Research On Collaborative Filtering Recommendation Algorithm Based On User Similarity

The rapid development of the Internet bring human into the information age. With vast amounts of information, internet users find it is difficult to quickly locate specific information they need, as a result of which a phenomenon of "information overload" appeared. However, with the appearance of personalized recommendation technology, users can be freed from the vast information retrieval. This new technology has become an important way to obtain information besides searching engine, which does not require users to specifically describe the need of personal information, performing extremely well when users do not have their particularly searching need. Personalized recommendation technology, an important way in dealing with information overload, has been widely used in the field of e-commerce and social networking.Collaborative filtering recommendation has been the most successful and most widely used technology in personalized recommendation because of its efficiency, simple algorithm and the capability in handling complex objects. It recommends items through obtaining users’historical data as well as analyzing and forecasting the interest of potential users. However, in actual use, collaborative filtering algorithm presents both problems of data sparsity and inaccurate recommendation. The following studies of this paper are to solve the problems.First, for the inaccurate recommendation, this paper present a solution namely user similarity algorithm which is based on weighted information entropy and users’rating feature. Calculating the similarity is the most crucial step in collaborative filtering. Traditional similarity algorithms are faced with an issue of data sparsity, therefore, the outcomes are always not conforming to reality when the similarity of users is calculated from sparse date. The similarity algorithm presented in this paper is much more realistic according to the result of experiment, it lowers MAE value in recommendation system and improves the recommendation accuracy through processing of user’s original ratings, and simplifying the algorithm, the similarity measured by the information entropy of users’ different rating values, and the comparison with traditional method through experiment.Second, for the data sparsity, this paper present a collaborative filtering algorithm based on users’characteristic and the feature of project. The traditional similarity algorithm generates its recommendation simply relied on a single user’s rating score, which is difficult to make accurate recommendations even fail to make recommendations when data is very sparse. To solve this problem, this paper present a new algorithm which is proved to be more capable than traditional method to generate relatively high recommendation accuracy when the traditional method fails to do reasonable recommendation through a similarity algorithm which is weighted on traditional similarity based on the attributes of project, fusing users’ activities when doing recommendation, and the experiments of comparing traditional methods.In this paper, data collection from MovieLens is used for the experimental test data. Experiments show that the proposed method can reduce the mean absolute error of prediction, it can improve the accuracy of recommendation...