<p>Anomaly detection in blockchain transaction networks is critical for ensuring the security and integrity of decentralized systems. Existing approaches often fall short in effectively capturing the intricate temporal dynamics and relational structures inherent in blockchain data. In this paper, we introduce Gated Temporal Knowledge Network (GTKNet), a framework designed to address these challenges by integrating temporal dynamics and relational graph structures through a gated mechanism. GTKNet leverages both temporal dependencies and spatial relationships within transaction networks, enabling it to identify anomalous patterns with high precision and robustness. Comprehensive evaluations on the Elliptic and Augmented Elliptic datasets demonstrate that GTKNet consistently outperforms representative traditional machine learning models and existing graph-based approaches across multiple metrics. Beyond superior performance, GTKNet exhibits robust performance and resilience under varying dataset sizes, making it well-suited for deployment in real-world blockchain environments. These attributes underscore its potential applicability to other domains requiring robust anomaly detection in large-scale and evolving graph structures, such as social network analysis and financial fraud detection.</p>