Inferring search category synonyms from user logs

• US 20070118512A1
• Filed: 11/22/2005
• Published: 05/24/2007
• Est. Priority Date: 11/22/2005
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Riley, Michael, Liu, Zhiyan
• Granted Patent: US 7,627,548 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3338 Query expansion
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing

Abstract

Systems and methods for inferring category synonyms from prior result data may include identifying prior query data including query information and category information relating to the prior result data; determining whether the query information is a synonym for the category information; and using the query information and the category information in subsequent search requests if it is determined that the query information is a synonym for the category information.

Claims

1. A method, comprising:

identifying prior query data including query information and category information relating to prior result data;

determining whether the category information is a synonym for the query information; and

using the query information and the category information in subsequent search requests if it is determined that the query information is a synonym for the category information.

2. The method of claim 1, wherein the prior query data relates to searches for business listing information.

3. The method of claim 1, wherein identifying prior query data further comprises:

determining whether the prior query data had received a previous indication of relevancy.

4. The method of claim 3, wherein the previous indication of relevancy includes receiving a predefined user action associated with the prior query data.

5. The method of claim 4, wherein the predefined user action includes receiving a user selection of a driving directions request.

6. The method of claim 1, wherein the prior query data includes query terms associated with a user query for business listing information, a resulting business name returned in response to the user query, and a resulting business category associated with the resulting business name.

7. The method of claim 1, further comprising:

determining whether the prior query data relates to a name query or a categorical query.

8. The method of claim 7, wherein determining whether the prior query data relates to a name query or a categorical query, further comprises:

determining whether the query terms relate to a request for a business name or a business category.

9. The method of claim 8, wherein determining whether the query terms relate to requests for a business name or a business category, further comprises determining a name perplexity for the query terms, wherein query terms having higher name perplexity are categorical queries and query terms having lower name perplexity are name queries.

10. The method of claim 6, further comprising:

determining whether the prior query data relates to a name query or a categorical query; and

determining whether the query terms are a candidate synonym of the resulting business category if it is determined that the query terms are a categorical query.

11. The method of claim 10, further comprising:

determining whether a query, category pair associated with the query data is a hyponym;

rejecting the query terms as a synonym of the resulting business category if it is determined that the query, category pair associated with the query data is a hyponym; and

accepting the query terms as a synonym of the resulting business category if it is determined that the query, category pair associated with the query data is not a hyponym.

12. The method of claim 11, wherein determining whether a query, category pair associated with the query data is a hyponym comprises:

calculating an F-measure for the query, category pair.

13. The method of claim 12, wherein the F-measure for the query, category pair is calculated in accordance with the expression:

$F<br><br>\text{-}<br><br>\mathrm{measure}<br><br>\left(\mathrm{query},\mathrm{category}\right)=\frac{P<br><br>\left(\mathrm{query},\mathrm{category}\right)}{\left(a\times <br><br>P<br><br>\left(\mathrm{category}\right)+\left(1-a\right)\times <br><br>P<br><br>\left(\mathrm{query}\right)\right)},\mathrm{where}<br><br>\text{\hspace{1em}<br><br>}<br><br>P<br><br>\left(n\right)$

denotes a probability of occurrence in the prior query data, and ‘

a’

is a predefined variable for balancing recall and precision.

14. The method of claim 13, wherein ‘

a’

is within the range of approximately 0.7 to 0.9.

15. The method of claim 13, wherein ‘

a’

is approximately 0.85.

16. A system, comprising:

means for determining whether query data and category data included within prior search results are category synonyms; and

means for incorporating the query data into subsequent search requests associated with the category data if it is determined that the query data and the category data are category synonyms.

17. The system of claim 16, wherein the means for determining whether query data and category data included within prior search results are category synonyms further comprises:

means for determining whether the prior search results are good search results; and

means for determining whether the query data and the category data included within the prior search results are category synonyms when it is determined that the prior search results are good search results.

18. The system of claim 17, wherein the means for determining whether the prior search results are good search results further comprises:

means for determining whether a desirable user action has occurred in association with the prior search results; and

means for determining that the prior search results are good search results when it is determined that the desirable user action has occurred.

19. The system of claim 17, further comprising:

means for determining whether the query data is name query data or categorical query data; and

means for determining whether the query data and the category data included within the prior search results are category synonyms when it is determined that the query data is categorical query data.

20. A device comprising:

logic to identify historical log data including at least one query, category pair associated with a prior search request;

logic to determine whether the query in the query, category pair is a name query or a categorical query; and

logic to determine whether the query is a synonym of the category when it is determined that the query is a categorical query.

21. The device of claim 20, wherein the logic to determine whether the query is a synonym of the category, further comprises logic to determine whether the query is a hyponym of the category.

22. A computer-readable medium containing instructions for controlling a processor to perform a method, comprising:

identifying prior search result data, where the prior search result data includes at least a query term and a result category;

determining whether the query term is a synonym for the result category; and

using both the query term and result category in performing subsequent searches associated with the result category when the query term is a synonym for the result category.

23. A method comprising:

receiving a search query from a client;

identifying a result category based on the received search query;

identifying category synonyms for the identified result category; and

performing a result search based on the received search query, the result category, and identified category synonyms.

24. The method of claim 23, wherein the performing a result search further comprises:

performing a business name search based on the received query;

performing a business name search based on the identified category synonyms; and

performing a result category search based on the identified result category.

25. The method of claim 24, wherein the performing a result category search based on the identified result category is an exact match search resulting in exact matches to identified result categories.