User query history expansion for improving language model adaptation

• US 9,299,342 B2
• Filed: 07/23/2015
• Issued: 03/29/2016
• Est. Priority Date: 09/23/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Chang, Shuangyu, Levit, Michael, Buntschuh, Bruce Melvin
• Primary Examiner(s): Desir, Pierre-Louis
• Assistant Examiner(s): THOMAS-HOMESCU, ANNE L
• Application Number:
  US14/807,106
• Publication Number:
  US 20150325237A1
• Time in Patent Office: 250 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24575 using context
• CPC Class Codes: G06F 16/9024 Graphs; Linked lists G06F16/9027 takes precedence
• CPC Class Codes: G06F 16/90324 using system suggestions
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/34 Adaptation of a single recogniser for parallel processing, e.g. by use of multiple processors or cloud computing

Abstract

Query history expansion may be provided. Upon receiving a spoken query from a user, an adapted language model may be applied to convert the spoken query to text. The adapted language model may comprise a plurality of queries interpolated from the user'"'"'s previous queries and queries associated with other users. The spoken query may be executed and the results of the spoken query may be provided to the user.

Claims

1. A computer-implemented method comprising:

receiving a spoken query from a user;

processing the spoken query using an adapted language model, wherein the adapted language model is created using data associated with the spoken query, previous queries of the user and previous queries of a plurality of other users that share at least one common characteristic with the previous queries of the user, wherein the common characteristic is data for a selected uniform resource locator that is included in the previous queries of the user and included in the previous queries of the plurality of other users; and

generating, using the adapted language model, a recognition result for the spoken query.

2. The computer-implemented method of claim 1, further comprising converting the spoken query to text, and wherein the converted text for the spoken query is processed by the adapted language model to generate the recognition result.

3. The computer-implemented method of claim 2, wherein the generating of the recognition result for the spoken query further comprises modifying the converted text of the spoken query based on the processing of the adapted language model.

4. The computer-implemented method of claim 1, further comprising generating a personalized language model for the user based on the adapted language model, and applying the personalized language model to process a subsequent query of the user.

5. The computer-implemented method of claim 1, wherein the applying of the adapted language model further comprises analyzing a click-query graph to identify the common characteristic between at least one previous query of the user and at least a previous query of another user, and using result data from the analyzing of the click-query graph to recognize the spoken query.

6. The computer-implemented method of claim 5, wherein the query-click graph is a bipartite graph providing analysis of previous query data in association with content corresponding to one or more selected uniform resource locators.

7. The computer-implemented method of claim 1, wherein the applying of the adapted language model further comprises identifying a cluster of users that is associated with the previous queries of the plurality of other users, and using additional query data of the cluster of users to expand the adapted language model to generate the recognition result for the spoken query.

8. The computer-implemented method of claim 1, wherein application of the adapted language model identifies queries that include the data for the selected uniform resource locator that corresponds with content having a lexical similarity to the previous queries of the user, wherein the identified queries are used to generate the recognition result for the spoken query, and wherein the lexical similarity is at least one selected from a group consisting of:

a lexical match between the previous queries of the user and the previous queries of the plurality of other users, a word class match between the set previous queries of the user and the previous queries of the plurality of other users, a phrase class match between the previous queries of the user and the previous queries of the plurality of other users, a match in latent attributes between the previous queries of the user and the previous queries of the plurality of other users, and a match in a query topic between the previous queries of the user and the previous queries of the plurality of other users.

9. A system comprising:

at least one processor; and

a memory operatively connected with the processor storing instructions, that when executed on the processor, cause the processor to execute operations that comprise;

receiving a spoken query from a user,

creating an adapted language model to process the spoken query, wherein the adapted language model is created by expanding a set of previous queries of the user by interpolating previous queries of a plurality of other users that share at least one common characteristic with the previous queries of the user, wherein the common characteristic is data for a selected uniform resource locator that is included in the previous queries of the user and included in the previous queries of the plurality of other users, and processing data associated with the spoken query by applying the adapted language model to generate a recognition result for the spoken query.

10. The system of claim 9, wherein the processing of the data associated with the spoken query further comprises converting the spoken query to text for subsequent processing.

11. The system of claim 9, wherein the operations further comprising generating a personalized language model for the user based on the adapted language model, and applying the personalized language model to process a subsequent query of the user.

12. The system of claim 9, wherein the processing comprises analyzing a click-query graph to identify the common characteristic between at least one previous query of the user and at least a previous query of another user, and using result data from the analyzing of the click-query graph to convert the spoken query.

13. The system of claim 12, wherein the query-click graph is a bipartite graph providing analysis of previous query data in association with the data for one or more selected uniform resource locators.

14. The system of claim 9, wherein the applying of the adapted language model further comprises identifying a cluster of users that is associated with the previous queries of the plurality of other users, and using additional query data of the cluster of users to expand the adapted language model in processing the data of the spoken query.

15. The system of claim 9, wherein the applying identifies a subset of queries that include the data for the selected uniform resource locator that corresponds with content having a lexical similarity to the previous queries of the user, wherein the identified subset of queries are used to generate a recognition result for the spoken query, and wherein the lexical similarity is at least one selected from a group consisting of:

a lexical match between the previous queries of the user and the previous queries of the plurality of other users, a word class match between the set previous queries of the user and the previous queries of the plurality of other users, a phrase class match between the previous queries of the user and the previous queries of the plurality of other users, a match in latent attributes between the previous queries of the user and the previous queries of the plurality of other users, and a match in a query topic between the previous queries of the user and the previous queries of the plurality of other users.

16. A computer-readable storage device which stores a set of instructions, which when executed on at least one processor, cause the processor to perform a method comprising:

receiving a spoken query from a user;

applying an adapted language model to convert the spoken query to text, wherein the adapted language model is created by expanding previous queries of the user by interpolating previous queries of a plurality of other users that share at least one common characteristic with the previous queries of the user, wherein the common characteristic is data for a selected uniform resource locator that is included in the previous queries of the user and included in the previous queries of the plurality of other users; and

converting the spoken query to text based on application of the adapted language model.

17. The computer-readable storage device of claim 16, wherein the method executed by the processor further comprises analyzing a click-query graph to identify the common characteristic between at least one previous query of the user and at least a previous query of another user, and converting the spoken query using result data from the analyzing of the click-query graph.

18. The computer-readable storage device of claim 17, wherein the query-click graph is a bipartite graph providing analysis of previous query data in association with content of for one or more selected uniform resource locators.

19. The computer-readable storage device of claim 16, wherein the method executed by the processor further comprises generating a personalized language model for the user based on the adapted language model, and applying the personalized language model to process a subsequent query of the user.

20. The computer-readable storage device of claim 16, wherein the applying of the adapted language model further comprises identifying a cluster of users that is associated with the previous queries of the plurality of other users, and using additional query data of the cluster of users to convert the spoken query.