Research On Recommendation Based On Network User Behavior

The rapid development of Internet technology has created the problem of information overload.Information overloading leads to the recommendation system being born.The recommendation system alleviates the problem of information overload to a certain extent,but its accuracy and effectiveness has been the focus of attention of the industry.With the rapid development of social networks and electronic business platform in recent years,researchers switch their attention to use the user behavior in the network to make the recommendation system better.To summaries,under the background of traditional recommendation it is of great significance to study the new recommendation strategy to improve the overall quality of the recommender system by using the network user behavior.The main research contents and innovations of this thesis are as follows:Firstly,through researching the current situation of recommendation technology and analyzing deeply the characteristics of user-based collaborative filtering recommendation,the traditional user-based collaborative filtering recommendation solve the problem of the imprecise of similarity calculating using single user attribution after introducing into the user multidimensional social attributes in the social network.Meanwhile the similarity calculating want the weighted attribution and the weight is subjective.As a result social-oriented multidimensional cloud similarity measurement method is proposed,which takes into account that multidimensional cloud model can be used to calculate multidimensional space attribute,and solve the problem of subjectivity in single and multiple attribute weight.Secondly,in the present recommendation scheme,almost all the schemes will regard score as a measure of user preference parameters.And different users has different score at the same degree of preference because of personality and individual differences.In order to solve this problem,a unified rating standard under the integrated cloud model are proposed,that the negative effect of rating standard could be solved.Finally,the long tail distribution in the vast amounts of data will affect the accuracy of the recommendation system.In order to make up for the defects of recommended methods in this aspect,we punish the extremely active users after the full analysis of user behavior of the long tail distribution,which aimed at correcting recommended deviation caused by abnormal active users.In summary,this thesis research the algorithm and application of collaborative filtering recommendation based on user behavior and proposal a hybrid recommendation algorithm for collaborative filtering recommendation and topN recommendation integrating user behavior in network.which enrich the study of the theory of collaborative filtering recommendation based on network.