GCoder: Improving Large Language Model for Generalized Graph Problem Solving

• DOI: 10.48550/arxiv.2410.19084
• Publication Date: 2024-10-24
• Authors: Qifan Zhang; Xiaobin Hong; Jianheng Tang; Nuo Chen; Yuhan Li; Wenzhong Li; Jing Tang; Jia Li
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Computation and Language (cs.CL)

Large Language Models (LLMs) have demonstrated strong reasoning abilities, making them suitable for complex tasks such as graph computation. Traditional steps paradigm problems is hindered by unverifiable steps, limited long-term reasoning, and poor generalization to variations. To overcome these limitations, we introduce GCoder, a code-based LLM designed enhance problem-solving in generalized computation problems. Our method involves constructing an extensive training dataset, GraphWild, featuring diverse formats algorithms. We employ multi-stage process, including Supervised Fine-Tuning (SFT) Reinforcement Learning from Compiler Feedback (RLCF), refine model capabilities. For unseen tasks, hybrid retrieval technique used augment performance. Experiments demonstrate that GCoder outperforms GPT-4o, with average accuracy improvement of 16.42% across various computational Furthermore, efficiently manages large-scale graphs millions nodes input formats, overcoming the limitations previous models focused on paradigm. This advancement paves way more intuitive effective using LLMs. Code data are available at here: https://github.com/Bklight999/WWW25-GCoder/tree/master.