AUTOMATIC CREATION OF COMPLEX CONVERSATIONAL NATURAL LANGUAGE CALL ROUTING SYSTEM FOR CALL CENTERS

• US 20110069822A1
• Filed: 09/24/2009
• Published: 03/24/2011
• Est. Priority Date: 09/24/2009
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kuo, Hong-Kwang Jeff, Jan, Ea-Ee, Lubensky, David M.
• Granted Patent: US 8,509,396 B2
• Time in Patent Office: Days
• US Class Current: 379/88.04
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A call routing system is created by receiving a set of initial target classes and a corresponding set of topic descriptions. Non-overlapping semantic tokens in the set of topic descriptions are identified. A set of clear target classes from the non-overlapping semantic tokens and the initial target classes is identified. Overlapping semantic tokens from the set of topic descriptions are identified. A set of vague classes is identified from the overlapping semantic tokens and the initial target classes. A set of disambiguation dialogues and a set of grammar prompts is generated according to the overlapping and non-overlapping semantic tokens. The call routing system is then created based on the set of clear target classes, the set of vague target classes, and the set of disambiguation dialogues.

Claims

1. A computer implemented method for creating a call routing system, the computer implemented method comprising:

receiving a set of initial target classes and a corresponding set of topic descriptions;

identifying non-overlapping semantic tokens the set of topic descriptions;

identifying a set of clear target classes from the non-overlapping semantic tokens and the set of initial target classes;

identifying overlapping semantic tokens from the set of topic descriptions;

identifying a set of vague classes from the overlapping semantic tokens and the initial target classes;

generating a set of disambiguation dialogues and a set of grammar prompts according to the non-overlapping semantic tokens; and

creating the call routing system based on the set of clear target classes, the set of vague target classes, and the set of disambiguation dialogues.

2. The computer implemented method of claim 1, further comprising:

Identifying the non-overlapping semantic tokens, wherein the non-overlapping semantic tokens are semantic tokens appearing in only one of the set of topic descriptions; and

identifying the overlapping semantic tokens from the set of topic descriptions, wherein the overlapping semantic tokens are semantic tokens appearing in a plurality of the set of topic descriptions;

identifying the set of vague classes from the overlapping semantic tokens and the set of initial target classes, wherein each of the set of vague classes corresponds to a plurality of the set of clear target classes; and

generating the set of disambiguation dialogues and the set of grammar prompts according to the non-overlapping semantic tokens, wherein the disambiguation dialogues is generated from the non-overlapping semantic tokens to distinguish among ones of the plurality of the set of clear target classes.

3. The computer implemented method of claim 1, the computer implemented method further comprising:

identifying the overlapping semantic tokens from the set of topic descriptions, wherein the overlapping semantic tokens are maximum length semantic tokens.

4. The computer implemented method of claim 1, wherein the step of generating the set of disambiguation dialogues and the set of grammar prompts according to the non-overlapping semantic tokens further comprises:

generating a summary that summarizes a user intent from a preceding query, wherein the summary is generated from the overlapping semantic tokens of a particular vague class of the set of vague classes; and

generating a proposed solution that enumerates possible solutions, wherein the possible solutions correspond to ones of the plurality of clear target classes for the particular vague class, and wherein the proposed solution is generated from the non-overlapping semantic tokens of the ones of the plurality of clear target classes for the particular vague class.

5. The computer implemented method of claim 1, further comprising:

identifying a set of clear target classes from the non-overlapping semantic tokens and the set of initial target classes, wherein each of the set of clear target classes corresponds to a final destination for routing a call within the call routing system, and wherein each of the set of clear target classes is a terminal node of a tree describing menu options of the call routing system.

6. The computer implemented method of claim 1 further comprising:

creating the call routing system based on the set of clear target classes, the set of vague classes, and the set of disambiguation dialogues, wherein the call routing system is an initial topic classification model.

7. A computer implemented method for routing a call through a call routing system, comprising:

receiving a spoken speech input from a user;

converting the spoken speech input into a converted input;

classifying the converted input into one of a set of clear target classes or one of a set of vague classes, wherein each of the set of vague classes comprises a plurality of the set of clear target classes;

responsive to classifying the converted input as one of the set of vague classes, presenting a set of disambiguation dialogues and a set of grammar prompts; and

responsive to classifying the converted input as one of the set of clear target classes, routing the call to a destination within the call routing system.

8. The computer implemented method of claim 7, further comprising:

converting the spoken speech input into the converted input, wherein the spoken speech is converted into the converted input using a statistically based classification scheme selected from the group consisting of K-nearest neighbor, Gaussian mixture modeling, support vector machines, vector quantization, hidden Markov modeling, and multivariate auto-regression modeling.

9. The computer implemented method of claim 7, further comprising:

classifying the converted input into one of the set of clear target classes clear target classes or one of a set of vague classes, wherein the set of clear target classes correspond to the non-overlapping semantic tokens identified from a set of topic descriptions, and wherein the set of vague classes correspond to overlapping semantic tokens identified from the set of topic descriptions, wherein each of the set of vague classes corresponds to a plurality of the clear target classes.

10. The computer implemented method of claim 7, wherein the step of presenting the set of disambiguation dialogues and the set of grammar prompts comprises:

presenting a summary that summarizes a user intent from a preceding query, wherein the summary is generated from overlapping semantic tokens of a particular vague class of the set of vague classes; and

presenting a proposed solution that enumerates possible solutions, wherein the possible solutions correspond to ones of the plurality of clear target classes for the particular vague target class, and wherein the proposed solution is generated from non-overlapping semantic tokens of the ones of the plurality of clear target classes for the particular vague target class.

11. The computer implemented method of claim 7, further comprising:

responsive to presenting the set of disambiguation dialogues and the set of grammar prompts, receiving a second spoken speech input from the user;

converting the second spoken speech input into a second converted input;

responsive to classifying the second converted input as one of the set of vague classes, presenting a second set of disambiguation dialogues and a second set of grammar prompts; and

responsive to classifying the second converted input as one of the set of clear target classes, routing the call to the destination within the call routing system.

12. The computer implemented method of claim 11, wherein the step of presenting the second set of disambiguation dialogues and the second set of grammar prompts further comprises:

responsive to classifying the second converted input as one of the set of vague classes, presenting the second set of disambiguation dialogues and the second set of grammar prompts, wherein the second set of disambiguation dialogues and the second set of grammar prompts leads to a maximum reduction of expected entropy of the set of vague classes.

13. A tangible computer storage type medium having a computer program product encoded thereon for creating a call routing system, the computer program product comprising:

computer usable instructions for receiving a set of initial target classes and a corresponding set of topic descriptions;

computer usable instructions for identifying non-overlapping semantic tokens from the set of topic descriptions;

computer usable instructions for identifying a set of clear target classes from the non-overlapping semantic tokens and the initial target classes;

computer usable instructions for identifying overlapping semantic tokens from the set of topic descriptions;

computer usable instructions for identifying a set of vague classes from the overlapping semantic tokens and the initial target classes;

computer usable instructions for generating a set of disambiguation dialogues and a set of grammar prompts according to the non-overlapping semantic tokens; and

computer usable instructions for creating the call routing system based on the set of clear target classes, the set of vague target classes, and the set of disambiguation dialogues.

14. The tangible computer storage type medium of claim 13, the computer program product further comprising:

computer usable instructions for identifying non-overlapping semantic tokens, wherein the non-overlapping semantic tokens are semantic tokens appearing in only one of the set of topic descriptions; and

computer usable instructions for identifying overlapping semantic tokens from the set of topic descriptions, wherein the overlapping semantic tokens are semantic tokens appearing in a plurality of the set of topic descriptions;

computer usable instructions for identifying a set of vague classes from the overlapping semantic tokens and the initial target classes, wherein the each of the set of vague classes corresponds to a plurality of the clear target classes; and

computer usable instructions for generating a set of disambiguation dialogues and a set of grammar prompts according to the non-overlapping semantic tokens, wherein the disambiguation dialogues is generated from the non-overlapping semantic tokens to distinguish among clear target classes of ones of the plurality of clear target classes.

15. The tangible computer storage type medium of claim 13, the computer program product further comprising:

computer usable instructions for identifying overlapping semantic tokens from the set of topic descriptions, wherein the overlapping semantic tokens are maximum length semantic tokens.

16. The tangible computer storage type medium of claim 13, wherein the computer usable instructions for generating the set of disambiguation dialogues and the set of grammar prompts according to the non-overlapping semantic tokens further comprises:

computer usable instructions for generating a summary that summarizes a user intent from a preceding query, wherein the summary is generated from the overlapping semantic tokens of a particular vague class of the set of vague classes; and

computer usable instructions for generating a proposed solution that enumerates possible solutions, wherein the possible solutions correspond to ones of the plurality of clear target classes for the particular vague target class, and wherein the proposed solution is generated from the non-overlapping semantic tokens of the ones of the plurality of clear target classes for the particular vague target class.

17. The tangible computer storage type medium of claim 13, the computer program product further comprising:

computer usable instructions for identifying a set of clear target classes from the non-overlapping semantic tokens and the initial target classes, wherein each of the set of clear target classes corresponds to a final destination for routing a call within the call routing system, and wherein each of the set of clear target classes is a terminal node of a tree describing menu options of the call routing system.

18. The tangible computer storage type medium of claim 13, the computer program product further comprising:

computer usable instructions for creating the call routing system based on the set of clear target classes, the set of vague target classes, and the set of disambiguation dialogues, wherein the call routing system is an initial topic classification model.

19. A data processing system comprising:

a bus;

a storage connected to the bus, the storage having a computer program product encoded thereon for creating a call routing system; and

a processor connected to the bus, wherein the processor executes the computer program product to receive a set of initial target classes and a corresponding set of topic descriptions;

to identify non-overlapping semantic tokens the set of topic descriptions;

to identify a set of clear target classes from the non-overlapping semantic tokens and the initial target classes;

to identify overlapping semantic tokens from the set of topic descriptions;

to identify a set of vague classes from the overlapping semantic tokens and the initial target classes;

to generate a set of disambiguation dialogues and a set of grammar prompts according to the non-overlapping semantic tokens; and

to create the call routing system based on the set of clear target classes, the set of vague target classes, and the set of disambiguation dialogues.

20. The data processing system of claim 19, wherein the processor further executes the computer program product:

to identify non-overlapping semantic tokens, wherein the non-overlapping semantic tokens are semantic tokens appearing in only one of the set of topic descriptions;

to identify overlapping semantic tokens from the set of topic descriptions, wherein the overlapping semantic tokens are semantic tokens appearing in a plurality of the set of topic descriptions;

to identify a set of vague classes from the overlapping semantic tokens and the initial target classes, wherein each of the set of vague classes corresponds to a plurality of the clear target classes; and

to generate a set of disambiguation dialogues and a set of grammar prompts according to the non-overlapping semantic tokens, wherein the disambiguation dialogues is generated from the non-overlapping semantic tokens to distinguish among the clear target classes of ones of the plurality of clear target classes.

21. The data processing system of claim 19, wherein the processor further executes the computer program product:

to identify overlapping semantic tokens from the set of topic descriptions, wherein the overlapping semantic tokens are maximum length semantic tokens.

22. The data processing system of claim 19, wherein the processor executing the computer program product to generate a set of disambiguation dialogues and a set of grammar prompts according to the non-overlapping semantic tokens further comprises the processor executing the computer program product:

to generate a summary that summarizes a user intent from a preceding query, wherein the summary is generated from the overlapping semantic tokens of a particular vague class of the set of vague classes; and

to generate a proposed solution that enumerates possible solutions, wherein the possible solutions correspond to ones of the plurality of clear target classes for the particular vague target class, and wherein the proposed solution is generated from the non-overlapping semantic tokens of the ones of the plurality of clear target classes for the particular vague target class.

23. The data processing system of claim 19, wherein the processor further executes the computer program product:

to identify a set of clear target classes from the non-overlapping semantic tokens and the initial target classes, wherein each of the set of clear target classes corresponds to a final destination for routing a call within the call routing system, and wherein each of the set of clear target classes is a terminal node of a tree describing menu options of the call routing system.

24. The data processing system of claim 19, wherein the processor further executes the computer program product:

to create the call routing system based on the set of clear target classes, the set of vague target classes, and the set of disambiguation dialogues, wherein the call routing system is an initial topic classification model.