Socio-Culturally Aware Evaluation Framework for LLM-Based Content Moderation
This work addresses content moderation for social media and LLMs by improving dataset diversity, though it is incremental as it builds on existing evaluation methods.
The authors tackled the problem of unreliable content moderation assessments due to underrepresented groups in datasets by proposing a socio-culturally aware evaluation framework and a scalable persona-based generation method for creating diverse datasets. Their analysis showed that these datasets provide broader perspectives and pose greater challenges for LLMs, especially smaller ones, in moderating diverse content.
With the growth of social media and large language models, content moderation has become crucial. Many existing datasets lack adequate representation of different groups, resulting in unreliable assessments. To tackle this, we propose a socio-culturally aware evaluation framework for LLM-driven content moderation and introduce a scalable method for creating diverse datasets using persona-based generation. Our analysis reveals that these datasets provide broader perspectives and pose greater challenges for LLMs than diversity-focused generation methods without personas. This challenge is especially pronounced in smaller LLMs, emphasizing the difficulties they encounter in moderating such diverse content.