Context-Based Classification for Link Data

In Web-based e-learning, an up-to-date catalogue of subject-specific Web resources can effectively offer inexperienced students with an advanced academic portal on the Web. To automatically construct such academic Web resource catalogue, a key issue is how to classify the collected Web pages. However, existing link-based classification methods treat all neighbors equally in the categorization of the target objects. In this paper, we propose a context-based classification approach that can scale well for noisy and heterogeneous link data such as Web pages. We quantitatively measure the contextually topical dependencies between linked objects using the dependence functions, which are then exploited to classify the target objects in the link structure. Experimental results show that the proposed classification model can better capture the link regularities and can facilitate better categorization of linked objects.