CCSG(Controlled Causal-Semantic Graph)
收藏资源简介:
CCSG数据集是一个合成图数据集,包含了可控的因果语义关系,用于分析LLM增强器与GNN模型之间的信息传递。该数据集基于维基百科条目,具有丰富的语义节点属性和可控的拓扑结构。数据集的构建旨在模拟复杂的语义关联,并允许对内部因果关系的精确操作,以便评估模型捕捉和表示关键信息的能力。通过互换干预方法,可以系统地分析LLM增强器与GNN模型之间的对应关系,揭示其内部逻辑结构。
The CCSG dataset is a synthetic graph dataset containing controllable causal semantic relations, designed to analyze information transfer between LLM augmenters and GNN models. Based on Wikipedia entries, this dataset boasts rich semantic node attributes and controllable topological structures. The dataset is constructed to simulate complex semantic associations, enabling precise manipulation of internal causal relations to evaluate models' capability in capturing and representing critical information. Through swap-based intervention approaches, the correspondence between LLM augmenters and GNN models can be systematically analyzed, uncovering their internal logical structures.
LLMEnhCausalMechanism数据集概述
基本信息
- 数据集名称:LLMEnhCausalMechanism
- 托管地址:https://github.com/WX4code/LLMEnhCausalMechanism
当前状态
- 代码尚未上传,README仅包含占位信息
备注
- 该数据集详情页面目前仅包含一个占位声明,无实质性内容

- 1LLM Enhancers for GNNs: An Analysis from the Perspective of Causal Mechanism Identification中国科学院软件研究所 · 2025年



