ROSA — Multi-agent System for Web Services Personalization

Abstract

Automatic and non-invasive web personalization seems to be a challenge for nowadays web sites. Many web mining techniques are used to achieve this goal. Since current web sites evolve constantly, web mining operations should be periodically repeated. A multi-agent architecture facilitates integration of different mining methods and permits the discovered knowledge to be verified and updated automatically. We propose ROSA (Remote Open Site Agents) -- a system that may be easily incorporated into an existing web site. It consists of multiple heterogeneous agents such as: User Session Monitor, Crawler, Content Miner, Usage Miner, Hyperlink Recommender, Banner Recommender, etc., that are responsible for specific web mining and personalization tasks. They integrate various mining techniques using common representation of documents in the vector space model in order to recommend hyperlinks and banners. Verification process is represented by a task graph. ROSA agents not only detect when their information should be verified, but they are also able to coordinate knowledge update operations (using method presented in this paper). The practical part describes the usage of FIPA- RDF0 and ACL languages.