Systems and methods for detecting sentiment-based topics

• US 8,356,025 B2
• Filed: 12/09/2009
• Issued: 01/15/2013
• Est. Priority Date: 12/09/2009
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Cai, Keke, Chen, Ying, Spangler, William Scott, Zhang, Li
• Primary Examiner(s): HASAN, SYED HAROON
• Application Number:
  US12/634,269
• Publication Number:
  US 20110137906A1
• Time in Patent Office: 1,133 Days
• Field of Search: None
• US Class Current: 707/708
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A method for analyzing sentiment comprising: collecting an object from an external content repository, the collected objects forming a content database; extracting a snippet related to the subject from the content database; calculating a sentiment score for the snippet; classifying the snippet into a sentiment category; creating sentiment taxonomy using the sentiment categories, the sentiment taxonomy classifying the snippets as positive, negative or neutral; identifying topic words within the sentiment taxonomy; classifying the topic words as a sentiment topic word candidates or a non-sentiment topic word candidate, filtering the non-sentiment topic word candidates; identifying the frequency of the non-sentiment topic words in each of the sentiment categories; identifying the importance of the non-sentiment topic word for each of the sentiment categories; and, ranking the topic word, wherein the rank is calculated by combining the frequency of the topic words in each of the categories with its importance.

Claims

1. A computer implemented method for analyzing sentiment concerning a subject, comprising:

collecting a plurality of objects from at least one external content repository, wherein the collected objects form a content database;

extracting snippets related to the subject from the content database;

calculating a sentiment score for at least one snippet that has been extracted;

for the at least one snippet for which the sentiment score has been calculated, classifying the snippet into at least one sentiment category;

creating a sentiment taxonomy using the sentiment categories, wherein the sentiment taxonomy classifies the snippets as positive, negative or neutral;

identifying topic words within the sentiment taxonomy;

classifying the topic words as sentiment topic words or non-sentiment topic words;

identifying the frequency of the non-sentiment topic words in each of the sentiment categories;

identifying the importance of the non-sentiment topic words in each of the sentiment categories; and

,

ranking the non-sentiment topic words, wherein the rank for a non-sentiment topic word is calculated by combining the frequency of the non-sentiment topic word in each of the sentiment categories with the importance of the non-sentiment topic word.

2. The computer implemented method as in claim 1 wherein one external content repository is one from a group consisting of:

a blog;

a message board;

a web page article; and

a comment feed.

3. The computer implemented method as in claim 1 wherein identifying the importance of the non-sentiment topic words in each of the sentiment categories comprises calculating for each non-sentiment topic word a Pointwise Mutual Information of the non-sentiment topic word.

4. The computer implemented method as in claim 1 wherein identifying the importance of the non-sentiment topic words in each of the sentiment categories comprises calculating for each non-sentiment topic word a word support value for the non-sentiment topic word.

5. The computer implemented method as in claim 4 wherein identifying the importance of the non-sentiment topic word in each of the sentiment categories further comprises calculating for each non-sentiment topic word a Pointwise Mutual Information value for the non-sentiment topic word.

6. A system for analyzing sentiment concerning a subject, comprising a processor and a storage coupled to the processor, the processor configured to:

collect a plurality of objects from at least one external content repository, wherein the collected objects form a content database;

extract snippets related to the subject from the content database;

calculate a sentiment score for at least one snippet that has been extracted;

for the at least one snippet for which the sentiment score has been calculated, classify the snippet into at least one sentiment category;

create a sentiment taxonomy using the sentiment categories, wherein the sentiment taxonomy classifies the snippets as positive, negative or neutral;

identify topic words within the sentiment taxonomy;

classify the topic words as sentiment topic words or non-sentiment topic words;

identify the frequency of the non-sentiment topic words in each of the sentiment categories;

identify the importance of the non-sentiment topic words in each of the sentiment categories; and

,

rank the non-sentiment topic words, wherein the rank for a non-sentiment topic word is calculated by combining the frequency of the non-sentiment topic word in each of the sentiment categories with the importance of the non-sentiment topic word.

7. The method system as in claim 6 wherein one external content repository is one from a group consisting of:

a blog;

a message board;

a web page article; and

a comment feed.

8. The system as in claim 6 wherein the processor is further configured, in identifying the importance of the non-sentiment topic words in each of the sentiment categories, to calculate for each non-sentiment topic word a Pointwise Mutual Information of the non-sentiment topic word.

9. The system as in claim 6 wherein the processor is further configured, in identifying the importance of the non-sentiment topic words in each of the sentiment categories, to calculate for each non-sentiment topic word a word support value for the non-sentiment topic word.

10. The system as in claim 6 wherein the processor is further configured, in identifying the importance of the non-sentiment topic word in each of the sentiment categories, to calculate for each non-sentiment topic word a Pointwise Mutual Information value for the non-sentiment topic word.

11. A computer program product comprising a non-transitory computer useable storage medium to store a computer readable program, wherein the computer readable program, when executed on a computer, causes the computer to perform for operations for determining analyzing sentiment concerning a subject, comprising:

collecting a plurality of objects from at least one external content repository, wherein the collected objects form a content database;

extracting snippets related to the subject from the content database;

calculating a sentiment score for at least one snippet that has been extracted;

for the at least one snippet for which the sentiment score has been calculated, classifying the snippet into at least one sentiment category;

creating a sentiment taxonomy using the sentiment categories, wherein the sentiment taxonomy classifies the snippets as positive, negative or neutral;

identifying topic words within the sentiment taxonomy;

classifying the topic words as sentiment topic words or non-sentiment topic words;

identifying the frequency of the non-sentiment topic words in each of the sentiment categories;

identifying the importance of the non-sentiment topic words in each of the sentiment categories; and

,

ranking the non-sentiment topic words, wherein the rank for a non-sentiment topic word is calculated by combining the frequency of the non-sentiment topic word in each of the sentiment categories with the importance of the non-sentiment topic word.

12. The computer program product as in claim 11 wherein one external content repository is one from a group consisting of:

a blog;

a message board;

a web page article; and

a comment feed.

13. The computer program product as in claim 11 wherein identifying the importance of the non-sentiment topic words in each of the sentiment categories comprises calculating for each non-sentiment topic word a Pointwise Mutual Information of the non-sentiment topic word.

14. The computer program product as in claim 11 wherein identifying the importance of the non-sentiment topic words in each of the sentiment categories comprises calculating for each non-sentiment topic word a word support value for the non-sentiment topic word.

15. The computer program product as in claim 11 wherein identifying the importance of the non-sentiment topic word in each of the sentiment categories further comprises calculating for each non-sentiment topic word a Pointwise Mutual Information value for the non-sentiment topic word.