From BERT to Qwen: Hate Detection across architectures
This addresses the challenge of accurately moderating hate speech on online platforms without over-censorship, though it is incremental as it compares existing model families.
The study tested whether ultra-large autoregressive LLMs improve hate-speech detection over classic bidirectional transformers on real-world online text, finding that LLMs showed modest gains in accuracy but with higher computational costs.
Online platforms struggle to curb hate speech without over-censoring legitimate discourse. Early bidirectional transformer encoders made big strides, but the arrival of ultra-large autoregressive LLMs promises deeper context-awareness. Whether this extra scale actually improves practical hate-speech detection on real-world text remains unverified. Our study puts this question to the test by benchmarking both model families, classic encoders and next-generation LLMs, on curated corpora of online interactions for hate-speech detection (Hate or No Hate).