Improved Abusive Comment Moderation with User Embeddings

TL;DR

This paper enhances abusive comment moderation by integrating user and user type embeddings into an RNN-based model, demonstrating significant performance improvements on a large Greek news portal dataset.

Contribution

The study introduces the use of user and user type embeddings in moderation models, showing their effectiveness in improving detection accuracy.

Findings

User embeddings yield the largest performance gains.

All proposed user-related features improve moderation accuracy.

The approach is validated on a large-scale real-world dataset.

Abstract

Experimenting with a dataset of approximately 1.6M user comments from a Greek news sports portal, we explore how a state of the art RNN-based moderation method can be improved by adding user embeddings, user type embeddings, user biases, or user type biases. We observe improvements in all cases, with user embeddings leading to the biggest performance gains.