Data Poisoning Attack Detection Model For Quality Of Service Aware Cloud API Recommender System

In the cloud era,cloud API is the best vehicle for intelligent interaction,open capabilities and data transfer.The rapid growth of cloud API brings sufficient resources for developers and numerous security threats,which include not only cyber-attacks such as distributed attacks and replay attacks,but also data poisoning attacks against cloud API recommender systems.Data attackers of cloud API affect the quality of service prediction performance and recommendation results by injecting false cloud API quality of service data into the cloud service platform,which seriously affects the credibility of the cloud service platform.In this paper,we explore the impact of data poisoning attacks on the robustness of cloud API quality of service prediction algorithms from the perspective of attack mechanism and explain ability,and propose a data poisoning attack detection model for quality of service aware cloud API recommendation systems,with the main work as follows.Firstly,research on the problem of robustness analysis of data poisoning attackoriented service quality prediction algorithm.There are few studies related to data poisoning attacks for cloud API quality of service prediction algorithms,and it is unclear whether data poisoning attacks can affect the accuracy of cloud API quality of service prediction algorithms.We consider quantifying the malicious user behaviors in cloud API quality of service awareness context,study the fake user generation methods of different kinds of data poisoning attacks,and realize simulated attacks based on different attack methods of quality of service poisoning data,so as to clarify the mechanism of poisoning attacks for cloud API quality of service prediction algorithms and realize robustness analysis of existing cloud API quality of service prediction algorithms at the same time.Secondly,research on the robustness interpretation model of cloud API quality of service prediction algorithm based on data features.The explanation of the mechanism of the impact of data poisoning attack on the quality of service prediction algorithm has not been explored,proposing to explain the impact of data poisoning attack on the quality of service prediction model from the perspective of quality of service data features using regression analysis,defining five quality of service data features as explanatory factors,adopting the modeling method based on regression model analysis,explaining the effect of data features on the cloud API-oriented quality of service prediction algorithm by significance test prediction algorithm,and gives the key data features that cause the fluctuation of prediction performance based on the analysis of experimental results.Finally,research on the data poisoning attack detection algorithm for cloud API service of quality prediction is investigated.The cloud API quality of service dataset is born from a dynamic open network environment,and the number of cloud API is often tens of times more than the number of countable users.With a small number of users,the traditional clustering detection and probability detection algorithms cannot fully learn the differences between fake users and real users,which leads to the low accuracy of the poisoning attack detection algorithm.User domain features are extracted by user similarity calculation and random wandering sampling.Feature embedding techniques are used to connect user domain features,hidden features and explanatory features to achieve a multiple embedding representation of users.On this basis,the false user detector is optimized by combining grid search to improve the detection of false users in a highdimensional sparse environment.