Extracting Signed Social Networks from Text

Most of the research on social networks has almost exclusively focused on positive links between entities. There are much more insights that we may gain by generalizing social networks to the signed case where both positive and negative edges are considered. One of the reasons why signed social networks have received less attention that networks based on positive links only is the lack of an explicit notion of negative relations in most social network applications. However, most such applications have text embedded in the social network. Applying linguistic analysis techniques to this text enables us to identify both positive and negative interactions. In this work, we propose a new method to automatically construct a signed social network from text. The resulting networks have a polarity associated with every edge. Edge polarity is a means for indicating a positive or negative affinity between two individuals. We apply the proposed method to a larger amount of online discussion posts. Experiments show that the proposed method is capable of constructing networks from text with high accuracy. We also connect out analysis to social psychology theories of signed network, namely the structural balance theory.