The paper presents a novel framework, CAFFEINE, aimed at improving response generation in long-term conversations by refining speaker personas. Traditional methods often yield uninformative persona sentences that limit response quality. CAFFEINE uses commonsense-based persona expansion to transform contradictory personas into context-rich speaker information. This approach enables better response generation in long-term conversations, providing a more human-like interaction. The authors claim to be the first to explore commonsense-based persona expansion in multi-session settings.
Publication date: 26 Jan 2024
Project Page: https://caffeine-15bbf.web.app/
Paper: https://arxiv.org/pdf/2401.14215
