Example-Based Frame Mapping Applied to Information Agents for Distributed Information Sources

摘要

When developing information agents used for collecting and integrating information from distributed in - formation sources, one must deal with different ontologies depending on the informahon source. The authors intro- duced Example-Based Frame Mapping (EBFM) that auto- matically defines correspondence of slots expressing the same attributes in the frame format, which is extensively used as the format for providing structured information. It is a method for automatic integration of different ontologies at the vocabulary level. In EBm, conceptually correspond - ing slots are obtained using expression examples of com- mon objects in different information sources. This involves identifying expression examples of the same objects within different information frame formats, and assigning corre- spondence on the basis of similarity of the slot values. To prevent assigning correspondence to different entities, the slots' correspondence relationships are derived after a hy- pothesis is generated using such clues as uniqueness, which is computed from the information theory and usage within the same class. The authors have manually extracted infor- mation in the frame format present on the WWW, and evaluated this method based on the extracted data. In the cases when common examples appeared, comparison with the results of manually assigned slot correspondences dem - onstrated accuracy of 99 or better, with a recall rate of 60 or higher.