Reproduce “You Impress Me: Dialogue Generation via Mutual Persona Perception” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qian Liu, Yihong Chen, Bei Chen, Jian-Guang Lou, Zixuan Chen, Bin Zhou, Dongmei Zhang
Researchers from Beihang University, UCL, Fudan University, and Microsoft Research developed P2BOT, a dialogue generation framework that explicitly models mutual persona perception between interlocutors. It achieved state-of-the-art results on the PERSONA-CHAT dataset, significantly improving dialogue quality, coherence, and persona consistency compared to baselines in both automatic and human evaluations.
Key insights
- Explicitly modeling mutual persona perception, integrated as a reward signal, significantly enhances dialogue agents' ability to generate consistent and engaging responses.
- The Receiver component, designed to quantify persona relevance from dialogue, effectively captures fine-grained correlations between utterances and persona facts, even for rephrased personas.
- Combining supervised pre-training with reinforcement learning, utilizing a composite reward function that includes language style, discourse coherence, and mutual persona perception, yields superior dialogue generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.