponding to an element of the second data set, each component of the each second ranking vector representing a degree to which the element is related to one of the categories,

assigning a link weight to each of the plurality of relationships of interconnected elements of the first and second data sets, said link weight being modifiable by interaction between the elements linked by the each relationship, wherein the ranking vectors are ranking vectors a and u, and each component a^j of the at least one ranking vector a is incremented in an interaction by an amount proportional to a difference between components a^j and u^j of the ranking vectors a and u, the amount being computed from κ

*(u^j−

a^j), κ

being an interaction coefficient;

computing a similarity quotient for the each relationship of the plurality of relationships, the similarity quotient being a measure of similarity between the elements linked by the each relationship and determined from the ranking vectors a and u, and

classifying the plurality of relationships based on values of the similarity quotients.

10. The method of claim 9, further comprising:

ranking the plurality of relationships based on values of the similarity quotients.

11. The method of claim 9, further comprising:

applying a clustering operation to the plurality of relationships to generate clusters of similarity quotients,

wherein the relationships are classified based on the clusters of similarity quotients.

12. The method of claim 11, wherein the plurality of relationships form a network of interconnected elements of the first and second data sets, the network including indirect links between elements of the same data set and between elements of different data sets;

and wherein the clustering operation is applied to a section of the network.

13. The method of claim 9, wherein each interaction between any pair of elements modifies at least one of the ranking vectors corresponding to the pair of interacting elements, the pair of interacting elements being 1) any two elements of the first data set, 2) any two elements of the second data set, or 3) an element of the first data set and an element of the second data set.

14. The method of claim 9, further comprising:

keeping a record of the at least one ranking vector a after each interaction of the interactions,

generating a trajectory of the at least one ranking vector a as a function of the interactions, and

predicting values of components of the at least one ranking vector a by extrapolating the trajectory to extrapolate the at least one ranking vector a and evaluating the extrapolated at least one ranking vector a.

15. The method of claim 9, wherein the interaction coefficient varies during an interaction.