Abstract
In order to meet the accuracy need of network users on information service, this paper introduces multi- agent feedback learning mechanism based on traditional collaborative filtering algorithms and proposes recommendation algorithm based on multi-agent. During the recommendation process, the algorithm, considering the recommended project as a recommended subject, infers autonomously whether the recommended project fits the target users. Meanwhile, the recommended subject can also improve recommendation precision by active learning based on user’s feedback. The recommendation algorithm can be finely applied to library recommender system and partial E-commerce recommender system.