Abstract: Large language model (LLM) has shown considerable potential in the automatic generation of unit test cases, yet limitations remain in covering complex structures and critical paths, resulting in gaps in coverage and accuracy. To address this issue, a path guidance iterative optimization strategy was proposed, integrating control flow graph analysis with a C-P-R (Context-Prompt-Response) template to enhance path guidance during test generation. In addition, a dual-verification mechanism was introduced to improve the reliability of generated cases, while iterative refinement was employed to progressively fill coverage blind spots. Experiments were conducted across multiple models and task scenarios, with systematic evaluations on key metrics including statement coverage, branch coverage, and assertion accuracy. Experimental results demonstrated that the proposed strategy achieved an average of 97% statement coverage and 94% branch coverage, while maintaining strong adaptability and stability across different models. These findings indicate that the combination of path guidance and iterative optimization significantly enhances the coverage and reliability of automatically generated test cases, providing a feasible new approach for automated testing and verification of complex software systems.

Key words: Large Language Model (LLM), unit test case generation, path guidance, iterative optimization, control flow graph analysis

摘要： 摘 要: 大语言模型（LLMs）在自动生成单元测试用例方面展现出显著潜力，但在覆盖复杂结构和关键路径时仍存在不足，导致生成结果在覆盖率与准确性上存在差距。为解决这一问题，提出一种路径引导与迭代优化策略，通过结合控制流图分析与C-P-R（Context-Prompt-Response）模板，强化测试用例生成过程中的路径引导能力。在此基础上，引入双重验证机制提升用例的可靠性，并通过迭代优化逐步弥补覆盖盲区。实验设计涵盖多种模型与任务场景，对语句覆盖率、分支覆盖率及断言准确率等关键指标进行了系统评估。实验结果显示，该策略能够在语句覆盖率上达到平均97%，在分支覆盖率上达到94%，同时在不同模型中均表现出较好的适应性与稳定性。研究表明，路径引导与迭代优化相结合能够有效提升测试用例生成的覆盖水平与可靠性，为复杂软件系统的自动化测试验证提供了可行的新思路。

关键词: 大语言模型, 单元测试用例生成, 路径引导, 迭代优化, 控制流图分析