ReposVul
收藏资源简介:
ReposVul是哈尔滨工业大学深圳研究院创建的第一个仓库级别的高质量漏洞数据集,包含6,134个CVE条目,代表236种CWE类型,跨越1,491个项目和四种编程语言。该数据集通过集成大型语言模型和静态分析工具,有效区分了与漏洞修复相关的代码变更,捕捉了漏洞间的调用关系,并通过基于跟踪的过滤模块识别了过时的补丁。ReposVul不仅提高了数据质量,还为软件安全领域提供了丰富的多粒度信息,支持从仓库级到行级的漏洞检测,旨在解决软件安全面临的挑战,提升漏洞检测的准确性和效率。
ReposVul is the first high-quality repository-level vulnerability dataset created by the Harbin Institute of Technology Shenzhen Research Institute. It contains 6,134 CVE entries, representing 236 CWE types, spanning 1,491 projects and four programming languages. This dataset integrates large language models and static analysis tools to effectively distinguish code changes related to vulnerability fixes, capture the calling relationships between vulnerabilities, and identify outdated patches via a trace-based filtering module. ReposVul not only improves data quality but also provides rich multi-granularity information for the field of software security, supporting vulnerability detection from repository-level down to line-level. It aims to address the challenges in software security and enhance the accuracy and efficiency of vulnerability detection.

- 1ReposVul: A Repository-Level High-Quality Vulnerability Dataset哈尔滨工业大学深圳研究院 · 2024年



