CoSQA+
收藏资源简介:
CoSQA+是由中山大学开发的一个高质量代码搜索数据集,旨在解决现有代码搜索数据集中存在的问题,如不切实际的查询、代码不匹配和单一代码对。该数据集包含103,000条高质量查询与多个匹配代码的配对,数据来源于CoSQA、StaQC和CodeSearchNet。创建过程中,利用了多种模型进行自动标注和代码生成,确保了数据集的高质量和多样性。CoSQA+特别适用于评估和提升多选择代码搜索的性能,通过引入新的评估指标MMRR,有效解决了传统评估方法的局限性。
CoSQA+ is a high-quality code search dataset developed by Sun Yat-sen University, which aims to address the prevalent issues in existing code search datasets, including unrealistic natural language queries, mismatched code snippets, and one-to-one single-code-per-query pairing limitations. This dataset includes 103,000 high-quality queries paired with multiple matching code snippets, and its source materials are drawn from CoSQA, StaQC, and CodeSearchNet. During the construction process, multiple models were utilized for automatic annotation and code generation to guarantee the high quality and diversity of the dataset. CoSQA+ is specifically designed for evaluating and improving the performance of multi-choice code search, and it effectively resolves the limitations of traditional evaluation methods by introducing the novel evaluation metric MMRR.

- 1CoSQA+: Enhancing Code Search Dataset with Matching Code中山大学 · 2024年



