Structure and retrieval mechanism of a minutes retrieval system

This paper describes structure of a minutes retrieval system, and analysis of person-topic relation using latent semantic indexing. This minutes retrieval system is developed for retrieving parts of minutes, and retrieving persons related to certain topics, which is named MRS (Minutes Retrieval System). This system treats the minutes of IOC (International Olympic Committees). The system handles minutes of 1980's, which describe with respect to Olympic game and related topics. The purposes for developing the system are to provide a flexible minutes retrieval system, to represent minutes in XML, and to obtain the relationships between individual committee members and their interests. To meet these purposes, at first, minutes are represented in XML. Some tags are defined according to the structure or rules of the original minutes. Moreover, in order to access and analyze the minutes, to make clear named entities is required. To understand the minutes precisely it is necessary to handle named entities suitably, because accurate understanding of named entities is required, and the same named entity appears in terms of different representations. Finally, it is meaningful to clarify who are interested in certain topics. To achieve this, speakers and their speech contents are analyzed. Co-occurrence relationships between speakers and contents are analyzed using latent semantic indexing. Their relationships are depicted in a space constructed by this indexing. Certain experiments were done in which topics were given as queries and persons related to these topics were retrieved.