Reproduce “Causal Discovery and Counterfactual Reasoning to Optimize Persuasive Dialogue Policies” 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
Donghuo Zeng, Roberto Legaspi, Yuewen Sun, Xinshuai Dong, Kazushi Ikeda, Peter Spirtes, Kun Zhang
This paper presents a framework that integrates causal discovery and counterfactual reasoning to optimize dialogue policies for persuasive outcomes. It identifies strategy-level causal mechanisms in conversations and generates realistic alternative scenarios, leading to an increase of $31.16 in predicted donations compared to ground truth data.
Key insights
- Causal discovery using GRaSP identified specific cause-effect relationships between persuader and persuadee strategies, such as "Logical appeal" often following persuadee strategies like "positive-to-inquiry".
- Integrating causal understanding from discovered relationships into counterfactual data generation (via BiCoGAN) produces more realistic and relevant alternative dialogue scenarios.
- Training a D3QN policy on a causally-enhanced counterfactual dataset leads to improved persuasion outcomes compared to training on randomly generated counterfactuals.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.