Online Harms

Risks and harms online are dangerous: misinformation, propaganda, bullying, abuse. We research data and methods for detecting and analysing harms.

Our work focuses on two major areas for automatic techniques - NLP for veracity/verification/rumour, and NLP for working with online abusive language.


Abusive Language Detection

ITU Computer Science - internal, 285.000 DKK, 2020

A project on automatic detection of bad online behaviour, including harassment, abusive language, and hate speech. The project goal is to build automatic systems and analyses for identifying this kind of behaviour, to be used to limit and address this problem by drawing attention to and allowing effective human moderation of these harmful social behaviours.


DFF, total 2,9M DKK, 2020-2022

Verif-AI’s central research question is: How can we scale AI-based verification of claims made online? It takes a machine learning and natural language processing approach to address this. The project researches multi-lingual fact extraction and verification. We identify multi-linguality as a scaling bottleneck in automated fact checking, and select fact verification as the route toward addressing propaganda and misinformation.


ITU Computer Science - internal, $43,000 USD, 2019-2020

While many pieces of misinformation can be spotted through their origin or comparison with an external source, emerging and fringe information is harder to ground and verify. This means that not all emerging claims can be checked, leading to a gap where misinformation and manipulation may exist undisturbed. This gap should be addressed.