Emotion-Classification-Tweets

Emotion Classification from Tweets Using TF-IDF and Logistic Regression

🧠 Emotion Classification in Tweets

This project focuses on building a multi-class emotion classifier for English-language tweets using a lightweight, interpretable NLP pipeline.

Instead of relying on deep learning or transformer-based models, this experiment uses TF-IDF vectorization and Logistic Regression to predict the emotion expressed in a tweet. It's simple, fast, and surprisingly effective — showing that classical methods still have strong value in NLP.

---

📌 Project Goal

• Detect the emotion behind a tweet (joy, sadness, anger, fear, love, or surprise)
• Use traditional NLP tools to keep the model interpretable and low-resource
• Serve as a baseline for future experimentation with deep learning

---

🗃️ Dataset

We use the mteb/emotion dataset from Hugging Face, which includes:

• ~20,000 tweets
• Six emotion labels: joy, sadness, anger, fear, love, surprise
• Pre-split into train, validation, and test sets

---

🔧 Methodology

1. Preprocessing
   Lowercasing, punctuation and stopword removal

2. Feature Engineering
   Text → TF-IDF representation

3. Model Training
   Logistic Regression classifier

4. Evaluation
   Accuracy, F1-score, and a confusion matrix

5. Visualization
   Emotion label distribution and confusion matrix heatmap

---

📊 Sample Confusion Matrix

![Confusion Matrix]

---

✅ Key Takeaways

• Achieved strong performance using only classical tools
• Model is fast to train, runs on CPUs, and is easy to understand
• Demonstrates that traditional NLP methods are still useful and relevant
• A solid starting point for more advanced models like BERT or RoBERTa

---

🛠️ How to Run

1. Clone the Repository

git clone https://github.com/isjustabhi/Emotion-Classification-Tweets.git
cd Emotion-Classification-Tweets