Topic kernelization for real-time conversation data

• US 10,606,954 B2
• Filed: 02/15/2018
• Issued: 03/31/2020
• Est. Priority Date: 02/15/2018
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Penrose, Andrew T., Dunne, Jonathan
• Primary Examiner(s): Flanders, Andrew C
• Application Number:
  US15/897,298
• Publication Number:
  US 20190251166A1
• Time in Patent Office: 775 Days
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06F 40/20 Natural language analysis semantic analysis of natural language G06F40/30
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Embodiments for text segmentation for topic modelling by a processor. Real-time conversation data may be analyzed and time intervals (e.g., inter-arrival times) between messages of the conversation data may be recorded. Each of the messages may be defined (and/or segmented) as burst segments or reflection segments according to the analyzing and recording. One or more topic modelling operations may be enhanced for text segmentation using the burst segments or reflection segments.

Claims

1. A method for text segmentation for topic modelling by a processor, comprising:

analyzing real-time conversation data, wherein time intervals between messages being received into the conversation data are recorded;

defining the messages as burst segments or reflection segments according to the analyzing;

wherein the burst segments comprise successive messages received into the conversation data within a first time interval and the reflection segments comprise multiple messages each received into the conversation data having an inter-arrival time outside the first time interval;

enhancing, using a machine learning mechanism, one or more topic modelling operations for text segmentation using the burst segments or reflection segments; and

presenting, via a display, a summary of the one or more topic modelling operations to a user according to an output of a text mining analysis implementing the one or more topic modelling operations enhanced by the machine learning mechanism.

2. The method of claim 1, further including using the burst segments or reflection segments to determine optimal topic model sizes.

3. The method of claim 1, further including selecting a topic cluster size using the one or more enhanced topic modelling operations.

4. The method of claim 1, further including selecting a topic term size using the one or more enhanced topic modelling operations.

5. The method of claim 1, further including determining a percentage of unique stop words.

6. The method of claim 1, further including altering a topic cluster size and a topic term size until a percentage of unique stop words are maximized.

7. A system for text segmentation for topic modelling in a computing environment, comprising:

one or more computers with executable instructions that when executed cause the system to;

analyze real-time conversation data, wherein time intervals between messages being received into the conversation data are recorded;

define the messages as burst segments or reflection segments according to the analyzing;

wherein the burst segments comprise successive messages received into the conversation data within a first time interval and the reflection segments comprise multiple messages each received into the conversation data having an inter-arrival time outside the first time interval;

enhance, using a machine learning mechanism, one or more topic modelling operations for text segmentation using the burst segments or reflection segments; and

present, via a display, a summary of the one or more topic modelling operations to a user according to an output of a text mining analysis implementing the one or more topic modelling operations enhanced by the machine learning mechanism.

8. The system of claim 7, wherein the executable instructions when executed cause the system to use the burst segments or reflection segments to determine optimal topic model sizes.

9. The system of claim 7, wherein the executable instructions when executed cause the system to select a topic cluster size using the one or more enhanced topic modelling operations.

10. The system of claim 7, wherein the executable instructions when executed cause the system to select a topic term size using the one or more enhanced topic modelling operations.

11. The system of claim 7, wherein the executable instructions when executed cause the system to determine a percentage of unique stop words.

12. The system of claim 7, wherein the executable instructions when executed cause the system to alter a topic cluster size and a topic term size until a percentage of unique stop words are maximized.

13. A computer program product for, by a processor, text segmentation for topic modelling, the computer program product comprising a non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising:

an executable portion that analyzes real-time conversation data, wherein time intervals between messages being received into the conversation data are recorded;

an executable portion that defines the messages as burst segments or reflection segments according to the analyzing;

wherein the burst segments comprise successive messages received into the conversation data within a first time interval and the reflection segments comprise multiple messages each received into the conversation data having an inter-arrival time outside the first time interval;

an executable portion that enhances, using a machine learning mechanism, one or more topic modelling operations for text segmentation using the burst segments or reflection segments; and

an executable portion that presents, via a display, a summary of the one or more topic modelling operations to a user according to an output of a text mining analysis implementing the one or more topic modelling operations enhanced by the machine learning mechanism.

14. The computer program product of claim 13, further including an executable portion that uses the burst segments or reflection segments to determine optimal topic model sizes.

15. The computer program product of claim 13, further including an executable portion that selects a topic cluster size using the one or more enhanced topic modelling operations.

16. The computer program product of claim 13, further including an executable portion that:

selects a topic term size; and

determine a percentage of unique stop words.

17. The computer program product of claim 13, further including an executable portion that alters a topic cluster size and a topic term size until a percentage of unique stop words are maximized.