Weakly-supervised Modeling of Language, Social, and Behavioral Abstractions for Microblog Political Discourse Classification
Social media microblogging platforms, specifically Twitter, have become highly influential and relevant to current political events. Such platforms allow politicians to communicate with the public as events are unfolding and shape public discourse on various issues. Furthermore, by selectively using framing techniques and political slogans, politicians are able to express or conceal their stances, as well as their underlying political ideologies and moral views on policies and issues. In this talk, I will present my proposals for overcoming the challenges associated with online political discourse analysis. Specifically, I will present my approach for identifying and modeling adaptable abstractions of language and behavior which can handle the ambiguous and context-independent nature of political tweets, as well as the dynamic nature of Twitter by reducing the need for expensive annotation. My works employ relational modeling of political social networks in combination with these abstractions to accurately predict and classify the ideological stances, policy frames, and moral foundations present in tweets. I will conclude the talk by discussing my future visions of porting my abstraction and modeling techniques beyond politicians to the general public, international analysis, and the study of current policy issues.
Kristen Johnson is a Ph.D. candidate in the Department of Computer Science at Purdue University advised by Prof. Dan Goldwasser. She is broadly interested in social computing - the application of machine learning and natural language processing techniques to solve computational social science problems. Her works have been published in ACL, COLING, and ICWSM. She has served as the computer science representative for the Purdue Women in Science Programs, a Local Sponsorships and Exhibits Chair for NAACL 2018, and the Widening NLP (WiNLP) 2019 Social Media Chair.