The politicization of AI semiconductor discourse: A longitudinal analysis of Korean media coverage (2023–2025)

This study examines the evolution of AI semiconductor media discourse in South Korea from January 2023 to June 2025, analyzing 49,117 news articles collected from the BigKinds database. Integrating four complementary computational methods—thematic categorization, Latent Dirichlet Allocation (LDA) topic modeling, keyword co-occurrence network analysis with Louvain community detection, and lexicon-based sentiment analysis—we identify a three-phase transformation in discourse structure. In the first phase (Stability, January 2023–July 2024), coverage centered on technological development and manufacturing competition, with US, technology, firm, and investment occupying the most central positions in the keyword network. The second phase (Anticipation, August–December 2024) was characterized by financial euphoria driven by NVIDIA's market surge and HBM demand, during which market-oriented narratives dominated and sentiment, having peaked in July 2024, began to decline. The third phase (Politicization, January–June 2025) marked a decisive shift toward political contestation: the Policy thematic category more than doubled from 7.61% to 17.49% across these empirically derived phases (|z| = 39.40, p < 0.001), the presidential-election topic grew nearly fourfold, and political keywords displaced technology keywords from central network positions. This politicization was accompanied by a sentiment decline that a formal shift-share decomposition attributes predominantly to within-topic change (82.5% of the decline) rather than to shifts in topic composition: political vocabulary infiltrated otherwise non-political coverage and depressed its tone, while market-focused articles remained affectively stable—evidence of lexical infiltration rather than simple topic substitution. These findings extend agenda-setting theory by demonstrating that the agenda setters for technology-intensive issues shift rapidly in response to geopolitical shocks, and offer a forward-looking framework for anticipating how AI semiconductor discourse may evolve as techno-geopolitical uncertainty intensifies. While the corpus is limited to mainstream institutional Korean news media—so that the lexical-infiltration mechanism is established for institutional news discourse rather than for the full public sphere—the study contributes, to our knowledge, the first longitudinal evidence that the politicization of a technology discourse can operate as a sub-topical, vocabulary-level process rather than as topic substitution.