燕山大学学报2025，Vol.49Issue(1)：44-54,11.DOI:10.3969/j.issn.1007-791X.2025.01.005
数据投毒下的云API服务质量预测模型鲁棒性分析与解释
Robust analysis and interpretation of cloud API QoS prediction model under data poisoning
摘要
Abstract
Accurate and reliable quality of service(QoS)prediction is the key to realize the successful application of QoS aware cloud API service recommendation and combination.However,existing studies usually assume that the QoS prediction model is reliable,ignoring the data poisoning attacks of malicious users on cloud API QoS model under the open network environment.From the view of the attack mechanism and interpretability,a robust analysis and interpretation method of cloud API QoS prediction model under data poisoning attack is proposed.Firstly,the malicious user behavior under cloud API QoS perception scenario is quantified,and the data poisoning attack against the QoS prediction models is realized using different attack modes.In addition,a robust analysis method considering the type,intensity and scale of the poisoning attack is presented in a unified framework.Secondly,five kinds of QoS data features are defined as explanatory factors,and a robust interpretation model of cloud API QoS prediction model is established by adopting the modeling method based on regression analysis.The experimental results can effectively clarify the mechanism and robustness of poisoning attack against cloud API QoS prediction models,explain and give the key data characteristics that cause the fluctuation of prediction performance under data poisoning,and provide the support for data poisoning attack protection.关键词
数据投毒/云API/服务质量预测/鲁棒性/可解释性Key words
data poisoning/cloud API/quality of service prediction/robustness/interpretability分类
计算机与自动化引用本文复制引用
乞文超,鲍泰宇,刘钰杰,申利民,陈真..数据投毒下的云API服务质量预测模型鲁棒性分析与解释[J].燕山大学学报,2025,49(1):44-54,11.基金项目
国家自然科学基金资助项目(62102348,62276226) （62102348,62276226）
中央引导地方科技发展资金项目(236Z0103G) （236Z0103G）
河北省自然科学基金资助项目(F2022203012) （F2022203012）
河北省教育厅高等学校科技计划项目(QN2020183) （QN2020183）
河北省创新能力提升计划项目(22567626H) （22567626H）
