Natural language interfaces significantly enhance the accessibility and usability of optimization modeling. Recent advances in large language models (LLMs) demonstrate potential in automatically translating textual problem descriptions into executable solver formulations. However, a key challenge for existing approaches is ensuring that the inferred formulation correctly implements the intended task, even if it may execute without errors. To address this, we introduce VeriSimpl, a solver LLM framework for robust natural language-to-optimization formalization.
Our approach is based on the idea of simplification-based verification, where the optimization solver generates simplified diagnostic queries about a candidate formulation, allowing the LLM to reason about the correctness of the formulation under fixed global contexts.
We present various simplification strategies concerning problem constraints and decision variables that enable local reasoning by the LLM.
Evaluations on a range of optimization benchmarks show that our approach consistently improves accuracy over existing methods while providing a novel high-precision self-verification signal.
Blogger's Review: The introduction of VeriSimpl offers a fresh perspective on natural language optimization modeling, significantly enhancing model accuracy and reliability through simplification verification strategies. This innovative approach not only optimizes the solving process but also provides a solid theoretical foundation for LLM applications in complex problems. Its self-verification mechanism is particularly noteworthy, opening new avenues for future research.