This paper analyzes the information retrieval techniques for relevance feedback, and the user interest feedback evaluation mechanisms are  studied. Using the agent technology, a relevance feedback model based on intelligent agents is given; and the user interest feedback evaluation mechanism of this model is experimented. Through the improved ID3 algorithms to learn the user interest and update the  user interest model,the personalized information retrieval is achieved. Experiments show that this feedback mechanism is an effective evaluation system,and the recall rate and the precision rate have obvious improvement to better meet the user's individual requirements.