多维在线跨语言Calling Network建模及其在可信国家电子税务软件中的实证应用

This project is conducted considering two new trustworthy requirements of national E-taxation software, namely user behavior trustworthy monitoring and trustworthy evaluation in software evolution. In our previous research, a model named as Calling Network (CN) was proposed to describe a software system’s runtime method call behaviors and structures. Software trustworthy evaluation and testing techniques based on CN model were also proposed. In this project, an advanced online CN model, named as MOC-CN, is studied to describe Multi-dimension, Cross-language interactive relations between software entities at different granularities. Software and user behavior patterns and their mapping relationship are investigated. Abnormal detection techniques on software and user behaviors are explored. Based on MOC-CN, the software behavior-based difference detection method is examined to find the effective differences among various versions of software, which is applied to guiding test case generation in software trustworthy testing. The trustworthy monitoring and evaluation systems will be developed for the individual income taxation systems in nine provinces, including Shaanxi, Hebei, Anhui, Guangxi, Shanghai etc. The trustworthy monitoring systems are deployed to monitor more than 900,000 users; and the evaluation systems are applied to supporting system updating about 3 months. Moreover, we will participate in the development of E-taxation software specification in “Golden Tax Project III”. This project would be an exemplary case to demonstrate the application and effectiveness of the research findings in the major research plan of trustworthy software.

针对国家电子税务软件新出现的可信需求：用户行为可信监控和系统版本演化可信评测，在现有软件调用网络模型Calling Network（CN）和软件可信评测与控制技术研究基础上，持续研究面向网络软件的MOC-CN模型，实现软件行为监控从单一系统离线建模，到多维（Multi-dimension）、在线（Online）、跨语言（Cross-language）建模的提升；挖掘软件行为与用户行为的模式特征和映射关系，研究软件和用户异常行为的实时监测和识别方法；研究基于MOC-CN的软件版本演化差异检测和差异引导的可信评测方法。以个人所得税征管系统为载体，研制电子税务软件可信监控和评测系统，在陕西、河北、安徽、广西、上海等9个省市开展示范应用，支持90万纳税人的行为监管，并协助税务部门维护软件升级3个月，参与国家金税工程三期电子税务软件行业规范的制定，在此基础上形成可信软件理论方法的典型应用案例。

本项目在软件调用模型Calling Network基础上，提出了软件多维在线跨语言调用网络模型（Multi-dimension Online Cross-language Calling Network, MOC-CN），实现了软件行为监控从单一系统离线建模，到多维（Multi-dimension）、在线（Online）、跨语言（Cross-language）建模的提升，实现对分布式软件系统、多语言系统、二进制程序、Android程序等不同软件系统的行为监控和调用结构分析。以MOC-CN模型为基础，提出了基于时序行为模式挖掘的软件及用户异常行为识别与控制方法；研究了基于MOC-CN的软件版本升级的差异性检测及其可信性评测。在理论研究的基础上，研制了Structure-CT 及Action-CT等一系列可信软件工具，并实际应用于金税三期工程管理决策分析平台和陕西省国家税务局自动化测试平台等项目，承担了项目中冗余代码检测、系统结构可信性评估、软件开发过程管理和可信性测试等任务，论证了研究成果的有效性和应用价值。.本项目研究过程中，课题组获得2015年教育部科技进步一等奖、2016年国家知识产权局“中国专利奖”优秀奖、2016 IEEE ISSRE最佳论文奖等。研究成果在IEEE TSE、IEEE TFS、JSS、ICSE、ASE等国际期刊和会议上发表论文20篇，授权国家发明专利7项，申请国际PCT专利4项，授权美国发明专利1项，培养博士4人，硕士8人。超额完成项目计划书所规定的各项目标。