The Research Of Collaborative Filtering Recommendation Algorithm Based On Complex Network

With the growing trends of social informatization and the rapid development of Internet technology,the amount of information from the Internet increases exponentially,which,meanwhile,results in severe information overload.Under this background,the way of extracting effective information from the mass becomes an urgent issue.Recommendation technology,as one of the most effective way of resolving the issue of information overload,is the hotspot of recent researches.Collaborative filtering technology has already been applied successfully and broadly in the sphere of personalized recommendation.However,there still exists several problems in the traditional collaborative filtering technology,such as sparsity,cold start,expandability and real-time recommendation.These problems reduce the accuracy and effectiveness of collaborative filtering recommendation,which also obstruct the development of recommendation technology.This paper attempts to improve accuracy of recommendation under the perspective of theory on group detecting of complex network and through the combining of group partition techonology and collaborative filtering recommendation technology.The major content of this paper is as follows:Collaborative filtering recommendation algorithm based on complex network.The users will be seen as nodes in the network and the complex network will be built on users'behaviors.The user network will be divided according to GN group partitioning algorithm and the user group forms.These user groups are neighborhood groups of collaborative filtering.Then similarity algorithm which,proposed by this paper,synthesizes user attributes distances and relationship strength,will be used to compute the similarity among users.This algorithm considers both user behavior preferences and user social attributes,so as to be more socially significant.The computation of similarity among users,combined with the nearest neighborhood group of target users by Top-K model,will be used as scores of recommendation.The research of collaborative filtering algorithm based on multilayer complex network.The user attribute and user behavior will be treated as two aspects of complex network to build a multilayer complex network.The user attribute hierarchy will be built on attribute influence and by attribute hierarchy will users be divided into different groups.Then behavior network will be built on behavior relationship among nodes in the network.The main way is to build network between two users who comment on the same movie,and the group partitioning algorithm will divide network into different groups.Then two partitioning methods will be used to form the preliminary neighborhood collection.Finally,Top-K model will be used to select the final collection in each neighborhood group,so as to get the final recommendation result.The result of experiment based on the dataset of Movielens,demonstrates that collaborative filtering recommendation in groups after multi-level partitioning on network nodes will improve the accuracy of recommendation.