Research On Microblog User Influence Evaluation Algorithm Based On Multi-criteria Decision Making

Weibo has the characteristics of high transmission speed, extensive source, and multi-angle, as a form of social media. It has become the main channel of information exchange and information sharing in people's daily lives, and has attracted wide attention of scholars both at home and abroad. It has great significance in user recommendation, information diffusion, public opinion monitoring, directed marketing and so on.First, based on the analysis of Sina Weibo's information diffusion characteristics, we construct a Weibo network model, and the Weibo network is divided into two networks, which are user relation network and the blog diffusion network. Second, to effectively avoid the influence of “zombie fans” and prevent users from forwarding or commenting their own blogs to improve influence, this paper extracts four indicators for evaluating user influence, which are the LeaderRank influence, the average retweeted number, the average commented number, and the average like number, from perspectives of following relationship between users and blog diffusion. Based on the four indicators for evaluating user influence, we use the Skyline method in the multi-criteria decision making. Then we construct a user influence evaluation model, and propose an influence evaluation algorithm named as WeiboLeaderRank.To investigate the effectiveness of the algorithm, we design and implement the Sina Weibo data collection system by using web crawler technique. Because the web server will redirect access request and prohibit user to access when it detects abnormal access request, which will affect the speed of the data collection, we use multi-account, simulation logging and multi-thread collection. The thread requests data using anonymous proxy server, and changes dynamically to request HTTP header information. What's more, we add an abnormally detection module to detect the abnormal situation in time and take the corresponding actions. We try to imitate the normal user access behavior to improve the efficiency of the data collection.WeiboLeaderRand is tested on the collected large scale datasets. The experimental results show that WeiboLeaderRand performs better than other representative user influence evaluation algorithms. The run time increases linearly with any increase in data size. Weibo LeaderRand is scalable to large scale true Weibo situation, and has a good real-time performance.