conversational-agent-stress-trauma

🤖 LSTM-Aware Emotion and Stress Detection Chatbot

This is a Colab-compatible Python chatbot that integrates:

It predicts the user’s emotion and stress level using both BERT and LSTM models and gives human-like responses.

🧠 Project Overview

The chatbot performs the following tasks:

  1. Takes user input via command line.
  2. Detects emotion using:
    • BERT model (bhadresh-savani/distilbert-base-uncased-emotion)
    • LSTM model trained on past conversations
  3. Infers stress level from detected emotion.
  4. Replies using BlenderBot (facebook/blenderbot-400M-distill).

📁 Files in This Repository

File Description
chatbot.py Full Python script to run the chatbot
lstm_model.h5 Pre-trained LSTM model (upload to Colab)
tokenizer.pkl Tokenizer fitted on the training dataset for LSTM
requirements.txt All necessary Python packages
example_inputs_outputs.txt Sample I/O showing how the chatbot works
.gitignore Ignore compiled files, model weights, cache, etc.

🚀 Setup Instructions

✅ Requirements

Install required packages using:

pip install -r requirements.txt

Or manually:

pip install transformers torch torchvision torchaudio tensorflow scikit-learn

▶️ How to Run (Colab or Local)

  1. Clone or download the repo.
  2. Upload the files lstm_model.h5 and tokenizer.pkl when prompted.
  3. Run the chatbot.py script:
python chatbot.py

In Google Colab, use files.upload() to upload the two files.

🧠 Emotion & Stress Detection Logic

Emotion Classes:

Stress Mapping:

Emotion Stress
anger High
fear High
sad High
neutral Low
happy Low
surprise Low

💬 Sample Interaction

You: I'm feeling down today.

 Bot: I'm here for you. Do you want to talk about it?
 BERT Emotion: sadness (0.92) → Stress: High
 LSTM Emotion: sad (0.88) → Stress: High

You: exit
 Bot: Take care! 💙

📌 Models Used

⚖️ License

This project is licensed under the MIT License.

🙋‍♀️ Author

Created by Harini Muthukumar

Feel free to star ⭐ the repository and contribute!

This site is open source. Improve this page.