Detecting Cross-Area Concept Links in HCI Using Persistent Homology

Identifying emerging research connections is essential for tracking how research fields evolve, yet many bibliometric methods rely on co-occurrence counts and therefore emphasize links only after they become common. We apply topological data analysis to 26 years of HCI literature (approximately 30,000 papers from CHI, UIST, CSCW, and UbiComp). For each year, we build a similarity-based concept complex from embeddings and compute persistent 1-cycles: rings of related concepts that remain unfilled, for example A–B–C–D–A with no diagonals, indicating semantic proximity without triangle closure. We cluster concepts into communities and treat cross-community edges appearing on a representative cycle as candidate “bridges,” while returning intermediate concepts for human inspection. We observe cycles that persist for more than a decade. Moreover, cycle death often precedes a sharp rise in direct co-occurrence between endpoints in subsequent years, typically by one to three years. We interpret these events as signals of structural readiness for integration that complement frequency-based metrics.