Direct Negative Opinions in Online Discussions

In this paper we investigate the impact of antagonism in online discussions. We define antagonism as a new class of textual opinions - direct sentiment towards the authors of previous comments. We detect the negative sentiment using aspect-based opinion mining techniques. We create a model of human behavior in online communities, based on the network topology and on the communication content. The model contains seven hypotheses, which validate two intuitions. The first intuition is that the content of the messages exchanged in an online community can separate good and insightful contributions from the rest. The second intuition is that there is a delay until the network stabilizes and until standard measures, such as betweenness centrality, can be used accurately. Taken together, these intuitions are a solid case for using the content of the communication along with network measures. We show that the sentiment within the messages, especially antagonism, can significantly alter the community perception. We use real world data, taken from the Slashdot1 discussion forum to validate our model. All the findings are accompanied by extremely significant t-test p-values.