The Naughtyformer: A Transformer Understands Offensive Humor

TL;DR

This paper introduces Naughtyformer, a Transformer-based model trained on a Reddit-filtered jokes dataset to classify humor subtypes, especially offensive jokes, demonstrating superior detection of offensiveness over existing methods.

Contribution

The paper presents a novel dataset and a Transformer model specifically designed for classifying humor subtypes, focusing on offensive humor detection.

Findings

Naughtyformer outperforms state-of-the-art methods in detecting offensive jokes.

The dataset enables nuanced classification of humor subtypes.

Transformer-based approach improves accuracy in humor subtype detection.

Abstract

Jokes are intentionally written to be funny, but not all jokes are created the same. Some jokes may be fit for a classroom of kindergarteners, but others are best reserved for a more mature audience. While recent work has shown impressive results on humor detection in text, here we instead investigate the more nuanced task of detecting humor subtypes, especially of the less innocent variety. To that end, we introduce a novel jokes dataset filtered from Reddit and solve the subtype classification task using a finetuned Transformer dubbed the Naughtyformer. Moreover, we show that our model is significantly better at detecting offensiveness in jokes compared to state-of-the-art methods.