一句话：提出融合静态分析、动态追踪、图神经网络与约束多目标优化及 LLM Agent 的单体自动微服务拆分方法。
该工作针对单体应用向微服务架构迁移中服务边界难以自动、高质量划分的问题，提出自动化工具 MicroWeaver。方法上，它构建了四阶段流水线：输入构建阶段结合静态代码分析与动态运行时追踪，抽取结构与语义图；微服务划分阶段使用图神经网络进行结构编码、配合语义嵌入，并在约束下做多目标优化；随后进行多维度评估与交互式可视化；此外还引入基于 LLM Agent 的迭代优化以精化划分结果。工件面向 Java 单体，覆盖 DayTrader、AcmeAir、JPetStore、Plants 与 TrainTicket 五个代表性基准系统，提供预处理图数据、预训练图编码器、约束优化求解器、Agent 模块以及复现脚本、配置与环境说明，以支持实验结果的可复现与透明验证，服务于智能软件架构拆分与云原生迁移研究。
Description for Zenodo: This repository contains the source code, datasets, experimental scripts, and supplementary materials for MicroWeaver, an automated tool for monolithic application microservice decomposition. MicroWeaver integrates static code analysis, dynamic runtime tracing, graph neural network-based structural encoding, semantic embedding, constrained multi-objective optimization, and LLM Agent-based iterative optimization to automatically partition Java monoliths into high-quality microservice architectures. The repository includes: Complete implementation of the four-stage pipeline: input construction, microservice partitioning, multi-dimensional evaluation, and interactive visualization. Preprocessed structural and semantic graph data for five representative benchmark systems: DayTrader, AcmeAir, JPetStore, Plants, and TrainTicket. Pre-trained graph encoders, constraint optimization solvers, and LLM Agent modules for result refinement. Reproduction scripts, configuration files, and environment specifications to support full experiment replication. All artifacts are organized to enable transparent validation of the proposed microservice decomposition approach, supporting reproducibility and further research in the field of intelligent software architecture decomposition and cloud-native migration.
@inproceedings{paperbot12537,
title = {MicroWeaver: A Constraint-Guided Structure–Semantic Aware Approach to Automated Microservice Decomposition},
author = {Huitian Xue and Kairui Zhang and Chenxi Zhang and Zheng Li and Chuan Luo and Yilong Yang and Tianyu Wo and Chunming Hu and Yanliang Li and Yang Zhou and Penghao Zhang and Tiezi Zhang and Renyu Yang},
booktitle = {ASE},
year = {2026},
doi = {10.1145/3832783.3834457}
}