|
--- |
|
title: Hate Speech Text Classifier |
|
emoji: 👁 |
|
colorFrom: green |
|
colorTo: red |
|
sdk: gradio |
|
sdk_version: 4.21.0 |
|
app_file: app.py |
|
pinned: false |
|
--- |
|
|
|
# Monitoring Harmful Text in Online Platforms |
|
|
|
## Overview |
|
This repository hosts the RandomForest classifier model designed for detecting harmful text AGAINST GROUPS. |
|
The model classifies text into one of three categories: "Offensive or Hateful", "Neutral or Ambiguous", and "Not Hate". |
|
Achieving an accuracy of 92.5%, this model was developed through the combination of three distinct datasets, ensuring robustness and reliability in varied contexts. |
|
It was presented at the prestigious annual Gulf Coast Conference & Expo on AI. |
|
|
|
Model Details |
|
Model Type: RandomForest Classifier |
|
Accuracy: 92.5% |
|
Labels: |
|
0: Neutral or Ambiguous |
|
1: Not Hate |
|
2: Offensive or Hateful |
|
Training Data: Augmented version of [this dataset](https://huggingface.co/datasets/TLeonidas/twitter-hate-speech-en-240ksamples) (279k+ rows) |
|
|
|
## Dataset |
|
The model was trained on a concatenated dataset from three distinct sources, curated to encompass a wide range of linguistic expressions and contexts. |
|
The datasets were preprocessed and balanced to ensure fair representation across all categories. |
|
|
|
## Performance |
|
The model achieves an accuracy of 92.5%, evaluated on a held-out test set. |
|
|
|
## Contributing |
|
We welcome contributions to improve the model and extend its applicability. |
