Proposal Computational Social Roles : Identify , Recommend and Configure Emergent Social Roles in Online Communities

Millions of people participate in online communities, exchange expertise and ideas, and collaborate to produce complex artifacts, such as Linux and Wikipedia. They engage in a variety of roles, which strongly influence the amount and types of work they do, and how they coordinate their activities. Better understanding members’ roles benefits members by clarifying how they should behave to participate effectively and also benefits the community overall by encouraging members to contribute in ways that best use their skills and interests. Social sciences have provided rich theoretic taxonomies of social roles within groups, while natural language processing techniques enable us to automate the identification of social roles in online communities. However, most social science work has focused on generic roles without accommodating the activities associated with tasks in specific contexts or automating the process of role identification. While there has been work to date about automatic role inference, identification of social roles has not had a corresponding strong emphasis in the language technologies community. A variety of methods were developed to extract specific “roles”, patterns, or components in different contexts, lacking of generalized definitions about what are roles and systematic methods about how to extract roles. Moreover, how roles change over time and how the awareness of roles influence role holders’ performance and the group production, have not been adequately researched in both fields. In this thesis, I advocate for both theories of social science and models of text analysis to better define roles, develop ways to extract roles, optimally recommend roles to users, and configure roles within the community. Concretely, I focus on four perspectives. The first work constructs profiles of users from what they do and with whom they interact in online cancer support groups, from which we then extract social roles in an unsupervised manner. The second work predicts when and how members transit from one role to another and examines how role contrast helps explain the occurrences of different transitions. Third, I model how the presence of different types of roles and their interaction with task level, group tenure, and group type, predicts group performances. The last perspective investigates whether making role expectation explicit increases group performance. I will produce both theoretical and computational results: this thesis will develop new algorithms to mine behavioral data to extract social roles and recommend roles, tasks, and groups to target an optimal group configuration; it can also advance the development of social science theories on how roles and role collaboration affect participation in online communities.