Modeling of Political Discourse on Twitter

Our works focus on the application of NLP methods for the analysis of political discourse on Twitter. Our guiding intuition is that modeling the language used on Twitter alone is not enough for the most accurate prediction possible. Therefore, we explore how weakly supervised models can be constructed to leverage both language and behaviors of politicians on Twitter to identify stance and framing patterns from the discourse. By incorporating behavioral features, such as similar temporal information, stances on current and future political issues, as well as the frames used to express these issues, can be determined with higher accuracy than what is possible with language based models alone.