面向软件缺陷数据的协同过滤抽样推荐算法

Sampling Recommendation Algorithm Based on Collaborative Filtering for Software Defect Data

基于没有一种抽样方法能在所有缺陷数据集上表现良好且为软件缺陷数据选择适用抽样方法是必要的这一前提,提出了一种面向软件缺陷数据的协同过滤抽样推荐算法.首先,在历史缺陷数据上对主流抽样方法进行排序,以得到在特定分类算法和度量指标下主流抽样方法的性能排序;然后,计算新缺陷数据和历史缺陷数据之间的杰卡德相似系数,以挖掘数据之间的相似性;最后,将抽样方法排名和数据相似性的信息结合起来构建一个推荐网络,利用协同过滤算法为新的软件缺陷数据推荐适用的抽样方法.通过Python对多个NASA缺陷数据集进行仿真实验,实验结果表明面向软件缺陷数据的协同过滤抽样推荐算法是可行和有效的.

Based on the fact that no single sampling method can be performed well on all defect data sets and it is necessary to select suitable sampling methods for software defect data,a sampling recommendation algorithm based on collaborative filtering for software defect data has been proposed.Firstly,the mainstream sampling methods are sorted on historical defect data to obtain the performance ranking of the mainstream sampling methods under specific classification algorithms and metrics.Secondly,the Jaccard similarity coefficient between the new defect data and the historical defect data is calculated to mine data similarity.And finally,the information of sampling method ranking and data similarity is combined to build a recommendation network,and the cooperative filtering algorithm is used to recommend the applicable sampling method for the new software defect data.The simulation experiment is carried out on multiple NASA defect data sets by using Python.The experimental results show that the sampling recommendation algorithm based on collaborative filtering for software defect data is feasible and effective.