Location and recency of developers’ activities in a noun-based approach to improve automatic bug assignment / Ramin Shokripour

Shokripour, Ramin (2015) Location and recency of developers’ activities in a noun-based approach to improve automatic bug assignment / Ramin Shokripour. PhD thesis, University of Malaya.

PDF (Full Text) Restricted to Repository staff only until 01 January 2018. Download (2410Kb)

Abstract

Bug assignment is one of the important activities in bug triaging that assigns valid bugs to the appropriate developers for fixing. The bug assignment process has high potential to reduce the effort and cost spent on software maintenance. In order to facilitate this process many automatic bug assignment approaches have been proposed which are mostly based, at least in part, on text analytics methods. Most of these methods recommend the appropriate developer(s) based on either the developers’ activity histories (called activity-based methods) or predicted location(s) for the new bug report (called location-based methods). On the other hand, some approaches in recent years considered metadata, which is associated to the data recorded in software repositories such as an associated component to the fixed bugs or the fixing time of the bugs. Absence of using the metadata into the text analytics of the automatic bug assignment approaches leads to a lack of using all potentials of the text analytics, which is one of the most efficient steps of the automatic bug assignment approaches. The aim of this thesis is using the metadata in the text analytics step of the automatic bug assignment process. Therefore, a new activity-based approach is proposed that uses time metadata in its text analytics step. Additionally, an activity-based approach is combined with a location prediction approach to provide a new location-based approach. Unlike the existing location-based approaches that use the general obtained information (e.g. authors of file) from the predicted location, the proposed location-based approach works based on the activity relevancy. Therefore, only the developers’ activities are considered that are relevant to the subject of the bug report. To achieve this goal, 1) the desired terms are selected according to the predicted location (term’s location metadata), and 2) the expertise of a developer is determined based on the time of using the terms (term’s time metadata). Additionally, the proposed approaches are restricted to deal with only the noun type of the text data which is extracted from information resources of the software projects to reduce the volume of dataset. The empirical experimentation is performed on a set of open-source projects and the results indicate that the accuracy of the activity-based approach outperforms the com- parative automatic bug assignment approaches which are based on the most popular text analytics methods (called baseline approaches). The accuracy is increased 6% to 47%. On the other hand, the proposed location-based approach improves the accuracy and effectiveness of the baseline approaches up to 60% and 40% respectively. In addition, the use of only nouns reduces the dataset size up to 52% along with improving the accuracy of the bug assignment up to 10%. The statistical analysis conducted on the obtained results indicates the significance of the improvement. According to the obtained results, the proposed approach is recommended to be used for determining the developers’ expertise in fixing a new bug report as it remarkably improves the accuracy of the automatic bug assignment.

Actions (For repository staff only : Login required)