Comparing One-Class Classification Algorithms for Finding Interesting Resources in Social Bookmarking Systems

Abstract

Social bookmarking systems are Web-based services that allow users to bookmark different type of resources, such as Web pages or pictures, annotate resources using keywords or tags and share their bookmarks with others. It has been argued that the main reason of the widespread success of these systems is the simplicity of organize resources using open-ended tags. The massive amount of user-generated content, however, poses the challenge for users of finding interesting resources to read as well as filtering information streams coming from social systems. The study presented in this paper aims at analyzing various types of one-class classifiers in their applicability to the problem of filtering interesting resources coming from social bookmarking systems. These algorithms were used to learn the user interests starting from different sources, such as the full-text of resources and their social tags. Experimental results using a dataset gathered from Del.icio.us collaborative system are reported, showing promising results for information filtering tasks.