Reproduce “Towards Persona-Based Empathetic Conversational Models” 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
Peixiang Zhong, Chen Zhang, Hao Wang, Yong Liu, Chunyan Miao
Researchers introduce the task of persona-based empathetic conversations and present a new large-scale dataset, PEC, collected from social media. The work develops CoBERT, a BERT-based response selection model with multi-hop co-attention, which achieves superior performance in selecting persona-aware empathetic responses and empirically demonstrates that persona has a more substantial impact on empathetic interactions compared to casual conversations.
Key insights
- Persona significantly influences how empathy is expressed, as empathetic responses from different persons in the PEC dataset were more distinct than random empathetic responses, suggesting individual 'styles'.
- Both first-order and second-order matching information captured by multi-hop co-attention are crucial for effectively integrating context and persona, with first-order being more critical.
- The impact of persona on conversational model performance is empirically found to be more pronounced in empathetic conversations than in non-empathetic (casual) interactions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.