🎥 YouTube Sentiment Insights

End-to-End MLOps Project for YouTube Comment Sentiment Analysis using a Chrome Extension, DVC, MLflow, AWS, and LightGBM

YouTube Sentiment Insights is an end-to-end Machine Learning + MLOps project that analyzes the sentiment of YouTube video comments and presents the results through a simple dashboard-like experience.

The project combines a traditional NLP/ML pipeline with MLOps best practices such as DVC for pipeline orchestration, MLflow for experiment tracking and model registry, AWS for remote artifact storage and deployment, and a Chrome extension frontend that interacts with the sentiment analysis API.

---

🎥 Demo

---

📌 Table of Contents

• Why This Project Exists
• Project Overview
• Features
• ML Workflow
• MLOps Workflow
• Tech Stack
• Project Structure
• How It Works
• Setup Instructions
• Future Improvements
• Acknowledgement
• Contact

---

❓ Why This Project Exists

YouTube comments contain a huge amount of user feedback, opinions, reactions, and sentiment. For creators, marketers, researchers, and even casual viewers, manually reading thousands of comments to understand the overall tone of a video is time-consuming and impractical.

Problems

• Videos often receive hundreds or thousands of comments
• Manually identifying whether the audience reaction is positive, negative, or neutral is inefficient
• There is no simple way to get a quick sentiment summary of a video’s comment section
• Building an ML solution is only half the task — making it reproducible, trackable, and deployable is equally important

✅ Solution

This project solves the problem by building a system that:

• Detects/open a YouTube video
• Extracts or sends comments for analysis
• Classifies comments into:
  • 1 → Positive
  • 0 → Neutral
  • -1 → Negative
• Produces sentiment outputs that can be consumed by a Chrome extension / frontend
• Uses DVC + MLflow + AWS + CI/CD to make the entire workflow production-oriented and reproducible

---

📖 Project Overview

This project follows a traditional Machine Learning NLP pipeline rather than an LLM-based approach.

The core idea is to train a sentiment classifier on a labeled Reddit sentiment dataset, then use that model to predict sentiment on YouTube comments through a lightweight application flow.

High-level workflow

1. Collect sentiment dataset for training
2. Clean and preprocess text
3. Train a sentiment classification model
4. Track experiments with MLflow
5. Version the ML pipeline with DVC
6. Register the trained model
7. Expose model predictions through a Flask API
8. Use a Chrome extension frontend to interact with the API for YouTube sentiment analysis
9. Dockerize + prepare AWS deployment + CI/CD workflow

---

🚀 Features

💬 YouTube Comment Sentiment Analysis

• Analyze comments from a YouTube video
• Predict sentiment for each comment
• Classify into:
  • Positive
  • Neutral
  • Negative

🧹 NLP Preprocessing Pipeline

• Missing value handling
• Duplicate removal
• Empty text filtering
• Lowercasing
• Stopword removal
• Lemmatization
• Text normalization

🧠 Traditional ML Model Training

• TF-IDF vectorization
• N-gram feature engineering
• LightGBM-based multi-class sentiment classifier
• Configurable hyperparameters using params.yaml

📊 Experiment Tracking with MLflow

• Log experiments and compare runs
• Track model performance
• Save artifacts remotely on AWS S3
• Register the final trained model in MLflow Model Registry

🔁 Reproducible DVC Pipeline

• Version-controlled ML pipeline
• Separate stages for:
  • Data ingestion
  • Data preprocessing
  • Model building
  • Model evaluation
• Easy reproducibility using DVC commands

☁️ AWS-based MLOps Setup

• MLflow tracking server hosted on EC2
• Artifacts stored in S3
• AWS CLI used to connect local development environment with AWS resources

🧩 Chrome Extension Frontend

• Chrome extension frontend for interacting with the sentiment analysis system
• Designed to work with YouTube and show sentiment insights to the user

🐳 Deployment Ready

• Dockerized application setup
• GitHub Actions CI/CD workflow included
• Ready for cloud deployment and automation

---

🛠 Tech Stack

Layer	Technology
Language	Python
ML / NLP	Scikit-learn, NLTK, LightGBM
Feature Engineering	TF-IDF Vectorizer
Experiment Tracking	MLflow
Pipeline Orchestration	DVC
Backend API	Flask
Frontend	Chrome Extension (HTML, CSS/JS, Manifest)
Cloud	AWS EC2, AWS S3, AWS CLI
CI/CD	GitHub Actions
Serialization	Pickle
Data Handling	Pandas, NumPy
Config Management	YAML

---

📂 Project Structure

Youtube-Sentiment-Insights/
│
├── .dvc/
├── .github/
│   └── workflows/
│       └── cicd.yaml
│
├── assets/
│   └── demo.gif
│
├── data/
│   └── ...
│
├── flask_app/
│   ├── app.py
│   └── test.py
│
├── notebooks/
│   └── ...
│
├── src/
│   ├── data/
│   │   ├── data_ingestion.py
│   │   └── data_preprocessing.py
│   │
│   └── model/
│       ├── model_building.py
│       ├── model_evaluation.py
│       ├── register_model.py
│       └── __init__.py
│
├── yt-chrome-plugin-frontend/
│   ├── manifest.json
│   ├── popup.html
│   └── popup.js
│
├── .dvcignore
├── .gitignore
├── confusion_matrix_Test Data.png
├── dvc.lock
├── dvc.yaml
├── experiment_info.json
├── lgbm_model.pkl
├── params.yaml
├── README.md
├── requirements.txt
├── setup.py
└── tfidf_vectorizer.pkl

---

🔄 How It Works

1. Train the Sentiment Model

The ML pipeline starts with the Reddit sentiment dataset.

• Ingest data
• Clean and preprocess comments
• Convert comments into TF-IDF features
• Train a LightGBM sentiment classifier
• Save the vectorizer and trained model

---

2. Track Experiments with MLflow

During training and evaluation:

• Metrics can be logged
• Artifacts can be stored
• The best model can be registered for downstream use

---

3. Serve the Model via Flask API

The Flask app loads:

• tfidf_vectorizer.pkl
• lgbm_model.pkl

Then it exposes prediction endpoints for inference.

---

4. Connect with Chrome Extension

The Chrome extension acts as the user-facing layer.

Typical flow:

• User opens a YouTube video
• Extension UI is triggered
• Comments are extracted / passed to backend
• Flask API predicts sentiment
• Results are shown back to the user

---

🧪 Model Development Journey

During experimentation, the project followed a practical model improvement path.

Baseline

• Random Forest baseline model
• Around 64% accuracy with 1000 features

Improvements Explored

• TF-IDF feature engineering
• Trigram features
• Tuning max_features
• Handling class imbalance using SMOTE
• Experimenting with stronger models such as:
  • XGBoost
  • Decision Trees
  • LightGBM

Final Direction

• Traditional NLP + TF-IDF pipeline
• LightGBM classifier
• MLOps pipeline for reproducibility and deployment

---

⚡ Setup Instructions

🔧 Prerequisites

Make sure you have the following installed:

• Python 3.9+ (recommended)
• Git
• pip
• DVC
• Docker (optional, for containerization)
• AWS CLI (if using remote MLflow / AWS setup)
• Google Chrome (for extension testing)

---

📥 1. Clone the Repository

git clone https://github.com/Umer2900/Youtube-Sentiment-Insights.git
cd Youtube-Sentiment-Insights

📦 2. Create Virtual Environment

Windows

python -m venv venv
venv\Scripts\activate

Mac/Linux

python3 -m venv venv
source venv/bin/activate

▶️ 3. Install Dependencies

pip install -r requirements.txt

▶️ 4. Run the DVC Pipeline

If the DVC pipeline is configured and dependencies are available:

dvc repro

▶️ 5. Run the Flask API

From the project root, run:

python flask_app/app.py

The API will typically run locally at: http://127.0.0.1:5000

🧩 6. Load the Chrome Extension

i. Open Chrome and go to:

chrome://extensions/

ii. Enable Developer mode

iii. Click Load unpacked

iv. Select the folder:

yt-chrome-plugin-frontend/

Now the extension can be tested in Chrome.

dvc repro

☁️ 7. AWS + MLflow Setup (Project Workflow)

If you want to replicate the MLOps setup used in this project, the general steps are:

Create AWS resources

• Create an IAM user
• Create an S3 bucket
• Launch an EC2 instance

Configure AWS CLI

Install and configure AWS CLI locally / on EC2:

aws configure

Add:

• AWS Access Key
• AWS Secret Key
• Region
• Output format

Start MLflow Server

Use the EC2 machine as the MLflow tracking server and S3 as artifact storage.

Note: Exact MLflow server commands may depend on your EC2 environment, bucket name, and tracking URI setup.

---

🙌 Acknowledgement

This project was built as an end-to-end MLOps + NLP + Deployment project covering:

• Traditional Machine Learning for NLP
• Text preprocessing and feature engineering
• Experiment tracking with MLflow
• DVC pipeline creation
• AWS integration
• Flask API development
• Chrome extension integration
• CI/CD and deployment concepts

It is designed as a hands-on project to demonstrate how a model moves from:

data → training → tracking → serving → frontend integration → deployment

---

📬 Contact

Mohammad Umer Jan
B.Tech CSE | Data Science Enthusiast | MLOps & Machine Learning Learner

• GitHub: Umer2900