A Multi-Indicator Framework for Science-Technology Convergence Analysis: Application to LLM-Robotics Integration and Research Priority Identification

This study presents a bibliometric framework that integrates convergence indicators, tech- nological maturity metrics, and research portfolio analysis to systematically map emerging research domains at the science–technology interface. The framework operationalizes three complementary analytical dimensions: Publication–Patent Topic Overlap (PTO), Convergence Intensity Index (CII), and strategic portfolio positioning through growth trajectory analysis. Applied to LLM–robotics literature spanning 2016–2025 (N = 3,054 documents comprising 2,960 publications and 94 patents), the analysis identifies four macro-level thematic clusters exhibiting exceptional growth rates (mean CAGR: 50.8%). Robot–Language Model Integration constitutes the dominant cluster (39.4% of corpus) with the highest growth velocity (69.2% CAGR), while Language Model Applications demonstrates the strongest measured science–technology conver- gence (CII = 0.457). Notwithstanding these findings, the analysis is substantially constrained by severe patent underrepresentation (3.1% vs. an empirically expected 30–40%) and coarse topic granularity (4 clusters vs. a methodologically preferable 15–20). These limitations are not peripheral caveats but structural constraints that directly compromise the construct validity of convergence indicators and narrow the analytical conclusions to feasibility demonstration rather than actionable niche identification. Nevertheless, the results reveal unambiguous field-wide acceleration in LLM-enabled robotics research—73% of Robot–Language Integration documents originate from 2023–2025—and expose a characteristic citation–growth inversion diagnostic of very recent domain emergence. The paper contributes both a replicable multi-indicator methodology and a transparent accounting of the empirical conditions under which convergence analysis yields reliable versus unreliable inference. Keywords: science–technology convergence; bibliometrics; large language models; robotics; patent analysis; topic modeling; research portfolio analysis; technology foresight