Ask Before You Optimize: Dynamic Pre-Formulation Clarification for Interactive Optimization
A research paper introduces OR-Clarify, a benchmark for pre-formulation clarification in optimization modeling with LLMs, and InterOPT, a two-stage framework that identifies unresolved formulation-critical gaps to guide questioning. In choice-based experiments, InterOPT substantially outperforms all baselines in exact slot recovery; in open-ended setting, it remains competitive.
Large language models are increasingly used to formulate optimization models from natural-language problem descriptions, but realistic operations research requests are often incomplete. The paper introduces OR-Clarify, a benchmark for pre-formulation clarification, and InterOPT, a two-stage framework that identifies unresolved formulation-critical gaps and uses them to guide whether to ask the next question or stop. In choice-based experiments, InterOPT substantially outperforms all baselines in exact slot recovery; in the open-ended setting, it remains competitive.
The InterOPT framework uses a two-stage process: first identifying unresolved formulation-critical gaps, then using those gaps to decide whether to ask another clarifying question or stop. The benchmark supports both open-ended and choice-based clarification, measuring slot recovery, stopping behavior, silent assumptions, and interaction cost. The reported substantial improvement in exact slot recovery in choice-based experiments suggests the approach effectively reduces silent assumptions and improves model completeness.
This research addresses a practical gap in applying LLMs to operations research: real-world problem statements are often incomplete, and agents must know when to ask for clarification before building a model. The benchmark and framework could influence how AI assistants for optimization are evaluated and built, potentially leading to more reliable interactive modeling tools in enterprise settings.
For enterprises using AI to translate business problems into optimization models, this work could reduce errors from silent assumptions and improve model accuracy, potentially lowering the cost of manual review and increasing trust in AI-generated models. It may also enable more autonomous deployment of optimization agents in operations research workflows.
Next observable signals include adoption of OR-Clarify in subsequent LLM agent evaluations, extensions of InterOPT to other domains requiring clarification, and integration of clarification mechanisms into commercial optimization modeling assistants. Further research may explore open-ended clarification performance and interaction cost trade-offs.