Studies on Relevant Document Retrieval and Summarization January, 2001 Doctor of Engineering

The goal of this dissertation is to develop methods for supporting and enhancing intellectual activities. For this purpose, firstly, we propose a method for retrieving relevant documents. Secondly, we propose a summarization method for multiple articles. Thirdly, we investigate a method of acquiring case frames from monolingual corpora. And, we present a word-sense disambiguation method required for acquisition of sequences of case elements. A great deal of information can be acquired nowadays thanks to the recent drastic progress of computer and network technologies. On the other hand, human physical abilities of information processing have been hardly augmented since the dawn of its history. Therefore, efficient and effective retrieval of information which we need from a great deal of information, and its quick understanding of the retrieved information, which may be realized by summarization, are quite important to make critical decisions in this information-oriented society. There are some kinds of information(documents, speech, images and video etc.). Among them, documents are quite important as the basis for intellectual activities. Most of earlier information retrieval studies suppose query-based systems, and they proposed language-independent methods. On the other hand, a different approach is possible by using the document itself as a query and by searching similar documents for the query. Some methods have been proposed for such retrieval, however, there have been few studies to investigate what unit of index terms should be employed for Japanese documents. Thus, in this dissertation, we propose a method of employing connections of morphemes as the unit of index