通胀叙事有向无环图标注数据集
收藏资源简介:
该数据集由汉堡大学与吕讷堡大学联合构建,聚焦新闻语料中的通胀因果叙事分析,采用有向无环图(DAG)结构标注事件节点与因果边。数据源自道琼斯新闻数据库的英文报道,通过定性内容分析法(QCA)迭代优化26类细粒度叙事子类别(供需/杂项)。研究提出基于克雷彭多夫α的图标注评估框架,旨在解决叙事理解中的人类标注变异(HLV)问题,为经济学与NLP领域提供结构化叙事分析工具。原始文本因版权限制未公开,但开源了图标注方法论实现。
This dataset was jointly constructed by the University of Hamburg and the Leuphana University of Lüneburg, focusing on causal narrative analysis of inflation in news corpora. It uses Directed Acyclic Graph (DAG) structures to annotate event nodes and causal edges. The data is sourced from English news reports in the Dow Jones News Database, where 26 fine-grained narrative subcategories (supply-demand/miscellaneous) were iteratively optimized via Qualitative Comparative Analysis (QCA). This study proposes a graph annotation evaluation framework based on Krippendorff's α, aiming to address the issue of Human Label Variation (HLV) in narrative understanding and providing structured narrative analysis tools for the fields of economics and Natural Language Processing (NLP). The original texts are not publicly available due to copyright restrictions, but the graph annotation methodology implementation is open-sourced.
- 1From Variance to Invariance: Qualitative Content Analysis for Narrative Graph Annotation汉堡大学·计算机科学系; 汉堡大学·社会经济系; 吕讷堡大学·人工智能与可解释性研究组 · 2026年



