LEARNING OPINION-RELATED PATTERNS FOR CONTEXTUAL AND DOMAIN-DEPENDENT OPINION DETECTION

• US 20140067370A1
• Filed: 08/31/2012
• Published: 03/06/2014
• Est. Priority Date: 08/31/2012
• Status: Abandoned Application

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Brun, Caroline
• Application Number:
  US13/600,329
• Publication Number:
  US 20140067370A1
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A method for extracting opinion-related patterns includes receiving a corpus of reviews, the reviews each including an explicit rating of a topic. The reviews are partitioned among a predefined plurality of classes, based on the ranking. Syntactic relations are identified in each review. The syntactic relations may each include an adjective and a noun. A set of patterns is generated, each of the patterns having at least one of the identified syntactic relations as an instance and the patterns clustered into a set of clusters based on a set of features. At least one of the features is based on occurrences, in the predefined classes, of the instances of the patterns. A polarity is assigned to ones of the clusters and propagated to patterns in the respective clusters. The polarity-labeled patterns can each be instantiated as a contextual rule for opinion mining.

Claims

1. A method for extracting opinion-related patterns, comprising:

receiving a corpus of reviews, each of the reviews in the corpus including an explicit rating of a topic;

partitioning at least a portion of the reviews among a predefined plurality of classes, based on the explicit ranking;

identifying syntactic relations in a text portion of each of the reviews in the portion of the reviews, each of the identified syntactic relations including a first term comprising an adjective and a second term comprising a noun, the adjective serving as a modifier or attribute of the respective noun;

generating a set of patterns, each of the patterns having at least one of the identified syntactic relations as an instance;

with a processor, clustering the patterns into a set of clusters based on a set of features, at least one of the features in the set of features being based on occurrences, in the predefined classes, of the instances of the patterns;

selecting a subset of the clusters and assigning a polarity to patterns in the respective clusters in the subset.

2. The method of claim 1, wherein the plurality of classes comprises at least three classes.

3. The method of claim 1, wherein each of the generated patterns specifies a lemma form of a specific noun and a specific adjective.

4. The method of claim 1, wherein the set of features comprises, for each of the classes, a respective feature which is based on a number of instances of the pattern which are identified in the text portions of the reviews in that class.

5. The method of claim 1, wherein the method further comprises, for reviews in the portion of reviews, identifying, in the text portion, instances of terms in a polar vocabulary, and wherein at least one of the features in the set of features is based on co-occurrence within a review of an instance of a pattern in the set of patterns and an identified instance of a term in a polar vocabulary.

6. The method of claim 5, wherein for each pattern, the set of features includes a first feature value based on a number of co-occurrences of an instance of the pattern with an instance of a positive polar term and a second feature value based on a number of co-occurrences of an instance of the pattern with an instance of a negative polar term.

7. The method of claim 1, wherein for each pattern, at least four feature values are assigned.

8. The method of claim 1, wherein at least 10 clusters are generated.

9. The method of claim 1, wherein the assigning a polarity to a plurality of the clusters comprises receiving a reviewer'"'"'s assignment of a polarity to each of a plurality of the clusters.

10. The method of claim 1 wherein the assigning a polarity comprises assigning a polarity from a predefined plurality of polarities.

11. The method of claim 1 wherein the assigning a polarity comprises assigning a positive polarity to at least a first of the plurality of the clusters and assigning a negative polarity to at least a second of the plurality of the clusters.

12. The method of claim 1, further comprising generating a contextual rule for each of the patterns which have been assigned polarity, the rule specifying that an instance of the pattern in text is to be assigned the polarity which has been assigned to the pattern.

13. The method of claim 12, further comprising inputting the contextual rules to an opinion detection system and applying the contextual rules during extraction of opinions from a text sample.

14. The method of claim 1, further comprising outputting the patterns and their assigned polarities to an opinion detection system.

15. The method of claim 1, wherein the topic comprises a type of product or type of service.

16. The method of claim 1, wherein the identifying syntactic relations in a text portion of each of the reviews in the portion of the reviews further comprises excluding, from the identified syntactic relations, syntactic relations in which the respective adjective is an instance of a term in a polar vocabulary of terms.

17. The method of claim 1, wherein there at least fifty patterns that are associated with a respective polarity.

18. A computer program product comprising a non-transitory computer-readable medium which stores instructions, which when implemented by a computer, perform the method of claim 1.

19. A system comprising memory which stores instructions for implementing the method of claim 1 and a processor in communication with the memory which implements the instructions.

20. An opinion detection system comprising:

memory which stores;

a set of contextual rules for identifying instances of the patterns generated by the method of claim 1 in a text sample and assigning a polarity to the instances based on the polarity assigned to the respective pattern, and

an opinion detection component which applies the rules to a text sample; and

a processor which implements the opinion detection component.

21. A system for generating contextual rules for opinion detection comprising:

a review classifier for partitioning reviews among a predefined plurality of classes, based on an explicit ranking of a topic associated with each of the reviews;

a parser for identifying syntactic relations in a text portion of each of the reviews in the portion of the reviews, each of the identified syntactic relations including a first term comprising an adjective and a second term comprising a noun, the adjective serving as a modifier or attribute of the respective noun;

a pattern extractor for generating a set of patterns, each of the generated patterns having at least one of the identified syntactic relations as an instance;

a clustering component for clustering the patterns into a set of clusters based on a set of features, at least one of the features in the set of features being based on occurrences, in the predefined classes, of the instances of the patterns;

a contextual rule generator for generating contextual rules for the patterns in a plurality of the clusters, to which a polarity has been assigned; and

a processor for implementing the review classifier, parser, pattern extractor, clustering component, and contextual rule generator.

22. The system of claim 21, further comprising a feature extractor, implemented by the processor, which extracts a value for each feature for each of the generated patterns.

23. The system of claim 21, further comprising an opinion detection component for detecting instances of opinions in the text portions based on a stored vocabulary of terms that have been assigned a respective polarity, at least one of the features being based on a frequency of co-occurrence of an instance of a pattern with a detected opinion instance.

24. A method for generating contextual rules, comprising:

receiving a corpus of documents, each of the documents in the corpus being associated with an explicit rating of a topic;

partitioning at least a portion of the documents among a predefined plurality of classes, based on the explicit ranking;

identifying opinion instances in the documents, each of the opinion instances comprising an instance of a term in an associated polar vocabulary;

identifying syntactic relations in the documents, each of the identified syntactic relations including a first term comprising an adjective that is not an instance of a term in the polar vocabulary and a second term comprising a noun, the adjective serving as a modifier or attribute of the respective noun;

generating a set of patterns, each of the patterns having at least one of the identified syntactic relations as an instance;

extracting features for each of the patterns, the extracted features including features based on occurrences, in the predefined classes, of the instances of the patterns;

with a processor, clustering the patterns into a set of clusters based on the extracted features;

generating contextual rules for the patterns in a plurality of the clusters for which a polarity has been assigned.