A study on user continuance intention for children’s intelligent companion robots based on BERTopic-PLS-SEM-fsQCA

The children’s intelligent companion robot market faces the industry-wide challenge of “high initial adoption yet low sustained retention.” Existing research has predominantly relied on single-method approaches, leaving the psychological mechanisms and configurational pathways driving user continuance intention insufficiently understood. This study constructs a sequential multi-method research framework integrating “BERTopic thematic mining—PLS-SEM linear validation—fsQCA configurational analysis.” Drawing on 6,405 valid online reviews from JD.com and 440 valid questionnaire responses, the study adopts an integrated theoretical framework combining the Expectation Confirmation Model (ECM) and the Value-based Adoption Model (VAM). BERTopic thematic mining identified six core demand dimensions from consumer reviews: perceived usefulness, perceived enjoyment, visual aesthetics, social/gifting value, technical reliability, and expectation confirmation. PLS-SEM analysis reveals that satisfaction is the strongest predictor of continuance intention ( β = 0.705, p < 0.001), explaining 50.3% of its variance ( R ² = 0.503). Perceived value ( R ² = 0.571) is jointly driven by expectation confirmation ( β = 0.230), perceived usefulness ( β = 0.197), technical reliability ( β = 0.163), social/gifting value ( β = 0.141), visual aesthetics ( β = 0.106), and perceived enjoyment ( β = 0.104). Notably, perceived enjoyment exhibits a distinctive dual-pathway mechanism, influencing continuance intention both indirectly through perceived value and directly through satisfaction ( β = 0.232, p < 0.001). Further fsQCA configuration analysis identified four configurations of high continuance intention (with a solution consistency rate of 0.811 and a coverage rate of 0.469), which can be grouped into three broader patterns based on their theoretical coherence: the “expectation confirmation-dominated” pattern (Configurations 1 and 2, with consistency rates of 0.855 and 0.839, respectively), the “full-dimensional attribute-driven” pattern (Configuration 3, consistency rate of 0.814), and the “Social-Hedonic Compensation” pattern (Configuration 4, consistency rate of 0.836), indicating substantial heterogeneity in continuance intention logic across different segments. Theoretically, this study extends the ECM-VAM framework to the context of children’s embodied intelligent products by introducing visual aesthetics and technical reliability as context-specific antecedents of perceived value, demonstrates that unstructured online review texts can serve as a valid source for the contextualized operationalization of theoretical constructs, and qualifies the linear assumptions of conventional net-effect models through a configurational perspective.