Final model's label
Pick a post and step through how a hate-speech filter reads it. Every number is from the real model.
No slurs appear on this page.
Model: TF-IDF + logistic regression, trained on 2017 tweets (Davidson et al., 2017) plus everyday tweets (OLID).
Limits: English only; reviewers assumed always right; dialect estimated from language, not identity. A research demo, not a tool for real posts.