Method and system for classifying users of a computer network

• US 20070150426A1
• Filed: 12/19/2006
• Published: 06/28/2007
• Est. Priority Date: 12/22/2005
• Status: Active Grant

• Current Assignee: Qnext Corp.
• Original Assignee: Qnext Corp.
• Inventors: Asher, Oren, Popovs, Anloonotn
• Granted Patent: US 7,873,584 B2
• Time in Patent Office: Days
• US Class Current: 706/20
• CPC Class Codes: G06F 18/24137 Distances to cluster centroïds
• CPC Class Codes: G06N 3/02 Neural networks

Abstract

A method of classifying an anonymous user communicating over a computer network, such as by age, gender, or personal interests, based on the content generated or otherwise communicated by the user. The method includes: (a) providing sets of reference attributes, wherein each set of reference attributes is associated with one of a number of personal profile classes; (b) analyzing the user content in order to extract a set of attributes corresponding to such content; (c) comparing the set of extracted attributes against the plural sets of reference attributes in order to find a close match; and (d) associating the user with the personal profile class of the matched set of reference attributes.

Claims

1. A method of identifying a user communicating over a computer network, comprising:

providing plural sets of reference attributes, wherein each set of reference attributes is associated with one of a plurality of personal profile classes;

analyzing text generated or provided by said user to the network in order to extract a set of attributes corresponding to said text;

comparing said set of extracted attributes against said plural sets of reference attributes in order to find a close match; and

associating said user with the personal profile class of said matched set of reference attributes.

2. A method according to claim 1, wherein said attributes include one or more of the following:

(a) spelling errors;

(b) grammatical errors;

(c) verb usage;

(d) adjective usage;

(e) use of emoticons; and

(f) usage of words from one or more topical vocabularies.

3. A method according to claim 1, wherein said attributes include the proportion of words in said text that are found in word vocabularies associated with each said personal profile class.

4. A method according to claim 3, including building said word vocabularies for each class by aggregating words from a plurality of known text samples.

5. A method according to claim 1, including:

provisioning a neural network, wherein the nodes thereof respectively represent said personal profile classes;

provisioning a plurality of training texts;

analyzing said training texts in order to extract a set of attributes corresponding to each said text;

training said neural network to establish weights for each said node based on said training text attributes; and

after the completion of said training, establishing said nodal weights as said set of reference attributes for each said personal profile class.

6. A method according to claim 5, wherein said training includes a self-organized mapping of said training text attributes to said nodes.

7. A method of identifying undesirable communications over a computer network between a first user and a second user thereof, said method comprising:

providing plural sets of reference attributes, wherein each set of reference attributes is associated with one of a plurality of personal profile classes;

providing pre-defined rules defining impermissible communication between various personal profile classes;

analyzing text generated or provided by said first user in order to extract a set of attributes corresponding to said first user text and analyzing text generated or provided by said second user in order to extract a set of attributes corresponding to said second user text;

comparing said set of extracted attributes for said first user text against said plural sets of reference attributes in order to find a close match and comparing said set of extracted attributes for said second user text against said plural sets of reference attributes in order to find a close match;

associating said first user with the personal profile class of said first matched set of reference attributes and associating said second user with the personal profile class of said second matched set of reference attributes;

signaling an alert in the event the personal profile class of said first user and the personal profile class of said second user contravene said rules.

8. A method according to claim 7, wherein said attributes include one or more of the following:

(a) spelling errors;

(b) grammatical errors;

(c) verb usage;

(d) adjective usage;

(e) use of emoticons; and

(f) usage of words from one or more topical vocabularies.

9. A method according to claim 7, wherein said attributes include the proportion of words in said text that are found in word vocabularies associated with each said personal profile class.

10. A method according to claim 9, including building said word vocabularies for each class by aggregating words from a plurality of known text samples.

11. A method according to claim 7, including:

provisioning a neural network, wherein the nodes thereof respectively represent said personal profile classes;

provisioning a plurality of training texts;

analyzing said training texts in order to extract a set of attributes corresponding to each said text;

training said neural network to establish weights for each said node based on said training text attributes; and

after the completion of said training, establishing said nodal weights as said set of reference attributes for each said personal profile class.

12. A method according to claim 11, wherein said training includes a self-organized mapping of said training text attributes to said nodes.

13. A method of ascertaining an approximate age of an anonymous user communicating over a computer network, said method comprising:

providing plural sets of reference attributes, wherein each set of reference attributes is associated with one of a plurality of age classes, and wherein said reference attributes for text include at least one of spelling errors and grammatical errors;

analyzing text or audio generated or provided by said user to the network in order to extract a set of attributes corresponding to said text or audio;

comparing said set of extracted attributes against said plural sets of reference attributes in order to find a close match; and

associating said user with the age class of said matched set of reference attributes.

14. A method according to claim 13, wherein said attributes include one or more of the following:

(a) verb usage;

(b) adjective usage;

(c) use of emoticons; and

(d) usage of words from one or more topical vocabularies; and

(e) usage or references to music or musical artists.

15. A method according to claim 13, wherein said attributes include the proportion of words in said text that are found in word vocabularies associated with each said personal profile class.

17. A method according to claim 13, including:

provisioning a neural network, wherein the nodes thereof respectively represent said personal profile classes;

provisioning a plurality of training texts;

analyzing said training texts in order to extract a set of attributes corresponding to each said text;

training said neural network to establish weights for each said node based on said training text attributes; and

after the completion of said training, establishing said nodal weights as said set of reference attributes for each said personal profile class.

18. A method according to claim 17, wherein said training includes a self-organized mapping of said training text attributes to said nodes.

16. A method according to claim 16, including building said word vocabularies for each class by aggregating words from a plurality of known text samples.