In this paper, we describe a peer to peer system that aims at enhancing the result ranking of web search engines by allowing a group of people to exchange, in an implicit way, their experiences in searching the web. Members of the target group are supposed to share the same information interests. The system functions as follows: Each user is associated with a personal assistant agent. Each agent saves in a log file all the interactions made by the user with search engines. This file is used to feed a local case-base. A case is composed of two parts: a problem part and a solution part. The problem part is composed of a description of a query as well as the ranked associated results returned by a search engine. The solution part is composed of the effective selection of results made by the user. When receiving an answer R for a query q, the agent tries to reorder the elements of R in function of the past experiences stored in his local case-base but also by asking peer agents to reorder the result set R. The paper describes 1) the internal architecture of the proposed agents, 2) the case-based reasoning cycle applied by each agent and 3) the collaboration protocol executed by the peer agents. First experimental results are also presented and commented.
展开▼