Friday, 3rd May - 1pm Peter Young : Seminar

Title: A Graph-Based Context-Aware Model to Understand Online Conversations

Abstract:

It is important to study and understand Internet debates because they often have consequences in the offline world, for better or worse. For debates that follow a tree whose nodes are arguments and whose directed edges are replies, we articulate two tasks: (1) the polarity prediction task, which classifiers whether a replying argument is pro or con, and (2) the misogynistic hate speech detection task, which classifies whether the comment submitted contains misogynistic hate speech. We present GraphNLI, a deep learning model that uses random walks to incorporate contextual information surrounding the node to be classified in a principled manner. The ability to incorporate context enables GraphNLI to outperform various other SOTA classifiers for both of these tasks (as of end-2023).

Link to Paper:

https://dl.acm.org/doi/abs/10.1145/3624579

Bio:

Peter Young is a data scientist working in fintech and a visiting researcher at King's College London. His research interests include computational argumentation, its relationship to cooperative game theory, and its application to understanding online conversations. Peter has studied mathematics, theoretical physics, computer science and quantitative finance in London and Manchester, and has received his PhD from King's College London in 2017.