METHOD OF ANALYZING CONVERSATIONAL TRANSCRIPTS

• US 20080181390A1
• Filed: 04/03/2008
• Published: 07/31/2008
• Est. Priority Date: 12/21/2006
• Status: Active Grant

• Current Assignee: ServiceNow Incorporated
• Original Assignee: International Business Machines Corporation
• Inventors: Kummamuru, Krishna
• Granted Patent: US 8,135,579 B2
• Time in Patent Office: Days
• US Class Current: 379/265.06
• CPC Class Codes: G06F 16/685 using automatically derived transcript of audio data, e.g. lyrics speech recognition G10L15/00
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models

Abstract

Analyzing transcripts of conversation between at least two users by receiving input information from a first user via a voice call, creating conversational transcripts from the information received from the first user, selecting at least one defined situation from a list of defined situations, identifying the selected situation in the conversational transcripts, identifying a set of procedural sequences by comparing the at least one identified situation in the conversational transcripts with knowledge derived from a corpus of historical conversational transcripts; and providing the set of procedural sequences to the first user.

Claims

1. A method for analyzing transcripts of conversation between at least two users, the method comprising:

receiving input information from a first user via a voice call;

creating conversational transcripts from the information received from the first user;

selecting at least one defined situation from a list of defined situations;

identifying the selected situation in the conversational transcripts;

identifying a set of procedural sequences by comparing the at least one identified situation in the conversational transcripts with knowledge derived from a corpus of historical conversational transcripts; and

providing the set of procedural sequences to the first user.

2. The method of claim 1, all the limitations of which are incorporated herein by reference, wherein knowledge derived from the corpus of historical conversational transcripts comprises

clustering textual segments in the corpus of historical conversational transcripts into groups of related textual segments;

representing each of the groups of related textual segments by a representative syntax;

forming a sequence of representative syntax for each historical conversational transcript in the corpus;

finding frequently occurring sub sequences from the sequences of representative syntax; and

finding distinct and long frequently occurring sub sequences from the frequently occurring sub sequences.

3. The method of claim 2, all the limitations of which are incorporated herein by reference, wherein clustering the textual segments in the corpus of historical conversational transcripts comprises

organizing the historical conversational transcripts into groups of conversational transcripts according to a defined criteria;

and

clustering the groups of conversational transcripts to obtain groups of related textual segment.

4. The method of claim 3, all the limitations of which are incorporated herein by reference, wherein clustering groups of conversational transcripts to obtain groups of related documents further comprises

generating at least two sets of documents from the groups of conversational transcripts, wherein each of the sets of documents corresponds to at least one of the users.

5. The method of claim 2, all the limitations of which are incorporated herein by reference, wherein each of the representative syntax comprises a segment of conversational transcripts belonging to each of the group of related textual segments.

6. The method of claim 2, all the limitations of which are incorporated herein by reference, wherein the representative syntax comprises a set of frequent and distinct words in each of the group of related textual segments.

7. The method of claim 2, all the limitations of which are incorporated herein by reference, wherein finding distinct and long frequently occurring sub sequences is performed by a clustering algorithm belonging to the class of leader clustering algorithm.

8. The method of claim 1, all the limitations of which are incorporated herein by reference, wherein the corpus of historical transcripts comprises a repository.

9. The method of claim 2, all the limitations of which are incorporated herein by reference, wherein clustering textual segments in the corpus of historical conversational transcripts into groups of related textual segments further comprises utilizing a set of configurable clustering parameters to cluster the textual segments.

10. The method of claim 9, all the limitations of which are incorporated herein by reference, wherein varying the parameters is based on normalized entropy values.

11. The method of claim 1, all the limitations of which are incorporated herein by reference, further comprising

prompting the second user on appropriate procedural sequences based on the identified situation wherein the second user is configured to provide the procedural sequence to the first user.

12. A method for creating a document of procedural sequences, the method comprising

receiving input information from a first user;

creating conversational transcripts from the information received from the first user;

identifying at least one defined situation in the list of defined situations from the conversational transcripts;

comparing the situations with knowledge derived from historical conversational transcripts in a corpus;

on negative determination

creating a document comprising a set of procedural sequences; and

adding the document created to a repository.

13. The method of claim 12, all the limitations of which are incorporated herein by reference, wherein knowledge derived from a corpus of historical conversational transcripts comprises

clustering textual segments in the corpus of historical conversational transcripts into groups of related textual segments;

representing each of the groups of related textual segments by a representative syntax;

forming a sequence of representative syntax for each historical conversational transcript in the corpus;

finding frequently occurring sub sequences from the sequences of representative syntax; and

finding distinct and long frequently occurring sub sequences from the frequently occurring sub sequences.

14. The method of claim 13, all the limitations of which are incorporated herein by reference, wherein clustering the textual segments in the corpus of historical conversational transcripts further comprises

organizing the historical conversational transcripts into groups of conversational transcripts according to a defined criteria;

and

clustering the groups of conversational transcripts to obtain groups of related textual segment.

15. The method of claim 14, all the limitations of which are incorporated herein by reference, wherein clustering groups of conversational transcripts to obtain groups of related documents further comprises

generating at least two sets of documents from the groups of conversational transcripts, wherein each of the sets of documents corresponds to at least one of the users.

16. The method of claim 13, all the limitations of which are incorporated herein by reference, wherein each of the representative syntax comprises a segment of conversational transcripts belonging to each of the group of related textual segments.

17. The method of claim 13, all the limitations of which are incorporated herein by reference, wherein each of the representative syntax comprises a set of frequent and distinct words in each of the group of related textual segments.

18. The method of claim 13, all the limitations of which are incorporated herein by reference, wherein finding distinct and long frequently occurring sub sequences is performed by a clustering algorithm belonging to the class of leader clustering algorithm.

19. The method of claim 12, all the limitations of which are incorporated herein by reference, wherein the corpus of historical transcripts comprises a repository.

20. The method of claim 13, all the limitations of which are incorporated herein by reference, wherein clustering textual segments in the corpus of historical conversational transcripts into groups of related textual segments comprises utilizing a set of configurable cluster parameters to cluster the textual segments based on normalized entropy values.

21. The method of claim 12, all the limitations of which are incorporated herein by reference, further comprising

prompting the second user on creating a set of procedural sequences based on the identified situation.