一句话：提出 GAER：一种无监督图表示学习方法，从源代码学习架构感知表示并聚类以恢复软件架构模块。
软件架构恢复对程序理解、维护与重构至关重要，尤其在大型系统中架构文档常不完整或过时。随着软件演化，架构漂移与侵蚀会使实现偏离预期设计，人工恢复架构模块费时且难以扩展。本文提出 GAER，一种无监督图表示学习方法。它从源代码为软件实体学习架构感知表示，再对这些表示进行聚类，从而恢复架构模块。摘要强调图表示学习可作为可扩展、数据驱动的软件架构恢复基础，但未提供具体实验细节，主要定位为方法论贡献。
Software architecture recovery is important for program understanding, maintenance, and re-engineering, particularly in large-scale systems where architectural documentation is often incomplete or outdated. As software evolves, architectural drift and erosion can cause the implemented system to diverge from its intended design, making manual recovery labor-intensive and difficult to scale. We present GAER, an unsupervised graph-based approach for software architecture recovery. The approach learns architecture-aware representations of software entities from source code and clusters them to recover architectural modules. GAER highlights the promise of graph representation learning as a scalable and data-driven foundation for software architecture recovery.