NLP-Emotion-Classification

Project Overview

This repository contains the implementation of a text classification project for identifying emotions (e.g., joy, sadness, fear, anger) in tweet texts. The project explores three deep learning approaches:

1. Fully Connected Neural Network (FCNN)
2. Recurrent Neural Network (RNN) with LSTM/GRU
3. Fine-tuned Transformer Model (e.g., BERT from HuggingFace)

The dataset consists of labeled tweet texts split into train.txt, test.txt, and validation.txt. The goal is to compare the performance of these models and analyze their effectiveness for emotion classification.

This project was developed as part of an NLP Exam by Alessandro Mencarelli, Mauricio Rodriguez, and Mario Zuna.

---

Repository Structure

NLP-Emotion-Classification/
│
├── data/
│   ├── train.txt
│   ├── test.txt
│   └── validation.txt
│
├── notebooks/
│   └── Emotion_Classification_Notebook.ipynb
│
├── models/
│   ├── fcnn_model.h5
│   ├── rnn_model.h5
│   └── transformer_model/
│
├── README.md
└── requirements.txt

---

Notebook

The main notebook, Emotion_Classification_Notebook.ipynb, contains:

• Data Preprocessing: Tokenization, padding, and encoding of tweet texts.
• Model Implementation:
  • Fully Connected Neural Network (FCNN)
  • Recurrent Neural Network (RNN) with LSTM/GRU
  • Fine-tuned Transformer Model (BERT)
• Results Comparison: Analysis of model performance and insights.

---

Requirements

To run the notebook, install the required dependencies:

pip install -r requirements.txt

---

How to Use

1. Clone the repository:

   git clone https://git.1-hub.cnmenca-lsn/NLP-Emotion-Classification.git
   cd NLP-Emotion-Classification

2. Install dependencies:

   pip install -r requirements.txt

3. Open the notebook:

   jupyter notebook notebooks/Emotion_Classification_Notebook.ipynb

4. Follow the instructions in the notebook to preprocess the data, train the models, and evaluate their performance.

---

Dataset

The dataset consists of tweet texts labeled with emotions (e.g., joy, sadness, fear, anger). It is split into:

• Train set: Used for training the models.
• Test set: Used for evaluating model performance during development.
• Validation set: Used for final performance evaluation to ensure no overfitting.

---

Models

1. Fully Connected Neural Network (FCNN):

   • A simple feedforward neural network for baseline performance.

2. Recurrent Neural Network (RNN):

   • Uses LSTM/GRU layers to capture sequential dependencies in text data.

3. Transformer Model (BERT):

   • Fine-tunes a pretrained BERT model from HuggingFace for state-of-the-art performance.

---

Results

• FCNN: Achieved 0.82 accuracy.
• RNN: Achieved 0.89 accuracy.
• Transformer (BERT): Achieved 0.92 accuracy, leveraging pretrained language representations for the best performance.

---

Happy coding! 🚀