Method and system for identifying user issues in forum posts based on discourse analysis

Patent metadata

Field	Value
Publication number	US-10120864-B2
Application number	US-201615083504-A
Country	US
Kind code	B2
Filing date	Mar 29, 2016
Priority date	Mar 29, 2016
Publication date	Nov 6, 2018
Grant date	Nov 6, 2018

Abstract

A method for categorizing an issue includes, for each of a plurality of categories of issue, providing at least one discourse pattern for identify text sequences that meet the discourse pattern. At least one of the discourse patterns specifies that an instance of a domain term in a domain term vocabulary be present in the text sequence for the pattern to be met. An issue is received which includes a text sequence. The text sequence is categorized based on which, if any, of the discourse patterns are met by the text sequence of the received issue. Information based on the categorization of the text sequence is output.

First claim

What is claimed is:

1. A method for categorizing an issue and outputting information based on the categorization, the method comprising:
for each of a plurality of categories of issue including an information request category and an anomaly category, providing a discourse model, the discourse models including a request discourse model which applies rules for identifying discourse patterns for identifying information requests in a text sequence and an anomaly discourse model which applies rules for identifying discourse patterns identifying anomalies in the text sequence, the discourse patterns configured for identify text sequences that meet the respective discourse pattern, at least one of the discourse patterns specifying that an instance of a domain term in a domain term vocabulary be present in the text sequence for the discourse pattern to be met,
wherein there are at least two discourse patterns provided for the anomaly category, the at least two discourse patterns for the anomaly category being selected from the group consisting of:
a pattern specifying that a term from a predefined set of deviance-related terms be in a syntactic dependency with a term in the domain term vocabulary;
a pattern specifying that a term classed as deviance be in a syntactic dependency with a term in the domain term vocabulary;
a pattern specifying that a deviance aspect indicator co-occur with a negative element, wherein the negative element is in a syntactic dependency with a term in the domain term vocabulary;
a pattern specifying that a deviance indicator of contrast co-occur with a deviance aspect indicator and a predicate or argument, wherein the predicate or argument is in a syntactic dependency with a term in the domain term vocabulary; and
a pattern specifying that a deviance indicator of contrast co-occur with deviance aspect indicator and a predicate or argument, wherein the predicate or argument is in a syntactic dependency with a term in the domain term vocabulary,

wherein the syntactic dependencies in the patterns are identified with rules for identifying:
a subject syntactic dependency, which identifies a noun or noun phrase and a verb for which the noun or noun phrase is the subject of that verb in a sentence;
an object syntactic dependency, which identifies a noun or noun phrase and a verb for which the noun or noun phrase is the object of that verb in a sentence;
a predicate syntactic dependency, which identifies a dependency between a predicate and a noun or noun phrase; and
a negation syntactic dependency, which identifies a syntactic relation between a noun or noun phrase and a negative term selected from a term set of negative terms;

receiving an issue comprising a text sequence;
with a processor:
processing the text sequence with a syntactic parser which identifies syntactic dependencies between tokens of the text sequence, the syntactic parser including rules which tag instances of the domain terms in the domain term vocabulary,
applying the discourse patterns, at least some of the discourse patterns requiring a syntactic dependency between a term and one of the domain terms, and
categorizing the text sequence based on which of the discourse patterns are met by the text sequence of the received issue; and

outputting information based on the categorization of the text sequence, wherein the outputting information comprises at least one of:
identifying a process for responding to the issue from a plurality of processes associated with different issue categories;
identifying a knowledge base from a plurality of knowledge bases for querying based on the categorization of the issue; and
querying a knowledge base or collection of web posts with the issue, wherein the query is also based on the categorization of the issue.

2. The system of claim 1 , wherein the categories of issue include an anomaly category having at least one discourse pattern for identifying a deviation from normal behavior of an instance of a domain term, expressed by the author of the issue.

3. The method of claim 1 , wherein the anomaly category is associated with a plurality of discourse patterns which are defined by the anomaly discourse model, the anomaly discourse model providing different discourse patterns with respective rules for identifying deviance.

4. The method of claim 1 , wherein the issue categories include a plurality of information request categories.

5. The method of claim 1 , wherein the information request category is selected from:
a how-to category;
a property category;
an explanation category; and
combinations thereof.

6. The method of claim 5 , wherein the information request category includes a how-to category, and the at least one discourse pattern is selected from:
a pattern specifying that a direct or indirect question contains the word how in combination with a 1st person subject;
a pattern specifying that a direct or indirect question contains the word how to or way,
a pattern specifying that a direct or indirect question containing an instance of the lemma do has an instance of the lemma what as its direct object; and
combinations thereof.

7. The method of claim 5 , wherein the information request category includes a property category, and the at least one discourse pattern is selected from:
a pattern specifying a direct or indirect yes-no question;
a pattern specifying a direct or indirect Wh-question, except when classed in the how-to category; and
combinations thereof.

8. The method of claim 5 , wherein the information request category includes an explanation category, and the at least one discourse pattern is selected from:
a pattern specifying a direct or indirect question containing an instance of the lemma why;
a pattern specifying a direct or indirect question where an instance of the lemma cause has a subject or an object; and
combinations thereof.

9. The method of claim 1 , further comprising applying rules for detection of direct and indirect questions in the text sequence.

10. A method for categorizing an issue and outputting information based on the categorization, the method comprising:
(I) for each of a plurality of categories of issue including an information request category and an anomaly category, providing a discourse model, the discourse models including a request discourse model which applies rules for identifying discourse patterns for identifying information requests in a text sequence and an anomaly discourse model which applies rules for identifying discourse patterns identifying anomalies in the text sequence, the discourse patterns configured for identify text sequences that meet the respective discourse pattern, at least one of the discourse patterns specifying that an instance of a domain term in a domain term vocabulary be present in the text sequence for the discourse pattern to be met;
(II) receiving an issue comprising a text sequence;
(III) with a processor, parsing the text sequence with a syntactic parser to:
(a) identify syntactic dependencies between tokens of the text sequence, the syntactic parser including rules which tag instances of the domain terms in the domain term vocabulary, and
(b) identify occurrences of syntactic dependencies between words of the text sequence from a predefined set of syntactic dependencies, at least some of the patterns specifying that the domain term be in a syntactic dependency, wherein the syntactic dependencies in the predefined set of syntactic dependencies are identified with rules, the rules including rules for identifying:
(i) a subject syntactic dependency, which identifies a noun or noun phrase and a verb for which the noun or noun phrase is the subject of that verb in a sentence,

Assignees

• Conduent Business Services Llc
• Conduent Business Services Llc

Inventors

• Sandor Ágnes
• Lagos Nikolaos
• Brun Caroline
• Vo Ngoc Phuoc An

Classifications

• G06F40/30

  Semantic analysis · CPC title

• G06F40/35 Primary

  Discourse or dialogue representation · CPC title

• G06F16/353 Primary

  into predefined classes · CPC title

• G06F40/253

  Grammatical analysis; Style critique · CPC title

• G06F40/211

  Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars · CPC title