Research Of Hybrid User Behavior Collaborative Filtering Recommendation Algorithm

In recent years,with the rapid development of mobile Internet and the popularity of smart devices,network application service become more and more rich.Followed by the outbreak of the amount of data,"Information big bang" era has arrived.People find it increasingly difficult to find what they are interested.In order to solve this problem,personalized recommendation system has been studied and applied in various fields.But the most commonly used recommendation systems only use the rating data,when the rating matrix is sparse,the amount of items two users rated is too little,the calculation of the user similarity become less reliable,prediction results are often unsatisfactory.An idea of get user's behavior information from the user's history rating data,as an auxiliary information to measure the user's similarity is proposed.And then correct errors in similarity calculation and improve the recommendation results.Particularly,define user comment index and user rating time distribution index to measure the user's behavior,and then calculate the rating behavior similarity between users.Put forward several hybrid models,integrate the user's behavior similarity and rating similarity to improve the results of recommendation system.For the proposed model,the data are from douban.com,and the comparison between the models is conducted on the prediction and the TopN recommendation problem.The results show that when the number of objects that two user evaluates is too small,the rating similarity can be replaced by the user's behavior similarity,making the calculation of similarity between users more accurate.Both the prediction and TopN recommendation results are better than traditional user-based collaborative filtering algorithm.