Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/blacksujit/youtube-analysis
This project is designed to classify YouTube comments as toxic or non-toxic using BERT (Bidirectional Encoder Representations from Transformers). By fine-tuning a pre-trained BERT model, we leverage state-of-the-art NLP capabilities to identify harmful content in online conversations.
https://github.com/blacksujit/youtube-analysis
bert bert-model bertmodel embeddings fine generator tuning wordcloud words
Last synced: 26 days ago
JSON representation
This project is designed to classify YouTube comments as toxic or non-toxic using BERT (Bidirectional Encoder Representations from Transformers). By fine-tuning a pre-trained BERT model, we leverage state-of-the-art NLP capabilities to identify harmful content in online conversations.
- Host: GitHub
- URL: https://github.com/blacksujit/youtube-analysis
- Owner: Blacksujit
- Created: 2024-09-13T18:26:52.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2024-10-10T20:46:49.000Z (3 months ago)
- Last Synced: 2024-12-01T12:19:34.190Z (26 days ago)
- Topics: bert, bert-model, bertmodel, embeddings, fine, generator, tuning, wordcloud, words
- Language: Jupyter Notebook
- Homepage:
- Size: 731 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: Readme.md
Awesome Lists containing this project
README
# 🎯 YouTube Toxic Comment Classification Using BERT
## 🚀 Overview
This project is designed to classify YouTube comments as **toxic** or **non-toxic** using **BERT** (Bidirectional Encoder Representations from Transformers). By fine-tuning a pre-trained BERT model, we leverage state-of-the-art NLP capabilities to identify harmful content in online conversations.
The model is trained to identify toxicity, which is crucial for creating safer and more respectful online platforms. This project can be extended for different toxicity levels or used in moderation tools.
---
## 🔎 Demo
Here's a quick demo of the toxicity classification tool:
---
## 🗂️ Table of Contents
1. [Dataset](#dataset)
2. [Installation](#installation)
3. [Data Preprocessing](#data-preprocessing)
4. [Model Architecture](#model-architecture)
5. [Training](#training)
6. [Evaluation](#evaluation)
7. [Results](#results)
8. [Usage](#usage)
9. [Future Improvements](#future-improvements)
10. [License](#license)
---
## 📊 Dataset
The dataset consists of YouTube comments with corresponding labels indicating whether the comment is **toxic** (`1`) or **non-toxic** (`0`). This binary classification problem is aimed at improving content moderation in online discussions.
- **Columns**:
- `comment_id`: Unique identifier for each comment.
- `content`: Text of the comment.
- `label`: `1` for toxic comments, `0` for non-toxic comments.
---
## 💻 Installation
To get started with this project, follow these instructions:
### Prerequisites
- Python 3.7+
- PyTorch 1.6+
- Hugging Face Transformers library
- CUDA-enabled GPU (optional but recommended)
### Setup
1. **Clone the Repository**:
```bash
git clone https://github.com/your-repository/youtube-toxic-comment-classification.git
cd youtube-toxic-comment-classification
```
---
## 🛠️ Data Preprocessing
The raw comments are preprocessed before being fed into the BERT model. This includes:
- Removing URLs, special characters, and extra spaces.
- Converting all text to lowercase.
- Tokenizing using BERT tokenizer, which converts the text into input tokens compatible with BERT.
## 🧠 Model Architecture:
BERT (Bidirectional Encoder Representations from Transformers)
This project uses BERT to handle the NLP task of toxic comment detection. BERT is a transformer-based model that understands the context of words in sentences, making it highly effective for text classification tasks.
## Key Components:
Tokenizer: Converts sentences into token IDs.
BERT for Sequence Classification: Pre-trained BERT model fine-tuned for binary classification.
## 📊 Results:
After fine-tuning BERT, we achieved the following performance metrics:
1.) Accuracy: ~90%
2.) F1 Score: ~0.85
3.) Precision: ~0.88
4.) Recall: ~0.83
## 📈 Future Improvements
1.) Multi-label Classification: Extend the model to classify different types of toxicity (e.g., hate speech, threats, etc.).
2.) Data Augmentation: Generate synthetic examples to address class imbalance.
3.) Model Optimization: Experiment with models like DistilBERT for faster performance.
4.) Multi-language Support: Expand to detect toxicity in different languages.
## 📜 License:
This project is licensed under the MIT License. See the LICENSE file for more details.