A Retrieval Method for Relevant Documents Employing Connective Information of Nouns

摘要

We propose a retrieval method of relevant Japanese documents by employing information on noun connections. The method is based on the vector space model, but the method employs noun connections as indexing terms. On the part of information retrieval, there is a problem that we can use various nouns to express a phenomenon. The problem causes a decline of precision in information retrieval. In addition, Japanese texts have many compound nouns and these nouns mny be hindrance to retrieve, because these nouns may cause fluctuation. In this paper, we propose employing noun connections as indexing terms to improve the precision. And we show that the recall raised by heuristics solving fluctuation of compound nouns. We carried out experiments with comparing a word-based method and the proposed method for a Japanese newspaper (The Nihon Keizai Shimbun). The experimental results show that the proposed method attains 85.9/100 F-measure on the average which is approximately 10/100 higher than that of a conventional word-based method.