(I) Much social media content is multimodal. Does one's multimodal model learn cross-modal interactions? It's hard to tell than one might think! (II) We consider controversial posts, meaning those that split the preferences of a community, receiving both significant positive and significant negative feedback. We predict community-specific controversy of social-media posts, examining domain transfer within an early detection paradigm utilizing both conversation content and conversation structure. (III) We consider how best to study the effect of comment deletion as used as a moderation strategy on social-media platforms. The standard interrupted time series analysis approach fails to distinguish the effect of having made a non-compliant comment from the effect of being subjected to moderator removal of that comment. We instead introduce a "delayed feedback" paradigm based on the observation that some users may remain active between the time when they posted the non-compliant comment and the time when that comment is deleted.
Joint work with Cristian Danescu-Niculescu-Mizil, Jack Hessel, Kumar Bhargav Srinivasan, Chenhao Tan
Lillian Lee is the Charles Roy Davis professor in the departments of computer science and of information science at Cornell University. Her research interests include natural language processing and computational social science. She is a AAAI Fellow, an ACL Fellow, an ACM Fellow, and a former Sloan Fellow. She received one of three inaugural awards for the Test of Time (2002-2012) Paper on Computational Linguistics (joint with Bo Pang), and best paper awards at NAACL 2004 (joint with Regina Barzilay) and the IJCAI 2016 Natural Language Processing meets Journalism workshop (joint with Liye Fu and Cristian Danescu-Niculescu-Mizil). Her co-authored work has received several mentions in the popular press, including The New York Times, NPR's All Things Considered, and NBC's The Today Show.