Natural Language to Optimization Modeling
Overview
The field studies how to turn written or multimodal requirements into decision variables, objectives, constraints, and executable models for operations research solvers. Approaches include structured intermediate representations and domain-specific languages, retrieval-augmented generation, multi-agent refinement, and post-training with solver execution or verification feedback. Benchmarks such as NL4OPT, MAMO, and IndustryOR test formulation accuracy and executability, while newer evaluations probe harder domains, industrial workflows, and computational efficiency. A central challenge is that code can run or match an objective value yet still misrepresent the intended constraints, especially when requirements are ambiguous or data must be grounded correctly.
Latest papers
76 papers14 papers in the last four weeks, up 40% on the four weeks before. 0.1% of all new papers.