Context-sensitive search using a deep learning model

Patent metadata

Field	Value
Publication number	US-9535960-B2
Application number	US-201414252703-A
Country	US
Kind code	B2
Filing date	Apr 14, 2014
Priority date	Apr 14, 2014
Publication date	Jan 3, 2017
Grant date	Jan 3, 2017

Abstract

A search engine is described herein for providing search results based on a context in which a query has been submitted, as expressed by context information. The search engine operates by ranking a plurality of documents based on a consideration of the query, and based, in part, on a context concept vector and a plurality of document concept vectors, both generated using a deep learning model (such as a deep neural network). The context concept vector is formed by a projection of the context information into a semantic space using the deep learning model. Each document concept vector is formed by a projection of document information, associated with a particular document, into the same semantic space using the deep learning model. The ranking operates by favoring documents that are relevant to the context within the semantic space, and disfavoring documents that are not relevant to the context.

First claim

What is claimed is:

1. A method, implemented by one or more computing devices, for performing a search, comprising:
receiving a query, and associated query information;
identifying context information associated with the query, pertaining to a context in which the query has been submitted, the context having a plurality of parts;
transforming the context information into a plurality of context concept vectors in a semantic space, using a deep learning model, each context concept vector representing one of the plurality of parts of the context;
receiving document information associated with a document, the document having a plurality of parts;
transforming the document information into a plurality of document concept vectors in the same semantic space, using the deep learning model, each document concept vector representing one of the plurality of parts of the document;
performing pairwise comparisons of the context concept vectors with the document concept vectors to produce a relevance measure, indicating a degree of a defined semantic relationship between the context and the document;
determining a ranking score for the document based at least on the relevance measure; and
providing a search result based on the ranking score.

2. The method of claim 1 , further comprising:
repeating said receiving a query, identifying context information, transforming the context information, receiving document information or a previously generated document concept vector, comparing, and determining for a plurality of documents, to generate a plurality of ranking scores for the documents; and
ranking the documents based on the ranking scores.

3. The method of claim 2 , further comprising identifying the plurality of documents based on the query, in a preliminary ranking operation.

4. The method of claim 1 , wherein the document pertains to an entity, and wherein the document information describes at least one aspect of the entity.

5. The method of claim 1 , wherein the context information describes textual content in proximity to the query, within a source document.

6. The method of claim 1 , wherein the context information describes at least one demographic characteristic associated with a user who has submitted the query.

7. The method of claim 1 , wherein the context information describes prior behavior exhibited by a user who has submitted the query.

8. The method of claim 7 , wherein the prior behavior describes previous queries submitted by the user and/or previous search selections made by the user.

9. The method of claim 1 , wherein the context information describes a current location of a user who has submitted the query.

10. The method of claim 1 , wherein the context information describes a time at which the query has been submitted.

11. The method of claim 1 , wherein the ranking score for the document is also generated based on the context concept vector and/or the document concept vector.

12. The method of claim 1 , further comprising transforming the query information into a query concept vector in the semantic space, using the deep learning model, wherein the ranking score for the document is further based on the query concept vector.

13. The method of claim 1 , wherein the deep learning model is a multilayered neural network.

14. The method of claim 1 , wherein each of said transforming operations comprises, with respect to an input vector associated with the context information or the document information:
transforming the input vector into a reduced-dimension vector; and
projecting, using a multilayered-neural network, the reduced-dimension vector into a concept vector, the concept vector being associated with either the context concept vector or the document concept vector.

15. The method of claim 1 , wherein the deep learning model is trained using click-through data such that a conditional likelihood of clicked documents, given respective contexts, is maximized, and the conditional likelihood of un-clicked documents, given the respective contexts, is reduced.

16. A computer readable storage device for storing computer readable instructions, the computer readable instructions when executed by one or more processing devices perform a method for providing a ranking framework, the method comprising:
receiving a query and associated query information;
identifying context information associated with the query, pertaining to a context in which the query has been submitted, the context having a plurality of parts;
transforming the context information into a plurality of context concept vectors in a semantic space, using a deep learning model, each context concept vector representing one of the plurality of parts of the context;
receiving document information associated with a document, the document having a plurality of parts;
transforming the document information into a plurality of document concept vectors in the same semantic space, using the deep learning model, each document concept vector representing one of the plurality of parts of the document;
performing pairwise comparisons of the context concept vectors with the document concept vectors to produce a relevance measure, indicating a degree of a defined semantic relationship between the context and the document;
determining a ranking score for the document based at least on the relevance measure; and
providing a search result based on the ranking score.

17. The computer readable storage device of claim 16 , wherein determining a ranking score for the document comprises determining a ranking score for the document based on the context concept vector and/or the document concept vector.

18. The computer readable storage device of claim 16 , wherein the deep learning model is a multilayered neural network.

19. At least one computing device which implements a search engine, comprising:
a processor; and
executable instructions operable by the processor, the executable instructions comprising a method for performing a search, the method comprising:
receiving a query and associated query information;
identifying context information associated with the query, pertaining to a context in which the query has been submitted, the context having a plurality of parts;
transforming the context information into a plurality of context concept vectors in a semantic space, using a deep learning model, each context concept vector representing one of the plurality of parts of the context;
receiving document information associated with a document, the document having a plurality of parts;
transforming the document information into a plurality of document concept vectors in the same semantic space, using the deep learning model, each document concept vector representing one of the plurality of parts of the document;
performing pairwise comparisons of the context concept vectors with the document concept vectors to produce a relevance measure, indicating a degree of a defined semantic relationship between the context and the document;
determining a ranking score for the document based at least on the relevance measure; and
providing a search result based on the ranking score.

20. The at least one computing device of claim 19 , wherein the deep learning model is a deep neural network.

Assignees

• Microsoft Corp

Inventors

• Guo Chenlei
• Gao Jianfeng
• Wang Ye-Yi
• Deng Li
• He Xiaodong

Classifications

• G06F16/24578 Primary

  using ranking · CPC title

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title

• G06F16/248 Primary

  Presentation of query results · CPC title

• G06N3/045 Primary

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title