在移动互联网大数据的背景下,带有发布时间、发布位置等特征标签的社交媒体数据在自然灾害应对中的关键作用受到了广泛的关注。选择我国2022年长江流域特大干旱为应用案例,以主流的互联网社交媒体——微博文本作为数据源,基于机器学习与人工智能算法,抓取干旱演化过程数据,深入分析旱灾舆情时空特征与主题特征。研究结果表明:① 社交媒体关注热度时间变化与干旱发生演化过程较为同步,7月份长江全流域干旱在微博的讨论数据较小,到8月上旬陡然上升,至8月中下旬达到顶峰,12月初基本归零。从旱灾发生演化过程上看,四川、重庆、江西等受旱严重省市微博讨论热度时间特征与当地水文径流数据时间变化整体呈“此消彼长”趋势;② 社交媒体关注热度空间特征与区域受旱严重程度分布基本吻合,四川、重庆、江西等高热度省市微博讨论占比超过50%,而云南、西藏、上海、青海等省市的微博讨论占比较低;③ 社交媒体干旱主题特征差异较大,江西、湖南两省主题热词为“鄱阳湖面积萎缩”、“水位下降”,四川、重庆两省市舆情热词是“山火”、“地震”、“农户减产”、“用电紧张”。伴随旱情发展演化过程,公众干旱舆情的情感倾向经历了从负向逐步趋向正向的过程。研究成果可为流域旱灾跟踪分析与社会公众抗旱动员提供技术支撑。

[Objective] In the context of big data from mobile internet, social media data with tags such as posting time and location has received widespread attention for its critical role in natural disaster response. In China, research on social attention and online public opinion regarding drought events remains limited, especially for the analysis of spatiotemporal and thematic characteristics of extreme drought events at the river basin scale, with no relevant reports yet. [Methods] This study used the 2022 extreme drought in China’s Yangtze River Basin as a representative case. Utilizing texts from Weibo, a mainstream social media platform in China, as data sources, this study used machine learning and artificial intelligence algorithms to collect Weibo text data throughout the drought progression process. The Latent Dirichlet Allocation (LDA) topic model was employed to perform term clustering and thematic characterization. Through this methodology, an in-depth mining of spatiotemporal and thematic characteristics of drought-related public opinion was conducted, along with sentiment analysis. [Results] (1) The temporal evolution of attention levels on social media was relatively synchronized with the progression of the drought event, with peak drought stage particularly prone to attracting heightened public attention. Across the entire Yangtze River Basin, drought-related discussions on social media remained relatively low in July 2022, rose dramatically in early August, peaked in mid-to-late August, gradually declined in mid-September, and returned to zero in early December. In terms of drought progression, an inverse correlation between the temporal variation characteristics of Weibo discussion level in severely affected provinces and municipalities including Sichuan, Chongqing, and Jiangxi and local hydrological flow data was observed. (2) The spatial characteristics of attention levels on social media basically matched the distribution of drought severity. The proportion of Weibo discussions in high-attention provinces and municipalities (e.g., Sichuan, Chongqing, and Jiangxi) exceeded 50%, reflecting widespread public concern about the drought and indirectly indicating severe socioeconomic impacts caused by the drought in these regions. In contrast, provinces and municipalities such as Yunnan, Tibet, Shanghai, and Qinghai showed relatively low levels of Weibo discussions. (3) The thematic characteristics of drought-related content on social media showed significant regional differences, with public attention levels being closely related to the severity of drought impacts. In Jiangxi and Hunan, key terms related to the drought were “shrinking of Poyang Lake” and “declining water levels” In Sichuan and Chongqing, key terms were secondary disasters such as “wildfires”, “earthquakes”, as well as drought-induced issues such as “reduced crop production by farmers” and “electricity supply shortages”. Other provinces primarily focused on “continuous high-temperature weather” and “meteorological drought”. As the drought progressed, the sentiment of public opinion on drought gradually transitioned from negative to positive. [Conclusion] Weibo texts serve as an effective data source for online public opinion analysis of sudden-onset disasters. The research findings can provide technical support for drought tracking analysis and mobilization efforts of the public for drought relief in river basins.