Implicit negative link detection on online political networks via matrix tri-factorizations

Political conversations have become a ubiquitous part of social media. When users interact and engage in discussions, there are usually two mediums available to them; textual conversations and platform-specific interactions such as like, share (Facebook) or retweet (Twitter). Major social media platforms do not facilitate users with negative interaction options. However, many political network analysis tasks rely on not only positive but also negative linkages. Thus, detecting implicit negative links is an important and a challenging task. In this work, we propose an unsupervised framework utilising positive interactions, sentiment cues, and socially balanced triads for detecting implicit negative links. We also present an online variant of it for streaming data tasks. We show the effectiveness of both frameworks with experiments on two annotated datasets of politician Twitter accounts. Our experiments show that the proposed frameworks outperform other well-known and proposed baselines. To illustrate the detected implicit negative links' contribution, we compare the community detection accuracies using unsigned and signed networks. Experimental results using detected negative links show superiority on the three datasets where the camps are known a priori. We also present qualitative evaluations of polarisation patterns between communities which are only possible in the presence of negative links.