ENTITY-RELATION BASED PASSAGE SCORING IN A QUESTION ANSWERING COMPUTER SYSTEM

• US 20160147763A1
• Filed: 03/11/2015
• Published: 05/26/2016
• Est. Priority Date: 11/20/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kalyanpur, Aditya A., Murdock, James W. IV
• Granted Patent: US 9,910,844 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries
• CPC Class Codes: G06F 16/288 Entity relationship models
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates

Abstract

According to an aspect, a query and a passage are parsed by a language parser to detect noun-centric phrases and verb-centric phrases in the query and the passage. Entities, including at least one untyped entity, are identified based on the noun-centric phrases and relations are identified based on the verb-centric phrases. Entity pairs are created that include an entity identified in the query and an entity identified in the passage, each pair satisfies a matching criteria. Relation pairs are created that include a relation identified in the query and a relation identified in the passage, each pair satisfies a matching criteria. A passage score that indicates the likelihood that an answer to the query is contained in the passage is determined based on the entity pairs, the matching criteria satisfied by each entity pair, the elation pairs, and the matching criteria satisfied by each relation pair.

Claims

1. A method comprising:

parsing, by a language parser, a query and a passage to detect noun-centric phrases and verb-centric phrases in the query and the passage;

identifying entities, for both of the query and the passage, based on the noun-centric phrases detected from the parsing, the entities including at least one untyped entity;

identifying relations, for both of the query and the passage, based on the verb-centric phrases detected from the parsing;

creating one or more entity pairs that each include an entity identified in the query and an entity identified in the passage, each entity pair satisfying a matching criteria with respect to entities of the entity pair;

creating one or more relation pairs that each include a relation identified in the query and a relation identified in the passage, each relation pair satisfying a matching criteria with respect to relations of the relation pair; and

determining, by a question answering computer system, a passage score for the passage that indicates a likelihood that an answer to the query is contained in the passage, the determining based on the one or more entity pairs, the matching criteria satisfied by each entity pair, the one or more relation pairs, and the matching criteria satisfied by each relation pair.

2. The method of claim 1, further comprising:

calculating, by the question answering computer system, for each of the entity pairs, an entity match confidence score that indicates a level of confidence that the entities of the entity pair match; and

calculating, by the question answering computer system, for each of the relation pairs, a relation match confidence score that indicates a level of confidence that the relations of the relation pair match,

wherein the determining is further based on the one or more entity match confidence scores and the one or more relation match confidence scores.

3. The method of claim 1, wherein at least one of the relations is an untyped relation.

4. The method of claim 1, wherein the language parser includes a part-of-speech parser and a dependency parser.

5. The method of claim 1, wherein the identifying of at least one of the relations is further based on an inference using a rule obtained from a knowledge base.

6. The method of claim 1, wherein each entity includes one or more terms.

7. The method of claim 1, wherein the matching criteria with respect to entities of the entity pair includes at least one of the entity identified in the query is the same as the entity identified in the passage, the entity identified in the query is a synonym of the entity identified in the passage, and the entity identified in the query is determined by latent semantic analysis (LSA) to be statistically similar to the entity identified in the passage.

8. The method of claim 1, wherein the matching criteria with respect to relations of the relation pair includes at least one of the relation identified in the query is the same as the relation identified in the passage, the relation identified in the query is a synonym of the relation identified in the passage, and the relation identified in the query is determined by LSA to be statistically similar to the relation identified in the passage.