GENERATING DISTRIBUTED WORD EMBEDDINGS USING STRUCTURED INFORMATION

• US 20160328386A1
• Filed: 06/14/2016
• Published: 11/10/2016
• Est. Priority Date: 05/08/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: CROSS, JAMES H. III, XIANG, BING, FAN, JAMES J., ZHOU, BOWEN
• Granted Patent: US 9,892,113 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri
• 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 computer program that uses structured information, such as syntactic and semantic information, as context for representing words and/or phrases as vectors, by performing the following steps: (i) receiving a first set of natural language text and a set of information pertaining to the first set of natural language text, where the information includes metadata and corresponding contextual information indicating a relationship between the metadata and the first set of natural language text; and (ii) generating a first vector representation for the first set of natural language text utilizing the metadata and its corresponding contextual information.

Claims

1. A method comprising:

receiving, by one or more processors, a first set of natural language text and a set of information pertaining to the first set of natural language text, where the information includes metadata and corresponding contextual information indicating a relationship between the metadata and the first set of natural language text; and

generating, by one or more processors, a first vector representation for the first set of natural language text utilizing the metadata and its corresponding contextual information.

2. The method of claim 1, wherein the generating of the first vector representation for the first set of natural language text comprises:

determining, by one or more processors, a substitute set of natural language text based, at least in part, on the metadata and its corresponding contextual information; and

generating, by one or more processors, the first vector representation for the first set of natural language text utilizing the substitute set of natural language text.

3. The method of claim 2, wherein the substitute set of natural language text includes the first set of natural language text.

4. The method of claim 2, wherein the substitute set of natural language text does not include the first set of natural language text.

5. The method of claim 2, wherein the generating of the first vector representation for the first set of natural language text utilizing the substitute set of natural language text further comprises:

generating, by one or more processors, a first initial vector representation from the first set of natural language text;

generating, by one or more processors, a second initial vector representation from the substitute set of natural language text; and

adding, by one or more processors, the first initial vector representation and the second initial vector representation to generate the first vector representation.

6. The method of claim 1, further comprising:

determining, by one or more processors, an amount of similarity between the first set of natural language text and a second set of natural language text by comparing the first vector representation for the first set of natural language text to a second vector representation for the second set of natural language text.

7. The method of claim 1, wherein the contextual information corresponding to the metadata indicates a syntactic relationship between the metadata and the first set of natural language text.

8. The method of claim 7, wherein the contextual information corresponding to the metadata includes at least one of the following:

a constituent parse tree and a dependency parse tree.

9. The method of claim 1, wherein the contextual information corresponding to the metadata indicates a semantic relationship between the metadata and the first set of natural language text.

10. The method of claim 9, wherein the metadata is a named entity type of at least a portion of the first set of natural language text.