Inferring relevant blocks on hyperlinked web page based on block-to-block similarity

Abstract

Internet users devote considerable time and effort to collecting information from the web. To do so efficiently, after following a hyperlink, a user must be able to rapidly determine whether the desired information is contained on the destination web page. In this paper, therefore, we propose a method called hyperlink referring block estimation HERB, which infers the existence and location of relevant contents on destination web pages. HERB utilises user context in web browsing, in particular, the selected hyperlink and the text around it. Through experiments simulating ordinary web browsing, we quantitatively investigate the effectiveness of HERB. Our experiments show that HERB can infer blocks relevant to a hyperlink with approximately 65% precision and 70% recall. Furthermore, we design two HERB implementations, namely, a web proxy and a web browser, and we present an overview of a web proxy prototype and an example use case.