Research And Implementation Of An Automatic Alarm Identification Method Oriented To Defect Detection Process

Static analysis tools can aid the developers detect the critical errors in software to some degree.However,challenges such as scalability and undecidability are likely to have impact on their precision and performances,preventing these tools from being widely adopted in practice.Recently,researchers have begun to utilize artificial intelligence techniques to improve the usability of these tools by automatically classifying false positive alarms,manual identification of which is laborious and time-consuming in software development processes.In order to make the defect features learned by the model consistent with the preconditions for the static analysis tools to generate alarms,this paper proposes an automatic alarm identification method for the defect detection process.The method first collects the relevant instruction sets that lead to the generation of alarm instances during the defect detection process,and extracts the grammatical structure information in the corresponding defect code segments based on the relevant instruction sets,and maps them into a sequence of tokens with contextual relationships.Then use word embedding technology to map the extracted token sequence into a high-dimensional vector representation,input the vector sequence into the neural network to generate and output defect semantic features,and finally use the feature to train the classifier,and output the possibility at which the alarm instance is identified as a false positive error.According to the process of automatic alarm identification,the system can be separated into three parts:extraction of relevant instruction sets based on fault pattern state machine,feature extraction based on relevant instruction sets,alarm automatic identification based on neural network.Design and complete these function parts one by one and implement them in a correct order,then a automatic alarm identification system can be constructed to process alarms in batch.In order to test the rationality and effectiveness of the automatic alarm identification method oriented to defect detect process,cross-project defect automatic identification experiments are carried out on five open source c projects and the results are compared with two baselines.The results present this method can improve the effectiveness in CPDI task,with an average accuracy reaching 71.08%.