The Research On Recommend Systems Based On Link Prediction In Social Networks

Over the past decade,with the rapid development of the social networks,research on the complex networks which constitutes the users and the relationship between the users has become new problem.Although the recommendation system can alleviate the social network information overload and information messiness to some degree,however,social networks recommendation system are different from traditional recommendation system.Traditional recommendation systems only focus on data instance itself,but the social network recommendation system also focus on link data at the same time.Social networks is a dynamic networks,as time goes on,there have been not only new users and links to join,but also have the old entity and links disappear,which determine the research in the social networks have big difference with common static networks.In that case,link prediction emerges at a historic moment.Link prediction in the social network is with known network nodes,the network structure and other related information,predicting the likelihood of the existence of links between two nodes.Obviously,as long as the prediction accuracy is high enough,it will inevitably enhance customer satisfaction and loyalty to social networking sites.In view of the complex situation of social networks,in this paper,establishing the social networks recommend system model based on link prediction,researching on some of the factors affecting the relationship between users,proposing combine algorithm based on link prediction.First of all,summary of the basic nature of the networks,which contains the average path length,clustering coefficient and degree distribution theory and so on,based on graph theory,introduces link prediction in social networks,building social network recommend system model based on link prediction.Introducing the common link prediction method based on node proximity,based on the relief analysis and based on probability in details.Secondly,through comparing with vertex degree,priority link characteristics,common neighbor,Adamic-Adar indicators and other indicators based on the structure of networks topology,choosing the proximity method based on common neighbor as networks topology proximity baseline method.In addition,each node properties of microblogging users are classified into background information,social information,micro blog,interactive information four categories,through the analysis of the four node properties of selected representative indicators.Then on this basis,defining four proximity algorithms based on node attributes: fans,attention,microblogging,retweet.Lastly,this paper proposes new algorithm---combine algorithm,which combined with the feature of user attributes and network topology structure of social network link prediction method.Based on networks structure characteristics(CN)and based on the proximity of microblogging users attribute(fans,attention,microblogging,retweet)in the simplest way of linear combination,proposed four kinds of combine algorithms.Experimental data using the crawl au data from sina weibo contain the user information and weibo two classes of 20 multiple attributes.The experimental results compared with the traditional link prediction algorithm,link prediction algorithm based combine algorithm is applied to social network recommendation system,covers the effective features of social networks,improves the accuracy of link prediction,superior to the effect of the traditional methods,which illustrate the proposed method is reasonable and effective.