The geography of digital and green (twin) firms in Germany

The twin transition, which combines green and digital innovation in economic activities, is increasingly central to policy agendas and is also receiving growing attention in regional research. However, accurately mapping green, digital and twin (both green and digital) economic activities across regions remains challenging, particularly due to data constraints. In this study, we advance this research frontier and present a geographic analysis of digital, green and twin economic activities in Germany, using a web-mined dataset of website texts from 678,381 firms, collected through web scraping in 2023. By processing over 44 million text paragraphs from these websites and applying a cosine similarity filter with green and AI-related terms, we filtered firms that are likely engaged in green, digital and twin activities. Based on this subset, 1437 text paragraphs were manually annotated to fine-tune two transformer models within a SetFit framework, accurately classifying firms as green, digital or both. We aggregate this firm-level data into hexagonal cells to reveal the geographic concentration of the twin transition in Germany. The final map shows a higher number of firms involved in green activities, widely spread across Germany, while AI activities are concentrated in urban centres. We identify 23,819 firms engaged in both green and digital activities, with major hubs like Berlin and Munich leading, and peripheral regions potentially being left behind. Our findings offer critical insights into the geography of the twin transition and highlight the need for policies that address potentially induced spatial inequalities.

双转型（twin transition）将绿色创新与数字创新融入经济活动之中，如今在政策议程中愈发占据核心地位，同时也日益受到区域研究领域的关注。然而，精准绘制各区域内绿色、数字及兼具两者属性的双转型经济活动的空间分布图谱仍颇具挑战，数据约束是其中的核心障碍。本研究推动了该领域的研究前沿，并基于2023年通过网络爬虫（web scraping）采集的678381家企业的网站文本挖掘数据集，对德国的数字、绿色及双转型经济活动开展地理空间分析。研究团队对上述网站的超4400万条文本段落进行处理，并结合绿色相关术语与人工智能（AI, Artificial Intelligence）相关术语构建余弦相似度（cosine similarity）筛选器，以此筛选出可能涉及绿色、数字及双转型活动的企业。基于筛选出的企业子集，研究人员对1437条文本段落进行人工标注，以在SetFit框架内微调两个Transformer模型，从而精准将企业划分为绿色型、数字型或兼具两者属性的类型。研究将企业层面的数据聚合至六边形格网中，以此揭示德国双转型经济活动的地理集聚特征。最终生成的空间图谱显示，参与绿色活动的企业数量更多，且广泛分布于德国全境；而人工智能相关活动则高度集聚于城市中心区域。研究共识别出23819家同时参与绿色与数字活动的企业，其主要集聚地包括柏林、慕尼黑等核心城市，而边缘区域则可能被甩在发展队列之外。本研究结果为双转型的空间地理分布研究提供了关键洞见，同时也凸显了制定政策以应对潜在空间不平等问题的必要性。