METHOD AND APPARATUS FOR CUSTOMIZING CONVERSATION AGENTS BASED ON USER CHARACTERISTICS

• US 20140317030A1
• Filed: 04/22/2013
• Published: 10/23/2014
• Est. Priority Date: 04/22/2013
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Palo Alto Research Center, Inc. (Xerox Holdings Corp.)
• Inventors: Shen, Jianqiang, Brdiczka, Oliver
• Granted Patent: US 9,177,318 B2
• Time in Patent Office: Days
• US Class Current: 706/12
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06Q 30/01 Customer relationship services

Abstract

A conversation-simulating system facilitates simulating an intelligent conversation with a human user. During operation, the system can receive a user-statement from the user during a simulated conversation, and generates a set of automatic-statements that each responds to the user-statement. The system then determines a set of behavior-characteristics for the user, and computes relevance scores for the automatic-statements based on the behavior-characteristics. Each relevance score indicates an outcome quality that the user is likely to perceive for the automatic-statement as a response to the user-statement. The system selects an automatic-statement that has a highest relevance score from the set of automatic-statements, and provides the selected automatic-statement to the user.

Claims

1. A computer-implemented method, comprising:

receiving, by a computing device, a user-statement from a user during a simulated conversation with the user;

determining a set of automatic-statements that each responds to the user-statement;

determining a set of behavior-characteristics associated with the user;

computing relevance scores for the automatic-statements based on the behavior-characteristics, wherein a respective relevance score indicates an outcome quality that the user is likely to perceive for the automatic-statement as a response to the user-statement;

selecting an automatic-statement that has a highest relevance score from the set of automatic-statements; and

providing the selected automatic-statement to the user.

2. The method of claim 1, wherein a respective behavior-characteristic indicates a personality-trait category, and indicates a numeric score for the corresponding personality-trait category.

3. The method of claim 2, wherein the personality-trait categories include one or more of:

openness to experiences;

conscientiousness;

extraversion;

agreeableness;

self-esteem;

novelty-seeking;

perfectionism;

rigidity;

impulsivity;

harm-avoidance;

disinhibition;

alexithymia;

neuroticism;

psychoticism; and

obsessionality.

4. The method of claim 2, further comprising:

identifying a personality-trait category to score for the user;

determining an automatic-response associated with the personality-trait category;

providing the automatic-response to the user;

responsive to receiving a response from the user for the automatic-response, computing a score for the personality-trait category for the user; and

updating the user'"'"'s behavior-characteristics to assign the computed score to the personality-trait category.

5. The method of claim 1, wherein determining the set of automatic-statements involves:

determining a query from the user'"'"'s user-statement;

identifying, from a response-template repository, one or more response templates that match at least one query attribute of the query, wherein the response-template repository includes a set of pre-generated response templates that are each associated with at least one corresponding query attribute; and

generating the set of automatic-statements based on the identified response templates.

6. The method of claim 1, wherein computing the relevance score for a respective automatic-statement involves computing a template-relevance score S_T using a response-prediction function:

S_T=ƒ

_T(T,C);

wherein T identifies a response template, wherein C indicates the user'"'"'s behavior-characteristics, and wherein S_T provides a score indicating how well response template T matches a conversation style for a user having behavior characteristics C.

7. The method of claim 6, further comprising:

providing a survey that indicates a set of conversation scripts to a plurality of users, and solicits a response rating for one or more conversation responses in the conversation script;

obtaining response ratings, from the users, for the conversation responses;

determining behavior characteristics for each of the plurality of users; and

training the response-prediction function based on the behavior preferences for the plurality of users, and the response ratings from each of the plurality of users.

8. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method, the method comprising:

receiving a user-statement from a user during a simulated conversation with the user;

determining a set of automatic-statements that each responds to the user-statement;

determining a set of behavior-characteristics associated with the user;

computing relevance scores for the automatic-statements based on the behavior-characteristics, wherein a respective relevance score indicates an outcome quality that the user is likely to perceive for the automatic-statement as a response to the user-statement;

selecting an automatic-statement that has a highest relevance score from the set of automatic-statements; and

providing the selected automatic-statement to the user.

9. The storage-medium of claim 8, wherein a respective behavior-characteristic indicates a personality-trait category, and indicates a numeric score for the corresponding personality-trait category.

10. The storage-medium of claim 9, wherein the personality-trait categories include one or more of:

openness to experiences;

conscientiousness;

extraversion;

agreeableness;

self-esteem;

novelty-seeking;

perfectionism;

rigidity;

impulsivity;

harm-avoidance;

disinhibition;

alexithymia;

neuroticism;

psychoticism; and

obsessionality.

11. The storage-medium of claim 9, wherein the method further comprises:

identifying a personality-trait category to score for the user;

determining an automatic-response associated with the personality-trait category;

providing the automatic-response to the user;

responsive to receiving a response from the user for the automatic-response, computing a score for the personality-trait category for the user; and

updating the user'"'"'s behavior-characteristics to assign the computed score to the personality-trait category.

12. The storage-medium of claim 8, wherein determining the set of automatic-statements involves:

determining a query from the user'"'"'s user-statement;

identifying, from a response-template repository, one or more response templates that match at least one query attribute of the query, wherein the response-template repository includes a set of pre-generated response templates that are each associated with at least one corresponding query attribute; and

generating the set of automatic-statements based on the identified response templates.

13. The storage-medium of claim 8, wherein computing the relevance score for a respective automatic-statement involves computing a template-relevance score S_T using a response-prediction function:

S_T=ƒ

_T(T,C);

wherein T identifies a response template, wherein C indicates the user'"'"'s behavior-characteristics, and wherein S_T provides a score indicating how well response template T matches a conversation style for a user having behavior characteristics C.

14. The storage-medium of claim 13, wherein the method further comprises:

providing a survey that indicates a set of conversation scripts to a plurality of users, and solicits a response rating for one or more conversation responses in the conversation script;

obtaining response ratings, from the users, for the conversation responses;

determining behavior characteristics for each of the plurality of users; and

training the response-prediction function based on the behavior preferences for the plurality of users, and the response ratings from each of the plurality of users.

15. An apparatus, comprising:

a communication module to receive a user-statement from a user during a simulated conversation with the user;

a response-generating module to determine a set of automatic-statements that each responds to the user-statement;

a response-scoring module to compute relevance scores for the automatic-statements based on behavior-characteristics associated with the user, wherein a respective relevance score indicates an outcome quality that the user is likely to perceive for the automatic-statement as a response to the user-statement; and

a response-selecting module to select an automatic-statement that has a highest relevance score from the set of automatic-statements;

wherein the communication module is further configured to provide the selected automatic-statement to the user.

16. The apparatus of claim 15, wherein a respective behavior-characteristic indicates a personality-trait category, and indicates a numeric score for the corresponding personality-trait category.

17. The apparatus of claim 16, further comprising a behavior-predicting module to:

identify a personality-trait category to score for the user;

determine an automatic-response associated with the personality-trait category;

provide the automatic-response to the user;

compute a score for the personality-trait category for the user, responsive to receiving a response from the user for the automatic-response; and

update the user'"'"'s behavior-characteristics to assign the computed score to the personality-trait category.

18. The apparatus of claim 15, wherein while determining the set of automatic-statements, the response-generating module is further configured to:

determine a query from the user'"'"'s user-statement;

identify, from a response-template repository, one or more response templates that match at least one query attribute of the query, wherein the response-template repository includes a set of pre-generated response templates that are each associated with at least one corresponding query attribute; and

generate the set of automatic-statements based on the identified response templates.

19. The apparatus of claim 15, wherein while computing the relevance score for a respective automatic-statement, the response-scoring module is further configured to compute a template-relevance score S_T using a response-prediction function:

S_T=ƒ

_T(T,C);

wherein T identifies a response template, wherein C indicates the user'"'"'s behavior-characteristics, and wherein S_T provides a score indicating how well response template T matches a conversation style for a user having behavior characteristics C.

20. The apparatus of claim 19, wherein the communication module is further configured to:

provide a survey that indicates a set of conversation scripts to a plurality of users, and solicits a response rating for one or more conversation responses in the conversation script; and

obtain response ratings, from the users, for the conversation responses; and

wherein the apparatus further comprises a prediction-training module to;

determine behavior characteristics for each of the plurality of users; and

train the response-prediction function based on the behavior preferences for the plurality of users, and the response ratings from each of the plurality of users.