A methodology for building information agents

Information agents are increasingly being used for efficient and precise information retrieval from the Internet. Most of them are handcrafted from scratch and can not easily be adapted to other search domains. Targeting on agents for searching semi-structured data, we propose a knowledgebased methodology for easy generation and adaptation of information agents. Agents can be built by adding domain knowledge and information source knowledge to a Web search framework. To adapt an information agent from one domain to another, one simply changes the knowledge bases. We introduce “Knowledge Units” as the basis for knowledge representation and information retrieval. Domain knowledge is represented as a set of Knowledge Units in Knowledge Unit Description Frames. The main function of the Web search framework is to parse Web pages to their structured pages via the recognition and grouping of Knowledge Units. Our methodology has been used to generate domain specific information agents for real estate advertisements, car advertisements and soccer scores.