Robust collective classification with contextual dependency network models

In order to exploit the dependencies in relational data to improve predictions, relational classification models often need to make simultaneous statistical judgments about the class labels for a set of related objects. Robustness has always been an important concern for such collective classification models since many real-world relational data such as Web pages are often accompanied with much noisy information. In this paper, we propose a contextual dependency network (CDN) model for classifying linked objects in the presence of noisy and irrelevant links. The CDN model makes use of a dependency function to characterize the contextual dependencies among linked objects so that it can effectively reduce the effect of irrelevant links on the classification. We show how to use the Gibbs inference framework over the CDN model for collective classification of multiple linked objects. The experiments show that the CDN model demonstrates relatively high robustness on datasets containing irrelevant links.