一句话：提出基于LLM的FLTestGen，利用失败测试为可疑方法生成缺陷相关测试，提升故障定位效果。
故障定位算法常依赖软件测试套件来识别故障，因此测试质量至关重要；然而实际中多数测试与故障无关，仅一小部分测试对定位有贡献。本文提出 FLTestGen，一种基于 LLM 的测试生成算法，利用失败测试为可疑方法构造新的、与缺陷相关的测试，从而为故障定位算法补充更多关于缺陷的信息。初步结果表明，FLTestGen 能够提升故障定位性能，且生成的测试中大多数也是缺陷相关测试。
Fault localization (FL) algorithms aim to identify faulty code elements in a software system. These algorithms often leverage the test suite of the software system to pinpoint faults. As such, the quality of a test suite plays a crucial role in successful suspicious code identification. However, in practice, most tests are not relevant to the fault, with only a subset of tests contributing to FL. This paper proposes FLTestGen, an LLM-based test generation algorithm that generates bug-relevant tests. FLTestGen uses failing tests to construct new tests for suspicious methods. The proposed algorithm complements FL algorithms by adding more information about the bug. Initial findings show that FLTestGen can improve FL performance, and a majority of the generated tests are also bug-relevant tests.