A Natural Language Processing Analysis of Newspapers Coverage of Hong Kong Protests Between 1998 and 2020

This article investigates how the SCMP, the China Daily-and western-based newspapers cover protests in Hong Kong in an effort to identify changes in journalistic practices between 1998 and 2020. It combines natural language processing (NLP) with a qualitative investigation of a novel corpus of newspaper articles spanning 22 years. It enlists topic modeling to contrast the treatment of protests in Hong Kong diachronically and across news sources. Through comparison of lexical frequency and lexical usage it showcases preferences and discrepancies in the use of protest-relevant keywords in the newspapers’ articles. Embedding neighborhood comparisons strengthens our understanding of how words are used differently between the SCMP, the China Daily and western-based newspapers, and also how the context of protest-related keywords may differ across news sources over time. Finally, computational sentiment analysis measures the tone and connotations of articles. The article fills a gap in the literature on Hong Kong media and its methodology broadens the application of NLP techniques to the social sciences.