关键词: 错误定位, 程序行为图, 图挖掘, 支持向量机

Abstract: The existing fault localization approaches only give a list of suspicious statements ranking in descending order. However, lacking of context information is hard for developers to understand the cause of failure. This paper presents a method that combines graph mining and support vector machine (SVM) to locate fault. First of all, program is modeled as weighted behavior graph according to the program information. Secondly, the frequent edges are extracted, and SVM is used to classify all executions. The suspicious degree of methods is measured by change of classification accuracy. The approach gets the bug-relevant function set for developers to locate fault finally. Experimental results show that using frequent edges as feature for classification keeps or improves the classification accuracy. Results show that this method is effective.

Key words: fault localization, program behavior graph, graph mining, support vector machine