COMP0087: Emotion AI Team Repository

Refining Empathy in Chatbots: Fine-Tuning, In-Context Learning, and the Critical Role of Human Evaluation

Project Overview

This repository contains the work of the Emotion AI team for the COMP0087: Statistical Natural Language Processing module at University College London for the term T2 of the academic year 23/24.

Directory Structure

• Baseline/: Initial models and baseline algorithms for the ESConv: DialoGPT and BlenderBot.
• ESConv Data/: Dataset of emotional support conversations for model training and test.
• Finetuning/: Scripts and results from fine-tuning our models for improved empathetic responses.
• In-Context Learning/: Experiments and findings on in-context learning to the chatbot understanding.
• Human Evaluation/: Methodology and results from human evaluation studies assessing chatbot performance.