Recent years have seen extraordinary breakthroughs in computational sentiment analysis and social network analysis. This research has helped to reveal that robust language understanding --- in dialogue, in text, on the Web --- depends on accurately identifying attitudes, emotions, and social relationships. However, missing from the current scientific and technical picture is a deep understanding of the ways in which sentiment affects, and is affected by, our interpersonal relationships and social networks. The central goal of this project is to fill this gap by developing algorithms, methods, and data sets for modeling sentiment as social and interpersonal. The approach balances fine-grained sociological and linguistic study, computational modeling, and large-scale data-mining of weblogs and other interactive social media.
This research aims to reshape the field of sentiment analysis by moving it towards inferential models of emotional language as situated in specific social contexts. It also provides social network analysis with rich features for characterizing social ties, thereby facilitating the identification of new latent structure in social networks. The work addresses a wide variety of social and political issues, including the extent of media polarization, the effects of bias and framing, and the role of emotional content in shaping the flow of information. It can also inform the creation of the next generation of communication tools --- software for virtual meetings and online collaboration that intelligently tracks users' evolving attitudes, social networking platforms that distinguish relationship types, and data-mining tools relevant for legal discovery, intelligent tutoring, and media analytics.