Identifying and Blocking Pornographic Content

Content filtering on the WWW is significant social issue, since children can easily access pornography and other illegal material with ease. In this study, a descriptive and predictive model of pornographic web page characteristics was developed, to assist with recognition of pornographic pages in a content filter. Support and confidence results from simple association rules suggest that using individual terms to identify pornographic web pages is useful for description, but unreliable for prediction. However, Bayesian classification provided 99.1% accuracy in classifying test pages from both pornographic and non-pornographic corpora. The challenges for multimodal filtering are discussed.