Research Of Improved Recommendation Algorithm Based On Time Effect And Changes In Users Interest

As the primary means to solve the information overload problems, recommendation systems usually predict the future behavior of the user by mining the behaviors of users, its essence is to contact users and items, recommend the items to whom may be interested in. Traditional recommendation algorithm usually works with getting user behaviors, establishing recommendation model, generating recommendation results. In the course of the recommendation, traditional methods less consider the impact effect of time effect and changes in users interest, which may leading to inaccurate and unreasonable results.To solve the problem above, based on the traditional collaborative filtering algorithm, a improved collaborative filtering algorithm with forgotten laws is proposed in this paper. The improved algorithm introduces time effect function and the concept of timeliness of collection to pre-filtering the data. Finally, by comparing the traditional collaborative filtering algorithm and the improved collaborative filtering algorithm based on time effect with experiments, we validate the conclusion that recommended accuracy has improved by using improved algorithm. In addition, in this paper, a improved random walk algorithm with time effect graph model is proposed. The improved algorithm lead into time effect factor and users interest changes, At last, it generates recommended results for target users by preference prediction. Comparative experimental results show that the improved algorithm get a more accurate recommendation accuracy after adding time effect. Finally, the traditional "user-item" model was improved in this paper, a new "item-user|period" model was proposed to adapt to changes in users interest and time effect. Then a path fusion algorithm was proposed base on this improved model to calculate the target user’s preferences for unknown items. Experiments show the feasibility of the improved algorithm and recommended accuracy has improved compared to the traditional collaborative filtering algorithm.