Research And Application Of Collaborative Filtering Recommendation Algorithm For User Interest Change

With the rapid development of the Internet,we have entered the era of "big data",and the huge amount of data generated every day also makes people face the serious problem of "information overload".The recommendation system can make personalized recommendations for users,bringing convenience and experience to users.Therefore,the research on recommender systems is very meaningful and valuable.As one of the mainstream recommendation algorithms,the collaborative filtering algorithm can calculate the similarity of the behavior information interacting between users and items to find out the relationship between users or items and then make recommendations.However,traditional similarity measurement formulas such as cosine similarity measurement can only be calculated by combining the common scores of users or items,and cannot fully use all the scoring information.In the real world,the dataset is very sparse,and the traditional similarity measurement formula will lead to inaccurate similarity calculation,thus affecting the accuracy of the recommendation system.At the same time,the traditional collaborative filtering recommendation algorithm does not consider the interests of users when making recommendations,and cannot make personalized recommendations for users’ interests,which ultimately affects the recommendation effect.Aiming at the problem of inaccurate similarity calculation in sparse datasets and the problem that traditional collaborative filtering algorithms do not consider user interests,this paper proposes a collaborative filtering recommendation algorithm oriented to changes in user interests.The main innovations of this paper include the following parts:(1)A hybrid similarity measurement algorithm combined with the temporal relationship of item ratings is proposed.This algorithm takes into account the temporal trend of item ratings and fully explores the time information of items rated by users.At the same time,the time-series relationship of item ratings is combined with common user ratings,which can make full use of the rating information of all users compared to traditional similarity measurement algorithms.(2)This paper proposes a collaborative filtering recommendation algorithm UICF for user interest changes,which considers the impact on user interests from two aspects: time context and item attribute characteristics.Combined with TF-IDF theory,a user interest model is constructed,and Ebbinghaus theory is introduced to simulate the decay trend of user interest,and a long-term and short-term interest decay model is proposed.(3)Based on the UICF algorithm proposed in this paper,we develop a movie recommendation platform that can make personalized recommendations for users’ interests.The platform is developed and implemented from the aspects of demand analysis,system design,database design,etc.At the same time,it integrates the UICF algorithm proposed in this paper,and can recommend movies individually according to the interests of users.Through experimental analysis on the Movielens-100 k dataset,compared with other similarity measurement algorithms,the hybrid similarity measurement algorithm proposed in this paper can more accurately calculate the similarity between items.Then,the influence of user interest on recommendation combination is considered in the proposed hybrid similarity measure algorithm.In the final recommendation,the user’s interest changes are fully considered.Compared with other recommendation algorithms,the UICF algorithm proposed in this paper can make personalized recommendations based on user interests.