Graph Sequential Self-Attention for Microservice Recommendation

Microservice workflows enable efficient management of complex business processes by clearly defining tasks and information flows. However, traditional Graph Neural Networks (GNNs) struggle to capture long-term dependencies and temporal information due to propagation limitations, often focusing only on short-term interests based on the last node of the workflow. In this paper, we propose the Graph Sequence Self- Attention (GSSAT) model to address these issues. GSSAT integrates spatial and temporal information by extracting spatial structures with GNNs and capturing temporal dependencies using sequence models, generating comprehensive vector representations for each node. Additionally, it considers both short-term interests from the last workflow node and long-term interests derived via a multi-head attention mechanism. By combining these interests, GSSAT enhances recommendation accuracy, effectively addressing workflow orchestration challenges.