New Survey Maps GNN Methods Across Entire Knowledge Graph Pipeline

Graph Neural Networks (GNNs) have become a cornerstone for modeling graph-structured data, yet a systematic review of their role across the entire Knowledge Graph (KG) technology pipeline was missing. A new survey accepted for publication in ACM Computing Surveys fills that gap. The authors—Chengcheng Sun, Jiayun Tian, Cheng Zhai, Zhixiao Wang, Yajie Song, Xiaobin Rui, Jian Zhang, and Philip S. Yu—propose a novel two-level taxonomy. The first level follows the KG pipeline: knowledge graph construction, knowledge graph embedding, knowledge reasoning, and knowledge graph applications. The second level categorizes methods by GNN model type, including Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and Heterogeneous Graph Neural Networks (HGNNs). This dual framework allows researchers to understand which GNN variant fits each stage of the KG lifecycle.

The survey systematically reviews state-of-the-art GNN-based models under this taxonomy, evaluating their advantages based on task characteristics (e.g., link prediction vs. relation extraction). It also summarizes strengths and limitations for each approach. Looking forward, the authors identify unresolved challenges such as scalability to massive KGs, dynamic graph handling, and integration with large language models. The paper provides a valuable roadmap for both newcomers and experienced researchers, with a linked GitHub repository tracking relevant papers. By unifying disparate threads of research, this survey serves as a definitive reference for advancing AI systems that rely on structured knowledge.