Features extraction for illicit web pages identifications using independent component analysis

Abstract

The illicit web content such as pornography, violence, gambling, etc. have greatly poUuted the mind of immature web users. Pornography perhaps is one of the biggest threats related to current children's and teenagers' healthy mental life. A proper way to identify iUicit web pages efficiently is highly desired. In this paper, we analyze the textual content of web pages such as pornography, gynecology, sex education and general business news using independent component analysis (ICA) algorithm. We establish three similar models which are principal component analysis (PCA) model, ICA model and PCA-ICA model as comparison. We evaluate the effectiveness of these proposed models using information retrieval measurement such as precision, recall, Fl and accuracy. Our experiment result shown that PCA and PCA-ICA models are capable to identify iUicit web pages correctly with overall performance above than 90%. The idea of this research would give researchers an insight into textual content-based for web pages categorization.Keywords -artificial neural network, independent component analysis, illicit web pages identification, principal component analysis, textual content analysis.