Bug Patterns Localization Based on Topic Model for Bugs in Program Loop

Statistic fault localization technique is a popular technique to perform fault localization via comparing the different statistics of failed executions and successful executions. They measure suspiciousness for each program component and rank them descending by their suspiciousness to assist programmers to identify root cause of failures. Unfortunately, since these techniques based on hit-spectrum information, which use binary flags to register whether a program component is covered or not during a program execution, their results are inherently limited, especially when the bug(s) occur in complex program loops. To resolve this issue, we propose a bug patterns localization technique based on topic model for the bugs in complex loops. We also conduct experiments involving 28 program versions contained bugs in program loops in Siemens program suite to compare the effectiveness of the proposed approach against Dstar. The results show that our approach is promising.