GraphThought
收藏资源简介:
GraphThought是一个新颖的框架,设计用于图组合优化问题,通过推理思维生成来微调大型语言模型。该框架由两个核心模块组成:选择器和构造器,它们协作生成特定任务的推理程序。GraphThought框架利用前向和后向元思维编程方法来获取算法知识,分别编码显式算法知识和隐式问题解决模式。该数据集是针对图组合优化问题设计的,旨在提高大型语言模型解决复杂任务时的推理和规划能力。
GraphThought is a novel framework tailored for graph combinatorial optimization problems, which fine-tunes large language models (LLMs) through generative reasoning thought processes. The framework comprises two core modules: a selector and a constructor, which collaborate to generate task-specific reasoning procedures. The GraphThought framework utilizes forward and backward meta-thinking programming methodologies to acquire algorithmic knowledge, where the former encodes explicit algorithmic knowledge and the latter encodes implicit problem-solving patterns. This dataset is designed for graph combinatorial optimization problems, with the goal of enhancing the reasoning and planning capabilities of LLMs when addressing complex tasks.

- 1GraphThought: Graph Combinatorial Optimization with Thought Generation华东师范大学计算机科学与技术学院 · 2025年



