oishooo/formality_classification

该数据集通过从多个来源采样构成，包括[ccdv/govreport-summarization]测试集、[MedRAG/pubmed]训练集、[osyvokon/pavlick-formality-scores]测试集和[HuggingFaceGECLM/REDDIT_comments]的tifu部分，这些来源代表了不同正式程度的文本片段。每个样本包括text（文本）、type（类型）、formality_label（正式度标签）和formality_explanation（正式度解释）字段。对于每个样本，使用大型语言模型[DeepSeek-V3]分配正式度标签并生成相应的解释。对于标签为neutral的样本，数据直接来源于现有的数据集，没有正式度解释。

The dataset is constructed by sampling from multiple sources — the test split of [ccdv/govreport-summarization], the train split of [MedRAG/pubmed], the test split of [osyvokon/pavlick-formality-scores], and the tifu split of [HuggingFaceGECLM/REDDIT_comments] — representing a diverse range of text snippets with varying levels of formality. Each sample includes the fields text, type, formality_label, and formality_explanation. A large language model [DeepSeek-V3] is used to assign a formality_label and generate a corresponding formality_explanation for each sample. However, for samples with the formality_label neutral, the data is sourced directly from an existing dataset without a formality explanation.