INFORMATION RETRIEVAL USING USER-GENERATED METADATA

• US 20100100543A1
• Filed: 10/21/2009
• Published: 04/22/2010
• Est. Priority Date: 10/22/2008
• Status: Active Grant

• Current Assignee: Webmynd Incorporated
• Original Assignee: Webmynd Incorporated
• Inventors: Brady, James
• Granted Patent: US 8,156,120 B2
• Time in Patent Office: Days
• US Class Current: 707/732
• CPC Class Codes: G06F 16/31 Indexing; Data structures therefor; Storage structures

Abstract

System, device and method for using user-generated metadata to arrive at a modified search index that emphasizes a relationship between documents selected by a user during a prior search session and salient terms of those documents. An initial search index is modified by adding a synthetic term and a synthetic document to terms and documents that are used to arrive at the elements of the index and by modifying the relevance scores to highlight one or more of the search terms, the synthetic term, and the synthetic document. Synthetic term ties a cluster of related documents together and synthetic document ties terms of these documents together. Synthetic term is not found in any other documents and synthetic document does not belong to any normal corpus of documents. Modified index aids in re-generating prior user choices because it contains artifacts reflecting associations that user perceived between various terms and documents.

Claims

1. A computer-implemented method for improving information retrieval in response to a query input by a user to a user computer during a browsing session, the method comprising:

collecting user-generated metadata indicative of user choices regarding documents returned responsive to a prior query;

storing the user-generated metadata on a storage medium;

modifying an initial search index with the user-generated metadata to obtain a modified search index, by a computer processor; and

storing the modified search index,

wherein the initial search index is an initial term-document matrix having relevance scores as elements, the relevance scores indicating relationship of salient terms of documents in a corpus to the documents in the corpus, and

wherein the modified search index is obtained by modifying the initial term-document matrix to highlight the relationship between documents related to the user-generated metadata and a subset of the salient terms.

2. The method of claim 1,

wherein the user-generated metadata includes a set of user-selected documents, a set of salient terms of the user-selected documents and measured user interest in each of the user-selected documents, and

wherein the user-selected documents consist of documents selected by one or more users during one or more browsing sessions.

3. The method of claim 1,

wherein the modified search index includes relevance scores pertaining to a synthetic term associating together user-selected documents,

wherein the modified search index includes relevance scores pertaining to a synthetic document associating together salient terms of the user-selected documents,

wherein the user-selected documents are selected from among discovered documents returned responsive to the prior query, and

wherein the synthetic document does not occur in the corpus and the synthetic term does not occur in the documents of the corpus.

4. The method of claim 1, further comprising:

applying the query to the modified search index; and

providing a set of discovered documents selected from the corpus to the user at a user interface responsive to the query.

5. The method of claim 4, wherein the set of discovered documents is further reflective of textual similarity.

6. The method of claim 1, further comprising:

collecting user interest data by monitoring prior browsing sessions; and

incorporating the user interest data in the obtaining of the modified search index by further modifying the initial term-document matrix according to the user interest data.

7. The method of claim 1, further comprising:

applying the modified search index to latent semantic indexing.

8. The method of claim 1, wherein the modifying of the initial search index with the user-generated metadata to obtain the modified search index, includes:

Identifying one or more first subsets of the initial term-document matrix from columns and rows corresponding to first documents and first salient terms; and

increasing value of elements of the first subsets to obtain a first modified term-document matrix,

wherein the modified term-document matrix is set equal to the first modified term-document matrix,

wherein the first documents are a cluster of related documents, and

wherein the first salient terms are a subset of the salient terms.

9. The method of claim 8,

wherein the first salient terms consist of search terms of the prior query.

10. The method of claim 8, wherein the modifying of the initial search index with the user-generated metadata to obtain the modified search index further includes:

synthesizing a synthetic term associating the first documents together;

obtaining relevance scores of the synthetic term with the documents of the corpus;

adding relevance scores for the synthetic term to the first modified term-document matrix to obtain a second modified term-document matrix; and

adding the synthetic term to the first salient terms,

wherein the modified term-document matrix is set equal to the second modified term-document matrix.

11. The method of claim 10, wherein the modifying of the initial search index with the user-generated metadata to obtain the modified search index further includes:

synthesizing a synthetic document associating the first salient terms together;

obtaining relevance scores of the synthetic document with the salient terms; and

adding relevance scores for the synthetic document to third term-document matrix to obtain a fourth term-document matrix,

wherein the modified term-document matrix is set equal to the third modified term-document matrix.

12. The method of claim 11,

wherein the synthetic term is synthesized to have relevance scores with the first documents that are higher than relevance scores of the synthetic term with other documents in the corpus, and

wherein the synthetic document is synthesized to have relevance scores with the first salient terms that are higher than relevance scores of the synthetic document with the other salient terms.

13. The method of claim 11, wherein the first salient terms consist of the search terms and the synthetic term.

14. The method of claim 1, wherein the user-generated metadata is stored at the user computer.

15. The method of claim 1, wherein the user-generated metadata is stored at a third party repository.

16. The method of claim 1, wherein the initial search index is obtained from an external source.

17. The method of claim 1, wherein the initial search index is internally formed.

18. The method of claim 1, wherein the user-generated metadata is indicative of user choices from previous browsing sessions of the user.

19. The method of claim 1, wherein the user-generated metadata is indicative of user choices from previous browsing sessions of the user and other users.

20. A computer-implemented method for information retrieval in response to a present query input by a user to a user computer, the method comprising:

obtaining an initial term-document matrix having relevance scores as elements, the relevance scores indicating relationship of salient terms of documents in a corpus to the documents in the corpus;

collecting and storing user-generated metadata identifying first documents related together through previous user selections or previous user interests or both;

synthesizing a synthetic term associating the first documents together;

obtaining relevance scores of the synthetic term with the documents in the corpus; and

adding the relevance scores of the synthetic term to the initial term-document matrix to obtain a modified term-document matrix.

21. The method of claim 20, wherein the synthetic term is synthesized to have a high relevance score with the first documents.

22. The method of claim 20, wherein the obtaining of the relevance scores of the synthetic term with the documents in the corpus includes setting the relevance scores of the synthetic term with the documents in the corpus.

23. The method of claim 20, further comprising:

receiving the present query;

applying the present query to the modified term-document matrix to obtain present query results; and

providing the present query results to the user.

24. The method of claim 20, further comprising:

applying the modified term-document matrix to latent semantic indexing.

25. A computer-implemented method for information retrieval in response to a present query input by a user to a user computer, the method comprising:

obtaining an initial term-document matrix having relevance scores as elements, the relevance scores indicating relationship of salient terms of documents in a corpus to the documents;

collecting and storing user-generated metadata identifying first documents related together through previous user selections or previous user interests or both;

identifying first salient terms as a subset of salient terms of the first documents;

synthesizing a synthetic document associating the first documents together through associating the first salient terms together;

obtaining relevance scores of the synthetic document with the salient terms of the documents in the corpus; and

adding the relevance scores of the synthetic document to the initial term-document matrix to obtain a modified term-document matrix.

26. The method of claim 25, wherein the synthetic document is synthesized to have a high relevance score with the first salient terms.

27. The method of claim 25, wherein the obtaining of the relevance scores of the synthetic document with the salient terms of the documents in the corpus includes setting the relevance scores of the synthetic document with the salient terms of the documents in the corpus.

28. The method of claim 25, wherein the first salient terms consist of search terms of the present query, the method further comprising:

receiving the present query;

applying the present query to the modified term-document matrix to obtain present query results; and

providing the present query results to the user.

29. The method of claim 25, further comprising:

applying the modified term-document matrix to latent semantic indexing.

30. A device for improving information retrieval in response to a query by a user, the device comprising:

means for receiving an initial search index showing relevance of documents in a corpus of searched document to salient terms of the documents;

means for collecting and storing user-generated metadata indicative of user choices during a document search session;

means for modifying the initial search index according to the user-generated metadata to obtain a modified search index;

means for applying the query to the modified search index; and

means for providing a set of documents discovered from the corpus of searched documents responsive to the query,

wherein user-selected documents are chosen by the user from among the discovered documents, and

wherein the modified search index is obtained by modifying the initial search index to emphasize a relationship between the user-selected documents and a subset of salient terms of the user-selected documents.

31. The device of claim 30, wherein the user-generated metadata indicates a set of salient terms of the user-selected documents and measured user interest in each of the user-selected documents in addition to the user-selected documents.

32. The device of claim 30, wherein the user-selected documents consist of documents selected by one or more users during one or more document search sessions.

33. The device of claim 30,

wherein the modified search index includes relevance scores pertaining to a synthetic document associating together the salient terms of the user-selected documents, and

wherein the synthetic document does not occur in the corpus.

34. The device of claim 33,

wherein the modified search index includes relevance scores pertaining to a synthetic term associating together the user-selected documents, and

wherein the synthetic term does not occur in the documents of the corpus.

40. The repository of claim 34, wherein the repository is a component of a user computer or a third party repository or a component of a search engine.

35. A repository for collecting user-generated metadata and generating a modified search index, the repository comprising:

an input and output interface for receiving an initial search index;

a storage medium for storing the initial search index and for storing user-generated metadata indicative of user choices during a browsing session; and

a processor for modifying the initial search index according to the user-generated metadata to generate the modified search index,

wherein the modified search index is generated by modifying the initial search index to emphasize a relationship between user-selected documents and a subset of salient terms of the user-selected documents, and

wherein the user-selected documents are chosen from among documents returned to a user during the browsing session.

36. The repository of claim 35,

wherein the user-generated metadata indicates a set of salient terms of the user-selected documents and measured user interest in each of the user-selected documents in addition to the user-selected documents.

37. The repository of claim 35,

wherein the user-selected documents consist of documents selected by one or more users during one or more browsing sessions.

38. The repository of claim 35,

wherein the modified search index includes relevance scores pertaining to a synthetic term associating together the user-selected documents, and

wherein the synthetic term does not occur in documents of a corpus of searched documents.

39. The repository of claim 38,

wherein the modified search index includes relevance scores pertaining to a synthetic document associating together the salient terms of the user-selected documents, and

wherein the synthetic document does not occur in the corpus of searched documents.