Abstract

We explore an effective approach for modeling and classifying Web sites in the World Wide Web. The aim of this work is to classify Web sites using features which are independent of size, structure and vocabulary. We establish Web site similarity based on search engine query hits, which convey document relevance and utility in direct relation to users' needs and interests. To achieve this, we use a generic Web site representation scheme over different feature spaces, built upon query traffic to the site's documents. For this task we extend, in a non-trivial way, our prior work using query-sets for single document representation. We discuss why this previous methodology is not scalable for a large set of heterogeneous Web sites. We show that our models achieve very compact Web site representations. Furthermore, our experiments on site classification show excellent performance and quality/dimensionality trade-off. In particular, we sustain a reduction in the feature space to 5% of the size of the bag-of-words representation, while achieving 99% precision in our classification experiments on DMOZ. © 2011 Springer-Verlag.