The Research On Noise Reduction And Defect Localization Based On Bug Report

During software development and maintenance phases, the bug tracking systems are widely used. The researchers use historical data obtained from the bug tracking system to predict various aspects of software, such as defect localization. However, research shows that about one third bug reports from the bug tracking system are nonbug and these noisy bug reports have potential impact on the validity of prediction models based on bug reports. Therefore, before the defect localization based on bug reports, we must manual preprocess bug reports extracted from the bug tracking system and try to remove those nonbug reports. In order to reduce the noisy bug reports and alleviate the burden of developers, this thesis proposes a multi-stage approach to automate the process of bug reports classification. In this thesis, we leverage bug reports from ten large open source projects maintained by various bug tracking systems to demonstrate the feasibility of the proposed multi-stage approach.After ensuring the purity of bug reports, the most important step of corrective maintenance is to localize the defect and then fix it. However, manual defect localization requires relatively high expertise and also increases the workload of software developers. This thesis presents an automatic approach for defect localization based on part-of-speech and invocation. Moreover, we present an approach to automatic adjust the weight of index terms in space vector model for increasing the weights of the nouns. To be more efficient, we create the invocation corpus to quickly and accurately obtain the invocation files. By contrast experiments, we demonstrate the effectiveness of proposed approach for defect localization.