Comment Toxicity Detection

Deep Learning-based multi-label toxicity detection system for online comments.
This project uses an LSTM model to detect various forms of toxic comments and provides a Streamlit app for real-time and bulk predictions.

---

Project Overview

Online communities and social media platforms have become integral parts of modern communication. However, the prevalence of toxic comments, which include harassment, hate speech, and offensive language, poses significant challenges to maintaining healthy discussions.

This project aims to build a deep learning-based comment toxicity detection system using Python. The model predicts the likelihood of a comment being toxic across multiple categories:

• toxic
• severe_toxic
• obscene
• threat
• insult
• identity_hate

By identifying toxic comments automatically, the system assists platform moderators in filtering, warning users, or initiating further review processes.

---

Project Type

• Type: EDA + Multi-label Classification

---

Features

• Text preprocessing (cleaning, tokenization, stopword removal)
• Multi-label classification using LSTM model
• Model evaluation: Accuracy, Loss, Confusion Matrix, Classification Report
• Real-time prediction using Streamlit app
• Bulk predictions from CSV files
• Pre-trained model (best_model.keras) included via Git LFS

---

Project Files

File/Folder	Description
best_model.keras	Trained LSTM model for toxicity detection (requires Git LFS)
tokenizer.pkl	Tokenizer used for text preprocessing
app.py	Streamlit application for real-time and bulk prediction
Comment_Toxicity.ipynb	Jupyter notebook with EDA, model training, evaluation, and markdown
data/	Example CSV files for training/test (optional)
README.md	Project overview and instructions

---

Getting Started

Prerequisites

• Python 3.8+
• Install dependencies:

pip install -r requirements.txt

• Git LFS (to download the model file):

git lfs install
git lfs track "*.keras"

Clone the Repository

git clone https://github.com/Vignesha-S/Comment-Toxicity-Detection.git
cd Comment-Toxicity-Detection
git lfs pull

---

Running the Streamlit App

streamlit run app.py

• Real-time prediction: Enter a comment in the text area and click Predict
• Bulk prediction: Upload a CSV file with a comment_text column

---

Usage Notes

• The model is trained for multi-label classification. Each comment can belong to multiple categories simultaneously.
• Predictions range from 0 to 1 for each label, representing the probability of that label.
• Ensure Git LFS is installed to download large files (best_model.keras)

---

Project Summary

This project demonstrates the complete workflow for building a deep learning solution for toxicity detection, from data preprocessing and EDA, to model training, evaluation, and deployment using Streamlit. Users can enter text or upload CSV files to get predictions in real-time.

The model helps online platforms identify and manage toxic comments, promoting healthier and safer online communities.

---

Note:

• The best_model.keras file is stored using Git LFS. Make sure to install Git LFS before cloning the repository.
• The training and test CSV files are large and may exceed GitHub’s standard file size limits. You can either download them via Git LFS or use smaller sample datasets for testing the Streamlit app.