SCALABLE USER INTENT MINING USING A MULTIMODAL RESTRICTED BOLTZMANN MACHINE

• US 20160188726A1
• Filed: 12/31/2014
• Published: 06/30/2016
• Est. Priority Date: 12/31/2014
• Status: Active Grant

• Current Assignee: TCL Research America, Inc. (TCL Technology Group Corp.)
• Original Assignee: TCL Research America, Inc. (TCL Technology Group Corp.)
• Inventors: SHANG, YUE, SONG, XIAOLI, LIU, MENGWEN, GUO, LIFAN, DING, WANYING, WANG, HAOHONG
• Granted Patent: US 9,910,930 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/9032 Query formulation

Abstract

A method for scalable user intent mining is provided. The method includes detecting named entities from a plurality of query logs in a public query log dataset and generating features of the plurality of query logs based on the detected named entities. The method also includes applying a multimodal restricted boltzmann machine (RBM) on the generated features of the plurality of query logs to train a public multimodal RBM and generating a plurality of public query representations. Further, the method includes receiving a search query from a user, determining whether there are a plurality of history queries of the user. When there is no history query, user intent is predicted using the public multimodal RBM. When there are the history queries, the public multimodal RBM is applied on the plurality of history queries to train a personalized multimodal RBM, and the user intent is predicted using the personalized multimodal RBM.

Claims

1. A method for scalable user intent mining, comprising:

detecting named entities from a plurality of query logs in a public query log dataset;

based on the detected named entities, generating corresponding features of the plurality of query logs;

applying a multimodal restricted boltzmann machine (RBM) on the corresponding features of the plurality of query logs to train a public multimodal RBM;

generating a plurality of public query representations;

receiving a search query from a user;

determining whether there are a plurality of history queries of the user;

when there is no history query of the user, predicting user intent using the public multimodal RBM; and

when there are the plurality of history queries of the user, applying the public multimodal RBM on the plurality of history queries of the user to train a personalized multimodal RBM, and predicting the user intent using the personalized multimodal RBM.

2. The method according to claim 1, wherein:

the plurality of history queries of the user are the latest M search queries of the user within a pre-defined period of time, wherein M is an integer greater than 1.

3. The method according to claim 1, wherein applying the public multimodal RBM on the plurality of history queries of the user to train a personalized multimodal RBM further includes:

detecting named entities from the plurality of history queries of the user;

based on the detected named entities, generating corresponding features of the plurality of history queries of the user;

generating at least one personal query representation of the plurality of history queries of the user using the public multimodal RBM;

based on the at least one personal query representation, comparing the at least one personal query representation with the plurality of public query representations to find one or more relevant users that share at least one similar interest with the user; and

training the public multimodal RBM using data from the one or more relevant users to generate the personalized multimodal RBM.

4. The method according to claim 1, further including:

based on the predicted user intent, presenting at least one search result to the user.

5. The method according to claim 3, wherein comparing the at least one personal query representation with the plurality of public query representations to find one or more relevant users that share at least one similar interest with the user further includes:

based on the at least one personal query representation, calculating similar named entities with the plurality of queries in the public query log dataset;

based on the calculated similar named entities, collecting identifications of sections of the plurality of queries with significant similar named entities, wherein the identifications of the sections are regarded as the relevant users having the at least one similar interest with the user; and

based on the collected identifications of the sections, building a new dataset including the plurality of queries from the one or more relevant users and the plurality of history queries from the user.

6. The method according to claim 3, wherein detecting named entities from the plurality of history queries of the user further includes:

refining n-grams from the public query log dataset using pointwise mutual information, wherein n is an integer less than or equal to 7.

7. The method according to claim 6, wherein:

provided that of two candidate named entities M₁ and M₂ are two random variables, the pointwise mutual information of M₁ and M₂ is defined by;

8. A system for scalable user intent mining, comprising:

a named entity detection unit configured to detect named entities from a plurality of query logs in a public query log dataset;

a feature generation unit configured to generate corresponding features of the plurality of query logs based on the detected named entities;

a public multimodal restricted boltzmann machine (RBM) learning unit configured to apply a multimodal RBM on the generated features of the plurality of query logs in the public query log dataset to generate a plurality of public query representations;

a personal model learning module configured to apply the multimodal RBM on a plurality of history queries of the user to train a personalized multimodal RBM; and

an intent mining module configured to predict user intent using one of the public multimodal RBM and the personalized multimodal RBM.

9. The system according to claim 8, wherein:

the plurality of history queries of the user are the latest M search queries of the user within a pre-defined period of time, wherein M is an integer greater than 1.

10. The system according to claim 8, wherein the personal model learning module further includes:

a named entity detection unit configured to detect named entities from the plurality of history queries of the user;

a feature generation unit configured to generate corresponding features of the plurality of history queries of the user based on the detected named entities;

a public multimodal RBM unit configured to generate personal query representations of the plurality of history queries of the user using the public multimodal RBM learned from the public query log dataset; and

a personalized multimodal RBM learning unit configured to train the personalized multimodal RBM using a plurality of queries for personality learning, wherein the plurality of queries for personality learning include a plurality of queries from one or more relevant users and the plurality of history queries.

11. The system according to claim 8, wherein:

based on the predicted user intent, at least one search result is presented to the user.

12. The system according to claim 8, wherein:

n-grams from the public query log dataset are refined by using pointwise mutual information, wherein n is an integer less than or equal to 7.

13. The system according to claim 11, wherein:

provided that two candidate named entities M₁ and M₂ are two random variables, the pointwise mutual information of M₁ and M₂ is defined by;

14. A computer readable storage medium storing computer-executable instructions to execute operations for scalable user intent mining, the computer-executable instructions comprising:

detecting named entities from a plurality of query logs in a public query log dataset;

based on the detected named entities, generating corresponding features of the plurality of query logs;

applying a multimodal restricted boltzmann machine (RBM) on the generated features of the plurality of query logs to train a public multimodal RBM;

generating a plurality of public query representations;

receiving a search query from a user;

determining whether there are a plurality of history queries of the user;

when there is no history query of the user, predicting user intent using the public multimodal RBM; and

when there are the plurality of history queries of the user, applying the public multimodal RBM on the plurality of history queries of the user to train a personalized multimodal RBM, and predicting the user intent using the personalized multimodal RBM.

15. The computer readable storage medium according to claim 14, wherein:

the plurality of history queries of the user are the latest M search queries of the user within a pre-defined period of time, wherein M is an integer greater than 1.

16. The computer readable storage medium according to claim 14, wherein:

based on the predicted user intent, at least one search result is presented to the user.

17. The computer readable storage medium according to claim 14, wherein:

n-grams from the public query log dataset are refined by using pointwise mutual information, wherein n is an integer less than or equal to 7.

18. The computer readable storage medium according to claim 17, wherein;

provided that two candidate named entities M₁ and M₂ are two random variables, the pointwise mutual information of M₁ and M₂ is defined by;