BERT-based Emotion Recognition Pipeline

Overview

This project provides a fully reproducible pipeline for text-based emotion detection using BERT and modern NLP preprocessing. Includes:

• Dataset merging
• Advanced preprocessing (including negation handling)
• Automatic negation-based data augmentation
• Model training and interactive/manual inference (CLI or web)

---

Features

• Unified data preparation script: Combines, preprocesses, and augments emotion datasets.
• Flexible CSV input: Accepts and merges standard emotion datasets (e.g. train.csv, test.csv, val.csv).
• Preprocessing: Lemmatization, proper stopword handling, explicit negation marking.
• Augmentation: Adds robust negated phrase samples for each emotion class to boost performance on tricky text.
• BERT workflow: Scripted, interactive, and web app interfaces for batch or real-time emotion detection.

---

Repository Structure

├── data/
│ ├── train.csv
│ ├── test.csv
│ ├── val.csv
│ ├── final_data.csv # Optional: raw/merged input
│ └── final_data_aug.csv # Output: run after prepare_data.py
├── prepare_data.py # Data pipeline: load, preprocess, augment
├── bert_emotion.py # Model training & CLI inference
├── requirements.txt
├── README.md
└── bert_emotion_model/
└── ... (saved model after training)

---

Setup

1. Install requirements

pip install -r requirements.txt

2. Download NLTK Data (first time only)

import nltk
nltk.download('punkt')
nltk.download('stopwords')
nltk.download('wordnet')
nltk.download('omw-1.4')

---

Data Preparation

Run the unified script to preprocess and augment data: python prepare_data.py

• Output: data/final_data_aug.csv for model training.

---

Model Training

Train the BERT emotion classifier interactively: python bert_emotion.py

• Choose 1 for model training.
• The trained model and tokenizer are saved in bert_emotion_model/.

---

Inference

Command-line emotion detection:

python bert_emotion.py

• Choose 2 for interactive detection.
• Type sentences and press Enter for results.

(Optional) Web App:

• If you have app.py (Streamlit), run:

streamlit run app.py

---

Requirements

pandas
nltk
torch
transformers
scikit-learn
accelerate

Add streamlit if you demo as a web app.

---

Tips

• Add or tune emotion labels in prepare_data.py as needed.
• Full GPU support if PyTorch w/ CUDA is installed.
• For other languages/datasets, adapt preprocessing and label mapping.

---

License

MIT License (or specify your license)

---

Questions? PRs and issues welcome!