Research Of Software Defect Prediction Based On Program Abstract Syntax

With the rapid development of the mobile Internet,competition in the mobile application market has become increasingly fierce.In addition,compared with the feasibility and innovativeness of software,its stability and reliability have gradually become the core competitiveness for mobile application enterprises.Software defect prediction(SDP)is referred to as an important technique to ensure software quality.It can effectively predict the defect existence trend and the distribution of faults in each software module,which assists the quality assurance team in reasonably allocating resources.Most of the existing SDP methods use data mining techniques to collect various software metrics from the source code,and then build defect predictive models based on these metrics to identify software modules that are more likely to be defective.However,since most software metrics are obtained by analyzing the code with statistical methods,they fail to reflect the rich semantic information and grammatical structure knowledge that can better reflect the discriminability between defective and non-defective modules.Moreover,in several SDP methods based on semantic information,it is usually necessary to convert the source code into an abstract syntax tree.But it is absence how to select nodes of abstract syntax tree to better express semantic information.Furthermore,due to the shortage of labeled data available for model training in the early stage of software development,cross-project defect prediction(CPDP)emerges as an essential need.But there are distribution discrepancies between different software projects,which makes the performance of the prediction model very unsatisfactory.In view of the above problems,this thesis has carried out a detailed study on software defect prediction.The main work is as follows:(1)Aiming to provide semantics information of software metrics,we combine with the characteristics of the abstract syntax tree to train a within-project defect prediction(WPDP)model based on convolutional neural network and attention mechanism,which extract semantic information from the abstract syntax tree.(2)To address the problem of node granularity selection,we propose a novel node granularity of abstract syntax tree,and then propose an improved WPDP model on this basis.(3)To address the problem of inconsistent data distribution in the CPDP model,we present a defect prediction framework under a cross-project environment,and implement a defect prediction system based on this to help testers detect the defects in the project.