Reproduce “Personalized Decision Modeling: Utility Optimization or Textualized-Symbolic Reasoning” 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
Yibo Zhao, Yang Zhao, Hongru Du, Hao Frank Yang
Researchers from Johns Hopkins University developed ATHENA, a two-stage framework that combines symbolic utility functions with Large Language Model-driven semantic adaptation to model personalized human decisions. This approach achieves over 6% higher accuracy and F1-score on real-world datasets like Swissmetro compared to leading baselines, while also providing interpretable decision logic.
Key insights
- The co-design of symbolic utility modeling and LLM-powered semantic adaptation effectively bridges the gap between interpretable, population-level theories and personalized, context-rich human choices.
- Group-level symbolic utility discovery provides a robust, interpretable foundation that significantly prunes the hypothesis space, while individual-level semantic adaptation is essential for capturing unique intra-group heterogeneity.
- Lightweight LLMs can achieve near-maximal performance within ATHENA's structured framework, indicating that architectural design, not just raw LLM capacity, drives effectiveness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.