System and method for question answering document retrieval

1. A question answering document retrieval system, implemented by a computer system, which executes document retrieval in accordance with a retrieval request in question form, comprising:

retrieval means for retrieving associated documents on the basis of a keyword extracted from a question serving as a retrieval request in question form from a user, the retrieval means extracting descriptions associated with the keyword and contained in the retrieved documents as first summaries, thereby acquiring document retrieval results including a list of the extracted first summaries;

question type determination means for determining a question type of the question by analyzing a semantic role of the question;

summary extraction means for extracting descriptions corresponding to answers matching the determined question type as second summaries from original-document data which forms documents indicated by the document retrieval results, the original-document data being included in original-document data stored in an original-document database and forming each document as a retrieval target, the summary extraction means acquiring a list of the extracted second summaries; and

an interface configured to display the document retrieval results and the list of the second summaries, as answers to the retrieval request from the user, on a display device which can be used by the user, the document retrieval results including the list of the first summaries,

wherein:

the retrieval means includes first score assignment means for calculating a first score representing a degree of relevance with a keyword extracted from the question for each document when associated documents are retrieved on the basis of the keyword, selection means for selecting associated documents on the basis of the first scores calculated by the first score assignment means, and first extraction means for extracting the first summaries from the documents selected by the selection means; and

the summary extraction means includes second score assignment means for calculating a second score representing a degree of relevance with a keyword extracted from the question for each candidate for a description corresponding to an answer matching the determined question type, and second extraction means for extracting corresponding descriptions as the second summaries on the basis of the second scores calculated by the second score assignment means.

2. The question answering document retrieval system according to claim 1 , wherein the interface includes a display order decision unit configured to decide a display order of the first summaries extracted by the first extraction means on the basis of the first scores corresponding to the documents containing the first summaries which are calculated by the first score assignment means, and decide a display order of the second summaries extracted by the second extraction means on the basis of the second scores corresponding to the second summaries which are calculated by the second score assignment means.

3. The question answering document retrieval system according to claim 1 , wherein:

the selection means selects associated documents on the basis of the first scores calculated by the first score assignment means in the order of the scores;

the first extraction means transfers, to the interface, document retrieval results including the list of the first summaries extracted from the documents selected in the order of the scores with the first summaries in the list being arranged in the order of the corresponding scores; and

the second extraction means transfers, to the interface, the list of the second summaries extracted on the basis of the second scores calculated by the second score assignment means with the second summaries in the list being arranged in the order of the corresponding scores.

4. The question answering document retrieval system according to claim 1 , further comprising a pattern determination dictionary which stores, for each predetermined question type, sentence structure information representing a sentence structure unique to the question type,

and wherein:

the summary extraction means includes means for extracting descriptions having a sentence structure unique to the determined question type as candidates for descriptions corresponding to answers matching the question type by pattern matching between the original-document data forming each document indicated by the document retrieval result and document structure information which is stored in the pattern determination dictionary and is unique to the determined question type; and

the second score assignment means calculates a score representing a degree of relevance with a keyword extracted from the question for each of the extracted candidates.

5. A method for question answering document retrieval which retrieves documents in accordance with a retrieval request in question form, comprising:

retrieving associated documents on the basis of a keyword extracted from a question serving as a retrieval request in question form from a user;

extracting descriptions associated with the keyword and contained in retrieved documents as first summaries;

acquiring document retrieval results including a list of the extracted first summaries;

determining a question type of the question by analyzing a semantic role of the question;

extracting descriptions corresponding to answers matching the determined question type as second summaries from original-document data, of original-document data stored in an original-document database and forming each document as a retrieval target, which form documents indicated by the document retrieval results; and

displaying the document retrieval results and the list of the extracted second summaries, as answers to the retrieval request from the user, on a display device which can be used by the user, the document retrieval results including the list of the acquired first summaries,

wherein:

the retrieving includes calculating a first score representing a degree of relevance with a keyword extracted from the question for each document when associated documents are retrieved on the basis of the keyword, and selecting, as retrieved documents, the associated documents on the basis of the calculated first scores; and

the extracting as the second summaries includes calculating a second score representing a degree of relevance with a keyword extracted from the question for each candidate for a description corresponding to an answer matching the determined question type, and extracting a corresponding description as the second summary on the basis of the calculated second score.

6. The method according to claim 5 , wherein the displaying includes deciding a display order of the extracted first summaries on the basis of the first scores corresponding to the documents containing the first summaries, and deciding a display order of the extracted second summaries on the basis of the second scores corresponding to the second summaries.

7. A computer program product tangibly embodied in a computer-readable storage medium for question answering document retrieval used to retrieve documents in accordance with a retrieval request in question form, comprising:

computer-readable code means for retrieving associated documents on the basis of a keyword extracted from a question serving as a retrieval request in question form from a user;

computer-readable code means for extracting descriptions associated with the keyword and contained in retrieved documents as first summaries;

computer-readable code means for acquiring document retrieval results including a list of the extracted first summaries;

computer-readable code means for determining a question type of the question by analyzing a semantic role of the question;

computer-readable code means for extracting descriptions corresponding to answers matching the determined question type as second summaries from original-document data which forms each document indicated by the document retrieval results, the original-document data being included in original-document data stored in an original-document database and forming each document as a retrieval target; and

computer-readable code means for displaying the document retrieval results and the list of the extracted second summaries, as answers to the retrieval request from the user, on a display device which can be used by the user, the document retrieval results including the list of the acquired first summaries,

wherein:

the retrieving includes calculating a first score representing a decree of relevance with a keyword extracted from the guest ion for each document when associated documents are retrieved on the basis of the keyword, and selecting, as retrieved documents, the associated documents on the basis of the calculated first scores; and

the extracting as the second summaries includes calculating a second score representing a degree of relevance with a keyword extracted from the question for each candidate for a description corresponding to an answer matching the determined question type, and extracting a corresponding description as the second summary on the basis of the calculated second score.