Document retrieval device

TL;DR: In this paper, the authors present a method to efficiently narrow down documents to obtain documents which are close to retrieval intention, and enable high-precision retrieval, by clearly showing the contents of an input sentence and a document to a user as the correspondence relation between partial concepts showing respective contents.

Abstract: PROBLEM TO BE SOLVED: To efficiently narrow down documents to obtain documents which are close to retrieval intention, and enable high-precision retrieval, by clearly showing the contents of an input sentence and a document to a user as the correspondence relation between partial concepts showing respective contents and linguistic features belonging to the respective concepts. SOLUTION: The input sentence described in a natural language is inputted from a user through an input/output control part 11. A feature classification part 21 classifies input features stored in an input feature storage part 15 so as to correspond to partial concepts constituting the input sentence, and gives classification names and stores them in an input feature classification storage part 17. Further, a feature classification part 21 similarly classifies document features stored in a document feature storage part 14 and stores them in a document feature classification storage part 16. A feature matching part 22 matches the document feature classifications and corresponding document features stored in the document feature storage part 16 against the input feature classifications and corresponding input features stored in the input feature classification storage part 17 to determine the order of the document, and stores a ranking table in a result storage part 18 and also shows it to the user.