Automatic Defect Classification: An Experience Applying Natural Language Processing

This paper presents initial results of a research effort that merges two approaches on Quality Management. On one hand, while working with process improvement initiatives involving defect causal analysis, the researcher were faced with the problem of measuring inter-rater reliability for a defect classification taxonomy in a software development organization. On the other hand, the researchers were looking for ways to guide defect correction effort by using information retrieval and natural language processing to cluster defect reports. This paper presents the results of applying the later approach to the inter-rater reliability problem. The results indicate that the machine learning approach scores at rates only achieved by senior members of the organization.