A Collaborative Filtering Recommendation System Combining User Context Information And Rating Preference

With the rapid development of cloud computing,big data and other technologies,”Internet +” will be further integrated into people's daily life and make a profound change in Life style,as a result,people are increasingly dependent on the internet.In the era of big data,people are faced with the problem of information overload,collaborative filtering recommendation technology is a powerful tool to solve this problem,but also to meet the user's personalized information needs.Collaborative filtering recommendation technology obtain the user's potential or future preferences by analyzing of all the user's historical behavior data,and then according to the user's preferences,filtering out from the mass of information.There is no need to analyze the content of information resources,and it can find the potential interest of users,be easily implemented,so has been widely studied and applied.However,in practical application,the information is endless,but users usually only focus on a small number of projects,eventually result in data sparsity becoming more and more serious,which seriously hindered the development of collaborative filtering technology and application.In this thesis,a similarity measurement method based on user context information is proposed,which considers the influence of user context information to user's similarity,and can reflect the correlation among users more accurately,so that the selection of nearest neighbor is more accurate.Secondly,a similarity measure method based on the degree of tendency of scoring is proposed.The algorithm introduces the concept of the degree of the score,and takes into account the influence of user's score,the user's common concern and the difference of the user's similarity.Then,the problem of inaccurate score prediction for data sparsity,a dynamic scoring method is proposed.The new scoring method takes into account the effect of user's nearest neighbor and the nearest neighbor of the project.Finally,the paper proposes a Collaborative filtering recommendation system combining user context information and rating preference.In this thesis,Using the Movie Lens-1M data-set provided by the University of Minnesota Grouplens research group,the data set is randomly divided into training data set and test data set according to 80% and 20%.The effect of the proposed algorithm is evaluated by using the average absolute error and the accuracy of recommendation.A total of four sets of experiments were designed to verify the performance of the proposed algorithm from four aspects: the similarity based on user context information,the similarity based on user's score,the dynamic score prediction and the effectiveness of CPCF algorithm.The results show that the proposed algorithm can effectively alleviate the data sparsity problem and improve the accuracy of prediction and recommendation accuracy.