The Research Of Collaborative Filtering Recommendation Based On User Preferences

Network resources are explosive growth,contributed to the problem of information overload.Recommendation technology has become one of the most effective ways to solve this problem.Personalized recommendation technology is widely used,especially collaborative filtering recommendation,is one of the most successful recommendation technology at present.It does this by analyzing the user's preferences that are highly similarity with the target user,and the project which neighbors like will be recommend to the target user.Although collaborative filtering has been a huge success,but there are some key problems such as data sparsity,scalability and groups recommendation problems that restrict its further development,so this paper focuses and in-depth study on these issues.The main research contents of this paper are as follows:1.Building a user preference matrix on items features by improved TF-IDF,thereby reducing the dimension of matrix.For the data sparsity problem of collaborative filtering algorithm,using project characteristics,at the same time,considering the influence of user interest drift of user preferences,to construct a user preference matrix on items features by improved TF-IDF,due to item features can show user's preferences from the aspects of content,and item features are far less than the items quantity,reducing dimension of user preference matrix on items features.2.For the scalability problem of collaborative filtering algorithm,an improved K-Means method will be put forward to cluster the user,the method by using the difference degrees between classes and in classes to determine the final K value to solve the problem which the K-Means method needs to artificially set the initial K value.And then computing the similarity between the target user and other users within the class,to get the nearest neighbor set,and to predict scores and recommendations,which improves the efficiency and scalability.3.In view of the traditional collaborative filtering algorithm is suitable for the personal recommendation,and does not apply to group recommendation.Therefore the recommendation method based on users' preferences model on items features and on ratings will be proposed to recommend for groups,according to the interaction among members of the group to build group score matrix and group preference matrix on items features.Based on the user's collaborative filtering algorithm,considering group as a virtual user,calculating the similarity of target group and other groups and predicting the score of the target group.4.Constructing a comprehensive similarity calculation method by using weighting factor method.The traditional similarity calculation is to use common score between the users,however similarity between users is not only with the user's rating information but also to the users' preference on items features.Both from different perspectives to reflect the user's preference.Therefore using weighting factors to combine the two,to calculate the comprehensive similarity.Finally,using the proposed method in this paper and several commonly used methods to experiment in the movielens data set,from three indicators which are MAE?Precision?Recall to verify the validity of the method this paper proposed,from a certain extent,to ease sparsity,scalability and groups recommended problem.