System and method for user modeling to enhance named entity recognition

• US 7,289,956 B2
• Filed: 05/27/2003
• Issued: 10/30/2007
• Est. Priority Date: 05/27/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Wang, Kuansan, Yu, Dong, Acero, Alejandro, Mau, Peter K. L., Mahajan, Milind
• Primary Examiner(s): Azad; Abul K.
• Application Number:
  US10/445,532
• Publication Number:
  US 20040243407A1
• Time in Patent Office: 1,617 Days
• Field of Search: None
• US Class Current: 704/240
• CPC Class Codes: G06F 40/295 Named entity recognition

Abstract

The present invention employs user modeling to model a user'"'"'s behavior patterns. The user'"'"'s behavior patterns are then used to influence named entity (NE) recognition.

Claims

1. A named entity (NE) list generator, comprising:

a NE identifier configured to receive information from a plurality of different sources and to identify NEs in the information, at least one of the plurality of different sources of the information including a plurality of NEs referred to by a given group of users; and

an analyzer calculating probabilities associated with the identified NEs to obtain a user group-specific NE list, specific to the given group of users, with associated probabilities.

2. The NE list generator of claim 1 wherein the given group of users comprises a single given user.

3. The NE list generator of claim 1 wherein the sensor is configured to identify NEs in the information based on a trigger input.

4. The NE list generator of claim 3 wherein at least one of the plurality of different sources comprises an application program.

5. The NE list generator of claim 4 wherein the trigger input comprises:

an application-generated trigger input, generated by the application program.

6. The NE list generator of claim 5 wherein the application comprises an electronic mail application and wherein the application-generated trigger comprises a send or receive event.

7. The NE list generator of claim 3 wherein the trigger input comprises a user-generated trigger input.

8. The NE list generator of claim 7 wherein at least one of the plurality of different sources comprises a word processing application and wherein the user-generated trigger input comprises a save document input.

9. The NE list generator of claim 3 wherein at least one of the sources comprises a speech recognition engine.

10. The NE list generator of claim 9 wherein the trigger input comprises a recognition result indication from the speech recognition engine.

11. The NE list generator of claim 3 wherein the trigger input comprises a timer input indicating that a desired time period has elapsed.

12. The NE list generator of claim 3 wherein the trigger input is based on a volume of information collected for processing by the NE identifier.

13. The NE list generator of claim 1 wherein the analyzer is configured to calculate the probabilities by weighting information based on the source of the information, and based on how closely the source is related to the given group of users.

14. The NE list generator of claim 13 wherein a first of the plurality of sources of information provides closely related information that is closely related to the given group of users and a second of the plurality of sources of information provides loosely related information that is more loosely related to the given group of users than the closely related information.

15. The NE list generator of claim 14 wherein the analyzer is configured to weight the closely related information more heavily in calculating the probabilities than the loosely related information.

16. The NE list generator of claim 1 wherein the NE list and probabilities are configured for integration into a plurality of different system components and wherein the analyzer calculates the probabilities based on system components in which the NE list and probabilities are integrated.

17. The NE list generator of claim 1 wherein the analyzer is configured to generate a probabilistic context free grammar from the NE list and probabilities.

18. The NE list generator of claim 1 wherein the analyzer is configured to cluster the identified NEs into groups.

19. The NE list generator of claim 18 wherein the analyzer is configured to generate a class-based language model from the NEs and probabilities and wherein the groups each correspond to a semantic class in the class-based language model.

20. The NE list generator of claim 1 wherein the analyzer is configured to generate a lexicon for a speech recognition engine based on the NEs and probabilities.

21. The NE list generator of claim 1 wherein the analyzer is configured to link the NE list and probabilities to controls.

22. A method of generating a named entity (NE) model, comprising:

receiving information from a source indicative of NE usage patterns of a given user, the NE usage patterns being indicative of a set of NEs referred to by the given user, wherein receiving information comprises receiving information from a plurality of different sources; and

generating a user-specific NE model, specific to the given user, based on the information received, wherein generating comprises identifying NFs in the information received and calculating a probability associated with each NP based on the information received, and wherein calculating the probability comprises calculating by weighting and information received differently based on the source from which it was received.

23. The method of claim 22 wherein generating a user-specific NE model comprises:

clustering identified NEs into one or more of a plurality of clusters.

24. The method of claim 23 wherein clustering comprises:

placing the NEs into clusters based on a source from which the NEs were identified.

25. The method of claim 23 wherein generating a user-specific NE model comprises:

generating a probabilistic context free grammar.

26. The method of claim 23 wherein generating a user-specific NE model comprises:

generating a class-based language model.

27. The method of claim 23 wherein generating a user-specific NE model comprises:

generating a lexicon for use in a speech recognition engine.

28. The method of claim 23 wherein generating a user-specific NE model comprises:

linking the user-specific NE model to controls.

29. A computer readable medium having stored thereon computer readable instructions which, when executed by the computer cause the computer to perform steps of:

receiving information from a plurality of different information sources, the information being indicative of a specific user'"'"'s usage of named entities (NEs);

generating a list of NEs in the information;

calculating probabilities associated with each NE in the list based on the information; and

integrating the NE list and probabilities into a computer component.

30. The computer readable medium of claim 29 wherein calculating probabilities comprises:

weighting the information based on how closely a source of the information is indicative of the specific user'"'"'s usage of NEs.