Energy security scholarship has expanded beyond fossil-fuel supply stability into a multi-dimensional discourse shaped by decarbonization, technology transitions, and geopolitical disruption. To characterize how scholarly attention propagates across themes, this study applies contextual topic modeling and time-series causal discovery to 6951 Scopus-indexed abstracts published in 2005Q1–2024Q4 (80 quarters). BERTopic identifies 29 interpretable topics, which are further organized into five macro-clusters using hierarchical clustering on semantic embeddings of topic summaries. Quarterly topic-prevalence series are then analyzed with Peter and Clark Momentary Conditional Independence (PCMCI) (maximum lag
; p < .01) as an exploratory tool for estimating directed, time-lagged conditional dependencies in discourse dynamics. Robustness is evaluated across five model specifications with progressively richer conditioning sets, including publication intensity and event indicators for the Paris Agreement (from 2015Q4) and the Russia–Ukraine war (from 2022Q1). Thirty-seven directed links remain significant across all specifications, indicating a densely coupled discourse system rather than isolated thematic silos. The water–food–energy nexus repeatedly appears as an upstream driver, while long-horizon links connect wind-integration discourse to geothermal development (lag 7) and coal-mining innovation (lag 8). A pre/post comparison (2005–14 vs 2015–24) shows relative re-weighting toward distributed integration, climate-risk framing, and data-driven forecasting, with relative dilution of biofuels and next-generation nuclear. The proposed framework provides reproducible lead–lag signals for agenda monitoring and anticipatory governance, while explicitly distinguishing discourse-level dependencies from real-world technological causation.
