基于上下文老虎机的智能选择蜕变关系方法  

An intelligent metamorphic relation selection method based on contextual bandits

作　　者：孙家泽 孔嵘 SUN Jiaze;KONG Rong(School of Computer Science and Technology,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;Shaanxi Key Laboratory of Network Data Intelligent Processing,Xi’an 710121,China)

机构地区：[1]西安邮电大学计算机学院,陕西西安710121 [2]陕西省网络数据智能处理重点实验室,陕西西安710121

出　　处：《西安邮电大学学报》2025年第2期93-98,共6页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省科技厅重点研发计划项目(2023-YBGY-030);西安市重点产业链核心技术攻关项目(人工智能领域)(23ZDCYJSGG0028-2022,2024JH-CLYB-0120)。

摘　　要：针对蜕变测试过程中难以快速且有效地选择蜕变关系的问题,提出一种基于上下文老虎机的智能选择蜕变关系方法。该方法在测试过程中先构造待选择的蜕变关系并提取源测试用例的特征向量,然后创建上下文老虎机,利用特征向量和决策算法智能选择每轮测试的蜕变关系,生成衍生用例进行测试。最后将测试结果反馈至老虎机,通过多轮训练筛选出最优蜕变关系。实验结果表明,相比于随机选择方法和Tetraband选择方法,所提方法提高了平均故障检测率,具有更强的系统缺陷检测能力。In response to the difficulty of quickly and effectively selecting metamorphic relations during the metamorphic testing process,a method for the intelligent selection of metamorphic relations based on contextual bandits is proposed.During the testing process,metamorphic relations are constructed for selection,and feature vectors of the source test cases are extracted.The contextual bandit is created to intelligently select the metamorphic relation for each round of testing by leveraging the feature vectors and decision-making algorithms,so as to generate derived test cases for testing.The testing results are then fed back to the bandit.Through multiple rounds of training,the optimal metamorphic relation is identified and refined.Experiment results demonstrate that,compared to the random selection and the Tetraband selection,the proposed method improves the average fault detection rate,and exhibits a stronger capability in identifying system defects.

关 键 词：蜕变测试 蜕变关系 上下文老虎机 图像识别 测试预言 

分 类 号：TP311.53[自动化与计算机技术—计算机软件与理论]