Earch On Influence In Microblog

With the development of Web2.0technology, various social media have sprung up in recent years. As a prevailing one, Microblog has been an important way in the daily mutual communications and information spread among people. Due to some unique characteristics Microblog owns, a growing number of researches have paid attention to it and among these researches, influence is a hot topic, including influence measurement and influence maximization. Even though some achievements have been obtained, there are still some problems. For example, among all the proposed algorithms to approximate the influence maximization, some are time-consuming, some inaccurate and some including too many assumptions. Regarding the influence measurement, there are indeed a certain number of novel perspectives deserving exploration. Therefore, this paper will focus on the influence in Microblog from both the two topics.According to the bidirectionality of user interactions, this paper first estimates the interactive degree between two users from retweeting strength, commenting intensity, mentioning density and keyword similarity these four factors. Then, considering that a follower’s contributions to the influences of his/her followees vary and depend greatly on the interactions between them, a novel influence measurement algorithm MBUserRank is put forward, which is the basis of the influence maximization. Further, to better approximate the activation probability and reduce the complexity, a modified algorithm MBRank based on MBUserRank is given. With the ranking results of MBRank on all the users in the network, some top ranked users are selected to constitute a candidate seed set. As this new heuristic is introduced, a novel influence maximization algorithm MBGreedy and its improvement MBCELF are proposed. Finally, a common problem in daily life whether the top-k ranked users always lead to the influence maximization is discussed and tuning for the candidate seed size is illustrated.Through the extensive experiments on Tencent Weibo dataset, the following conclusions are made. First, the new proposed influence measurement algorithm MBUserRank is really competent in giving a better ranking result closer to reality. Second, combining both the advantages of greedy-based and heuristic-based algorithms, MBCELF can obtain a good approximate to the optimal solution with a nice speed. Third, the top-k ranked users could not always result in the influence maximization.