Automatic generation of software test cases fusing requirement semantics and function call chains

To address the issue of low test coverage caused by implicit test scenarios such as missing boundary conditions and exception handling in software requirement documents， a code-enhanced requirement analysis method is proposed. The code-requirement associations are established through the proposed method which is based on semantic vector similarity and LLM verification， and five types of implicit test scenarios （boundary conditions， error handling， resource management， state transitions and performance stress） are extracted from function call chains to enhance requirement descriptions， and the enhanced requirements are decomposed into test function points and scenario-driven test cases are generated. The results of experiments on open-source projects show that compared with the baseline method by using LLM directly， the significant improvements are achieved by using CERA method in comprehensive test quality and test requirement coverage， which maintains higher API test accuracy. The effectiveness of three core components： scenario extraction， two-stage matching strategy and BERT-based rough screening is verified through ablation experiments. The good adaptability on both parsing libraries and embedded systems is demonstrated by the results of proposed method applied that is particularly suitable for third-party testing and acceptance testing scenarios.