OPOR-BENCH
收藏资源简介:
OPOR-BENCH是由中国传媒大学等机构联合构建的事件中心化基准数据集,专为自动化在线舆情报告生成任务设计。该数据集涵盖463个危机事件(2012-2025年),每个事件包含多源文档(平均19.1篇新闻文章和400.8条社交媒体帖子)及结构化参考摘要,总令牌数平均每事件超过32K,数据来源于权威数据库(如EM-DAT)和公共平台(如Wikipedia、X/Twitter)。数据集通过混合标注流程构建,结合LLM自动化框架与人工专家标注,确保了时间线、事实属性和社交媒体作者分类的高质量。其应用旨在解决危机管理中多源信息整合的挑战,支持政府和企业快速生成结构化舆情报告,以提升应急响应效率。
OPOR-BENCH is an event-centric benchmark dataset jointly constructed by Communication University of China and other institutions, specifically designed for the task of automated online public opinion report generation. This dataset covers 463 crisis events spanning from 2012 to 2025, with each event containing multi-source documents (averaging 19.1 news articles and 400.8 social media posts) and structured reference summaries, and an average total of over 32,000 tokens per event. The data is sourced from authoritative databases such as EM-DAT and public platforms including Wikipedia and X/Twitter. The dataset is built through a hybrid annotation pipeline that combines an LLM-powered automated framework with manual expert annotations, ensuring high-quality annotations for timelines, factual attributes, and social media author classification. Its applications aim to address the challenge of multi-source information integration in crisis management, supporting governments and enterprises to rapidly generate structured public opinion reports and thereby improving emergency response efficiency.

- 1OPOR-Bench: Evaluating Large Language Models on Online Public Opinion Report Generation中国传媒大学、哈尔滨工业大学、中国铁道科学研究院集团有限公司、圣克拉拉大学 · 2025年



