2273. How LLM Reshaped Complex OR Problem Solving?
Invited abstract in session TD-34: Advancements of OR-analytics in statistics, machine learning and data science 5, stream Advancements of OR-analytics in statistics, machine learning and data science.
Tuesday, 14:30-16:00Room: Michael Sadler LG10
Authors (first author is the speaker)
| 1. | YUAN WANG
|
| school of business, Singapore University of Social Sciences |
Abstract
Operations Research (OR) seeks to develop mathematical models for addressing complex decision-making problems encountered across various industry sectors. To streamline the process and minimize reliance on domain-specific modeling experts, NL4Opt (Natural Language for Optimization) (Ramamonjison et al., 2022a) has recently been introduced as a promising yet challenging task within the field of Natural Language Processing (NLP). The goal of NL4Opt is to convert textual descriptions of OR problems into mathematical formulations suitable for optimization solvers. In this paper, we investigate the automatic modeling and programming of complex OR problems derived from real-world industrial demands. We propose the first solution based on large language models (LLMs), termed Chain-of-Experts (CoE), a novel multi-agent cooperative framework designed to improve reasoning capabilities. In this framework, each agent is designated a specific role and equipped with domain-specific knowledge related to Operations Research (OR). Additionally, we introduce a "conductor" that coordinates the agents using a mechanism of forward thought construction and backward reflection. To further advance OR research and foster community engagement, we develop a benchmark dataset, ComplexOR, which comprises intricate OR problems. Experimental results demonstrate that CoE significantly outperforms the current state-of-the-art LLM-based approaches.
Keywords
- Artificial Intelligence
- Decision Support Systems
- Complexity and Approximation
Status: accepted
Back to the list of papers