Cognitive-affective configuration of university students' continuance intention in online learning

Most studies of continuance intention in online learning ask which psychological variables predict it. This study asks a different question: how are those variables arranged in relation to one another? Drawing on TAM, ECM, and a Control-Value-informed perspective, we treat continuance intention as a cognitive-affective configuration rather than as a belief-to-intention pathway dominated by a single variable.

Survey data from 421 university students were analysed using five comparative models: BiLSTM, BiLSTM-Attention, CNN-BiLSTM, Transformer, and Random Forest.

The retained four-construct solution showed high internal consistency and strong convergent validity, although discriminant validity required cautious interpretation, particularly for the relationship between Affective Appraisal and Continuance Intention. Predictive performance was consistently strong across models (mean AUC = 0.936-0.954). Although the Friedman test detected overall variation in AUC, no pairwise comparison survived multiple-comparison correction. The BiLSTM-Attention model showed strong illustrative performance, but was interpreted as an exploratory computational tool rather than as evidence of temporal, recursive, or causal psychological structure.

The findings suggest that continuance intention in online learning is associated with a closely connected pattern of usefulness, trust, and affective appraisal rather than with usefulness alone. Practically, the results imply that platform design should extend beyond functional optimisation to include trust-building and support for positive evaluative experience.