Topic Modeling of Central Asian BRI Debt Discourse Based on TF-IDF-NMF Model

Abstract

Global media have generated extensive discussions over BRI investment financing and debt sustainability, and debt risk has evolved into a controversial mainstream discourse in international news. Targeting the implicit thematic distribution and semantic correlation of BRI debt risk reports, this study integrates strategic narrative theory and computational communication methods to quantitatively analyze 907 English news samples collected from 2013 to June 2026. This paper adopts TF-IDF vectorization and NMF algorithm for unsupervised topic modeling, adopting 1–2 gram combinations and 5,000 maximum feature words to extract 15 latent thematic clusters, and implements keyword frequency statistics and co-occurrence network analysis to quantify semantic links of core terms. Quantitative results indicate that development cooperation articles account for 24.3% (220 pieces), while independent debt risk reports only occupy 5.8% (53 pieces); debt keywords show obvious co-occurrence with geopolitical vocabularies, and negative-sentiment news rises sharply after 2023 but remains secondary to neutral and positive coverage. Computational statistics prove that media coverage is dominated by development cooperation narratives, and debt risk discourse is embedded in investment and infrastructure topics without forming an independent interpretive framework.