A learning multi-agent system for personalized information filtering

A multi-agent hybrid learning approach to the problem of personalized information filtering is proposed in this paper. There are four agents in the multi-agent model. The problem is modeled as Monte Carlo reinforcement learning. Our proposed algorithm is modified Monte Carlo method combined with features of unsupervised suffix tree clustering and supervised backpropagation network. We argue that this proposed approach could precisely capture the user's interest without repeatedly asking for his/her explicit rates and converge to the user's interest quickly. A conclusion is drawn that our approach is efficient, precise and converges more quickly compared with existing approaches. A prototype system is being developed.