dipta007/decomposeRL-tiny-judge

该数据集是一个多配置的自然语言处理数据集，专注于评估文本响应的质量，包括答案正确性、原子性（如检查表、基础性、是否为问题、无连词、单焦点和可验证性）、覆盖范围和问题可回答性等任务。每个配置包含文本、标签（整数或浮点数表示分类或评分）、文本哈希、声明哈希、提取的响应、来源运行、提示和原始响应等特征。数据被分为训练、验证和测试集，并提供平衡版本以处理类别不平衡。数据集规模较大，适用于机器学习模型训练和评估，特别是在自然语言理解和生成领域。

This is a multi-configuration natural language processing (NLP) dataset focused on evaluating the quality of textual responses. Its evaluation tasks include answer correctness, atomicity (covering checklist adherence, fundamentality, whether the content is a question, absence of conjunctions, single focus, and verifiability), coverage, and question answerability. Each configuration contains features such as text, labels (integers or floats for classification or scoring), text hash, claim hash, extracted responses, source run, prompt, and original response. The dataset is split into training, validation, and test sets, and balanced versions are provided to mitigate class imbalance. With a large scale, this dataset is suitable for training and evaluating machine learning models, particularly in the domains of natural language understanding (NLU) and natural language generation (NLG).