Dynamic Model Adaptive to User Interest Drift Based on Cluster and Nearest Neighbors

A recommendation system provides personalized recommendations on products and services to users. In the traditional recommendation system, the user interest is regarded as constant over time, while in fact, the user interest changes over time. Hence, tracking the user interest drift becomes key in designing the dynamic recommendation system. However, it is a challenge to find an accurate and effective method that can predict the user interest drift. To solve the prediction problem of the user interest drift, this paper adopts clustering and time impact factor matrix to monitor the degree of user interest drift in the class and more accurately predict an item's rating. We add a time impact factor to the original baseline estimates and use the linear regression to predict the user interest drift. Our comparative experiments are conducted on three big data sets: MovieLens100K, MovieLens1M, and MovieLens10M. The experimental results show that our proposed approach can efficiently improve the prediction accuracy.