Web sensitive text filtering by combining semantics and statistics

Web sensitive information is defined as texts, pictures and other forms of information which contain erotic content on Web. How to filter this harmful information attracts researchers´ interests. In order to keep Web content safe, governments have also given great support on the research on this problem. This paper first briefly review recent developments in Web sensitive information filtering then the statistic and semantic features of sensitive texts are analyzed and represented by a CNN-like word net. Finally a novel method which combines semantics and statistics is proposed to filter sensitive text on Web. Experimental results have demonstrated the proposed method´s promising performance.