Page 298 - 《软件学报》2021年第7期

2216 Journal of Software 软件学报 Vol.32, No.7, July 2021

(a) PDEPC4 (b) AR6PC4
Fig.8 The performance variation of misclassification loss under different weights
based on given prior distribution difference and MMD
图 8 在给定先验分布差异和最大均值差异权重情况下,分类损失在不同权重下的性能变化

总体而言,如果选择了合理的参数配置,就可以保证模型逐渐收敛至一个较好的性能,该模型可以有效地捕
获源项目与目标项目的边缘概率分布并将其映射到新的子空间中,在该子空间中,这两个项目同时拥有相似的
边缘概率分布的条件概率分布.

5 总 结

异构缺陷预测方法尝试解决异构特征之间的跨项目预测问题,该类方法具有非常强大的实用性.本文提出
了一种基于变分自编码器的异构缺陷预测特征表示方法.针对现有异构缺陷预测方法,并不能很好地学习源项
目与目标项目之间的隐式特征表示、拟合其中的分布信息等问题,本文基于变分自编码器,并结合最大均值差
异对提取源项目与目标项目之间共性特征的方法进行了研究,通过进一步引入判别网络学习在隐式特征表示
下的条件概率分布,可以有效地验证本文所提模型的特征抽象能力,通过在大量缺陷数据集的多组跨项目预测
实验以及与多种缺陷预测方法比较,验证了本文所提出的异构缺陷预测方法不仅可以有效地学习两个项目之
间的缺陷分布信息,还可以进一步提升缺陷预测的性能.

References:
[1] Hall T, Beecham S, Bowes D, Gray D, Counsell S. A systematic literature review on fault prediction performance in software
engineering. IEEE Trans. on Software Engineering, 2011,38(6):1276–1304.
[2] Chen X, Gu Q, Liu WS, Liu WS, Liu SL, Ni C. Survey of static software defect prediction. Ruan Jian Xue Bao/Journal of Software,
2016,27(1):1–25 (in Chinese with English abstract). http://www.jos.org.cn/1000-9825/4923.htm [doi: 10.13328/j.cnki.jos.004923]
[3] D’Ambros M, Lanza M, Robbes R. Evaluating defect prediction approaches: A benchmark and an extensive comparison. Empirical
Software Engineering, 2012,17(4-5):531–577.
[4] Lee T, Nam J, Han D, Kim S. Developer micro interaction metrics for software defect prediction. IEEE Trans. on Software
Engineering, 2016,42(11):1015–1035.
[5] Menzies T, Greenwald J, Frank A. Data mining static code attributes to learn defect predictors. IEEE Trans. on Software
Engineering, 2006,33(1):2–13.
[6] Zimmermann T, Nagappan N, Gall H, Giger E, Murphy B. Cross-project defect prediction: A large scale experiment on data vs.
domain vs. process. In: Proc. of the 7th Joint Meeting of the European Software Engineering Conf. and the ACM SIGSOFT Symp.
on the Foundations of Software Engineering. New York: Association for Computing Machinery, 2009. 91–100.