Techniques to perform relative ranking for search results

• US 7,974,974 B2
• Filed: 03/20/2008
• Issued: 07/05/2011
• Est. Priority Date: 03/20/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Taylor, Michael, Robertson, Stephen, Meyerzon, Dmitriy, Tankovich, Vladimir
• Primary Examiner(s): Mofiz; Apu M
• Assistant Examiner(s): Le; Hung D
• Application Number:
  US12/051,847
• Publication Number:
  US 20090240680A1
• Time in Patent Office: 1,202 Days
• Field of Search: None
• US Class Current: 707/729
• CPC Class Codes: G06F 16/24578 using ranking

Abstract

Techniques to perform relative ranking for search results are described. An apparatus may include an enhanced search component operative to receive a search query and provide ranked search results responsive to the search query. The enhanced search component may comprise a resource search module operative to search for resources using multiple search terms from the search query, and output a set of resources having some or all of the search terms. The enhanced search component may also comprise a proximity generation module communicatively coupled to the resource search module, the proximity generation module operative to receive the set of resources, retrieve search term position information for each resource, and generate a proximity feature value based on the search term position information. The enhanced search component may further comprise a resource ranking module communicatively coupled to the resource search module and the proximity generation module, the resource ranking module to receive the proximity feature values, and rank the resources based in part on the proximity feature values. Other embodiments are described and claimed.

Claims

1. A method, comprising:

searching for resources in response to a search query with a number of search terms;

retrieving search term position information for each resource;

generating with a computing device a proximity feature value based on the search term position information using an unordered cost function, wherein the unordered cost function does not require search terms to be in the same order as in the search query and wherein the unordered cost function is proportional to a number of different search terms in a chunk of the resource and inversely proportional to a length of the chunk and the number of search terms in the search query; and

ranking the resources based on the proximity feature value.

2. The method of claim 1, the search term position information to represent a position for each of the multiple search terms within a single resource.

3. The method of claim 1, the proximity feature value to represent a distance between positions for two or more of the search terms within a single resource.

4. The method of claim 1, comprising retrieving the search term position information for each resource from a proximity database.

5. The method of claim 1, comprising extracting the search term position information from each resource to retrieve the search term position information.

6. The method of claim 1, comprising ranking the resources based on the proximity feature value using a neural network algorithm.

7. The method of claim 6, comprising ranking the resources based on the proximity feature value and at least one ranking feature comprising BM25F, Click Distance, URL Depth, File Type priors, or language of the document.

8. An article comprising a storage medium containing instructions that if executed by a processor, cause the processor to:

search for resources in response to a search query with multiple a number of search terms;

retrieve search term position information for each resource;

generate a proximity feature value based on the search term position information using an unordered cost function, wherein the unordered cost function does not require search terms to be in the same order as in the search query and wherein the unordered cost function is proportional to a number of different search terms in a chunk of the resource and inversely proportional to a length of the chunk and the number of search terms in the search query; and

rank the resources based on the proximity feature value.

9. The article of claim 8, further comprising instructions that if executed enable the system to retrieve the search term position information for each resource from a proximity database.

10. The article of claim 8, further comprising instructions that if executed enable the system to extract the search term position information from each resource to retrieve the search term position information.

11. The article of claim 8, further comprising instructions that if executed enable the system to rank the resources based on the proximity feature value using a neural network algorithm.

12. The article of claim 8, further comprising instructions that if executed enable the system to:

rank the resources using a first ranking algorithm and a first set of ranking criteria to form a first set of ranked resources; and

rank a subset of the first set of ranked resources using a second ranking algorithm and a second set of ranking criteria including the proximity feature values to form a second set of ranked resources.

13. An apparatus, comprising:

at least one processor for executing computer readable instructions;

at least one computer readable storage medium storing the computer readable instructions that when executed by the at least one processor provide;

an enhanced search component operative to receive a search query and provide a ranked search results responsive to the search query, the enhanced search component comprising;

a resource search module operative to search for resources using a number of search terms from the search query, and output a set of resources having some or all of the search terms;

a proximity generation module communicatively coupled to the resource search module, the proximity generation module operative to receive the set of resources, retrieve search term position information for each resource, and generate a proximity feature value based on the search term position information using an unordered cost function, wherein the unordered cost function does not require search terms to be in the same order as in the search query and wherein the unordered cost function is proportional to a number of different search terms in a chunk of the resource and inversely proportional to a length of the chunk and the number of search terms in the search query; and

a resource ranking module communicatively coupled to the resource search module and the proximity generation module, the resource ranking module to receive the proximity feature values, and rank the resources based in part on the proximity feature values.

14. The apparatus of claim 13, the search term position information to represent a position for each of the multiple search terms within a single resource, and the proximity feature value to represent a distance between positions for two or more of the search terms within the single resource.

15. The apparatus of claim 13, comprising a proximity database communicatively coupled to the proximity generation module, the proximity database operative to store search term position information for each resource.

16. The apparatus of claim 13, the resource ranking module operative to rank the resources using multiple ranking algorithms, the first ranking algorithm to rank the resources using a first set of ranking criteria to form a first set of ranked resources, and the second ranking algorithm to rank a subset of the first set of ranked resources using a second set of ranking criteria including the proximity feature values to form a second set of ranked resources.