Question answering system and question answering processing method

• US 20050086045A1
• Filed: 06/03/2004
• Published: 04/21/2005
• Est. Priority Date: 10/17/2003
• Status: Active Grant

• Current Assignee: National Institute of Information and Communications Technology
• Original Assignee: National Institute of Information and Communications Technology
• Inventors: Murata, Masaki
• Granted Patent: US 7,444,279 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/20 Natural language analysis semantic analysis of natural language G06F40/30

Abstract

A question answering system estimates an answer type of an answer from a input question statement, extracts keywords from the question statement and retrieves a document database, extracts language expressions (answer candidates) from the extracted document data, assigns evaluation points thereto. When there are a plurality of answer candidates having the same language expression, the system sorts those evaluation points in descending ranking order of evaluation, calculates values of evaluation points using such weighting that the value to be processed for each evaluation point diminishes, regards the sum total of those values as the evaluation point of the answer candidate, and outputs the answer candidate whose counted evaluation point is equal to or greater than a predetermined evaluation as an answer.

Claims

1. A question answering system which receives question statement data expressed in a natural language and outputs an answer to the question statement data from a document data group which is an answer retrieval target, the system comprising:

answer type estimating means for analyzing language expressions of the question statement data entered and estimating an answer type which is a type of language expression that can be an answer to the question statement data;

document retrieving means for extracting keywords from the question statement data, retrieving and extracting document data including the keywords from the document data group;

answer candidate evaluation point calculating means for extracting language expressions that can be the answer from the document data as answer candidates and assigning evaluation points to the answer candidates;

answer candidate evaluation point counting means for sorting and ranking, when evaluation points are counted for each answer candidate having the same language expression, evaluation points of answer candidates having the same language expression in descending order of evaluation ranking, calculating values of the evaluation points using such weighting that the value to be processed for each evaluation point diminishes in descending ranking order of evaluation points and regarding the sum total of the values as the evaluation point of the answer candidate having the same language expression; and

answer outputting means for outputting answer candidates whose evaluation point counted by the answer candidate evaluation point counting means is equal to or higher than a predetermined evaluation as an answer.

2. The question answering system according to claim 1, wherein when counting evaluation points for each answer candidate having the same language expression, the answer candidate evaluation point counting means calculates values of the evaluation points of the answer candidates having the same language expression and sorted and ranked in descending ranking order of evaluation using a formula 1;

Score=Σ

1/(i+b)/(i+b−

1)*Point_i  

 

(1)

where Σ

is the sum total when i takes a value of 1 to n;

n is the number of answer candidates having the same language expression; and

Point_i (1≦

i≦

n) is the ranking order of evaluation points of the answer candidates;

from which each sum total of the values of the answer candidates is obtained as the representative evaluation point of answer candidate having the same language expression.

3. The question answering system according to claim 2, wherein the answer candidate evaluation point counting means calculates the evaluation points using the formula 1 in which b takes a value from 0.1 to 0.5.

4. The question answering system according to claim 1, wherein when counting evaluation points for each answer candidate having the same language expression, the answer candidate evaluation point counting means calculates values of the evaluation points of the answer candidates having the same language expression and sorted and ranked in descending ranking order of evaluation using a formula 2;

Score=Σ

k<sup>i−

1</sup>*Point_i  

 

(2)

where Σ

is the sum total when i takes a value of 1 to n;

n is the number of answer candidates having the same language expression; and

Point_i (1≦

i≦

n) is the ranking order of evaluation points of the answer candidates;

from which each sum total of the values of the answer candidates is obtained as the representative evaluation point of answer candidate having the same language expression.

5. The question answering system according to claim 4, wherein the answer candidate evaluation point counting means calculates the evaluation points using the formula 2 where k takes a value from 0.1 to 0.5.

6. A question answering processing method which receives question statement data expressed in a natural language and outputs an answer to the question statement data from a document data group which is an answer retrieval target, the method comprising:

an answer type estimation processing process of analyzing language expressions of the question statement data entered and estimating an answer type which is a type of language expression that can be an answer to the question statement data;

a document retrieval processing process of extracting keywords from the question statement data, retrieving and extracting document data including the keywords from the document data group;

an answer candidate evaluation point calculation processing process of extracting language expressions that can be the answer from the document data as answer candidates and assigning evaluation points to the answer candidates;

an answer candidate evaluation point counting processing process of sorting and ranking, when evaluation points are counted for each answer candidate having the same language expression, evaluation points of answer candidates having the same language expression in descending order of evaluation ranking, calculating values of the evaluation points using such weighting that the value to be processed for each evaluation point diminishes in descending ranking order of evaluation points and regarding the sum total of the values as the evaluation point of the answer candidate having the same language expression; and

an answer output processing process of outputting answer candidates whose evaluation point counted by the answer candidate evaluation point counting means is equal to or higher than a predetermined evaluation as an answer.

7. The question answering processing method according to claim 6, wherein when evaluation points are counted for each answer candidate having the same language expression in the answer candidate evaluation point counting processing process, evaluation points of answer candidates having the same language expression are sorted and ranked in descending ranking order of evaluation, and a formula 1;

Score=Σ

1/(i+b)/(i+b−

1)*Point_i  

 

(1)

where Σ

is the sum total when i takes a value of 1 to n;

n is the number of answer candidates having the same language expression; and

Pointi (1≦

i≦

n) is the ranking order of evaluation points of the answer candidates;

which is calculated to obtain each sum total of the values of the answer candidates as the representative evaluation point of answer candidate having the same language expression.

8. The question answering processing method according to claim 6, wherein when evaluation points are counted for each answer candidate having the same language expression in the answer candidate evaluation point counting processing process, evaluation points of answer candidates having the same language expression are sorted and ranked in descending ranking order of evaluation, and a formula 2;

Score=Σ

k<sup>i−

1</sup>*Point_i  

 

(2)

where Σ

is the sum total when i takes a value of 1 to n;

n is the number of answer candidates having the same language expression; and

Point_i (1≦

i≦

n) is the ranking order of evaluation points of the answer candidates;

which is calculated whereto obtain each sum total of the values of the answer candidates as the representative evaluation point of answer candidate having the same language expression.