Fig. 1: Framework of this study.

a Schematic of novel technology applications for flood loss perception in post-flood response. b TSEM framework overview. TSEM framework developed human-perceived multisectoral flood losses and human-perceived Technology Adoption indexes (TAIs) from social media and revealed the effects of technology in multiple sectors against flooding. The multisectoral temporal corpus and four technical temporal corpora were filtered from the original microblog data. The MFLPs, including Casualty Loss Perception (CALP), Transport Loss Perception (TRLP), Communication Loss Perception (CMLP), Water and Electricity supply Loss Perception (WELP), and Building Loss Perception (BDLP), were estimated in five sensitive sectors on Weibo. Technology Adoption Indexes (TAIs, e.g., UAV, LSD, Amap, and DW) were constructed via hourly technology-related word frequency statistics. Structural equation models were built to reveal the relationship between technological adoption and flood loss perception in various sectors. TSEM (Fig. 1b) incorporates two core modules: text data processing module and structural equation modeling module. Text processing module completed the construction of the multisectoral perceived flood loss index, four technology adoption indexes, and Rescue Effort Effectiveness (REE). REE is quantified by pre-trained BERT and a multi-layer perceptron (MLP). By analyzing the topics of four technological corpora, seven hypotheses are proposed. Then, in the SEM module, these indexes and hypotheses obtained from the text module are used for structural equation modeling. The appropriate models were selected by comparing the performance indicators (e.g., Chi-square, GFI, and RMSEA) of the models to identify the best-fitting models. After determining the optimal model, the SEM results were used to compare and evaluate technological effects across sectors. The TSEM process is detailed in the Methods section.