Research On User Behavior Based Collaborative Filtering Algorithms

The increasing information on the Internet offers a lot of convenience for users.At the same time,it quietly brings the problems of "information overloading" and "information disorientation".In order to efficiently obtain information resources and effectively solve these problems,personalized information recommendation system has been proposed,and has become one of the most promising methods in information technology.As one of the most common recommendation algorithms,collaborative filtering(CF)has become a research hotspot and has achieved success in Web 2.0 applications.However,problems exist when CF is used in the real application,such as data sparsity,system scalability and recommendation diversity,etc.In the thesis,to deal with those problems,three kinds of improved algorithms are proposed to carry out the related research work.The main work is as follows:Traditional algorithms using user ratings as input exist serious data sparsity.Aiming at the problem,user behavior records are selected as the input data,behavior similarity is used to measure the similarity relations among the users or items,and a behavior similarity based collaborative filtering algorithm is proposed.In order to calculate behavior similarity in a better way,behavior sequence based method and graph based method are designed.The results of the simulation experiments prove that proposed algorithm can alleviate the negative effect of sparsity problem,and improve the recommendation quality.In some real applications,users do not always wish to have system recommending a similar type of items,but related items in other kinds.A behavior influence based collaborative filtering algorithm is proposed.Mining the influence among user behaviors,and using the influence can effectively predict the further behaviors.The feasibility of proposed algorithm is proved by experiments.Providing different and personalized recommendation by the system has a positive and important impact on improving user experience.Therefore,user context is introduced into the first two algorithms in order to improve the temporal diversity of recommendations,and a users' context based collaborative filtering algorithm is proposed.The results of the simulation experiments show that the proposed algorithm's recommendation list is more diversiform,and the accuracy is acceptable.