Reproduce “DeepRule: An Integrated Framework for Automated Business Rule Generation via Deep Predictive Modeling and Hybrid Search Optimization” 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
Yusen Wu, Xiaotie Deng
DeepRule, an integrated framework from Peking University, addresses the optimality-operationality gap in retail assortment and pricing by combining LLM-driven knowledge fusion, game-theoretic optimization, and interpretable symbolic regression. The system achieved approximately 2% higher profit compared to competitive baselines in a real-world setting, while generating auditable business rules.
Key insights
- Integrating LLMs as meta-optimizers for symbolic regression significantly enhances the efficiency and effectiveness of discovering interpretable business rules.
- The general analytical and reasoning capabilities of current-generation LLMs are sufficient to drive performance gains in the framework, rather than relying on specialized mathematical competencies of a single model.
- Decoupling function structure generation (via LLMs) from parameter optimization (via gradient-based methods) helps manage the complexity of finding optimal, interpretable strategies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.