Takaya Ueda

Example-based frame mapping for heterogeneous information agents

Communication among heterogeneous information agents is often difficult because they often use different structures to express the same information. A common ontology could resolve this problem. However, it would be very difficult to build such an ontology for dynamic information spaces such as the WWW. Thus, automatic reconciliation of differences is necessary to permit multiagent communication within an open environment where new expressions tend to emerge in communication. We propose a technique that enables automatic mapping of equivalent slots in different frames. The proposed technique, Example-Based Frame Mapping (EBFM), generates hypotheses of equivalent slots, then selects a plausible hypothesis by comparing similarities within the examples inferred to express the same object. EBFM solves the essential problems that arise in frame mapping (e.g., synonyms and homonyms) by using only the available information without requiring any domain specific heuristic knowledge, and can achieve more than 99% accuracy.

An information retrieval system based on personal viewpoints for everyday use

We propose an information retrieval system, Fit, to support daily information gathering and utilization. Fit can adapt to shifts in personal viewpoint, which is essential for everyday use. Using simple folders for the representation of the viewpoint eases the adaptation, but browsing all the folders becomes difficult. Fit facilitates efficient gathering and effective utilization by features with two information retrieval technologies, the vector space model method and the Boolean model method, which are arranged for everyday use. An evaluation of the system involving reading a newspaper, saving articles and retrieving them demonstrates the effectiveness of the features.