Software Bug Report Quality Detection Research Based On Convolutional Neural Network

With the continuous expansion of software scale and continuous improvement of software complexity,software maintainers use bug reports to record software bugs and guide bug fixing.In real software maintenance,relevant personnel discovered the existence of a large number of low-quality bug reports.These low-quality bug reports are invalid for bug fixing,and they hinder the development of software maintenance.Invalid bug reports usually meet two problems: 1)the information in the bug report is incomplete;2)the bug cannot be reproduced according to the description of the bug report.In order to detect invalid bug reports,several methods based on feature engineering have been proposed in recent years.However,these methods have high requirements for the integrity of the bug report fields,so they may encounter obstacles in feature extraction,which may lead to failure of the method.Automatically detecting valid bug reports and discovering the valid patterns has certain research significance.Duplicate bug report is a special type of invalid bug report.Due to different expression habits,different reporters will use different expressions to describe the same bug.Duplicate bug reports are usually complete and reproducible,so other methods are needed for detecting.A large number of methods have been proposed for the detection of duplicate bug reports,but their performance needs to be improved.This paper learns the correlation between bug reports based on convolutional neural network,so as to detect duplicate bug reports more accurately.The main work of this paper is summarized as follows:1)This paper introduces the research of bug report quality,valid bug report detection and duplicate bug report detection,and analyzes the shortcomings of previous valid bug report detection methods and duplicate bug report detection methods.2)In order to solve the problem of valid bug report detection,an effective valid bug report detection model based on convolutional neural network is proposed.This model overcomes the problem of difficult feature extraction,and only uses the summary and description of the bug report to detect valid bug reports more efficiently.3)In order to fundamentally solve the problem of invalid bug reports,based on the trained valid bug report detection model,the de-convolution technique is used to extract the valid bug report patterns.Bug reporters are guided to submit valid bug reports based on these patterns.4)A duplicate bug report detection model based on dual-channel convolutional neural network is proposed.This method combines two bug reports into a dual-channel matrix representing a pair of bug reports,so that the model can extract the correlation features between the two bug reports and achieve the goal of accurately detecting duplicate bug reports.5)A large number of experiments have been conducted on many large-scale open source data sets such as Open Office,Eclipse,Mozilla,etc.,and the effectiveness of the valid bug report detection and duplicate bug report detection methods proposed in this paper is proved by analyzing the experimental results.