Papers by Ivo Verhoeven
A (More) Realistic Evaluation Setup for Generalisation of Community Models on Malicious Content Detection (2024.findings-naacl)
Copied to clipboard
| Challenge: | despite the performance of community models for malicious content detection, misinformation and hate speech continue to propagate on social media networks. |
| Approach: | They propose a new evaluation setup for community models for malicious content detection based on a few-shot subgraph sampling approach to test generalisation of models using local explorations of a larger graph. |
| Outcome: | The proposed evaluation setup outperforms existing models on real-world graphs on a training graph. |