Compositional Vector Space Models for Improved Bug Localization

摘要: Software developers and maintainers often need to locate code units responsible for a particular bug. A number of Information Retrieval (IR) techniques have been proposed map natural language bug descriptions the associated units. The vector space model (VSM) with standard tf-idf weighting scheme (VSM natural), has shown outperform nine other state-of-the-art IR techniques. However, there are multiple VSM variants different schemes, their relative performance differs software systems. Based on this observation, we propose compose various variants, modelling composition as an optimization problem. We genetic algorithm (GA) based approach explore possible compositions output heuristically near-optimal composite model. evaluated our against several baselines thousands reports from AspectJ, Eclipse, SWT. On average, composite) improves hit at 5 (Hit@5), mean average precision (MAP), reciprocal rank (MRR) scores by 18.4%, 20.6%, 10.5% respectively. also integrate compositional AmaLgam, which is state-of-art localization technique. resultant named AmaLgam can improve Hit@5, MAP, MRR 8.0%, 14.4% 6.5%