Mining translations of web queries from web click-through data

• US 8,543,580 B2
• Filed: 12/23/2008
• Issued: 09/24/2013
• Est. Priority Date: 12/23/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Chen, Weizhu, Chen, Zheng
• Primary Examiner(s): Kerzhner, Aleksandr
• Application Number:
  US12/342,098
• Publication Number:
  US 20100161642A1
• Time in Patent Office: 1,736 Days
• Field of Search: 707/748, 707/999.004
• US Class Current: 707/748
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 2216/03 Data mining

Abstract

Methods and technologies providing translations of web queries based on an analysis of user behavior in click-through data. These methods and technologies generates large-scale and timely query translation pairs guided by a small set of seed word pairs from a dictionary, without relying on additional knowledge or complex models.

Claims

1. A method performed on a computing device and for generating a query translation pair, the method comprising:

establishing a plurality of seed query pairs that each comprise a source language query and a target language query that have the same meaning but are in different languages, where the source language query is different for each of the plurality of seed query pairs;

finding a plurality of URL pairs from click-through data, the plurality of URL pairs based on the seed query pair;

extracting URL pair patterns from ones of the plurality of URL pairs having a similarity score above a particular threshold;

extracting from the click-through data bilingual URL pairs identified by the URL pair patterns;

identifying in the click-through data candidate query pairs corresponding to the bilingual URL pairs; and

filtering the candidate query pairs based on a confidence score resulting in a query translation pair where queries of the query translation pair are translations of each other.

2. The method of claim 1 wherein the similarity score is calculated as

3. The method of claim 1 wherein the seed query pair is taken least in part from a bilingual dictionary.

4. The method of claim 1 wherein each of the plurality of URL pairs is found by identifying the most-clicked URL associated with each query of one of the plurality of seed query pairs.

5. The method of claim 1 wherein the extracted bilingual URL pairs follow at least one pattern identified by the URL pair patterns.

6. The method of claim 1 wherein a candidate query pair of the candidate query pairs is formed using queries of a click-through data pair in which a URL of each of the click-through data pair matches a unique one of another URL of a particular bilingual URL pair of the bilingual URL pairs.

7. The method of claim 1 wherein the filtering removes noisy candidate query pairs.

8. At least one computer storage device storing computer-executable instructions that, when executed by a computing device, cause the computing device to perform actions for generating a query translation pair, the actions comprising:

establishing a plurality of seed query pairs that each comprise a source language query and a target language query that have the same meaning but are in different languages, where the source language query is different for each of the plurality of seed query pairs;

finding a plurality of URL pairs from click-through data, the plurality of URL pairs based on the seed query pair;

extracting URL pair patterns from ones of the plurality of URL pairs having a similarity score above a particular threshold;

extracting from the click-through data bilingual URL pairs identified by the URL pair patterns;

identifying in the click-through data candidate query pairs corresponding to the bilingual URL pairs; and

filtering the candidate query pairs based on a confidence score resulting in a query translation pair where queries of the query translation pair are translations of each other.

9. The at least one computer storage device of claim 8 where the similarity score is calculated as

10. The at least one computer storage device of claim 8 where the seed query pair is taken least in part from a bilingual dictionary.

11. The at least one computer storage device of claim 8 where each of the plurality of URL pairs is found by identifying the most-clicked URL associated with each query of one of the plurality of seed query pairs.

12. The at least one computer storage device of claim 8 where the extracted bilingual URL pairs follow at least one pattern identified by the URL pair patterns.

13. The at least one computer storage device of claim 8 where a candidate query pair of the candidate query pairs is formed using queries of a click-through data pair in which a URL of each of the click-through data pair matches a unique one of another URL of a particular bilingual URL pair of the bilingual URL pairs.

14. A computing device and at least one software module together configured for generating a query translation pair based on:

establishing a plurality of seed query pairs that each comprise a source language query and a target language query that have the same meaning but are in different languages, where the source language query is different for each of the plurality of seed query pairs;

finding a plurality of URL pairs from click-through data, the plurality of URL pairs based on the seed query pair;

extracting URL pair patterns from ones of the plurality of URL pairs having a similarity score above a particular threshold;

extracting from the click-through data bilingual URL pairs identified by the URL pair patterns;

identifying in the click-through data candidate query pairs corresponding to the bilingual URL pairs; and

filtering the candidate query pairs based on a confidence score resulting in a query translation pair where queries of the query translation pair are translations of each other.

15. The computing device and at least one software module of claim 14 where the similarity score is calculated as

16. The computing device and at least one software module of claim 14 where the seed query pair is taken least in part from a bilingual dictionary.

17. The computing device and at least one software module of claim 14 where each of the plurality of URL pairs is found by identifying the most-clicked URL associated with each query of one of the plurality of seed query pairs.

18. The computing device and at least one software module of claim 14 where the extracted bilingual URL pairs follow at least one pattern identified by the URL pair patterns.

19. The computing device and at least one software module of claim 14 where a candidate query pair of the candidate query pairs is formed using queries of a click-through data pair in which a URL of each of the click-through data pair matches a unique one of another URL of a particular bilingual URL pair of the bilingual URL pairs.

20. The computing device and at least one software module of claim 14 where the filtering removes noisy candidate query pairs.