Library and Information Science

In these days the online full-text databases are increasing, but these full-text databases are difficult to retrieve, because recall is higher than bibliographic databases, and precision is so lower. There are cases where we don’t always read whole paper, but use one or a few part of an article. So this paper presents an approach to retrieve the relevant parts of a document by using paragraphs of individual documents. Sample documents are 49 articles in Japanese about information retrieval and natural language processing studies. The retrieval technique used in this retrieve experiment is the vector space model. As a result, the higher precision and recall were shown by using the words in chapter titles or section headings to retrieve the relevant paragraphs.