Abstract: In order to evaluate the risk of software defects, this paper proposes an approach for defects evaluation based on complex network analysis. In this approach, a network model is used to express the relationship between entities, and the methods in source code and the call relations between them are extracted as nodes and directed edges respectively to construct the source code network. Then the importance of root cause nodes with respect to the whole network is computed by the betweenness centrality and PageRank, which is used to assess the severity of defects. The experimental results show that the approach has a good performance in evaluating inside software severe defects, which can help maintainers pay more attention to implicit severe defects and provide helpful reference to the program repair, thus supporting software evolution.

Key words: complex network, severe defects, defects evaluation, betweenness centrality, PageRank

摘要： 为了评估软件缺陷的风险，提出了一种基于复杂网络分析的软件缺陷评估方法。该方法首先用一个网络模型表达程序实体之间的关系，将源代码中的方法抽象为节点，方法间的调用关系抽象为有向边，以此构造程序源代码网络；然后分别用介数算法和PageRank算法计算造成软件缺陷的方法节点在源代码全局网络中的地位，由此评估缺陷的风险高低。实验结果表明，该方法在评估内部高危缺陷时有较好的效果，有助于提高软件开发维护人员对一些隐蔽高危缺陷的关注度，进而为后续修复缺陷与软件演化提供有益的线索。

关键词: 复杂网络, 高危缺陷, 缺陷评估, 介数, PageRank