Cardoso Ana Sofia, da Silva Catarina, Soriano-Redondo Andrea, Jarić Ivan, Batel Susana, Santos João Andrade, Jorge Alípio, Vaz Ana Sofia
CIBIO, Centro de Investigação em Biodiversidade e Recursos Genéticos, InBIO Laboratório Associado, Campus de Vairão, Universidade do Porto, Vairão, 4485-661, Portugal.
Departamento de Biologia, Faculdade de Ciências, Universidade do Porto, Porto, 4099-002, Portugal.
Sci Rep. 2025 Apr 28;15(1):14924. doi: 10.1038/s41598-025-97441-1.
Social media has become a popular stage for people's views over climate change. Monitoring how climate change is perceived on social media is relevant for informed decision-making. This work advances the way social media users' perceptions and reactions towards climate change can be understood over time, by implementing a scalable methodological framework grounded on natural language processing. The framework was tested in over 1771 thousand X/Twitter posts of Spanish, Portuguese, and English discourses from Southwestern Europe. The employed models were successful (i.e., > 84% success rate) in detecting relevant climate change posts. The methodology detected specific climate phenomena in users' discourse, coinciding with the occurrence of major climatic events in the test area (e.g., wildfires, storms). The classification of sentiments, emotions, and irony was also efficient, with evaluation metrics ranging from 71 to 92%. Most users' reactions were neutral (> 35%) or negative (> 39%), mostly associated to sentiments of anger and sadness over climate impacts. Almost a quarter of posts showed ironic content, reflecting the common use of irony in social media communication. Our exploratory study holds potential to support climate decisions based on deep learning tools from monitoring people's perceptions towards climate issues in the online space.
社交媒体已成为人们表达对气候变化看法的热门平台。监测社交媒体上人们对气候变化的认知情况，对于做出明智决策具有重要意义。这项工作通过实施一个基于自然语言处理的可扩展方法框架，改进了随着时间推移理解社交媒体用户对气候变化的认知和反应的方式。该框架在来自欧洲西南部的西班牙语、葡萄牙语和英语语篇的超过177.1万条X/Twitter帖子中进行了测试。所采用的模型在检测相关气候变化帖子方面取得了成功（即成功率>84%）。该方法在用户话语中检测到了特定的气候现象，这与测试区域内主要气候事件的发生情况相吻合（例如野火、风暴）。情感、情绪和讽刺的分类也很有效，评估指标在71%至92%之间。大多数用户的反应是中性的（>35%）或负面的（>39%），主要与对气候影响的愤怒和悲伤情绪相关。近四分之一的帖子显示出讽刺内容，这反映了社交媒体交流中讽刺的普遍使用。我们的探索性研究有潜力通过深度学习工具来支持气候决策，这些工具可用于监测人们在网络空间中对气候问题的认知。
