Typologies analysis of AI Innovation Strategies in Korean Public Institutions: : Combining Semantic Network and Cluster Analysis

• In Event: Performance and its Discontents: Contemporary Challenges in Public Management

Abstract

Under an aggressive national drive for Artificial Intelligence Transformation (AX), Korean public institutions have rapidly embedded AI across their workflows—yet whether this transition produces substantive public value or merely symbolic compliance remains unverified. Moving beyond binary adoption metrics, this study derives an empirical typology of AI-innovation adoption and compares its public-value performance, realizing a three-phase design that combines semantic network analysis (SNA), cluster analysis, and an analysis of variance (ANOVA). Using the AI-innovation sections of 71 institutions' 2025 management-performance reports, we (1) map how institutions textually conceptualize the AI–outcome relationship, (2) cluster institutions into types, and (3) test how the types differ on public-value indicators. The SNA reveals a discourse organized around a central “data” hub and four communities (data analytics, worker safety, generative-AI governance, citizen-service automation). Clustering yields four adoption types: an Internal-Efficiency type, a Citizen-Service (access) type, a small Inclusive-Comprehensive type, and a dominant Governance-Trust type. The types differ significantly in their public-value profile, value breadth, governance maturity, and AX intensity (ANOVA, p < .01), but not in their substantive measurement rate (p = .069)—indicating that broader, better-governed types are not necessarily more substantive in producing measured outcomes. The findings caution against a uniform AX mandate and support tailored, function- and capacity-matched AI adoption portfolios.

H1. Efficiency-oriented value claims will constitute the modal and most frequent category of reported AI performance, significantly exceeding equity-, access-, and trust-oriented claims.
H2. Institutions whose statutory mission is citizen welfare and social service will report a higher proportion of equity- and access-oriented claims than infrastructure- or finance-oriented institutions.
H3. Greater AI-governance maturity (a Chief AI Officer, a dedicated unit, an AI committee, a published ethics policy) will be associated with a broader public-value repertoire—more balanced coverage across all four dimensions rather than efficiency alone.
H4. Civic-trust value claims will be the least frequently quantified dimension, appearing more often as rhetorical commitments than as measured outcomes.

Data & Analysis Plan
H1 is tested with a chi-square goodness-of-fit test against a uniform distribution and paired Wilcoxon signed-rank tests of efficiency versus each other dimension. H2 is tested with a one-sided Mann–Whitney U test comparing the combined equity-plus-access proportion of welfare institutions against all others. H3 is tested with a Spearman rank correlation between governance maturity and value breadth, corroborated by the governance profile of the clusters. H4 is tested with a two-proportion z-test comparing the quantified share of trust versus efficiency claims.

Authors

• Hyung Jun Park, Sungkyunkwan University
  Presenting Author

• seeun park, Sungkyunkwan university
  Non-Presenting Co-Author

• Minchang LEE, Chosun University
  Non-Presenting Co-Author

• Jeongwon Lee, Sungkyunkwan University
  Non-Presenting Co-Author

• Sohyun Park, Sungkyunkwan University
  Non-Presenting Co-Author