AI Insight
This study examined continuous learning behavior among 2,133 Chinese university students using intelligent technology-supported classrooms over three years, analyzing 70,280 observations through Hidden Markov Models. The research identified three distinct learning persistence states and found that learning attention and satisfaction were most strongly associated with continuous learning, while affective identification and immersion showed unexpected negative or inconsistent relationships. The findings suggest that continuous learning is a dynamic, state-dependent process rather than a stable learner trait.
Why it matters
The results challenge assumptions about which psychological factors support sustained learning in technology-enhanced educational environments, suggesting that institutions should prioritize interventions that enhance attention management and satisfaction rather than solely focusing on emotional engagement or immersion when designing intelligent learning systems.
Understand the Science
IntroductionThis study investigates undergraduates’ continuous learning behavior in intelligent-technology-supported university classrooms. Continuous learning behavior is conceptualized as observable persistence across repeated learning occasions. Drawing on Expectation-Confirmation Theory (ECT) and the Stimulus-Organism-Response (S-O-R) framework, the study examines how technology-enabled instructional and social features are associated with learners’ affective and cognitive processes and their subsequent learning-state transitions.MethodsData were obtained from 2,133 undergraduates at M University across the 2021–2023 College and University Student Survey (CCSS) waves, comprising 70,280 recorded observations, with an average of 32.95 observations per participant. Survey-derived and system-captured indicators were analyzed using Hidden Markov Models (HMMs) to identify latent continuous learning states and transition patterns. Models containing one to eight states were compared, and Support Vector Regression (SVR) was used exploratorily to examine the associations of affective identification, immersion, satisfaction, and learning attention with the identified learning states.ResultsThe three-state HMM yielded the lowest Bayesian Information Criterion (BIC = 3,326.39) and was therefore selected as the optimal specification. The three latent states were interpreted as emerging persistence, established persistence, and high persistence. Exploratory SVR results showed that learning attention had the strongest positive normalized association (0.751), followed by satisfaction (0.434). Affective identification showed a negative overall association (-0.123), although a small positive coefficient was observed in State 2 (0.027), providing only limited state-dependent support for H1. Immersion was negatively associated overall (-0.187) and showed a particularly strong negative coefficient in State 3 (-0.473); thus, H2 was not supported as a uniform positive relationship.DiscussionThe findings suggest that continuous learning behavior is better understood as a dynamic, state-dependent process rather than as a fixed learner characteristic or developmental stage. Learning attention and satisfaction appear to be more consistently associated with persistent learning states than affective identification or immersion. Given the observational design and single-institution sample, the findings should be interpreted as associational rather than causal.