Abstract
The thesis proposes a multi-agent system that could reason over domain ontology could help retrieve information better than keyword matching search engines. We build a multi-agent system for digital library (MADL) that consists of librarian agent, thesaurus agent, ontology agent, and information gathering agents to carry out the tasks. Through the cooperation among these agents and the utilization of lexical thesaurus and domain ontology, MADL can transform user’s simple structured natural language into query schemas, which are designed for domain and language specific. And with the derived query schema, Information Gathering Agents could extract information from a specific information source. Currently we have built an experimental system for a small set of questions in a school domain. We compare the result of search with MADL’s aid and the result of a pure full-text search engine. The comparison shows that the experiment system performs better recall and precision than a pure full-text search engine.