Real-time data mining

• US 9,558,267 B2
• Filed: 02/11/2011
• Issued: 01/31/2017
• Est. Priority Date: 02/11/2011
• Status: Active Grant

• Current Assignee: Maplebear Inc.
• Original Assignee: International Business Machines Corporation
• Inventors: Dasgupta, Koustuv, Banerjee, Nilanjan, Chakraborty, Dipanjan, Mittal, Sumit, Nagar, Seema, Joshi, Anupam, Rai, Angshu
• Primary Examiner(s): TRAN, ANHTAI V
• Application Number:
  US13/025,482
• Publication Number:
  US 20120209852A1
• Time in Patent Office: 2,181 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification

Abstract

A significant recent trend in the internet and mobile telephony has been the dominance of user generated content. As such, in mobile technology have permitted users to upload content onto the internet, whereby sites provide an easily accessible and manageable medium for users to share their thoughts and form a portal for media-rich exchanges. It has been found that much of what is exchanged by users in such settings is context-sensitive, ranging from users'"'"' moods and opinions, to communication about users'"'"' plans. Broadly contemplated herein, in accordance with at least one embodiment of the invention, is the employment of data mining in information repositories settings to efficiently classify an information stream in real-time and thereby discern user intent.

Claims

1. A computer-implemented method comprising:

using a processor to mine user related content information, wherein the information is mined from an information repository;

filtering the mined user related content information from the information repository, wherein the filtering comprises identifying a subset of the mined user related content information comprising information related to a predetermined category;

identifying, using a cosine similarity measure, a plurality of words having a similarity to a seed set of words by analyzing the subset, using a plurality of analyzers, wherein each analyzer is configured to capture a plurality of representational variations, from the information repository, related to the seed set of words;

classifying, based on the analyzing, the plurality of representational variations, wherein the classifying comprises ranking the filtered user related content information;

combining the classified plurality of representational variations of the user related content information from each of the plurality of analyzers, wherein the combining comprises identifying a relevancy of a representational variation to a user intent based upon the ranking of the filtered user related content information; and

training a classifier for characterizing real-time intention content from information repositories using the combined plurality of representational variations.

2. The method as claimed in claim 1, further comprising pre-processing the user related content information mined from an information repository prior to said analyzing.

3. The method as claimed in claim 2, wherein said pre-processing comprises extracting predetermined relevant user related content information from the user related content information repository.

4. The method as claimed in claim 1, wherein the mined user related content information comprises at least one of a message or e-mail or information stored in a repository or an instant message or a short message or telephone call or a browser.

5. The method as claimed in claim 1, further comprising testing the mined user related content information for satisfaction of at least one predetermined category.

6. The method as claimed in claim 5, further comprising at least one negative determination of the at least one predetermined category.

7. The method as claimed in claim 1, wherein said analyzing comprises analyzing in parallel the mined user related content information from the information repository by each of the plurality of analyzers.

8. The method as claimed in claim 1, wherein said combining comprises assigning weights to the classified plurality of representational variations of the user related content information from each of the plurality of analyzers.

9. The method as claimed in claim 8, wherein said combining further comprises summing the weighted classified plurality of representational variations of the user related content information from each of the plurality of analyzers.

10. The method as claimed in claim 1, wherein said assigning of a final classification comprises comparing the summed weights of the classified plurality of representational variations of the user related content information to a threshold.

11. The method as claimed in claim 10, wherein said assigning of a final classification comprises ascertaining a user intent in connection with the plurality of representational variations of the user related content information.

12. The method as claimed in claim 1, wherein one or more of the plurality of analyzers is taken from the group consisting essentially of a co-occurrence-driven analyzer, a POS-driven analyzer, a rule-based analyzer, a dependency grammar-based analyzer and a Δ

-TFIDF-based analyzer.

13. The method as claimed in claim 1, wherein one of the plurality of analyzers comprises a Δ

-TFIDF-based analyzer.

14. A system comprising:

at least a processor and a memory configured for;

mining user related content information, wherein the information is mined from an information repository;

filtering the mined user related content information from the information repository, wherein the filtering comprises identifying a subset of the mined user related content information comprising information related to a predetermined category;

identifying, using a cosine similarity measure, a plurality of words having a similarity to a seed set of words by analyzing the subset using a plurality of analyzers, wherein each analyzer is configured to capture a plurality of representational variations, from the information repository, related to the seed set of words;

classifying, based on the analyzing, the plurality of representational variations, wherein the classifying comprises ranking the filtered user related content information;

combining the classified plurality of representational variations of the user content information from each of the plurality of analyzers, wherein the combining comprises identifying a relevancy of a representational variation to a user intent based upon the ranking of the filtered user related content information; and

training a classifier for characterizing real-time intention content from information repositories using the combined plurality of representational variations.

15. The system as claimed in claim 14, further comprising pre-processing the user related content information prior to said analyzing, and wherein said pre-processing comprises extracting predetermined relevant user related information from the user related content information repository.

16. The system as claimed in claim 14, wherein the mined user related content information comprises at least one of a message or e-mail or information stored in a repository or an instant message or a short message or telephone call or a browser.

17. The system as claimed in claim 14, further comprising testing the mined user related content information for satisfaction of at least one predetermined category, and further comprising at least one negative determination of the at least one predetermined category.

18. The system as claimed in claim 14, wherein said analyzing comprises analyzing in parallel the mined user related content information from the information repository by each of a plurality of analyzers, and wherein said combining comprises assigning weights to the classified plurality of representational variations of the user related content information from each of the plurality of analyzers, and wherein said combining further comprises summing the weighted classified plurality of representational variations of the user related content information from each of the plurality of analyzers.

19. The system as claimed in claim 14, wherein said assigning of a final classification comprises comparing the summed weights of the classified plurality of representational variations of the user related content to a threshold, and wherein said assigning of a final classification comprises ascertaining a user intent in connection with the plurality of representational variations of the user related content information.

20. The system as claimed in claim 14, wherein one or more of the plurality of analyzers is taken from the group consisting essentially of a co-occurrence-driven analyzer, a POS-driven analyzer, a rule-based analyzer, a dependency grammar-based analyzer and a Δ

-TFIDF-based analyzer.