ABSTRACT
This study investigates how fans with different levels of community engagement behave and use language in Sina Weibo’s Super Topic Communities, integrating psycholinguistic analysis with machine learning. Analysing data from 9248 users, we compared high-level and low-level fans across behavioural indicators, simplified Chinese linguistic inquiry and word count categories, and fine-grained sentiment dimensions. Results show that highly engaged fans disclose more personal information, maintain substantially larger social networks, and produce richer content, suggesting stronger identification with their fan communities. Linguistically, they use more achievement-related words, motion verbs, quantifiers and first-person plural pronouns—patterns may reflecting collectivist orientations and enhanced group identity. Sentiment analysis reveals that high-level fans express more intense emotions, including both positive (joy, goodness) and negative (sadness, disgust) feelings, suggesting greater emotional involvement. Machine learning models trained on combined behavioural and linguistic features accurately predict users’ community levels (support vector machine accuracy 83.8%, area under the receiver operating characteristic curve 0.917), outperforming models using linguistic features alone. These findings advance understanding of how online language and digital footprints signal social identity and emotional investment in fandom. They also highlight fans’ evolving role in idol cultivation within Eastern collectivist cultures, offering practical implications for community management and marketing strategies.