Replication Data for: PAPEA – A Modular Pipeline for the Automation of Protest Event Analysis

Protest event analysis is the core method to understand spatial patterns and temporal dynamics of protest. It has been used widely for the empirical analysis and theory building on social movements and contentious politics. However, the method is time- and resource intensive because it usually depends on manual annotation. Its application is thus mostly limited to selected national newspapers or newswires, often sampling newspaper issues or days to reduce the amount of data to annotate. Advances in Natural Language Processing (NLP) have provided Large Language Models (LLM) as powerful tools for identifying and classifying relevant text segments. As we will show in this paper, using these tools, the automated classification of protest events and of political event data more broadly can reach levels of accuracy comparable to humans, while reducing necessary annotation time by several orders of magnitude. We propose a modular pipeline for the automation of PEA based on various fine-tuned LLMs. Our pipeline uses publicly available models and tools and can thus be easily adapted and extended. With this pipeline, we get from newspaper articles to PEA datasets with high levels of precision without human intervention, preparing the ground for an almost real-time analysis of protest dynamics. We illustrate the potential of PAPEA with the use case of a large data set of German language local newspaper articles.抗议事件分析（Protest Event Analysis, PEA）是理解抗议活动空间格局与时间动态的核心方法，已广泛应用于社会运动与抗争政治的实证分析与理论建构。然而，该方法高度依赖人工标注，耗时耗力，因此其应用多局限于精选的全国性报纸或新闻专线，通常通过抽样报纸期数或日期以缩减待标注数据规模。自然语言处理（Natural Language Processing, NLP）技术的发展催生了大语言模型（Large Language Model, LLM），成为识别与分类相关文本片段的有力工具。正如本文将展示的，借助此类工具，抗议事件乃至更广泛政治事件数据的自动化分类，可达到媲美人工的准确率，同时将所需标注时间缩减数个数量级。我们基于多款微调后的大语言模型，提出了一套用于抗议事件分析自动化的模块化流程。本流程采用公开可用的模型与工具，因此可轻松适配与拓展。借助该流程，我们可在无需人工干预的情况下，从报纸文章直接生成高精度的抗议事件分析数据集，为抗议动态的近乎实时分析奠定基础。我们以大型德语地方报纸文章数据集为应用案例，展示了PAPEA的应用潜力。