User profiles of young breast cancer survivors on Chinese social media: machine learning-based text mining analysis study.

PURPOSE: Social media is vital for improving healthcare access, especially among disadvantaged groups. During the COVID-19 pandemic, young breast cancer survivors (YBCSs) in China increasingly relied on social media for health information, shaping their experiences and needs. However, little is known about how their online behaviors changed during such crises. This study examines the characteristics and health needs of Chinese YBCSs on social media during the pandemic, providing evidence to inform public health management in emergencies. METHODS: We used web crawlers to collect 6,415 breast cancer-related posts from Sina Weibo and Zhihu (November 30, 2017-May 31, 2022). Posts were filtered using operational criteria, combining manual screening and machine learning models. Text mining and natural language processing were applied to construct multidimensional user profiles across three pandemic phases: pre-pandemic, outbreak, and normalization. RESULTS: In total, 2,640 posts from YBCSs were included for analysis. YBCSs' online activity increased markedly during the outbreak (from 0.37 to 2.22 posts/hour), with peak engagement during leisure times but shorter active durations. Content shifted from treatment-focused discussions to collective encouragement and pandemic-related topics, then returned to disease management in the normalization phase. Sentiment was generally positive, with fluctuations during the outbreak and stabilization in the normalization phase (sentiment index 0.17-0.38). CONCLUSION: This study underscores the significant impact of the COVID-19 pandemic on YBCSs, highlighting shifts in temporal routines, content priorities, and emotions. The findings redefine the perspective on managing healthy lives of these vulnerable and fragile groups in the post-crises era, and emphasize the urgent need for timely health information support and equality in healthcare through social media platforms with machine-learning approaches.