A Generative AI Framework for Longitudinal Semantic Structure in Geotagged Digital Discourse

This repository provides anonymized replication materials for the manuscript: "A Generative AI Framework for Longitudinal Semantic Structure in Geotagged Digital Discourse." The replication package documents the full analytical workflow used to transform geotagged social media discourse into scale-aware semantic macrothemes validated through explicit spatial and temporal criteria. The repository includes: • Fully documented code implementing all analytical stages • Environment specifications for computational reproducibility • Aggregated and anonymized outputs sufficient to reproduce all tables and figures • Data dictionary and variable documentation Raw tweet texts are not redistributed due to institutional data-use restrictions. The original dataset is available via the Harvard Center for Geographic Analysis (CGA) GeoTweet Archive v2.0, subject to institutional approval. The workflow follows six sequential stages: 1. Data extraction and preprocessing 2. Embedding generation 3. Clustering and structural stability assessment 4. Language model–assisted semantic synthesis 5. Human validation and macrotheme construction 6. Spatiotemporal evaluation and spatial robustness testing The replication materials support transparency, reproducibility, and independent verification of all reported results.