Splog Filtering Based on Writing Consistency
Wei Liu, Songbo Tan, Hongbo Xu, Lihong Wang · 2008
Splog is the key challenge in the access of blogosphere. Existing splog-filtering methods are restricted to the way for traditional web spam filtering, without considering the characteristics of blogs. Inspired by the observation that fake writers (writers of splogs) have striking higher consistent writing behavior than real writers (writers of legitimate blogs), we propose to detect splogs by distinguishing fake writers from real writers. To measure how consistent the writing behavior is, we propose the consistency-based features derived from writing interval, writing structure and writing topic. Then we designed a splog-filtering system which can use the consistency-based features effectively and flexibly. The experimental results on Blog06 data set show that, proposed measure can effectively detect splogs, reaching an accuracy of 90%. Compared with content-based methods, our approach can get a comparable accuracy with fewer features and smaller train set, indicating that writing consistency represents the essential difference between splogs and blogs.