The PSI3 Agent Recommender System

This paper presents a multi-agent system (MAS) that implements a recommender system, essentially using collaborative filtering techniques. The design of the MAS is flexible to support the implementation of different filtering strategies and to control the global behavior of the system and its users. It has been applied in the PSI3 project to implement a personalized information dissemination service. Personalization requires some feedback from users in order to obtain some rating of their satisfaction with the information they receive. But, in order to make this effective, it has to be done with minimum disturbance to them. One way to achieve such purpose is by associating a personal agent for each user (to build and manage the user's profile) and a community agent for each group of users (to manage the dissemination and evaluation of information in the group, improve system performance, and prevent malicious behavior).