GASN: a graph attention and self-attention network for blockchain service reliability prediction

Abstract With the rapid evolution of cloud-based service-oriented computing, Blockchain as a Service (BaaS) has become a promising paradigm for deploying decentralized applications on cloud infrastructures. However, the increasing number of blockchain services offered by heterogeneous peer nodes introduces significant challenges in selecting reliable services with superior Quality of Service (QoS). Moreover, the scarcity of personalized QoS data and the high sparsity of existing reliability datasets further complicate accurate service prediction. To address these challenges, this paper proposes GASN (Graph Attention and Self-Attention Network), a hybrid deep learning framework for cloud-based blockchain service reliability prediction. The proposed architecture integrates a multi-layer graph attention mechanism with a self-attention module to capture both inter-service dependencies and global contextual features, thereby enhancing feature representation and mitigating data sparsity. Extensive experiments on a large-scale real-world blockchain service dataset demonstrate that GASN consistently outperforms state-of-the-art benchmark models, reducing the prediction error (MAE and RMSE) by up to 32.6% and 22.9%, respectively, providing an effective solution for reliability-aware blockchain service selection in cloud environments.