In many cases the subjects of document retrieval systems are whole documents. In a document like a thesis, multiple topics can be identified that all harmonize with the major. It is becoming more and more important to support retrieving not only main topics but also"supplementary" topics. In this paper we propose a new document retrieval method using the vector space model. It is based on the relationships among the topics of the document substructures. It gives precise document vectors even with very long documents containing several topics. We introduced the 'sequential' and 'include' query pattern for topic substructures. A uset can specify his/her demand with its context. We also built a prototype system, and conducted some experiments with a document set, Master's these of Information Science at NAIST, to demonstrate the effectiveness of our method.
展开▼
