Research On Recommender System Based On User Contribution Weight

In order to satisfy the information needs of users in the big data era,the personalized recommender system has been widely used.Recommender System is an important tool for users to discovery information of their interest,it is also an important tool to overcome information overload problem.The main idea of recommender system is to make recommendation by analyzing users' historical behaviors.Collaborative filtering is the most successful and widely used method in personalized recommendation service since it is simple and effective.The key point is to find similar users or items through the user-item rating matrix.However,most traditional similarity methods only compute the similarity based on the users' co-rated scores.In addition,they are not very suitable in a sparse data environment.To solve this problem,this thesis proposes two new methods.We can obtain the symmetrical user contribution weight by the traditional similarity.In fact,the user's recommendation contribution is different.The first measure based on traditional similarity considers the above all factors that neighborhood users have rated and that not rated by the target user.It also considers the impact of the proportion of co-ratings.Because the number of co-rated items is a lot,it means that the two users are similar.The second similarity method is based on Bhattacharyya coefficient,which uses all users' rating information for items,not only can obtain similar interest feature of users through the user's rating behavior,but also can obtain the correlation between the items that the users have rated.Meanwhile,the new method also takes into account each user's rating preference,since different users have different rating habits.Considering more relevant factor about user similarity can select more appropriate neighborhood for the target users,and lead to efficiently improve the recommendations.With experiments on two real data sets,the results show that our proposed method outperforms the other state-of-the-art similarity metrics.