A Biased Support Vector Machine approach to web filtering

Abstract

Web filtering is an inductive process which automatically builds a filter by learning the description of user interest from a set of pre-assigned web pages, and uses the filter to assign unprocessed web pages. In web filtering, content similarity analysis is the core problem, the automatic-learning and relativity-analysis abilities of machine learning algorithms help solve the above problems and make ML useful in web filtering. While in practical applications, different filtering task implies different userinterest and thus implies different filtering result. This work studies how to adjust the web filtering results to be more fit for the user interest. The web filtering result are divided into three categories: relative pages, similar pages and homologous pages according to different user interest. A Biased Support Vector Machine (BSVM) algorithm, which imports a stimulant function, uses training examples distribution n+/n_ and a user-adaptable parameter k to deal imbalancedly different classes of the pre-assigned pages, is introduced to adjust the filtering result to be best fit for the user interest. Experiments show that BSVM can greatly improve the web filtering performance. © Springer-Verlag Berlin Heidelberg 2005.