<p>Vulnerability has become a central but conceptually diffuse term across the human and social sciences, used to designate heterogeneous risks, conditions, and causal processes. This conceptual dispersion hinders cumulative theorizing and limits the research’s capacity to inform policy and intervention. This article presents a reproducible, multi-resolution mapping of how vulnerability is organized in the scholarly literature, based on a Natural Language Processing pipeline that combines contextual keyword extraction, static keyword aggregation, data-driven dimensionality reduction, and hierarchical agglomerative clustering with Ward linkage. A Scopus search conducted in May 2025 identified 12,450 English-language journal articles in social science, psychology, arts and humanities, economics, and business with “vulnerability” in the title. The resulting dendrogram is reported across the full range <i>k</i> = [2, 12], and a further analysis inspects the specific <i>k</i> = 7 cut, at which seven semantic clusters emerge: psychopathology, health, rights and violence, climate and rural livelihood, floods and natural disasters, security and infrastructure, economic poverty, and pandemic and migration. Cluster quality is assessed through a multi-metric validation envelope that combines geometric fit with topic coherence. Three VosViewer keyword co-occurrence maps trace how the lexical landscape evolves across historically meaningful periods. Since 2000, climate- and disaster-related vulnerability has expanded sharply, while psychopathology-oriented research has lost relative prominence. The contribution is positioned as a methodologically transparent instrument for critical inquiry into how a contested concept is organized and how that organization changes over time.</p>