HCI Lab Negotiation Agent (Nego-interface)

This repository contains a Streamlit-based interface for online negotiation experiments in HCI Lab at Gachon University. Participants take the role of buyer or seller and negotiate with an LLM-based agent. The system logs the dialogue, outcomes, and subjective evaluations for research analysis.

Note: Due to the study context and target population, all in-app prompts and negotiation content are written in Korean.

Main Features

• Streamlit web UI (setup screen + chat screen)
• LangGraph-based negotiation workflow
  • setup → negotiator → logger
• Multiple prompting strategies
  • Baseline – basic agent
  • CoT_previous – CoT + ICL
  • CoT_upgrade – CoT + ICL + few-shot + RAG + stronger negotiation strategy guidance
• RAG tool (policy_search_tool) for searching Naver Pay refund/exchange policies
• Firebase logging of negotiation traces and metadata
• Post-negotiation human coding + subjective satisfaction questionnaire UI

Tech Stack

• Python 3.10+
• Streamlit
• LangChain / LangGraph
• OpenAI / Anthropic LLMs (e.g., gpt-4o, Claude 3 Sonnet)
• ChromaDB (vector DB for Naver Pay policy)
• Firebase Admin (for research data logging)

All required packages are listed in requirements.txt.

How to Run (minimal)

This project is under active research development; details may change. For now, a minimal way to start the app is:

pip install -r requirements.txt
streamlit run app.py

External services (Firebase, RAG DB, API keys, etc.) are configured directly in the code (e.g., core/helpers.py, tools/rag_tools.py) and may evolve without being fully reflected in this README.

Logs & Research Data

• By default, negotiation results are stored in Firebase Firestore (collection negotiation_results), including:
  • full dialogue, per-turn metadata (speaker, utterance, agent "thought", tool calls)
  • buyer/seller goals, scores, rewards, human-coded outcome, survey responses
  • optional Pareto plot image as base64 (pareto_image field)
• Firebase requires a valid serviceAccountKey.json and a reachable Firestore project; if initialization fails, results are not saved.
• A legacy CSV export helper exists but is not used by default in the current UI.

Notes

• This codebase is a research prototype, not a production-ready system.
• For real user studies, please carefully review your IRB/ethics requirements, data retention policy, and any personal data handling.