“Cannot Stop, Yet Resent It”: How Short-Form Video Recommendation Algorithms Shape Users’ Continuance Intention

Short-form video platforms have become deeply embedded in everyday digital life, where algorithmic recommendation enhances user experience but also raises concerns about invasiveness and manipulativeness. Drawing on the Cognitive–Affective–Conative (CAC) framework, this study examines the relationships between users’ cognitive evaluations of recommendation algorithms and continuance intention through algorithm dependence and algorithm aversion, while also identifying configurational pathways associated with high continuance intention. Based on 547 valid responses from short-form video users in China, we measured personalization, diversity, invasiveness, manipulativeness, algorithm dependence, algorithm aversion, and continuance intention. PLS-SEM was used to test direct and mediating effects, and fsQCA was employed to identify sufficient configurations associated with high continuance intention. The results show that personalization and diversity positively predict continuance intention and exert significant indirect effects through algorithm dependence, whereas manipulativeness negatively predicts continuance intention; invasiveness and algorithm aversion show no significant direct effects. The fsQCA results identify three dependence-maintenance pathways—experience-reinforced dependence, risk-buffered dependence, and risk-tolerant dependence—with algorithm dependence serving as a core condition across all four sufficient configurations. Notably, algorithm dependence and algorithm aversion coexist in two risk-tolerant configurations associated with high continuance intention. By distinguishing algorithm dependence from algorithm aversion, this study refines the affective dimension of the CAC framework, clarifies their differentiated roles in continuance intention, and provides configurational insights for responsible recommendation-system governance.