Analysis of news trends of UNESCO World Heritage sites in Korea through text embedding and keyword networks

Although there has been a rise in text mining-based research on World Heritage, trend analysis highlighting the evolution of social perceptions and discourses related to World Heritage remains insufficient. In this study, we examine a trend of contextual relationship between World Heritage sites in Korea and their keywords through text embedding and network analysis on news data from 1990 to 2023. First, we find contextual clusters by embedding each news article with Sentence Transformer. The dendrograms generated for 5-year spans reveal coherent clusters on monuments, groups of buildings, and sites, with several small transient clusters. Furthermore, we demonstrate the evolution of contextual clusters by detecting communities in the keyword networks built upon word2vec. We find coherent communities about inscription, evaluation, and diplomacy, with transient communities on hot topics such as Jeju Volcanic Island, and Korean Tidal Flats. This study suggests a novel trend analysis framework combining text embedding and keyword networks and provides insights for future policy-making by highlighting the limitations of media coverage related to World Heritage.