🚨 Hate Speech & Offensive Content Detection in Social Media

A social media prototype that detects and flags hate speech or offensive content in uploaded videos using AI + NLP. The system ensures a safe user experience by moderating user-generated content in real time.

---

📌 Problem Statement

With the rise of social media, hate speech and offensive content have become a major concern. This project addresses the problem by automatically analyzing uploaded videos, detecting harmful speech, and ensuring flagged content is securely handled.

---

📂 Project Overview

This project is a web application that allows users to upload and watch short videos, similar to TikTok or Instagram Reels. The core of the project is an AI-powered content moderation system that automatically detects and flags videos containing hate speech or offensive language.

The system uses a multi-stage process:

1. Video Transcription: The audio from uploaded videos is transcribed into text using OpenAI's Whisper model.
2. Toxicity Analysis: The transcribed text is then analyzed by a custom-trained NLP model to identify toxic content.
3. Content Flagging: Videos that are identified as containing toxic content are flagged and separated from the main feed.

🛠️ Project Architecture

The project is built with a Django backend and a simple HTML/CSS/JavaScript frontend. The main components are:

• Django Backend:
  • censorAI: The main Django project directory.
  • store: A Django app that handles user authentication, video uploads, and other core functionalities.
  • media: Directory where uploaded videos are stored.
  • db.sqlite3: The SQLite database for the project.
• AI/ML:
  • ml.py: A script for training the toxic comment classification model.
  • inferml.py: A script that uses the trained model to perform inference on transcribed text.
  • process.py: A script that orchestrates the video processing pipeline (transcription and inference).
  • models/: Directory where the trained model and tokenizer are saved.
  • datasets/: Directory containing the training and testing data.
• Frontend:
  • templates/: Directory containing the HTML templates for the web pages.
  • static/: Directory containing the CSS and JavaScript files.

---

Due to large file size the datasets and Saved AI model, copy it directly from the drive link given below. the folders must be directly in the main folder.

https://drive.google.com/file/d/1WoGpNhltqSwVn0U_hI4UkULTnXivO0n3/view?usp=sharing

directly extract it, in the main folder

dir/datasets/ dir/saved_models/

⚙️ Installation

To set up and run this project on your local machine, follow these steps:

1. Clone the repository:

   git clone https://github.com/your-username/censorAI.git
   cd censorAI

2. Create and activate a virtual environment:

   python -m venv env
   source env/bin/activate  # On Windows, use `env\Scripts\activate`

3. Install the required dependencies:

   pip install -r requirements.txt

   (Note: A requirements.txt file is not yet present. I will add one in a future step.)

4. Apply the database migrations:

   python manage.py migrate

5. Run the development server:

   python manage.py runserver

   The application will be available at http://127.0.0.1:8000/.

---

📊 Dataset

The AI model is trained on the "Toxic Comment Classification Challenge" dataset from Kaggle. This dataset contains a large number of Wikipedia comments which have been labeled by human raters for toxic behavior. The types of toxicity are:

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

The dataset is located in the datasets/ directory and is split into train.csv and test.csv.

---

🧠 Training the Model

To train the toxic comment classification model, you can run the ml.py script. This script will:

1. Load the training data from datasets/train.csv.
2. Preprocess the text data.
3. Build and compile a TensorFlow/Keras model.
4. Train the model on the preprocessed data.
5. Save the trained model and tokenizer to the models/ directory.

To run the training script:

python ml.py

(Note: The ml.py script might need to be adjusted to match the final project structure.)

---

🤖 Using the AI

The AI model is used to filter and flag videos that contain hate speech or offensive content. Here's how it works:

1. Video Upload: A user uploads a video through the web interface.
2. Processing: The process.py script is triggered, which does the following:
   • Transcribe Audio: It uses the inferml.py script to transcribe the audio from the video into text using the Whisper model.
   • Analyze Text: The transcribed text is then fed into the trained toxic comment classification model to get a toxicity score.
3. Flagging: If the toxicity score exceeds a certain threshold, the video is flagged as is_flagged = True in the database. Flagged videos are not shown in the main feed.

---

📝 Examples

Input: A video file (e.g., my_video.mp4) containing speech.

Processing Steps:

1. The audio is transcribed to text: "This is a test video. I hope you like it."
2. The text is analyzed for toxicity. For the example above, the model would likely return a low toxicity score.
3. If the transcribed text was "You are a stupid idiot, I hate you", the model would return a high toxicity score for toxic and insult.

Output:

• If Safe: The video is saved to the database with is_flagged = False and is visible to other users.
• If Toxic: The video is saved to the database with is_flagged = True and is hidden from the main feed.

---

🚀 Features

• 🎥 Video upload & reel-based browsing
• 🔑 Secure login with OTP validation
• 🤖 Whisper + NLP pipeline for transcript analysis
• 🚫 Automatic hate speech & offensive content detection with dynamic thresholding.
• 👍 Like/Dislike functionality for videos.
• 💬 Commenting functionality for videos.
• 🔒 Secure section for flagged videos (Not Implemented Yet)

---