Question answering device, question answering method, and question answering program

• US 8,983,977 B2
• Filed: 02/20/2007
• Issued: 03/17/2015
• Est. Priority Date: 03/01/2006
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Ishikawa, Kai, Akamine, Susumu
• Primary Examiner(s): BURKE, JEFF A
• Application Number:
  US12/281,350
• Publication Number:
  US 20090012926A1
• Time in Patent Office: 2,947 Days
• Field of Search: 707748-750, 707/769, 707/771
• US Class Current: 707/750
• CPC Class Codes: G06F 16/243 Natural language query formulation
• CPC Class Codes: G06F 16/3325 Reformulation based on results of preceding query
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3334 Selection or weighting of terms from queries, including natural language queries
• CPC Class Codes: G06F 16/3335 Syntactic pre-processing, e.g. stopword elimination, stemming

Abstract

A question-answering device, a question-answering method, and a question-answering program that can obtain an answer to an inputted query with high probability are described. A score calculation element 305 determines a matching degree between the group of the style and the topic of an inputted query and the group of the style and the topic of the query of question-answer pairs. A search result presentation element 306 narrows the question-answer pairs on the basis of the matching degree.

Claims

1. A question-answering device comprising:

a question-answer pair storing unit that stores a plurality of question-answer pairs composed of reference queries and their answers;

an analysis result storing unit that stores groups of a style representing the type of a question of each reference query of said reference queries and a topic representing an object of a question of each reference query;

an input unit that receives a query;

an analysis pattern storing unit that stores a rule used to decide groups of a style and a topic of the query received in the input unit;

an analysis unit that obtains the style and the topic from the query by analyzing the query received in the input unit on a basis of the rule stored in the analysis pattern storing unit, and decides the groups of the style and the topic of the query;

a calculation unit that determines a matching degree between each group of the style and the topic of each reference query stored in the analysis result storing unit and each group of the style and the topic of the query received in the input unit decided by the analysis unit, and

calculates a search score based on the matching degree and information about a frequency of appearance of a keyword in the query received in the input unit in each reference query and each reference query'"'"'s answer; and

a narrowing unit that narrows search results of the query received in the input unit from the plurality of question-answer pairs stored in the question-answer pair storing unit into only question-answer pairs, of which the question contents match the query received in the input unit, on a basis of the search score calculated by the calculation unit,

wherein the calculation unit calculates the search score by an operation including multiplying the matching degree by information about the frequency of appearance of the key word in the query received in the input unit in each reference query and each reference query'"'"'s answer.

2. The question-answering device according to claim 1, further comprising:

a retrieval unit that extracts the keyword from the query received in the input unit, and performs a keyword search on the plurality of question-answer pairs stored in the question-answer pair storing unit as search objects by using the extracted keyword,

wherein the calculation unit reads out the groups of the style and the topic of each reference query, which is included in the search results of the retrieval unit, from the analysis result storing unit, and determines a matching degree between each group of the style and the topic of the read-out reference query and the group of the style and the topic of the query decided in the analysis unit.

3. The question-answering device according to claim 1, further comprising:

a question-content matching degree table storing unit that stores a question-content matching degree table showing the matching degree between two groups of the styles and the topics,

wherein the calculation unit determines a matching degree between each group of the style and the topic of each reference query and the group of the style and the topic of the query determined in the analysis unit, by using the question-content matching degree table.

4. The question-answering device according to claim 1,

wherein the calculation unit calculates a score that represents a matching degree between the keyword of the query received in the input unit and a keyword of each of the question-answer pairs, and

the narrowing unit narrows the search results of the query received by the input unit from the plurality of question-answer pairs stored in the question-answer pair storing unit, on the basis of the score calculated in the calculation unit and the matching degree determined in the calculation unit.

5. The question-answering device according to claim 1, wherein the information about the frequency of appearance of the key word is calculated of the key word by TF-IDF method.

6. A question-answering device comprising:

a question-answer pair storing means that stores a plurality of question-answer pairs composed of reference queries and their answers;

an analysis result storing means that stores groups of a style representing a type of a question of each reference query of said reference queries and a topic representing an object of a question of each reference query;

an input means that receives a query;

an analysis pattern storing means that stores a rule used to decide the groups of the style and the topic of the query received in the input means;

an analysis means that obtains the style and the topic from the query by analyzing the query received in the input means on a basis of the rule stored in the analysis pattern storing means, and decides the groups of the style and the topic of the query;

a calculation means that determines a matching degree between each group of the style and the topic of each reference query stored in the analysis result storing means and each group of the style and the topic of the query received in the input means decided by the analysis means, and

calculates a search score based on the matching degree and information about a frequency of appearance of a keyword in the query received in the input means in each reference query and each reference query'"'"'s answer; and

a narrowing means that narrows search results of the query received in the input means from the plurality of question-answer pairs stored in the question-answer pair storing means into only question-answer pairs, of which the question contents match the query received in the input means, on a basis of the search score calculated by the calculation means,

wherein the calculation means calculates the search score by an operation including multiplying the matching degree by information about the frequency of appearance of the key word in the query received in the input unit in each reference query and each reference query'"'"'s answer.

7. The question-answering device according to claim 6, wherein the information about the frequency of appearance of the key word is calculated of the key word by TF-IDF method.

8. A question-answering method performed by a question-answering device comprising:

an input step of receiving a query;

an analysis step of obtaining a style and a topic from the query by analyzing the query received in the input step on a basis of a rule stored in an analysis pattern storing unit, and deciding the groups of the style and the topic of the query;

a calculation step of determining a matching degree between each group of the style and the topic of a reference query stored in an analysis result storing unit and the group of the style and the topic of the query received in the input step, and

calculating a search score based on the matching degree and information about a frequency of appearance of a keyword in the query received in the input step each reference query and each reference query'"'"'s answer; and

a narrowing step of narrowing search results of the query from a plurality of question-answer pairs stored in a question-answer pair storing unit into only question-answer pairs, of which the question contents match the query, on a basis of the search score calculated,

wherein the question-answering device includes the question-answer pair storing unit that stores the plurality of question-answer pairs composed of the reference queries and their answers, the analysis result storing unit that stores groups of the style representing a type of the question of each reference query and the topic representing an object of the question of each reference query, and the analysis pattern storing unit that stores a rule used to decide the groups of the style and the topic of the query,

wherein the calculation step comprises calculating the search score by an operation including multiplying the matching degree by information about the frequency of appearance of the key word in the query received in the input unit in each reference query and each reference query'"'"'s answer.

9. The question-answering method according to claim 8, further comprising:

an extraction step of extracting a keyword from the query; and

a search step of performing a keyword search on the plurality of question-answer pairs as search objects, that is stored in the question-answer pair storing unit, by using the extracted keyword,

wherein in the calculation step, the groups of the style and the topic of each reference query, which is included in the search results of the search step, are read out from the analysis result storing unit; and

a matching degree between each group of the style and the topic of the read-out reference query and the group of the style and the topic of the query is determined.

10. The question-answering method according to claim 8,

wherein the question-answering device further includes a question-content matching degree table storing unit that stores a question-content matching degree table showing the matching degree between two groups of the styles and the topics, and

in the calculation step, a matching degree between each group of the style and the topic of each reference query and the group of the style and the topic of the query is determined by using the question-content matching degree table.

11. The question-answering method according to claim 8,

wherein in the calculation step, a score that represents a matching degree between a keyword of the query and a keyword of each of the question-answer pairs is calculated, and

in the narrowing step, the search results of the query are narrowed from the plurality of question-answer pairs stored in the question-answer pair storing unit, on the basis of the score and the matching degree.

12. The question-answering method according to claim 8, wherein the information about the frequency of appearance of the key word is calculated of the key word by TF-IDF method.

13. A non-transitory recording medium recording thereon a question-answering program for performing a question-answering process by execution on a computer, the computer being connected to a question-answer pair storing unit that stores a plurality of question-answer pairs composed of reference queries and their answers, an analysis result storing unit that stores groups of a style representing a type of a question of each reference query of said reference queries and a topic representing an object of the question of each reference query, and an analysis pattern storing unit that stores a rule used to decide the groups of the style and the topic of the query,

wherein the question-answering process includes:

an input process for receiving a query;

an analysis process for obtaining a style and a topic from the query by analyzing the query received in the input process on a basis of the rule stored in the analysis pattern storing unit, and deciding the groups of the style and the topic of the query;

a calculation process for determining a matching degree between each group of the style and the topic of each reference query stored in the analysis result storing unit and the group of the style and the topic of the query received in the input process, and

calculating a search score based on the matching degree and information about a frequency of appearance of a keyword in the query received in the input process in each reference query and each reference query'"'"'s answer; and

a narrowing process for narrowing search results of the query from the plurality of question-answer unit stored in the question-answer pair storing unit into only question-answer pairs, of which the question contents match the query, on a basis of the search score calculated,

wherein said calculation process calculates the search score by an operation including multiplying the matching degree by information about the frequency of appearance of the key word in the query received in the input unit in each reference query and each reference query'"'"'s answer.

14. The non-transitory recording medium according to claim 13, wherein the information about the frequency of appearance of the key word is calculated of the key word by TF-IDF method.