Microblog Friend Recommendation Algorithm Based On Social Relationship And Temporal-topic

With the popularity of the Microblog social network, more and more people prefer to use Microblog platform to get information and express opinions. However, with the sharp increase number of users, Microblog's information is critical overload, it's very difficult for users to find the information they are interested in. In the usually, Message received by a user mainly depend on whom the use follows. Thus, recommending users with similar interests may improve the quality for information receiving and extend their own social circles. How to recommend high quality following friends is always a difficulty of Microblog personalized service.Currently, among all Microblog friend recommender system, mixed friend recommended algorithms can't comprehensive analysis various features of Microblog data and can't dynamic analysis user interest preferences. To solve these problems, a friend recommendation algorithm based on social relation and temporal-topic is proposed in the paper. In the recommender model based on social relation, we extracted the characteristics of user's age, gender and social activities from the Microblog data source, and then introduced into the latent matrix factorization model which is suitable for Microblog friend recommender, optimize the model and get the user intimacy matrix, finally computing user's social similarity. In the recommender model based on temporal-topic, according to the time window divide the user's microblogging content into many segments; In each period, aggregate the Microblog text base on every user into a user document, and use the Latent Dirichlet Allocation(LDA) model get the users – themes feature, then use Jensen-Shannon divergence compute interest similarity, finally for each period use time decay function predict the final user interest similarity. Then, combine the social similarity with the interest similarity and using collaborative filtering algorithm recommend Top-k friend to a target user.We had conducted an experiment on the real sina weibo data sets and verified that friend recommendation algorithm based on social relation and temporal-topic has better recommendation effect. In the mean average precision(MAP) is superior to the existing hybrid friend recommendation algorithm.