Sentiment Mining and Visualization of Covid-19-Related Weibo Dataset

The COVID-19 pandemic profoundly disrupted daily life worldwide, eliciting diverse emotional responses as individuals faced the uncertainties of the crisis. Social media platforms like Weibo became critical outlets for public expression, capturing a rich spectrum of sentiments that evolved with key events and policy changes. This study utilizes Weibo data to conduct fine-grained sentiment analysis, addressing the limitations of traditional models that classify emotions into broad categories (positive, negative, neutral), which fail to capture the complexity of emotions during a prolonged health crisis. By comparing the performance of three advanced NLP models-BERT-LSTM, XLNet, and RoBERTa-this research identifies BERT-LSTM as a promising candidate for detecting temporal dependencies and nuanced emotions, including fear, anger, sadness, disgust, gratitude, surprise, and optimism. Leveraging BERT-LSTM, we annotated unlabeled data to build a comprehensive dataset, which was then used to develop a dynamic visualization system. This system illustrates sentiment trends over time, offering public health officials and policymakers actionable insights into emotional responses to critical events and policy developments. By revealing emotional responses to critical events, policy changes, and health developments, the system supports data-driven decision-making and enhances understanding of public mood dynamics. The findings highlight the value of fine-grained sentiment analysis and dynamic visualization in navigating public health crises, offering a framework that can be adapted for future crises or real-time sentiment monitoring efforts.