Chinese Public's Discourse and Emotional Responses Regarding Japan's Nuclear Wastewater Discharge on Social Media: A Content Analysis of Sina Weibo Data

Abstract

Abstract Background: The global discourse surrounding the Japanese government's decision to discharge nuclear wastewater from the Fukushima Daiichi nuclear power plant into the ocean has attracted substantial international attention and fervent debates, notably across various social media platforms. This study aims to systematically investigate and analyze the subjects of discourse as well as the emotional inclinations expressed by the public prior to and subsequent to Japan's official declaration regarding the release of nuclear wastewater into the ocean (spanning from April 1, 2021 to May 30, 2021). Methods: Employing a Python-based web crawler, we extracted a database comprising 139364 Sina Weibo microblogs from April 2021 to May 2021 pertained to the incident of Japanese nuclear wastewater discharge. This study demonstrates how to combine human and natural language processing (NLP) machine analysis, using TF-IDF (Term Frequency-Inverse Document Frequency) improved latent Dirichlet assignment (LDA) topic modeling and dictionary-based unsupervised learning to analyze seven segmented emotions of netizens at different stages of public opinion development. Results: Based on the public opinion life cycle theory, we find that the dynamics of netizens' public opinion about Japan's discharge of nuclear wastewater are divided into three stages: incubation period, outbreak period and recession period. The NLP method discovered six topic trend: political statement, government accusations/netizens discussions, nuclear pollution and environmental hazards, netizens expression of dissatisfaction, appeal for science popularization, netizens outcry over the drinkability of nuclear wastewater. The sentimental analysis revealed a notable prevalence of negative emotions among individuals in relation to the incident, with negativity constituting 65% and positivity representing 35% of the emotional spectrum. Subsequent to an emotional categorization, it was discerned that the sentiment of "Disappointed" exhibited the highest proportion. Conclusion: This study conclusively demonstrates that the approach we use here is capable of effectively reducing large amounts of community feedback (e.g., blog posts, social media data) through NLP and ensuring contextualization and rich human interpretation. Further, detecting and assessing the interests and concerns of social media users in real time can help relevant administrative agencies adapt to genuine public concerns and enable timely response, guidance and oversight.