The field of graph learning and fraud detection is rapidly evolving, with a focus on developing innovative methods for self-supervised learning, node influence detection, and anomaly detection. Recent research has highlighted the importance of dataset size and task difficulty in evaluating the performance of graph contrastive learning methods. Additionally, there is a growing interest in developing more robust and classifier-agnostic complexity measures for knowledge graph link prediction evaluation.

Noteworthy papers in this area include: Graph Contrastive Learning versus Untrained Baselines: The Role of Dataset Size, which investigates the role of dataset size in graph contrastive learning. Contrastive clustering based on regular equivalence for influential node identification in complex networks, which proposes a novel deep unsupervised framework for influential node identification. KGBERT4Eth: A Feature-Complete Transformer Powered by Knowledge Graph for Multi-Task Ethereum Fraud Detection, which introduces a feature-complete pre-training encoder for Ethereum fraud detection. LMAE4Eth: Generalizable and Robust Ethereum Fraud Detection by Exploring Transaction Semantics and Masked Graph Embedding, which proposes a multi-view learning framework for Ethereum fraud detection. Enhancing Self-Supervised Speaker Verification Using Similarity-Connected Graphs and GCN, which improves self-supervised speaker verification using similarity-connected graphs and GCN.