Henghui Xu

GESR introduces a graph-based framework that reconstructs edge semantics from local structural context to detect stealthy malicious communications using only benign training data, achieving high performance on standard datasets.