NATURAL LANGUAGE UNDERSTANDING (NLU) PROCESSING BASED ON USER-SPECIFIED INTERESTS

• US 20160098393A1
• Filed: 10/01/2014
• Published: 04/07/2016
• Est. Priority Date: 10/01/2014
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Hebert, Matthieu
• Granted Patent: US 10,817,672 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/685 using automatically derived transcript of audio data, e.g. lyrics speech recognition G10L15/00
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models

Abstract

Methods and apparatus for natural language understanding (NLU) processing based on user-specified interests. Information specifying a weight for each of a plurality of domains is received via a user interface. The plurality of domains each relates to a potential area of interest for the user, and the weight for a domain from among the plurality of domains indicates a level of interest for the user in the domain. A ranking classifier used to rank NLU hypotheses generated by an NLU engine is trained using training data from which features are, at least in part, based on the information specifying a weight for each of the plurality of domains.

Claims

1. A method of training a ranking classifier used by a natural language understanding (NLU) engine to rank NLU hypotheses generated by the NLU engine for a user of a computer system, the method comprising acts of:

receiving, via a user interface, information specifying a weight for each of a plurality of domains, wherein the plurality of domains each relates to a potential area of interest for the user, wherein the weight for a domain from among the plurality of domains indicates a level of interest for the user in the domain; and

training the ranking classifier using training data from which features are, at least in part, based on the information specifying a weight for each of the plurality of domains.

2. The method of claim 1, further comprising an act of:

tracking usage information from the user; and

retraining the ranking classifier based, at least in part, on the tracked usage information.

3. The method of claim 1, wherein training the ranking classifier comprises:

providing the training data as a set of annotated exemplars, wherein each exemplar in the set is associated with a plurality of feature sets, wherein each of the plurality of feature sets comprises values including one of the plurality of domains, the weight for the domain, and a target value.

4. The method of claim 3, wherein the values in each of the plurality of feature sets further includes ranking information determined by a speech recognizer and/or an NLU engine.

5. The method of claim 4, wherein the values for each of the plurality of feature sets further includes a language model score for the exemplar determined by the speech recognizer.

6. A non-transitory computer-readable storage medium encoded with a plurality of instructions that, when executed by a computer, performs a method, the method comprising an act of:

processing, using a natural language understanding (NLU) engine, a textual representation for an input provided by a user to determine an NLU output, wherein the processing the textual representation is performed based, at least in part, on user-specific information associating a weight with each of a plurality of domains, wherein the plurality of domains each relates to a potential area of interest for the user, and wherein the weight associated with the each of the plurality of domains indicates a level of interest for the user in the corresponding domain.

7. The computer-readable storage medium of claim 6, wherein the method further comprises acts of:

receiving speech input; and

processing the speech input with a speech recognizer to determine at least a portion of the textual representation.

8. The computer-readable storage medium of claim 6, wherein the processing the textual representation is further performed based, at least in part, on at least one domain-specific language model score.

9. The computer-readable storage medium of claim 6, wherein processing the textual representation comprises:

determining at least one NLU hypothesis for the textual representation, wherein each of the at least one NLU hypothesis is associated with a domain for the textual representation;

classifying the at least one hypothesis using user-specified information; and

selecting a hypothesis of the at least one hypothesis, wherein the selecting is based, at least in part on the classification of the at least one hypothesis.

10. The computer-readable storage medium of claim 9, further comprising acts of:

receiving speech input;

processing the speech input with a speech recognizer to determine one or more recognition results; and

providing at least a portion of the at least one of the one or more recognition results to the NLU engine as the textual representation.

11. The computer-readable storage medium of claim 10, wherein the one or more recognition results comprise N-best results for the speech input.

12. The computer-readable storage medium of claim 6, further comprising an act of:

initiating at least one action on a mobile device, wherein the at least one action is determined based, at least in part, on the NLU output.

13. The computer-readable storage medium of claim 6, wherein at least one of the plurality of domains is associated with a weight that is user independent.

14. A computer system, comprising:

a user interface configured to receive information specifying a weight for each of a plurality of domains, wherein the plurality of domains each relates to a potential area of interest for the user, wherein the weight for a domain from among a plurality of domains indicates a level of interest for the user in the domain; and

at least one processor programmed to;

train a ranking classifier using training data from which features are, at least in part, based on the information specifying a weight for each of the plurality of domains.

15. The computer system of claim 14, wherein specifying a weight for each of a plurality of domains comprises:

specifying a weight for a first domain of the at least one of the plurality of domains based on the information received via the user interface; and

specifying a weight for a second domain of the at least one of the plurality of domains based, at least in part, on the weight specified for the first domain.

16. The computer system of claim 15, wherein specifying the weight for the second domain of the at least one of the plurality of domains is performed without receiving user input directly specifying the weight for the second domain.

17. The computer system of claim 14, wherein the at least one processor is further programmed to:

track usage information from the user; and

retrain the ranking classifier based, at least in part, on the tracked usage information.

18. The computer system of claim 14, wherein training the ranking classifier comprises:

providing the training data as a set of annotated exemplars, wherein each exemplar in the set is associated with a plurality of feature sets, wherein each of the plurality of feature sets comprises values including one of the plurality of domains, the weight for the domain, and a target value.

19. The computer system of claim 18, wherein the values in each of the plurality of feature sets further includes ranking information determined by a speech recognizer and/or an NLU engine.

20. The computer system of claim 19, wherein the values for each of the plurality of feature sets further includes a language model score for the exemplar determined by the speech recognizer.