The study presents a new framework, ‘CAFFEINE’, to resolve issues in long-term conversation. It focuses on refining contradictory personas into sentences that contain rich speaker information based on their contextual backgrounds. As the first to explore commonsense-based persona expansion in multi-session settings, CAFFEINE leverages large language models to iteratively refine the contradictory personas, facilitating better response generation in long-term conversations.
Publication date: 26 Jan 2024
Project Page: https://caffeine-15bbf.web.app/
Paper: https://arxiv.org/pdf/2401.14215
