一句话：提出基于统计的无监督方法，从软件运维数据中提取潜在结构，以支撑异常检测等下游任务。
现代软件系统在提供服务时往往消费数据，并产生大量监控数据，这些数据对系统运维与分析至关重要。然而，运维数据常是非结构化、半结构化或结构未知的，数据结构提取因此成为理解数据特征并自动化异常检测等下游任务的关键步骤。现有方法存在局限，例如需要人工标注。为此，作者提出新的无监督、基于统计的结构提取方法，用于发现软件运维数据中的潜在结构。其贡献覆盖数据工程与软件分析的多个方面，对学术界和工业界均具有重要意义。摘要未给出具体评估结果。
Software often consumes data to provide intelligent services and produces a large volume of monitoring data. Such data play a crucial role in the operations and analysis of modern software systems. However, these data are often unstructured, semi-structured, or have unknown structures. Data structure extraction is a critical step for understanding data characteristics and automating downstream tasks, such as anomaly detection. To address the limitations of existing data structure extraction approaches (e.g., requiring manual labeling), I propose novel methods for unsupervised, statistic-based structure extraction. The contributions of this research span multiple aspects of data engineering and software analysis, with significant implications for both academia and industry.