Systems and methods that employ a distributional analysis on a query log to improve search results

• US 20040254920A1
• Filed: 06/16/2003
• Published: 12/16/2004
• Est. Priority Date: 06/16/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Brill, Eric D., Carmichael, Philip F.
• Granted Patent: US 7,617,202 B2
• Time in Patent Office: Days
• US Class Current: 707/3
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9536 Search customisation based on social or collaborative filtering
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

The present invention provides systems and methods that employ a statistical distributional analysis to improve content search engine search results. In particular, a substring and/or a string sequence distributional algorithm can be applied to a set of queries to generate a distributional characteristic (e.g., a profile) for the set of queries, wherein the set is selected from a plurality of queries stored on a query log. Typically, the queries are selected based on a substring of interest and/or an identification of a user initiating searches. The distributional characteristic can then be employed to determine a distributional similarity measure that can be utilized in connection with a search to facilitate search results via providing a mechanism to determine synonymous search terms, spelling corrections/variations, and facilitate collaborative filtering, for example. Thus, the present invention employs a novel technique that mines and employs previous queries to enhance the query search results.

Claims

1. A system that facilitates analyzing a search query log, comprising:

a component that obtains a set of queries from the query log, and

a distributional component that defines a respective distributional characteristic for the set of queries.

2. The system of claim 1, the distributional characteristic generated from one of a substring distribution algorithm and a string sequence distribution algorithm.

3. The system of claim 1, the set of queries selected based in part on one or more of a distribution algorithm, a global unique identification (GUID), a substring, and a string.

4. The system of claim 1, further comprising a processing component that generates a statistic for the queries in the set of queries.

5. The system of claim 4, the statistic comprising a frequency of occurrence for the queries in the set of queries.

6. The system of claim 5, the distributional component employing the frequency of occurrence to generate a substring distributional characteristic.

7. The system of claim 6, the substring distributional characteristic generated for a plurality of users.

8. The system of claim 1, the distributional characteristic being a string sequence distributional characteristic.

9. The system of claim 8, the string sequence distributional characteristic associated with an individual user.

10. The system of claim 1, further comprising employing the distributional characteristic to generate a distributional similarity measure.

11. The system of claim 1, the query log including at least previously executed query searches and user information associated with the previously executed searches.

12. The system of claim 1, employed in connection with a search engine.

13. A system that employs distributional analysis on a query log to facilitate improving search engine query results, comprising:

a component that obtains a set of queries;

a profiling component that generates a distributional profile for the set of queries, and a similarity component that employs the distributional profile to generate a distributional similarity.

14. The system of claim 13, the component obtaining the set of queries by mining queries from the query log, based in part on at least one of a distributional algorithm, a substring, a string and a user identification.

15. The system of claim 13, the query log including one or more of previously executed query searches, user information associated with the previously executed searches, an indication as to whether the search was successful, and a probable reason why a search failed.

16. The system of claim 13, the profiling component further comprising determining the frequency of occurrence for respective search queries, and employing the frequency of occurrence to generate a substring distributional profile.

17. The system of claim 13, the distributional profile comprising a string sequence distributional profile.

18. The system of claim 13, the similarity component employing one of a cosine distance, a Jaccard coefficient, an entropy-based measure, a divergence measure and a relative separation measure to generate the distributional similarity.

19. The system of claim 13, the distributional similarity employed to provide at least one of a synonymous search term, a spelling correction, and a spelling variation.

20. The system of claim 13, the distributional similarity employed to facilitate collaborative filtering.

21. The system of claim 13, the distributional similarity employed to facilitate collaborative filtering to render search results with a greater degree of correlation to the content being searched, determining what queries are failing, and determine why queries are failing.

22. The system of claim 13, the distributional similarity employed to provide information regarding correlated searches to facilitate anticipating a user'"'"'s needs.

23. The system of claim 13, the distributional similarity employed to enhance the presentation of the search results to the user.

24. The system of claim 13, the distributional profile further employed to determine at least one of an alternative query(s) and an additional query(s) that can be of interest, wherein the alternative query(s) and the additional query(s) can be provided to the user.

25. The system of claim 13, further comprising a user interface (UI).

26. The system of claim 25, the UI comprising one of a graphical user interface (GUI) and a command-line interface.

27. The system of claim 25, the UI being a web browser.

28. The system of claim 26, the GUI comprising one or more of a text region, a graphic regions, a dialogue box, a static control, a drop-down-menu, a list box, a pop-up menu, an edit control, a combo box, a radio button, a check box, a push button, a vertical scroll bar, and a horizontal scroll bar.

29. A method to analyze a search query log, comprising:

obtaining a set of queries from the query log, and

generating a distributional profile for the queries.

30. The method of claim 29, generating the distributional profile from one of a substring distribution algorithm and a string sequence distribution algorithm.

31. The method of claim 29, the distributional profile comprising one of a substring distribution profile and a string sequence distribution profile.

32. The method of claim 29, the set of queries selected based in part on one ore more of the distribution algorithm, a substring, a string, a keyword, and a global unique identification (GUID).

33. The method of claim 29, further comprising generating a frequency of occurrence for the queries in the set of queries, and employing the frequency of occurrence to generate the distributional profile.

34. The method of claim 29, employed in connection with a search engine.

35. A method that employs distributional analysis to facilitate search engine searches, comprising:

mining a query log for a set of queries;

generating a distributional characteristic for the set of queries, and

generating a distributional similarity from the distributional characteristic.

36. The method of claim 35, the distributional characteristic comprising one of a substring distributional characteristic and a string sequence distributional characteristic.

37. The method of claim 35, employed in connection with a search engine to provide a mechanism to determine synonymous search terms, spelling corrections, and spelling variations, facilitate collaborative filtering, correlated searches, anticipating a user'"'"'s needs, and enhance a subsequent search in order to return results more correlated to the content being searched by the user.

38. The method of claim 35, the distributional similarity based on one of a cosine distance, a Jaccard coefficient, an entropy-based measure, a divergence measure and a relative separation measure.

39. The method of claim 35, the set of queries selected based in part on one or more of a distribution algorithm, a global unique identification (GUID), a substring, and a string.

40. The method of claim 35, further comprising generating a frequency of occurrence for the queries in the set of queries, and employing the frequency of occurrence to generate a substring distributional characteristic.

41. A data packet transmitted between two or more computer components to employ a distributional analysis technique to improve search query results, comprising:

a component that constructs a set of queries from a query log, based in part on one of a distributional algorithm, a substring, and a unique user identification, a component that generates a distributional profile for the set of queries, and a component that employs the distributional profile to generate a distributional similarity.

42. A computer readable medium storing computer executable components that employs distributional analysis on a query log to improve search query results, comprising:

a component that obtains a set of queries from the query log based on criteria associated with a distributional algorithm;

a component that employs the distributional algorithm to generate a distributional profile for the set of queries, and

a component that employs the distributional profile to generate a distributional similarity that can be utilized to improve content search query results.

43. A system that employs distributional analysis with query searches, comprising:

means for obtaining a set of queries from the query log, and

means for generating a distributional characteristic for the set of queries.

44. The method of claim 43, further comprising means for generating a distributional similarity from the distributional characteristic.