价值观内容类别标签数据集

价值观内容类别标签数据集主要面向主流价值观特征提取与内容识别研究、模型训练需求建设，基于国内主流新闻平台的真实新闻数据和基于现有标签词生成的虚拟新闻数据产生，主要记录了新闻文本片段及其对应的价值观标签词等数据，数据量约190,722条。其中，raw_data.json包含人工筛查的14,535条新闻文本片段，涵盖新闻、军事、财经、科技、文化等多个领域，每条数据包含“Instruction”和“output”两个字段，分别对应文本数据和抽取到的标签词；infer_data.json则包含176,187条新闻文本，用于模型推理阶段。该数据集通过结合真实与虚拟数据，确保了数据的丰富性和代表性，为构建高精度的价值观内容识别模型提供了坚实基础。The Value-Oriented Content Category Label Dataset is primarily designed for research on mainstream value feature extraction and content recognition, as well as meeting the needs of model training. It is constructed based on real news data from domestic mainstream news platforms and virtual news data generated using existing label words, mainly recording data such as news text snippets and their corresponding value-oriented label words, with a total volume of approximately 190,722 entries. Among them, raw_data.json contains 14,535 manually screened news text snippets covering multiple fields including news, military affairs, finance, technology, culture and others. Each entry in this file has two fields: "Instruction" and "output", which correspond to the text data and the extracted label words respectively; infer_data.json, on the other hand, contains 176,187 news text snippets for the model inference stage. By combining real and virtual data, this dataset ensures the richness and representativeness of the data, providing a solid foundation for building high-precision value-oriented content recognition models.

搜集汇总

背景与挑战

背景概述

该数据集面向主流价值观特征提取与内容识别研究与模型训练，基于真实与虚拟新闻数据构建，包含约19万条记录，涵盖多个领域。数据分为人工筛查的文本片段和用于推理的新闻文本，为价值观内容识别模型提供了丰富且具代表性的基础。