Research And Realization On Cycled Drift Interest Model

Face to the Information Explosion caused by the Internet, the users' attention appears to be more and more precious, and people are frustrated to the overflowing information. User interest model and its corresponding personalized recommendation service have recommended the customers' interested information to them. It eased the conflict between the limited attention and the huge information. Currently, user interest model as the basis of personalized recommendation service is becoming the research focus for many researchers study.The psychology proved that the user's interest tends to occur with a variety of volatile changes under the influence of internal and external environmental factors. Such changes of Interest are to be called User Interest Drift. How to accurately predict interest drift and updates the user interest model timely, how to modeling considering the current interest and original interest, how to make CDIM-based demonstration system to provide high quality recommendation services, these problems are to be researched in this paper.There are two steps as follows: the first is the proposition of cycled drift interest model-CDIM. The second is the designation of CDIM-based demonstration system.The focus of this paper is a new drift interest model- cycled drift interest model. Based on the research of users' cycle drift interest, it proposed several definitions and rules in the paper. Updating the user interest model and predicting interest drift by the definitions and rules.During the modeling, firstly interest is classified as long-term drift interest and short-term drift interest, secondly update the original model by the cycle definition and rules. There are different updating mechanisms to the different drift interest. Short-term interest uses the sliding window, and the long-term interest uses the adaptive optimization window and the concept-based progressive forget mechanism, and as a long-term interest in the window of time, the paper also proposes a new method - the natural unbalance of time Window.The paper proposed a CDIM-based demonstration system. Through the demonstration system, it proved the effective of CDIM on the side of user personalization recommendation quality. This paper designed the related demo system to CDIM applications. Through the demonstration system it obtained the experimental data between the different models. The experimental data proved that the CDIM has higher accuracythan the same two DIM (drift interest model).We validated the validity of our proposed model through experiments, and comparewith present drift interest model on recommendation quality. The results showed us that our model outperformed the existing drift model in recommendation precise rate and recall rate, which contribute the excellent performance of personalized recommendation system.