CoDesc

CoDesc是由孟加拉工程技术大学创建的一个大规模代码-描述并行数据集，包含420万个Java方法及其自然语言描述。该数据集通过广泛的分析和噪声去除处理，提高了数据质量。CoDesc主要用于代码摘要和代码搜索任务，通过预训练和微调设置，显著提升了代码搜索性能，并达到了代码摘要的新技术水平。此外，CoDesc还支持构建预训练语言模型，为Java语言的下游任务提供支持。CoDesc is a large-scale code-description parallel dataset created by the Bangladesh University of Engineering and Technology. It contains 4.2 million Java methods paired with their corresponding natural language descriptions. This dataset has undergone extensive analysis and noise removal processing to improve data quality. CoDesc is primarily utilized for code summarization and code search tasks. When applied in pre-training and fine-tuning scenarios, it significantly enhances code search performance and achieves state-of-the-art results for code summarization tasks. Furthermore, CoDesc also supports the construction of pre-trained language models, providing support for downstream tasks related to the Java programming language.

搜集汇总

背景与挑战

背景概述

CoDesc是一个大型并行数据集，包含420万个Java源代码及其对应的自然语言描述，适用于代码搜索和代码摘要任务。该数据集通过整合多个来源（如CodeSearchNet、FunCom等）并去除噪声构建而成，提供了预处理代码和训练示例，以支持相关研究。