Collaborative Filtering Recommender System Based On Social Network

With the rapid development of Internet and computer,the personalized recommendation technology has been widely used in many areas such as movies,e-commerce,music and so on due to the growth of massive information.Collaborative filtering recommender algorithm is the most popular in personalized recommendation technology,which can be used to search for similar users or objects based on the user's historical behavior data,and then make a scoring prediction and recommendation.Although collaborative filtering recommender algorithm has been successfully applied to many areas,there are still problems such as cold start-up of new users,data problems about data sparseness and the change of user interests.At the same time,social network has grown so fast that social network has spread across the Internet in just a few decades.The effective use of the information in social network plays an important role on the improvement of final recommendation accuracy.In terms of these problems,the thesis focuses on the studies of the new user cold start and the data sparseness problem.The improved K-prototypes algorithm and collaborative filtering recommendation algorithm based on social network are proposed.Finally,a collaborative filtering recommendation system based on social network is established.The main research contents and achievements of this thesis are summarized as follows.Firstly,an improved K-prototypes clustering algorithm is proposed.Aiming at the problem of new user cold start in collaborative filtering recommendation algorithm,an improved K-prototypes clustering algorithm based on different attribute weights is proposed in the algorithm.Users with similar user attributes are clustered into one class,the users who are closest to the target users in the same category are recommended to the target users,and the TOP-N items in these users are recommended to the target users.The improved K-prototypes algorithm is applied to the GroupLens data set,which is improved by 5%compared with the traditional K-prototypes algorithm recommendation.Secondly,a collaborative filtering recommendation algorithm based on social network is proposed.In social network,people with similar interests tend to interact with each other,and interests can also be spread on social networks.Therefore,the concept of interactive similarity is put forward in this thesis,which is used to describe the relationship between the social network users.The interactive similarity not only makes full use of the social network relationship among users to improve the accuracy of recommendation,and also improves the data sparseness problem.The proposed collaborative filtering recommendation algorithm based on social network is applied to Epinions data set,and the MAE(mean absolute error)is reduced by 6%compared with the traditional collaborative filtering recommendation algorithm based on social network.Thirdly,a collaborative filtering recommendation system based on social network is proposed.In view of the proposed method,a complete collaborative filtering recommendation system based on social network is constructed.In this system,the improved K-prototypes clustering algorithm is recommended for new users and recommend using collaborative filtering recommendations algorithm based on social networks for older users.Compared with the traditional recommendation method,the proposed accuracy can be improved effectively.