集中式协作过滤算法中 ,服务器的负荷过大且成为瓶颈环节 ,该文研究了分布式算法。Agent只有局部视角 ,算法以有限朋友列表和信任度为基础。文中将信任度的控制规则分为比例、积分、微分规则。通过仿真实验研究了比例、积分规则对系统全局性能的影响。实验结果表明 ,各个Agent通过自适应学习 ,逐渐与自己的朋友形成了恰当的信任联系 ,该连接强度反映了合作的密切程度。同时 ,所有这些连接强度构成的加权连接图 ,反映了多 Agent的宏观聚类效果和自组织现象 In the centralized collaborative filtering algorithm, the server is overloaded and becomes a bottleneck. This paper studies distributed algorithms. Agent only partial perspective, the algorithm based on a limited list of friends and trust. The article will be divided into control rules of trust, proportion, integral, differential rules. Through the simulation experiment, the influence of ratio and integral rules on the global performance of the system is studied. Experimental results show that each agent gradually forms an appropriate trust relationship with his friends through adaptive learning, and the strength of the connection reflects the close cooperation. At the same time, the weighted connection diagrams formed by all these connection intensities reflect the macro-clustering effect and self-organization phenomenon of multi-agent