Sports, crisis, and social media: a Twitter-based exploration of the Tokyo Olympics in the COVID-19 era

The global landscape underwent a transformative shift due to the unexpected arrival of the COVID-19 pandemic, disrupting various aspects of life, including the staging of sports events. Notably, the Tokyo Olympics, a renowned international sports spectacle, was compelled to postpone its grand event from 2020 to 2021. In the following year, during the pandemic's severe second wave, Japan witnessed fervent local protests against the Tokyo Olympics' continuation, leading to numerous athletes opting out due to health concerns. Concurrently, demands to cancel the Olympics gathered momentum, resonating across social media platforms, particularly Twitter, celebrated for its real-time information dissemination. The CUP Framework was employed for qualitative and quantitative analysis of Twitter data. Qualitatively, emotion analysis explored the prevailing emotional context before the Tokyo Olympics' commencement. Quantitatively, the focus was on identifying specific entities and targets discussed in Twitter mentions and annotations, with preliminary hashtag analysis revealing #StopTokyoOlympic and #canceltheTokyoOlympics as prominent symbols of dissent, leading to the collection of over 100,000 relevant tweets from January to June 2021. A deep learning approach, employing transfer learning via the cross-lingual language model robustly optimized BERT pretraining approach (XLM-RoBERTa), was the analytical engine, with the model trained using a Reddit-based standard emotion dataset from CrowdFlower. This endeavor culminated in a cross-lingual multi-class emotion classifier system, achieving an impressive 82.35% classification accuracy, signifying a significant advancement in AI-based emotion classification methodology's efficacy.