Probabilistic graphical model for detecting spammers in microblog websites
Zhongming Han, Ke Yang, Fengmin Xu, Dagao Duan · International Journal of Embedded Systems · 2015
Most recently, methods of identifying spammers in microblog websites exploit user attributes, such as their tweets, registration time and number of friends, which perform poorly on a real dataset. In this paper, we thoroughly analysed various user attributes and behaviours. Then we propose a probabilistic graphical model named spammers probabilistic graphical model (SPGM) in which user profile features are treated as the input variables while behaviour features act as the observed variables and the probability of a user being a spammer is a hidden variable of SPGM model. Collecting data from Sina Weibo and Twitter, we build two experimental datasets to evaluate our model. We define different user features for these two real-life datasets and compare the performance of SPGM model running on the experimental datasets. We also compare our SPGM model with baseline SVM model. Experimental results prove that SPGM model outperform SVM model on effectively detecting microblog spammers.