HateCheckHIn: Evaluating Hindi Hate Speech Detection Models

TL;DR

HateCheckHIn is a new evaluation dataset designed to diagnose and improve multilingual hate speech detection models, focusing on Hindi and code-mixed social media content, by providing targeted functionalities beyond standard metrics.

Contribution

The paper introduces HateCheckHIn, a diagnostic evaluation dataset with functionalities for analyzing multilingual hate speech models, highlighting their specific failure modes.

Findings

Transformer-based m-BERT model shows specific failure patterns.

Perspective API's performance varies across different hate speech functionalities.

HateCheckHIn enables targeted diagnostics for multilingual hate speech detection.

Abstract

Due to the sheer volume of online hate, the AI and NLP communities have started building models to detect such hateful content. Recently, multilingual hate is a major emerging challenge for automated detection where code-mixing or more than one language have been used for conversation in social media. Typically, hate speech detection models are evaluated by measuring their performance on the held-out test data using metrics such as accuracy and F1-score. While these metrics are useful, it becomes difficult to identify using them where the model is failing, and how to resolve it. To enable more targeted diagnostic insights of such multilingual hate speech models, we introduce a set of functionalities for the purpose of evaluation. We have been inspired to design this kind of functionalities based on real-world conversation on social media. Considering Hindi as a base language, we craft test cases for each functionality. We name our evaluation dataset HateCheckHIn. To illustrate the utility of these functionalities , we test state-of-the-art transformer based m-BERT model and the Perspective API.