Mark Last

In this paper we describe a new approach to classification of web documents. Most web classification methods are based on the vector space document representation of information retrieval. Recently the graph based web document representation model was shown to outperform the traditional vector representation using k-Nearest Neighbor (k-NN) classification algorithm. Here we suggest a new hybrid approach to web document classification built upon both, graph and vector representations. K-NN algorithm and three benchmark document collections were used to compare this method to graph and vector based methods separately. Results demonstrate that we succeed in most cases to outperform graph and vector approaches in terms of classification accuracy along with a significant reduction in classification time.