CONGRA
收藏资源简介:
CONGRA是由香港大学系统安全实验室创建的一个大规模冲突分级基准数据集,旨在评估软件合并工具在不同复杂度冲突场景下的性能。该数据集包含44,948个冲突案例,来源于34个真实世界的开源项目,涵盖C、C++、Java和Python等多种编程语言。数据集通过代码操作分类方法,将冲突分为七个类别,以反映冲突的复杂性。创建过程包括从GitHub收集项目历史合并场景,通过轻量级语法树分析提取代码操作,并根据这些操作对冲突进行分类。CONGRA主要应用于自动冲突解决任务,旨在解决大规模项目中手动合并冲突的高成本问题,并评估大型语言模型在冲突解决中的表现。
CONGRA is a large-scale conflict grading benchmark dataset developed by the System Security Laboratory of the University of Hong Kong, designed to evaluate the performance of software merge tools in conflict scenarios with varying complexities. This dataset includes 44,948 conflict cases sourced from 34 real-world open-source projects, covering multiple programming languages such as C, C++, Java, and Python. Using code operation classification methods, it categorizes conflicts into seven categories to reflect their complexity levels. The dataset creation process involves collecting historical merge scenarios of projects from GitHub, extracting code operations via lightweight syntax tree analysis, and classifying conflicts based on these operations. CONGRA is primarily applied to automatic conflict resolution tasks, aiming to resolve the high costs associated with manually resolving merge conflicts in large-scale projects, and to evaluate the performance of large language models (LLMs) in conflict resolution.

- 1CONGRA: Benchmarking Automatic Conflict Resolution香港大学 · 2024年



