Modeling Message Roles and Influence in Q&A Forums

We are modeling roles of individual messages and participants in Q&A discussion forums. In this paper, we present a mixed network model that represents message exchanges and message influences within a discussion thread. We first model individual message roles and thread-level user roles using discussion content features. We then combine the resulting message roles and the user roles to generate the overall influence network. Message influences and their aggregation over the network are analyzed using B-centrality measures. We use the results in identifying the most influential message in answering the initial question of the thread.