Head: A Modular Emotion-Aware Middleware for LLM-Powered Chatbots

🌟 Overview

Head is a modular middleware architecture designed to make Large Language Models (LLMs) emotionally aware. Instead of retraining or fine‑tuning large models, Head acts as a plug‑and‑play component that analyzes user input, extracts emotional signals, and passes emotion‑tagged metadata to any downstream chatbot model.

This repository contains:

• 🧠 Head Module — A BERT-based multi‑label emotion classifier trained on GoEmotions.
• 💬 Emotion-Aware Chat Application — A Streamlit UI that integrates the Head module with a locally running LLM using Ollama.
• ⚙️ Middleware Architecture — Clean, modular design for easy integration into any chatbot system.

Our goal is to create more empathetic, context‑aware, and human‑aligned conversational agents.

---

🚀 Features

• Multi-label Emotion Classification Powered by a fine‑tuned BERT model on GoEmotions (simplified variant, 28 emotions).

• Probabilistic Emotion Outputs Each input message returns emotion scores with threshold-based multi-label selection.

• Plug-and-Play Middleware Head can sit between any input and chatbot model.

• Interactive Streamlit UI Complete frontend for testing and demonstrating emotion-aware interactions.

---

🧩 Architecture

User → Head (Emotion Extractor) → Emotion Tags → LLM (Ollama) → Chatbot Response

🔹 Head Module

• Built on AutoModelForSequenceClassification from HuggingFace.
• Multi-label BCEWithLogits setup.
• Thresholding per emotion class.

🔹 LLM Response Generator

• Uses Hugging Face Chat Completion APIs.
• Injects emotion tags + user text into the prompt.

🔹 Why Middleware?

• No need to fine‑tune LLMs.
• Works with any model.
• Modular, interpretable, and expandable (audio/video/physiological modalities later).

---

🖥️ Streamlit Demo

The app provides:

• Input text box
• Predicted emotions + probabilities
• Emotion-aware LLM response
• Debug panel for prompt inspection

Run with:

streamlit run app.py

---

📦 Installation

1. Clone the repo

git clone https://github.com/2RAJARYAN/EmotionalChat_web.git
cd EmotionalChat_web

2. Create a vene

python -m venv .venv

3. Install dependencies

pip install -r requirements.txt

4. Setup .streamlit/secrets.toml

• Create a Folder name .streamlit
• A File called secrets.toml
• This store the HF_TOKEN, EMOTION_MODEL_ID, LLM_MODEL_ID

---

🧠 Future Scope

• Multimodal Head (text + audio + video)
• Novel emotion decoding algorithms
• Benchmark suite for emotional coherence and empathy
• Publication in NLP/affective computing venues

---

📄 License

MIT License.

---

🤝 Contributing

Pull requests are welcome! For major changes, please open an issue first.

---

⭐ Acknowledgements

• Google’s GoEmotions dataset
• HuggingFace Transformers

---

If you like this project, consider giving us a ⭐ on GitHub!