摘要
Abstract
Against the backdrop of large-scale construction of water conservancy infrastructure,the whole-lifecycle quality management of projects is confronting new challenges.Existing risk assessment systems rely excessively on expert judgment mechanisms,with inherent defects of strong subjectivity,high resource consumption,and single evaluation dimension.To address the aforementioned issues,this study innovatively constructs a semi-supervised assessment model integrating dual-feature encoding.Taking quality supervision reports as the dataset,a co-training mechanism is adopted to realize risk level prediction under the condition of limited labeled data.Empirical analysis demonstrates that,under the condition of incomplete data labels,the proposed method increases the F1 score by 19.8%compared with traditional methods,and is particularly suitable for the automated analysis of unstructured texts such as supervi sion logs and inspection reports,which provides a practical and implementable technical pathway for the digital transformation of industry supervision.关键词
水利基础设施/质量管控/非结构化数据处理/机器学习/风险量化Key words
water conservancy infrastructure/quality control/unstructured data processing/machine learning/risk quantification分类
建筑与水利
