Incorporation of Japanese Information Retrieval Method Using Dependency Relationship into Probabilistic Retrieval

Abstract

Our Japanese information retrieval method using both dependency relation-ship between words and their semantic information, which we call DRB method, handles a frame that consists of a noun and a verb governing the noun as a key to judge whether or not a document is relevant to a query. This method proved that using dependency relationship in information retrieval is highly effective in terms of Precision. On the other hand, it is often too exacting to retrieve an appropriate number of documents, making Recall plunge. In order to solve this problem, we incorporate DRB method into a probabilistic method so that it scores and ranks documents with statistical information. In the experiments we compare performance of the incorporated methods, (A) Respective Frame matching with 2-Poisson Model and (B) Frame matching with 2-Poisson Model, with other four information retrieval methods, (C) 2-Poisson Model, (D) NEAR matching, (E) AND matching and (F) Frame matching, by utilizing BMIR-J21. Respective Frame matching retrieves every document that has an identical frame with at least one of a query’s frames while Frame matching retrieves documents that have the identical frames with the query’s frames.