Seminars & Colloquia

Vibhor Agarwal

University of Surrey

"Understanding Online Conversations and Tackling Online Harms with Graph-based Context-aware Methods"

Friday September 08, 2023 10:00 AM
Location: None, EB2 NCSU Centennial Campus
Zoom Meeting Info
(Visitor parking instructions)

This talk is part of the AI in Society


Abstract: Online forums that allow for participatory engagement between users have been transformative for the public discussion of many important issues. However, such conversations can sometimes escalate into full-blown exchanges of hate and misinformation. Existing approaches in NLP just consider single comments or replies for detecting online hate. But in online conversations, comments and replies may be based on external context beyond the immediately relevant information. We proposed GraphNLI, a novel graph-based deep learning model that incorporates a wider conversational context in a principled manner. This additional conversational context is useful for downstream NLP tasks such as polarity prediction, hate speech detection, etc. for understanding online conversations.




For subjective tasks such as hate speech detection, we generally rely on majority voting of human annotators to get ground truth labels for datasets. However, it is hard to achieve high agreement among annotators in subjective tasks. In our ICWSM 2023 paper, we therefore developed AnnoBERT, a novel architecture integrating annotators’ different perspectives and label text with a transformer-based model to capture the subjectivity and therefore, improve hate speech detection.

Short Bio: Vibhor Agarwal is a PhD Researcher at University of Surrey, UK and currently a Visiting Researcher at Georgia Tech. His research interests lie at the intersection of Natural Language Processing and Graph Machine Learning to model and understand online conversations in social media. In his recent research published in Web Conference 2022 and ACM Transactions on the Web 2023, he showed that conversational context captured through GraphNLI improves the detection of online harms such as hate speech, misinformation, etc. in social media conversations. Besides, he is also working on detecting and mitigating online harms due to large language models such as hallucination, misinformation, etc. In his free time, he loves to go for hiking, travelling, and playing badminton.

Special Instructions: Zoom only.

Host: Munindar Singh, CSC

Back to Seminar Listings
Back to Colloquia Home Page