Research On Asymmetric Collaborative Filtering Algorithms Based On Similar Rating Items And Users' Common Preference

The calculation of similarity between users is the key point in the process of neighbor-based collaborative filtering algorithms(CF),which determines the neighbors of the target user,affects the ratings of items and ultimately affects the list of recommendations that users can see.The traditional recommender algorithms only consider the common rating items between users and try to capture the similarity between users with the distance between ratings or the linear relationship between ratings of two users.These algorithms cannot fully utilize the information in ratings.In the case of new user cold start in consumer recommender systems,due to the lack of common rating items,it is hard for these methods to capture users' preference.This paper try to expand the basic definition of the common rating items in traditional CF,extending common rating items to similar rating items,aiming at alleviating the problem of the lack of common rating items in the case of cold start.By combing the similar rating items with the common users' preference from the ratings,an asymmetric similar rating items and common users' preference based collaborative filtering algorithm is proposed in this paper.The proposed method in this paper can alleviate new user cold start problem for new users with a few rating information in consumer recommender systems.The main work of this paper is as follows:(1)Introduce the research significance of the recommendation algorithms and analyze the importance and the current research status of collaborative filtering and the similarity methods.(2)Introduce mainstream recommendation algorithms and analyze the advantages and disadvantages of these methods.(3)Analyze the reasons for the cold start problems in recommendations and the common solutions for this problem.Introduce the influence of asymmetry between users,and explain the necessity of the asymmetric factors.(4)Analyze the similarity methods in traditional collaborative filtering algorithms,and introduced merits and demerits of these method.An asymmetric user common preference based collaborative filtering algorithm is proposed.(5)Explore the possible reasons for the problems according to the analysis of the similarity methods in CF,and analyze the reason for the poor effect of these method in the case of cold start.Based on this,the definition of the common rating items is expanded to similar rating items.Combined with the user' common preference based asymmetric CF,the similar rating items and users' common preference based recommendation algorithm is proposed.(6)Perform the experiments using MovieLens-100 K dataset.Compared with another eight algorithms,the experimental results show that the proposed method can achieve better recommendation results in non-cold start conditions.In the cases of different degrees of cold start,the proposed method could make full use of the users' limited rating information,and capture the users' preferences,resulting reasonable recommendations for users.The performance of the proposed method works very well.