We propose scalable, linear-time models and algorithms for online social network analysis that remedy limitations of current state-of-the-art models by creating a capability for tracking, predicting affiliations, and roles of participants in and across online communities. The model goes beyond simple clustering and community detection by using more of the message in the media, i.e. by taking into account the semantic, pragmatic and temporal content of computer-mediated-communication by an individual and within a community. Using parallel models of individual participants and groups, we propose to augment previous algorithms by incorporating a multi-dimensional network representation incorporating attitudes of participants and groups toward entities, issues, beliefs, and other participants. Further, individual participant models will represent their roles in the community based on the nature of their online interactions. This approach will reveal social ties among group participants and the relative strength of their group affiliations. The constructed representations will reveal user-specific and group-prevalent themes, sentiments, activities, and roles. This will allow us to predict patterns of group formation and dissolution, and to predict an individual participant?s likelihood of initiating or maintaining an affiliation with a group based on a mathematical comparison of that individual?s profile with the group?s profile.
Keywords: Social media, social network analysis, sentiment analysis, influence modeling, group evolution, group affiliation, online communities, text mining
