基于混合学习策略的多Agent信息过滤系统

Multi-Agent system for information filtering based on hybrid learning approach

针对用户个性化服务的要求,给出了一种基于混合学习策略和BP神经网络的多Agent信息过滤系统实现方案。系统采用蒙特卡罗强化学习算法进行多Agent协作学习,同时运用三层BP神经网络计算用户的满意度,根据算出的满意度对用户兴趣模型进行更新。本系统中用户无须反复提供显示反馈,由Agent跟踪并记录用户的浏览行为而得到用户的隐式反馈信息,从而减轻了用户的负担。

With the requirement of user＇s specific information service, a framework of multi-Agent system for information filtering, based on hybrid leaming approach and BP neural network, was proposed in this paper. This system aimed at helping the user obtain the relevant information precisely. The Modified Monte-Carlo method was used to collaborate multi-Agent system, and the intelligent Agent was exploited to watch the user＇s behaviors. Besides, the BP neural network was applied to learn the user＇s interests in this system. A conclusion is drawn that this system is able to obtain the user＇s interests precisely without the user＇s repeatedly feedback, so as to reduce the labor of the user.