More Than a Feeling: Dataset on media sentiment regarding the Berlin Stock Exchange

Data appendix to: Lino Wehrheim/Janos Borst-Graetz/Bernhard Liebl/Manuel Burghardt/Mark Spoerer: More than a Feeling. Introducing an NLP-Based Media Sentiment Index for the Berlin Stock Exchange, 1872–1930, in: Historical Methods: A Journal of Quantitative and Interdisciplinary History 58 (2025), S: 139–159. https://doi.org/10.1080/01615440.2025.2506427. Financial economists and psychologists agree that collective sentiment plays a crucial role in financial markets. We present our newly created data for a daily aspect-based index that captures the sentiment at the Berlin Stock Exchange from 1872 to 1930, a time when Berlin was the key financial market in Germany. This index is based on market reports published every trading day in the Berliner Börsen-Zeitung, which give a verbal description of the sentiment among market participants. Due to daily publication and the long observation period, our corpus consists of about 18,000 market reports. To derive sentiment values, we apply a combination of expert annotation and machine learning. Methdologically, this data covers two novel aspects: First, we focus on both historical and highly domain-specific language, a dual challenge that thus far has rarely been addressed. As there are many similar historical sources, such as reports by chambers of commerce, our solutions will be helpful for future research. Second, we address a characteristic but neglected feature of financial texts that might be relevant also in a broader sentiment analysis context. Particularly, we focus on aspect- and entity-based sentiment analysis. Note: The data was generated in the project „More Than a Feeling: Media Sentiment as a Mirror of Investors’ Expectations at the Berlin Stock Exchange, 1872-1930“附属于以下文献的数据：Lino Wehrheim、Janos Borst-Graetz、Bernhard Liebl、Manuel Burghardt、Mark Spoerer：《超越情绪：构建1872-1930年柏林证券交易所基于自然语言处理（Natural Language Processing）的媒体情感指数》，刊载于《历史方法：定量与跨学科历史期刊》（Historical Methods: A Journal of Quantitative and Interdisciplinary History）2025年第58期，页码139-159，DOI：https://doi.org/10.1080/01615440.2025.2506427。 金融经济学家与心理学家均认为，集体情绪在金融市场中发挥着至关重要的作用。本数据集介绍了我们全新构建的每日维度情感指数相关数据，该指数可捕捉1872年至1930年柏林证券交易所的市场情绪——彼时柏林乃是德国核心金融市场。 该指数的数据来源为每日交易时段刊载于《柏林交易所报》（Berliner Börsen-Zeitung）的市场报告，此类报告以文字形式描述了市场参与者的情绪状态。 由于采用每日发布的资料且观测周期漫长，本语料库共包含约18000篇市场报告。 为获取情感数值，我们结合了专家标注与机器学习两种方法。 从方法论层面而言，本数据集包含两项创新维度：其一，我们同时兼顾了历史文本与高度领域专属的语言表达，这一双重挑战此前极少被相关研究攻克；鉴于当前存在大量同类历史文献（例如商会发布的报告），我们提出的解决方案可为后续相关研究提供参考。 其二，我们针对金融文本的一项典型却被忽视的特征展开研究，该特征在更广泛的情感分析场景中亦具备研究价值。 具体而言，我们聚焦于基于维度与实体的情感分析（aspect- and entity-based sentiment analysis）。 注：本数据集生成自「超越情绪：媒体情感作为1872-1930年柏林证券交易所投资者预期的镜像」项目。