Supporting material for the research titled "Global advances in extreme heat research: An LLM-based quantitative review"

Supporting material for the research titled “Global advances in extreme heat research: An LLM-based quantitative review” This supporting material provides the files and models used for data processing, topic modeling, relevance checking, and named entity extraction in this study. It consists of the following components: 1. Original data: This folder contains the raw text corpus used in the study, including titles, abstracts, keywords, metadata, and other original information extracted before processing. The dataset was compiled in November 2025 from three major academic databases: Web of Science (WoS), PubMed, and Scopus. It includes article titles, abstracts, journal names, author names, author affiliations, author keywords, article keywords, and other relevant metadata. 2. Processed data: This folder includes the cleaned and filtered text corpus together with the corresponding sentence embeddings (embeddings.npy) generated after preprocessing, deduplication, and keyword-based screening. 3. all-MiniLM-L6-v2_embeddings_model: This folder contains the sentence embeddings generated using the all-MiniLM-L6-v2 model from the SentenceTransformers library. These embeddings serve as semantic input for BERTopic and provide 384-dimensional dense vectors optimized for clustering and semantic topic discovery. 4. BERTopic_parameters.xlsx: This file contains all parameter settings used for topic modeling with BERTopic v0.17.3, including: -UMAP dimensionality reduction -HDBSCAN clustering -c-TF-IDF weighting -MMR-based topic representation 5. SciBERT_model: This folder contains the SciBERT model used to evaluate the semantic relevance between scientific texts and the topics identified by BERTopic. SciBERT is trained on 1.14 million scientific papers from Semantic Scholar and is well suited for scientific similarity assessment and relevance checking. 6. BERT_NER_model: This folder contains the dbmdz/bert-large-cased-finetuned-conll03-english model used for named entity recognition (NER) in this study. The model, fine-tuned on the CoNLL-2003 English NER dataset, is used to extract multiple categories of entities that support data validation and topic interpretation. Its applications include, but are not limited to: - identifying study-area related entities (e.g., countries, regions, cities), - extracting key scientific entities (e.g., climate terms, variables, processes), - assisting in checking the consistency between extracted entities and topic modeling outputs. References: Shi, M., Luo, M. & Wu, S. Global advances in extreme heat research: An LLM-based quantitative review. Environment International. In review, 2025.