一句话：提出缺陷感知的 GUI 测试工具 BugHunter，用 LLM 加检索增强从相似应用 bug 报告生成探索路径，提升缺陷检测效率。
现有 GUI 自动化测试方法主要分析软件内部结构来覆盖更多页面，而不探索可能触发缺陷的目标路径，导致缺陷检测效率低。本文提出缺陷感知的 GUI 自动化测试方法并实现工具 BugHunter，利用大语言模型与检索增强技术，从相似应用的历史 bug 报告中获取外部知识，动态调整测试过程并生成探索路径。方法首先构建高质量 bug 知识库，再通过信息检索找到相关报告，并引入局部与全局路径规划机制，以应对软件功能和 UI 设计差异以及在线报告行为模糊或步骤缺失的问题。作者在 71 个应用、121 个缺陷上评估 BugHunter，并与 16 个先进基线比较，结果显示其缺陷检测效果比最佳基线提升 60%。
Software with similar functions often encounters similar bugs. However, current automated GUI testing methods mainly generate test scripts by analyzing the internal structure of software to cover more pages, without exploring targeted paths that may trigger bugs, resulting in low bug detection efficiency. Considering that bug reports on open source platforms can provide external knowledge for testing, we propose a bug-aware automated GUI testing method and develop an automated tool, BugHunter, which uses a large language model and retrieval augmentation techniques to dynamically adjust the testing process and generate exploration paths based on bug reports from similar applications, improving bug detection efficiency. First, build a high-quality bug knowledge base from historical bug reports. Then use information retrieval to retrieve relevant reports from this bug knowledge base, and generate test paths based on bug reports of similar applications. BugHunter also introduces local and global path planning mechanisms to handle differences in software functionality and UI design, as well as ambiguous behavior or missing steps in online bug reports. We evaluate BugHunter on 121 bugs across 71 apps and compare its performance against 16 state-of-the-art baselines. BugHunter achieves 60% improvement in bug detection over the best baseline. GitHub link: https://github.com/test7766987/BugHunter. Demo video: https://www.youtube.com/watch?v=jgqNwg1AkhA
@inproceedings{paperbot11635,
title = {BugHunter: An Automated Tool for Bug-Aware GUI Testing via Retrieval Augmentation},
author = {Zhe Liu and Mengzhuo Chen and Chunyang Chen and Junjie Wang and Xu Xiang and Yujiao Yuan and Qing Wang},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3774748.3787602}
}