Know Your LLM with Code Smell: Role-Aware Code Smell Summaries Corpus

This dataset is derived from the CodeXGLUE benchmark, specifically from its code summarization(Code-to-Text) subset featuring Java methods. The original path can be found in the CodeXGLUE dataset's `Code-To-Text\/java` folder. We curated and refined a portion of this dataset to focus on code snippets exhibiting code-smell characteristics (e.g., long methods, mysterious method names, deep nesting), aiming to explore how Large Language Models (LLMs) handle code with design issues when explaining code.