Unveiling viral marketing dynamics in online social networks: insights from China's otome games and explainable predictive modeling

Purpose – The prominence of online social networks (OSNs) has made them ideal platforms for viral marketing. Otome games efficiently use OSNs for viral campaigns, making them a valuable case for studying viral marketing strategies. This study identified key elements of viral marketing posts to inform the operational strategies of commercial accounts. Design/methodology/approach – This study analyzed China's sizable otome game market using operational behavioral data from the top five games, compiled into the OtomeVM dataset. Following the knowledge discovery in databases framework, this study identified key characteristics of the top 25% most viral posts and proposed the ViralGD model, a multimodal machine learning model for virality prediction. The model's decision logic was further interpreted through a global surrogate method to ensure transparency. Findings – This study identified lottery mechanisms as significant predictors of post virality, with optimal performance observed for content combining short videos of 180 s or less and long-text descriptions of 175 characters or more. Commercial social media accounts often produced emotionally homogeneous content, with emotions having minimal impact on the viral spread of their posts. Originality/value – This study is the first to apply large-scale real-world data and data mining to uncover overlooked patterns in viral marketing, offering theoretical insights into defining virality and refining campaign design. Practically, the iterative application of the ViralGD model uncovers high-impact features that boost content diffusion, effectively guiding operators in selecting optimal improvements.