Visual features in genre classification of html

Automatic genre classification historically has focused on extracting textual features from documents. In this research, we investigate whether visual features of HTML documents can improve the classification of fine grained genres. Three different sets of features were compared on a genre classification task in the e-commerce domain - one with just textual features, one with HTML features added, and a third with additional visual features. Our experiments show that adding HTML and visual features provides much better classification than textual features alone.