|
|
--- |
|
|
datasets: |
|
|
- google/jigsaw_toxicity_pred |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
--- |
|
|
# Multi-Label Hate Speech Classifier |
|
|
|
|
|
## Overview |
|
|
The **Multi-Label Hate Speech Classifier** is a machine learning model designed to detect and categorize multiple forms of hate speech within textual data. It leverages a OneVsRest Logistic Regression classifier combined with TF-IDF vectorization to analyze and classify text into multiple labels simultaneously. |
|
|
|
|
|
## Features |
|
|
- **Multi-Label Detection:** Assigns multiple hate speech categories to a single piece of text. |
|
|
- **Supported Categories:** |
|
|
- **toxic** |
|
|
- **obscene** |
|
|
- **insult** |
|
|
- **threat** |
|
|
- **identity_hate** |
|
|
- **Custom Thresholds:** Optimized thresholds are applied to each label to balance precision and recall. |
|
|
|
|
|
## Model Architecture |
|
|
- **Text Vectorization:** Utilizes TF-IDF (Term Frequency-Inverse Document Frequency) to convert raw text into a numerical format. |
|
|
- **Classifier:** Implements a OneVsRest Logistic Regression approach for multi-label classification. |
|
|
- **Training Process:** Trained on a balanced dataset with pre-processed text to achieve robust performance across all categories. |
|
|
|
|
|
## Setup & Installation |
|
|
|
|
|
### Requirements |
|
|
- Python 3.x |
|
|
- Dependencies: |
|
|
- `numpy` |
|
|
- `pandas` |
|
|
- `scikit-learn` |
|
|
- `joblib` |
