Research Of Real-time Recommendation System Based On Dynamic Clustering Algorithm And Attribute Information

Collaborative filtering is an effective method to help users find their interested items or services in e-commerce,such as Tmall,Amazon.The development of recommendation algorithms has been focused on the improvement of recommendation quality.For one thing,users could get favorite items quickly;and for another thing,it can promote the income of merchants.It showed that,many factors could lead to poor recommendation,such as overload information,outdated information,spare data and so on.In order to improve the accuracy of recommendation,two algorithms are proposed in this paper.They are Personal Recommender System based on User Interest Community in Social Network model and Dynamic clustering collaborative filtering recommendation algorithm based on multi-layer network,respectively.(1)Personal Recommender System based on User Interest Community in Social Network model.Users' interests will gradually change over the time.Some interests which are pretty popular in today's life may disappear in the following years.As a result,compared with new information,outdated information plays a weaker role in recommendation than the recent score information.In order to make full use of this outdated information,an efficient time weighted collaborative filtering algorithm is proposed in this paper.In our presented recommendation algorithm,interest changing over time is fully mined.Firstly,combining with rounding-forgetting function,a new type of time weighted score matrix is constructed.The newfound matrix could reflect many users' interests.Then,the users and items with higher correlation are clustered into the same community according to difference equations.Stable same state values mean the users who have similar interests,so they are assigned into the same community.Finally,the real-time prediction results are obtained by dynamic similarity measurement.Effectiveness of our proposed algorithm is shown by providing extensive experimental evaluations which are based on several real datasets with different scales.Diverse comparing results with several related better methods are given to prove the efficiency of our algorithm.(2)Dynamic clustering collaborative filtering recommendation algorithm based on multi-layer networks.With the rapid development of Internet economy,an increasing number of people are devoting themselves in recommendation algorithms.Traditional collaborative filtering algorithms only depended on rating information or attribution information.However,most of them were considered by a single-layer network perspective,which destroyed the original hierarchy of data and resulted in sparse matrix and poor timeliness.In order to solve these problems and improve the accuracy of recommendation further,Dynamic clustering collaborative filtering recommendation algorithm based on multi-layer networks is put forward in the paper.Firstly,user attribution information and item attribution information are respectively used to construct user network and item network.Secondly,new hierarchical clustering method is come up with,which separates users into different community according to difference equations.Finally,score prediction and Top-N recommendation are obtained by similarity in each community.A lot of experiments are executed in several real datasets;and the effectiveness of our algorithm is verified.