Method and apparatus for determining decision points for streaming conversational data

• US 7,904,399 B2
• Filed: 11/15/2007
• Issued: 03/08/2011
• Est. Priority Date: 11/15/2007
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Faruquie, Tanveer A, Ramakrishnan, Ganesh, Subramaniam, L. Venkata
• Primary Examiner(s): VINCENT, DAVID ROBERT
• Application Number:
  US11/940,551
• Publication Number:
  US 20090132442A1
• Time in Patent Office: 1,209 Days
• Field of Search: 706/12, 706/20, 706/62, 704/231, 704/236, 704/240, 704/245, 704/246, 704250-255
• US Class Current: 706/20
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding

Abstract

A method for determining a decision point in real-time for a data stream from a conversation includes receiving streaming conversational data; and determining when to classify the streaming conversational data, using a measure of certainty, by performing certainty calculations at a plurality of time instances during the conversation and by selecting a decision point in response to the certainty calculations, the decision point not being based on a fixed window of conversational data but being based on accumulated conversational data available at different ones of the plurality of time instances. Systems and computer program products are also provided.

Claims

1. A computer-implemented method for determining a decision point in real-time for a data stream from a conversation, the method comprising:

receiving, by a computing device, streaming conversational data; and

determining, by said computing device, when to classify the streaming conversational data into one of thematic categories and labels from a predefined set, using

a measure of certainty, by performing certainty calculations at a plurality of time instances during the conversation and

by selecting a decision point in response to the certainty calculations, the decision point being based on accumulated conversational data available at different ones of the plurality of time instances.

2. The method in accordance with claim 1, wherein the certainty calculations comprise entropy calculations.

3. The method in accordance with claim 1, wherein multiple decisions are made at one or more of the time instances before the completion of the conversation, and wherein respective decisions have respective different decision points which can be different from one another and which are based on respective certainty calculations.

4. The method in accordance with claim 2, and further comprising training, by said computing device, a classifier and engineering features based on labeled information.

5. The method in accordance with claim 3, wherein the labeled information comprises outcomes of previous conversations which need not have the decision points.

6. The method in accordance with claim 3, wherein a single conversation is labeled with a plurality of different labels without specifying portions of the conversation as being associated with any of the labels.

7. The method in accordance with claim 1, and further comprising training, by said computing device, a classifier using a portion of a conversation.

8. The method in accordance with claim 1, and further comprising training, by said computing device, a classifier using a portion of a conversation chosen in response to the selected decision point.

9. The method in accordance with claim 1, wherein the conversation is a call center conversation, and wherein the method further comprises outputting, by said computing device, information to assist a human in decision-making at the selected decision point.

10. The method in accordance with claim 1, wherein the conversation is a call center conversation, and wherein the method further comprises taking automated action, by said computing device, at the selected decision point.

11. The method in accordance with claim 1, wherein the conversation is a call center conversation, wherein the conversation is scripted and has a flow structure, and wherein knowledge of the script is used in selecting a decision point.

12. A system comprising:

a computer including a network adapter configured to receive connections from external computers, the computer being configured

to receive streaming text from a conversation and

to determine when to classify the streaming text into one of thematic categories and labels from a predefined set, using

a measure of certainty, by performing entropy calculations at a plurality of time instances during receipt of the streaming text, and

by selecting a decision point in response to the entropy calculations, the decision point being based on text accumulated at the respective time instances, and

to perform a classification operation at the decision point, prior to completion of the conversation.

13. The system in accordance with claim 12, and configured to make multiple decisions at one or more of the time instances before the completion of the conversation, and wherein respective decisions have respective different decision points which can be different from one another and which are based on respective entropy calculations.

14. The system in accordance with claim 12, and further comprising a classifier configured to be trained based on labeled information that comprises outcomes of previous conversations.

15. The system in accordance with claim 12, and comprising an output device, wherein the system is configured to output classification information, using the output device, to assist a human at the selected decision point.

16. A non-transitory computer program medium having a computer readable program, wherein the computer readable program when executed on a computer causes the computer to:

determine when to classify streaming conversational data into one of thematic categories and labels from a predefined set, using

a measure of certainty, by performing certainty calculations at a plurality of time instances during the conversation and

by selecting a decision point in response to the certainty calculations, the decision point being based on accumulated conversational data available at different ones of the plurality of time instances.

17. The non-transitory computer program medium in accordance with claim 16, wherein the certainty calculations comprise entropy calculations.

18. The non-transitory computer program medium in accordance with claim 16, wherein multiple decisions are made at one or more of the time instances before the completion of the conversation, and wherein respective decisions have respective different decision points which can be different from one another and which are based on respective certainty calculations.