🧠 Sentiment Classification Using Sentence Transformers

~ This project implements an embedding-based sentiment classification system that classifies Twitter tweets into Positive, Negative, or Neutral sentiments using Sentence Transformer embeddings and a machine learning classifier.

~ Unlike API-based solutions, this system works fully offline, making it efficient, scalable, and reproducible.

📌 Problem Statement

~ Social media platforms generate millions of posts daily, making manual sentiment analysis impractical. Understanding public sentiment helps brands, governments, and organizations make informed decisions.

🎯 Objective

~ The goal of this project is to build a sentiment classifier using:

1. Text preprocessing and cleaning

2. Transformer-based semantic embeddings

3. A machine learning classification model

📊 Dataset

~ Dataset: Twitter Tweets Sentiment Dataset ~ Size: ~27,000 tweets ~ Columns:

- textID

- text

- selected_text

- sentiment

~ Sentiment Labels:

 - Positive
  
 - Negative
  
 - Neutral

🔗 Dataset Link:

https://www.kaggle.com/datasets/yasserh/twitter-tweets-sentiment-dataset

🛠️ Technologies Used

1. Python
2. Pandas, NumPy
3. NLTK (text preprocessing)
4. Sentence Transformers (all-MiniLM-L6-v2)
5. Scikit-learn (Logistic Regression)
6. Matplotlib, Seaborn
7. WordCloud
8. VS Code (Jupyter Notebook)

🧠 Embedding Model Used

~ We use the lightweight transformer model:

all-MiniLM-L6-v2

~ Key Features:

384-dimensional semantic embeddings

~ Captures contextual meaning of sentences

Fast and lightweight

~ Works completely offline

~ No API dependency

🔄 Project Workflow

1. Exploratory Data Analysis (EDA)

2. Text preprocessing and cleaning

3. Word cloud visualization

4. Embedding generation using Sentence Transformers

5. Train-test split

6. Model training using Logistic Regression

7. Model evaluation using classification metrics

8. Custom tweet sentiment prediction

📈 Results

~ The model successfully classifies tweets into positive, negative, and neutral categories.

~ Transformer-based embeddings capture contextual meaning effectively.

~ The classifier performs strongly on short social media texts.

~ Custom user-defined tweets were accurately classified.

🧪 Sample Predictions

"I absolutely love this new phone!" → Positive

"This service is horrible and frustrating" → Negative

"The event happened yesterday" → Neutral

🚀 How to Run the Project

1️⃣ Clone the repository git clone https://github.com/coderShreyIn/Sentiment_Classification_Using_Embeddings.git cd Sentiment_Classification_Using_Embeddings 2️⃣ Install dependencies pip install -r requirements.txt 3️⃣ Run the Jupyter Notebook

Open in VS Code or Jupyter:

jupyter notebook

Run all cells sequentially.

📦 No API Key Required

This project uses offline Sentence Transformers, so:

❌ No Gemini API key needed

❌ No rate limits

❌ No internet dependency after first model download

📊 Model Performance

• Logistic Regression trained on transformer embeddings

• High accuracy on multi-class sentiment classification

• Good generalization on unseen tweets

⚠️ Limitations

• Sarcasm detection is challenging

• Very short texts may reduce classification confidence

• Mixed sentiment sentences can cause ambiguity

🔮 Future Improvements

• Fine-tune a transformer model directly for sentiment classification

• Add confidence score display

• Implement hyperparameter tuning

• Deploy as a web app (Streamlit/Flask)

• Add real-time tweet streaming analysis

🌍 Real-World Applications

• Social media sentiment monitoring
• Brand reputation analysis
• Customer feedback analytics
• Political opinion mining
• Product review classification
• Chatbot emotion detection

👨‍💻 Author

~ Shrey Dak

AI & Machine Learning Enthusiast GitHub: https://github.com/coderShreyIn

⭐ If You Found This Useful

~ Please consider giving this repository a ⭐ on GitHub!