Implicit links based kernel to enrich Support Vector Machine for web page classification

Support Vector Machine (SVM) is a powerful classifier used widely in textual and web classification. It tries to find an hyperplane that separates positive and negative data, maximizes the margin. SVM is a classifier that is based on a kernel whose choice is very critical. We propose in this paper an implicit links based Gaussian kernel that uses an implicit links based distance. This kernel helps enrich SVM for web page classification by involving users' intuitive judgments in the classification. We tested our approach on four subsets of the Open Directory Project (ODP). Results show that implicit links based kernel helps bringing improvements on SVM's results.

I. Introduction

Nowadays, the web represents one of the principal data sources. Its size is growing permanently, which makes the task of locating relevant information in it much difficult. Due to this enormous number of web pages, the use of the information contained within the web becomes a tedious task. Thus, the need for methods and tools that facilitate the exploitation of the web's data is now more noticeable.