Neural Word Decomposition Models for Abusive Language Detection

TL;DR

This paper investigates how character, subword, and byte pair encoding models improve the robustness of abusive language detection in social media text, especially under domain shifts and text variations.

Contribution

It compares various word decomposition techniques and demonstrates their effectiveness in enhancing large pretrained models for abusive language detection.

Findings

Character, subword, and BPE models improve robustness to spelling and typological variations.

Combining decomposition techniques enhances detection accuracy.

Pretrained models fine-tuned with these techniques perform well across different social media datasets.

Abstract

User generated text on social media often suffers from a lot of undesired characteristics including hatespeech, abusive language, insults etc. that are targeted to attack or abuse a specific group of people. Often such text is written differently compared to traditional text such as news involving either explicit mention of abusive words, obfuscated words and typological errors or implicit abuse i.e., indicating or targeting negative stereotypes. Thus, processing this text poses several robustness challenges when we apply natural language processing techniques developed for traditional text. For example, using word or token based models to process such text can treat two spelling variants of a word as two different words. Following recent work, we analyze how character, subword and byte pair encoding (BPE) models can be aid some of the challenges posed by user generated text. In our work, we analyze the effectiveness of each of the above techniques, compare and contrast various word decomposition techniques when used in combination with others. We experiment with finetuning large pretrained language models, and demonstrate their robustness to domain shift by studying Wikipedia attack, toxicity and Twitter hatespeech datasets