Multi-Class Detection of Abusive Language Using Automated Machine Learning

Abstract

Abusive language detection online is a daunting task for moderators. We propose Automated Machine Learning (Auto-ML) to semi-automate abusive language detection and to assist moderators. In this paper, we show that multi-class classification powered by Auto-ML is successful in detecting abusive language in English and German as well as and better than the state-ofthe-art machine learning models. We also highlight how we combatted the imbalanced data problem in our data-sets through feature selection and undersampling methods. We propose Auto-ML as a promising approach to the field of abusive language detection, especially for small companies who may have little machine learning knowledge and computing resources.

No Time Like the Present: Effects of Language Change on Automated Comment Moderation

“…Finding the optimal ML configuration for a problem usually involves a repetitive and time-consuming process of testing different models, hyperparameters, preprocessing techniques, and feature engineering strategies. The goal of Auto-ML is to automate much of this workflow and reduce the developer's bias towards prioritizing specific models or configurations over others [48,49].…”