Research On Followee Recommendation Model Over Microblogging Systems

With the emergence of Web2.0, microbloggings systems, such as Twitter and SinaWeibo, have attracted more and more users’ attention. Different from traditional onlinesocial network, in the community of microblogging system, users are following orfollowed by others, which forms a relationship. As a platform for information sharing andpropagation, the most fundamental and unique service provided by microblogging systemis that it can be used as a tool for users to post short messages (called as tweets) whateverthey want. In every user’s homepage, the recent tweets from all his/her followees are listedchronologically. As to followee recommendation in microblogging systems, it is thusforemost to recommend most relevant followees to users so that they can benefit mostfrom the tweets information gathered from their followees.However, since the extremely large population in the community of microbloggingsystems, it is a challenge to make precious followee recommendation for a user.According to the unique feature that microblogging system is a combination of newsmedia and social network, we design a model called NDCG-LFM based on latent factormodel. The method takes both users’ content information and their social relationshipinformation into consideration, while utilizing NDCG as the objective function tooptimize model parameters. Moreover, since the objective function of traditional latentfactor model is not target at getting good top-k recommendation results, we design a newcriterion as the objective function. Considering the discontinuity of the new objectivefunction, we further propose a smoothed version of the objective function, thus thefollowee recommended model can provide high-quality of top-k recommendation results.Based on the online and offline experiment results, NDCG-LFM, mentioned in thiswork greatly outperforms other state-of-the-art followee recommendation schemes.Compared with the most successful recommender system called Twittermender, ouralgorithm makes prominent improvement with46.8%in precision and32.8%in NDCGrespectively.