DMI Webinar: "What Sota does not tell: Insights and challenges of online abusive language detection"

“What Sota does not tell: Insights and challenges of online abusive language detection”

12 October 2020, 12:30-13:30 CEST

ABSTRACT: 

Social media messages are often written to attack specific groups of users based on their religion, ethnicity or social status, and they can be particularly threatening to vulnerable users such as teenagers. It is therefore very important to develop reliable, unbiased and robust detection systems to support stakeholders in fighting online hatred. Although several systems obtained very good classification results (> 90 F1) on the English hate speech task at the last Offenseval campaign, the problem is far from being solved. In my talk, I will discuss which issues still affect the development of abusive language detection systems, from discrepancies in annotation guidelines to disregarded contextual information and cross-lingual differences. I will also present ongoing work aimed at (partly) mitigating the above issues.

SPEAKER: 

Sara Tonelli is the head of the Digital Humanities research group at Fondazione Bruno Kessler in Trento, and adjunct professor of Language Interfaces at the Dept. of Psychology and Cognitive Science, University of Trento. She got a PhD in Language Sciences from Università Ca’ Foscari in 2010, after which she joined FBK first as a post-doc in the Natural Language Processing group and then as a tenured researcher leading the newly founded DH group. She has been involved in several national and European projects dealing with historical archives, event and temporal information extraction and political stance detection and, more recently, with social media monitoring and hate speech detection. Her research interests are highly interdisciplinary, trying to apply and adapt advanced text analysis approaches to social sciences and historical investigation.

The talks will be held online. For more information, write to dmi@unibocconi.it.