Research And Implementation Of Personalized Followee Recommendation In Microblog Based On Social Networks

In the era of "big data",technological innovation on Internet has brought the growing popularity of social networks.Microblog is one of the major social platforms for people to share information and communicate with others.Everyday there are countless hot spots in microblog and it has attracted millions of users.People can share their mood and can also choose their interested items to follow.People can retweet other users' tweets and make comments to them and can use"@" to mention others.These social interactions enrich people's experiences and make microblog more diversified.But it's also an era of information overload,so it is very difficult for users to find the item that interests them facing such massive information of items.Recommender system is the bridge between users and items.It can provide active recommendations for users according to users' preferences and it has been applied to many fields.Collaborative filtering algorithm is regarded as the most popular and widely used algorithm in personalized recommendation systems,but it relies heavily on user's historical preferences.In microblog,there are no direct ratings from users to items,so we can't just apply collaborative filtering to the problem of followee recommendation directly.Social behaviors,social trust,the influence of neighborhood and latent factor model can greatly improve the performance of recommendation.The thesis firstly studies on the development of recommender system and personalized followee recommendation,gives an introduction to collaboratiove filtering algorithm.Then we compare the differences of the two kind of algorithms and introduce the challenges that faced.Through the Tencent Weibo dataset,we analyze the characteristics,social maps as well as user relations of social networks.Aiming at the problem of followee recommendation in microblog,we redefine some terms and build model on different social behaviors.Besides,the system's overall process,technology platform,system environment is also introduced.Our main contributions are mainly as follows:1.We propose a social similarity-based Top-N recommendation algorithm.We calculate similarities based on follow behaviors,interaction behaviors and historical recommendation records to find the nearest neighbors.According to the neighbor's interests,we make recommendations for users.We compare their precision,recall,and F1-measure using the real word datasets.The algorithm is paralleled on the Hadoop platform using MapReduce,which improves the efficiency of the algorithm.2.We propose a followee recommendation algorithm fusing social trust and latent factor model.First,implicit trust is calculated based on user interaction behaviors while explicit trust is based on the direct connections between users;Then an extended trust matrix is constructed combining both implicit trust and explicit trust.Finally,we utilize both the extended trust and ratings and apply matrix factorization techniques to build the model.Experiments on KDD Cup 2012 dataset demonstrate that our approach achieves better performance in terms of RMSE and MSE.