Applicability Analysis and Ensemble Application of BERT with TF-IDF, TextRank, MMR, and LDA for Topic Classification Based on Flood-Related VGI

Volunteered geographic information (VGI) plays an increasingly crucial role in flash floods. However, topic classification and spatiotemporal analysis are complicated by the various expressions lengths of social media textual data. This paper conducted applicability on bidirectional encoder representation from transformers (BERT) four traditional methods, TextRank, term frequency–inverse document frequency (TF-IDF), maximal marginal relevance (MMR), linear discriminant (LDA), results show that for user type, BERT performs best Government Affairs Microblog, whereas LDA-BERT We Media Microblog. As text length, TF-IDF-BERT works better texts with a length <70 >140 words, 70–140 words. For evolution pattern, study suggests Henan rainstorm, topics follow general pattern “situation-tips-rescue”. Moreover, this detected hotspot “Metro Line 5” related to rainstorm discovered topical focus spatially shifts Zhengzhou, first Xinxiang, then Hebi, showing remarkable tendency south north, which was same as report issued authorities. integrated multi-methods improve overall accuracy Sina microblogs, facilitating flooding.