Performant relevance improvements in search query results

• US 20060253428A1
• Filed: 05/06/2005
• Published: 11/09/2006
• Est. Priority Date: 05/06/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Ratnaparkhi, Adwait, Liu, Jun, Katariya, Sanjeev, Yao, Qi, Green, Bradley R.
• Granted Patent: US 7,529,736 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 16/9538 Presentation of query results
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching

Abstract

Property store information and an aggregation of a plurality of ranking mechanisms, including a learning mechanism, are leveraged to provide performant query results with increased user relevancy. The learning mechanism permits query feedback to be accepted to facilitate in optimizing user relevance. This mechanism can also be incorporated with traditional Information Retrieval (IR) components, each supplying independent ranking to a relevance aggregation function that determines relevancy at a high level. This precludes diminishing the value of query feedback that occurs when the data is fed into traditional IR algorithms. By allowing the query feedback to maintain its proper weighting and utilizing scope and bias capabilities of the property store information, relevance increases in a highly performant manner.

Claims

1. A system that facilitates obtaining search query results, comprising:

a plurality of search query result ranking components, including a learning component, that receive a search query and separately determine search query results for the search query utilizing, at least in part, a property store; and

a relevance component that determines an optimum search query result relevancy ranking via aggregation of the separately determined search query results.

2. The system of claim 1, the relevance component employs a weighted summation to facilitate in the aggregation of the separately determined search query results.

3. The system of claim 1, the relevance component employs a product summation based on selection of an á

priori probabilistic relationship.

4. The system of claim 3, the product summation based on naï

ve Bayesian modeling and/or N-gram feature modeling.

5. The system of claim 1, the plurality of search query result ranking components including a traditional information retrieval (IR) component.

6. The system of claim 5, the traditional information retrieval (IR) component utilizes an OKAPI-based algorithm to facilitate in determining its search query results.

7. The system of claim 1, the learning component utilizes user search query feedback to facilitate in determining its search query results.

8. The system of claim 7, the user search query feedback is comprised of explicit user search query feedback, implicit user search query feedback, and/or user click-through of search query results data.

9. The system of claim 8, the explicit user search query feedback comprising user responses to a search query result relevancy prompt.

10. The system of claim 8, the implicit user search query feedback comprising observed heuristics.

11. A method for facilitating determination of search query results, comprising:

receiving a search query;

separately determining search query results for the search query via a plurality of search query result ranking processes, including a learning process, that utilize, at least in part, a property store to bias and/or scope the search query result ranking processes; and

determining an optimum search query result relevancy ranking via aggregation of the separately determined search query results.

12. The method of claim 11, aggregation of the separately determined search query results comprising:

determining a first set of highest ranking n results in terms of OKAPI ranking, where n is an integer from one to infinity;

selecting scores that are coerced via a best bets procedure and moving them to a top of the first set;

determining a second set of highest ranking n results in terms of a learning relevance score;

merging the first and second sets of highest ranking n results via one-to-P interleaving, where P is an integer from one to infinity;

removing duplicate results; and

returning a first n results of the merged set.

13. The method of claim 11 further comprising:

employing a weighted summation to facilitate in the aggregation of the separately determined search query results.

14. The method of claim 11 further comprising:

employing a product summation based on selection of an a priori probabilistic relationship.

15. The method of claim 14, the product summation based on naï

ve Bayesian modeling and/or N-gram feature modeling.

16. The method of claim 11 further comprising:

utilizing user search query feedback to facilitate in determining learning search query results.

17. The method of claim 16, the user search query feedback is comprised of explicit user search query feedback, implicit user search query feedback, and/or click-through of search query results by a user.

18. The method of claim 11 further comprising:

employing a learning process that utilizes a learning model generated from, at least in part, direct query feedback, aggregated question clustering feedback, and/or aggregated query concept feedback.

19. The method of claim 11, the property store comprising a property database and/or a property-based index.

20. A system that facilitates obtaining search query results, comprising:

means for receiving a search query;

means for separately determining search query results for the search query via a plurality of search query result ranking means, including a learning means, that employ, at least in part, a property store; and

means for determining an optimum search query result relevancy ranking via aggregation of the separately determined search query results.