Abstract: Feature location is the process that confirms the potential influence in software evolution. Feature location is a recognized critical factor that decides success or failure in evolution, and how to use domain knowledge to promote accuracy in feature location becomes an important problem in current research. This method extracts the features in the software source code, and analyzes the features by topic model, then inputs update report as a query to locate which source code has been changed. This paper makes experiments with open source software, then compares experimental results with the benchmark of the open source software, correlation results indicate that this method has higher accuracy and universality, and can verify software evolution. The average of recall can achieve 69.16% and 100%, the average of precision can achieve 1.28% and 2.43%, the average of harmonic mean can achieve 2.50% and 4.72%. The experimental results show that the performance is better than that of baseline.

Key words: software evolution, feature location, topic model, domain knowledge

摘要： 确定演化活动潜在影响的过程称为特征定位。特征定位已经被公认为影响软件演化项目成败的一个关键因素，如何利用程序的领域知识促进特征定位的准确性已经成为当前研究的一个重要问题。该方法提取出软件源代码中的特征，并对提取后的特征进行主题分析，然后通过输入查询语句定位出被修改的源代码。利用现有的开源软件进行实验，并将实验结果与对应开源软件的Benchmark进行对比，结果表明所提出方法的精确度有所提高，可以进行软件特征的定位。实验结果中，平均查全率达到69.16%和100%，平均查准率达到1.28%和2.43%，平均调和平均数达到2.50%和4.72%，性能较对比方法有较大的提高。

关键词: 软件演化, 特征定位, 主题模型, 领域知识