Abstract
Dream research has traditionally relied on small samples and retrospective self-reports, which are vulnerable to memory decay. The present study combines big data with natural language processing to investigate dreams that users spontaneously shared on Weibo, a major Chinese social media platform. Using the keywords “dream of,” “dream about,” and “in a dream,” we collected posts from April of each year between 2018 and 2023 and, after five rounds of screening, obtained 418,961 dream reports. We analyzed the demographic characteristics of the posters, the thematic content of the dreams (via word-frequency analysis and a co-occurrence semantic network), and the emotional tone of the reports (via lexicon-based sentiment scoring). Women shared dreams more often than men; users aged 20–35 predominated; posting peaked in the early morning. Six broad themes emerged—family, school, romantic relationships, work, danger, and body—covering a broader range but occurring at lower frequencies than typical-dream themes assessed by questionnaire. Themes related to danger and death moved in the opposite direction to the national urban unemployment rate after 2019, suggesting that dreams may buffer waking emotional states. Negative emotion scores consistently exceeded positive scores, indicating that negative content dominates dream sharing. We discuss the implications of these findings for understanding the emotion-regulation function of dreams and outline directions for future research.