Diversifying user comments on news articles

Abstract

In this paper we present an approach for diversifying user comments on news articles. In our proposed framework, we analyse user comments w.r.t. four different criteria in order to extract the respective diversification dimensions in the form of feature vectors. These criteria involve content similarity, sentiment expressed within comments, article's named entities also found within comments and commenting behavior of the respective users. Then, we apply diversification on comments, utilizing the extracted features vectors. The outcome of this process is a subset of the initial comments that contains heterogeneous comments, representing different aspects of the news article, different sentiments expressed, as well as different user categories, w.r.t. their commenting behavior. We perform a preliminary qualitative analysis showing that the diversity criteria we introduce result in distinctively diverse subsets of comments, as opposed to a baseline of diversifying comments only w.r.t. to their content (textual similarity). We also present a prototype system that implements our diversification framework on news articles comments. © 2012 Springer-Verlag.