Selective retrieval metasearch engine

• US 20020165860A1
• Filed: 06/29/2001
• Published: 11/07/2002
• Est. Priority Date: 05/07/2001
• Status: Abandoned Application

• Current Assignee: NEC Research Institute, Inc. (NEC Corporation)
• Original Assignee: NEC Research Institute, Inc. (NEC Corporation)
• Inventors: Glover, Eric, Lawrence, Stephen R.
• Application Number:
  US09/896,338
• Publication Number:
  US 20020165860A1
• Time in Patent Office: Days
• US Class Current: 707/5
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 16/9532 Query formulation
• CPC Class Codes: G06F 16/9538 Presentation of query results

Abstract

A selective retrieval metasearch engine uses relevance estimation and confidence computation to select documents for which additional information is to be obtained. The additional information is used to update relevance estimation for the selected documents. A selective retrieval metasearch engine improves execution time, resource usage, throughput, and/or result quality.

Claims

1. A selective retrieval metasearch engine comprising:

means for accepting a search query;

means for sending the search query to at least one search engine and for retrieving results of the search query from the at least one search engine;

means for estimating the relevance of each result retrieved;

means for computing a confidence of the relevance estimation for each result retrieved;

means for selecting results using the computed confidence of the relevance estimation;

means for obtaining additional information about the selected results;

means for updating the relevance estimation based on the additional information obtained for each selected result;

means for ranking the results retrieved based on the relevance estimation of each result retrieved; and

means for returning the ranked results.

2. A selective retrieval metasearch engine as set forth in claim 1, further comprising means for transforming the search query before sending the search query to at least one search engine.

3. A selective retrieval metasearch engine as set forth in claim 1, where the search query comprises at least one keyword.

4. A selective retrieval metasearch engine as set forth in claim 1, where the search query comprises additional information.

5. A selective retrieval metasearch engine as set forth in claim 1, where the search query comprises at least one keyword and additional information.

6. A selective retrieval metasearch engine as set forth in claim 1, where said means for obtaining additional information about the selected results includes retrieving the current contents of the selected results.

7. A selective retrieval metasearch engine as set forth in claim 1, where said means for obtaining additional information about the selected results includes obtaining information selected from the group consisting of link statistics, word statistics, and other document statistics.

8. A selective retrieval metasearch engine as set forth in claim 1, where said means for estimating the relevance of each result includes similarity measures means.

9. A selective retrieval metasearch engine as set forth in claim 1, where said means for estimating the relevance of each result includes machine learning means.

10. A selective retrieval metasearch engine as set forth in claim 9, where said means for estimating the relevance of each result includes a neural network.

11. A selective retrieval metasearch engine as set forth in claim 9, where said means for estimating the relevance of each result includes a support vector machine.

12. A selective retrieval metasearch engine as set forth in claim 1, where said means for computing a confidence includes using information provided by the at least one search engine.

13. A selective retrieval metasearch engine as set forth in claim 1, where said means for computing a confidence includes using similarity measures.

14. A selective retrieval metasearch engine as set forth in claim 1, where said means for computing a confidence includes using machine learning means.

15. A selective retrieval metasearch engine as set forth in claim 14, where said means for computing a confidence includes using a neural network

16. A selective retrieval metasearch engine as set forth in claim 14, where said means for computing a confidence includes using a support vector machine

17. A selective retrieval metasearch engine as set forth in claim 1, where said means for computing a confidence includes estimating an accuracy of classifying the result.

18. A selective retrieval metasearch engine as set forth in claim 1, where said means for selecting results includes means for comparing the confidence with a threshold.

19. A selective retrieval metasearch engine as set forth in claim 18, where said means for selecting results further comprises dynamically altering the threshold based on system load.

20. A selective retrieval metasearch engine as set forth in claim 18, where said means for selecting results further comprises dynamically altering the threshold based on user preference.

21. A selective retrieval metasearch engine as set forth in claim 18, where the threshold is based on the estimated relevance.

22. A selective retrieval metasearch engine as set forth in claim 18, where the threshold is based on relevance estimation for results that have already been estimated.

23. A selective retrieval metasearch engine as set forth in claim 1, where said means for returning results to the user presents initial results based on initial relevance estimations, and the relevance and rank of documents are updated as additional information about the selected results are obtained.

24. A selective retrieval metasearch engine as set forth in claim 23, where said means for obtaining additional information about the selected results obtains additional information from the selected results which is most expected to improve overall results of the metasearch engine.

25. A selective retrieval metasearch engine as set forth in claim 1, where said means for returning the ranked results comprises returning the ranked results to a user.

26. A selective retrieval metasearch engine as set forth in claim 1, where said means for returning the ranked results comprises storing the ranked results.

27. A selective retrieval metasearch engine as set forth in claim 1, where said means for returning the ranked results comprises further processing the ranked results.

29. A method of performing selective retrieval metasearch as set forth in claim 28, further comprising transforming the search query before sending the search query to at least one search engine.

30. A method of performing selective retrieval metasearch as set forth in claim 28, where the search query comprises at least one keyword.

31. A method of performing selective retrieval metasearch as set forth in claim 28, where the search query comprises additional information.

32. A method of performing selective retrieval metasearch as set forth in claim 28, where the search query comprises at least one keyword and additional information.

33. A method of performing selective retrieval metasearch as set forth in claim 27, where said obtaining additional information about the selected results includes retrieving the current contents of the selected results.

34. A method of performing selective retrieval metasearch as set forth in claim 28, where said obtaining additional information about the selected results includes obtaining information selected from the group consisting of link statistics, word statistics, and other document statistics.

35. A method of performing selective retrieval metasearch as set forth in claim 28, where said estimating the relevance of each result includes using similarity measures.

36. A method of performing selective retrieval metasearch as set forth in claim 28, where said estimating the relevance of each result includes using machine learning.

37. A method of performing selective retrieval metasearch as set forth in claim 36, where said estimating the relevance of each result includes using a neural network.

38. A method of performing selective retrieval metasearch as set forth in claim 36, where said estimating the relevance of each result includes using a support vector machine.

39. A method of performing selective retrieval metasearch as set forth in claim 28, where said computing a confidence includes using information provided by the at least one search engine.

40. A method of performing selective retrieval metasearch as set forth in claim 28, where said computing a confidence includes using information provided by similarity measures.

41. A method of performing selective retrieval metasearch as set forth in claim 28, where said computing a confidence includes using machine learning means.

42. A method of performing selective retrieval metasearch as set forth in claim 41, where said computing a confidence includes using a neural network.

43. A method of performing selective retrieval metasearch as set forth in claim 41, where said computing a confidence includes using a support vector machine.

44. A method of performing selective retrieval metasearch as set forth in claim 28, where said computing a confidence includes estimating an accuracy of classifying the result.

45. A method of performing selective retrieval metasearch as set forth in claim 28, where said selecting results includes comparing the confidence with a threshold.

46. A method of performing selective retrieval metasearch as set forth in claim 43, where said selecting results further comprises dynamically altering the threshold based on system load.

47. A method of performing selective retrieval metasearch as set forth in claim 43, where said selecting results further comprises dynamically altering the threshold based on user preference.

48. A method of performing selective retrieval metasearch as set forth in claim 43, where the threshold is based on the estimated relevance.

49. A method of performing selective retrieval metasearch as set forth in claim 43, where the threshold is based on relevance estimation for results that have already been estimated.

50. A method of performing selective retrieval metasearch as set forth in claim 28, where said returning results to the user presents initial results based on initial relevance estimations, and the relevance and rank of documents are updated as additional information about the selected results are obtained.

51. A method of performing selective retrieval metasearch as set forth in claim 50, where said obtaining additional information about the selected results obtains additional information from the selected results which is most expected to improve overall results of the metasearch engine.

52. A method of performing selective retrieval metasearch as set forth in claim 28, where said returning the ranked results comprises storing the ranked results.

53. A method of performing selective retrieval metasearch as set forth in claim 28, where said returning the ranked results comprises further processing the ranked results.

54. A method of performing selective retrieval metasearch as set forth in claim 28, where said means for returning the ranked results comprises returning the ranked result to a user.

55. A method of performing selective retrieval metasearch as set forth in claim 28, where said computing a confidence, said selecting results, said obtaining additional information, and said updating the relevance estimation are repeated a plurality of times.

28. A method of performing selective retrieval comprising the steps of:

accepting a search query;

sending the search query to at least one search engine and retrieving results of the search query from the at least one search engine;

estimating the relevance of each result retrieved;

computing a confidence of the relevance estimation for each result retrieved;

selecting results using the computed confidence of the relevance estimation;

obtaining additional information about the selected results;

updating the relevance estimation based on the additional information obtained for each selected result;

ranking the results retrieved based on the relevance estimation of each result retrieved; and

returning the ranked results.