Learning from the Worst: Dynamically Generated Datasets to Improve Online Hate Detection

TL;DR

This paper introduces a dynamic, human-in-the-loop dataset creation process that significantly improves online hate detection models by providing challenging, well-labeled data and demonstrating enhanced robustness and accuracy.

Contribution

The paper presents a novel dynamic data generation method with multiple rounds of annotation, resulting in a large, challenging dataset that enhances hate detection model performance.

Findings

Models trained on later data rounds perform better on test sets.

The approach increases model robustness against adversarial examples.

The dataset includes detailed labels for hate type and target.

Abstract

We present a human-and-model-in-the-loop process for dynamically generating datasets and training better performing and more robust hate detection models. We provide a new dataset of ~40,000 entries, generated and labelled by trained annotators over four rounds of dynamic data creation. It includes ~15,000 challenging perturbations and each hateful entry has fine-grained labels for the type and target of hate. Hateful entries make up 54% of the dataset, which is substantially higher than comparable datasets. We show that model performance is substantially improved using this approach. Models trained on later rounds of data collection perform better on test sets and are harder for annotators to trick. They also perform better on HateCheck, a suite of functional tests for online hate detection. We provide the code, dataset and annotation guidelines for other researchers to use. Accepted at ACL 2021.