Identifying Roles in Social Networks using Linguistic Analysis.

Abstract

Social media sites have been significantly growing in the past few years. This resulted in the emergence of several communities of communicating groups, and a huge amount of text exchanged between members of those groups. In our work, we study how linguistic analysis techniques can be used for understanding the implicit relations that develop in on-line communities. We use this understanding to develop models that explain the processes that govern language use and how it reveals the formation of social relations. We study the relation between language choices and attitude between participants and how they may lead to or reveal antagonisms and rifts in social groups. Both positive (friendly) and negative (antagonistic) relations exist between individuals in communicating communities. Negative relations have received very little attention, when compared to positive relations, because of the lack of an explicit notion of labeling negative relations in most social computing applications. We alleviate this problem by studying text exchanged between participants to mine their attitude. Another important aspect of our research is the study of influence in discussions and how it affects participants’ discourse. In any debate or discussion, there are certain types of persons who influence other people and affect their ideas and rhetoric. We rely on natural language processing techniques to find implicit connections between individuals that model this influence. We couple this with network analysis techniques for identifying the most authoritative or salient entities. We also study how salience evolves over time. Our work is uniquely characterized by combining linguistic features and network analysis to reveal social roles in different communities. The methods we developed can find several interesting areas of applications. For example, they can be used for identifying authoritative sources in social media, finding influential people in communities, mining attitude toward events and topics, detecting rifts and subgroup formation, summarizing different viewpoints with respect to some topic or entity, and many other such applications.