Identifying Duplicate Bug Records Using Word2Vec Prediction with Software Risk Analysis

Abstract

Reporting duplicated bugs in bug reports have serious productivity consequences on software projects. The fewer reporting of duplicated bugs, the better software maturity processes are set between the internal software stakeholders. Automated identification of the duplicated category through bug reports could enhance risk identification approaches during the software life cycle. In this paper, we propose two different similarity measures to identify duplicated bugs using the word-embedding (Word2Vec) natural language processing technique through Tensorflow tool. We conduct a comparison experiment on two related bug records descriptions from eight different software components from the Mozilla Core dataset. We choose different sentence types through the duplicated bug category records to compare and discuss each component's accuracy results and identify whether the proposed module will be able to detect the related records. Using an earlier work, this paper calculates software risk values from duplication records and from bug-fix time prediction for the components that have not been identified as duplicated by the Word2Vec approach. The study results show maximum precision accuracy of 99.89% for the components that have been identified correctly as duplicated by the used approach. Additionally, we found that 66% of the software components that were excluded from the bug duplication proposed module showed an increase in software risk values.