A Multi-Agent Framework for Personalized Information Filtering

As today the amount of accessible information is overwhelming, the intelligent and personalized filtering of available information is a great challenge. The main problems are that the relevant information is spread over a big number of sources and useful information is hidden under the huge amount of useless data. To cope with this problem several filtering and information query strategies have been developed but they are usually specialized on a bounded problem and do not take into account the individual preferences of the user. Moreover most search engines rate every document separately and do consider the relationship between the documents in the result set. In this paper we present a multi-agent system that integrates heterogeneous information sources, a big number of filtering and rating strategies as well as strategies for combining ratings from different agents and optimizing the filter result set according to the individual user preferences. In the framework each information source, filtering strategy and optimization strategy is presented as an intelligent agent so that the system is open and extendable at runtime. The framework monitors the resource demand of each agent as well as the availability of system resources for choosing the most adequate agents according to the requested response time. User feedback is collected and used for optimizing the filtering strategies and for learning in which context which strategy performs best. The filtering framework provides the basis for the Personalized Information System. The first evaluation results show that the filtering framework provides better results and that new filtering strategies can be seamlessly integrated