一种面向软件配置管理制品的层次分类方法

发布时间：2018-08-17 10:07

【摘要】：配置管理工具(configuration management tool,简称CMT)作为运维自动化的组成部分,是实现开发运维一体化(development and operations,简称Dev Ops)的重要支撑技术.当前,互联网开源社区中存在数量众多的CMT脚本制品,但是缺乏有效的层次分类管理,给快速检索和高效利用CMT脚本制品带来困难.针对该问题,提出一种面向CMT制品的基于在线非结构化描述文档分析的层次分类方法.该方法利用标签共现性关系(tag co-occurrence)建立层次类别体系,基于描述属性特征,实现对CMT制品的层次分类器;并使用混合的样本划分方式针对数据倾斜问题进行了改进.对超过11 000例训练数据和1 000例测试数据进行实验,结果表明:改进的样本划分方式得到的最佳查准率、查全率、调和平均值分别达到0.81、0.88、0.85,较传统方式查全率提高0.15,调和平均值提高0.06.该结果验证了层次分类方法的有效性.
[Abstract]:As a part of operation and maintenance automation, configuration management tool (configuration management tool, (CMT) is an important supporting technology for developing integrated (development and operations, for Dev Ops). At present, there are a large number of CMT scripts in the open source community of the Internet, but the lack of effective hierarchical classification management makes it difficult to quickly retrieve and efficiently utilize CMT scripting products. To solve this problem, a hierarchical classification method based on online unstructured description document analysis for CMT products is proposed. This method uses label co-occurrence relation (tag co-occurrence) to establish a hierarchical classification system. Based on the description of attribute features, the hierarchical classifier of CMT products is implemented, and the mixed sample partitioning method is used to improve the problem of data skew. More than 11,000 training data and 1,000 test data were tested. The results showed that the best precision and recall rate obtained from the improved sample partition method, The harmonic average value reached 0.81g 0.88m 0.85, which was 0.15% higher than that of the traditional method, and 0.06% higher than that of the traditional method. The results show that the hierarchical classification method is effective.
【作者单位】：中国科学院软件研究所;中国科学院大学;计算机科学国家重点实验室(中国科学院软件研究所);
【基金】：国家自然科学基金(61402453) 国家重点研发计划(2016YFB1000803)~~
【分类号】：TP311.5

【相似文献】