Classifying Web Pages With Visual Features

Abstract

Abstract:To automatically classify and process web pages, current systems use the textual content of those pages, including both the displayed content and the underlying (HTML) code. However, a very important feature of a web page is its visual appearance. In this paper, we show that using generic visual features we can classify the web pages for several different types of tasks. The features used in this document are simple color and edge histograms, Gabor and texture features. These were extracted using an off-the-shelf visual feature extraction method. In three experiments, we classify web pages based on their aesthetic value, their recency and the type of website. Results show that these simple, global visual features already produce good classification results. We also introduce an online tool that uses the trained classifiers to assess new web pages.

A Large Visual, Qualitative and Quantitative Dataset of Web Pages

“…DeBoer et al [1] use a tiny dataset, only visual and collected for categorization within four classes (news, hotels, conferences, and celebrities). These categories are quite different from each other, so the categorization problem is of less complexity.…”

“…This is usually done by analyzing both the textual content and underlying HTML code. However, the visual appearance is also an important part of a Web page, and many topics have a distinctive visual appearance, e.g., Web design blogs have a highly designed visual appearance, whereas newspaper sites will have a lot of text and images [1].…”

“…Internet is commonly associated with the Web, both terms are intimately related but are very different concepts. The first one refers to the big network of networks, i.e., the infrastructure, whereas the second one refers to the content, formed by Web sites, which are a collection of Web pages on a specific topic and linked to each other [1].…”