Crying in the algorithm: modeling academic stress via multilayer topic construction and ERA effect.

Amid intensifying educational competition and societal expectations, academic stress has emerged as a multidimensional force influencing student mental health. While prior research has explored individual and institutional factors, limited attention has been paid to how learners semantically construct and express academic stress in digital environments. Addressing this gap, this study introduces an innovative multilayered topic modeling framework that integrates BERTopic and Latent Dirichlet Allocation (LDA), enabling a semantic, data-driven analysis of 33,827 user-generated comments related to academic pressure on social media. Grounded in Multilevel Stress Theory, the analysis identifies six interrelated topics reflecting the interplay of individual, situational, and structural stressors. Drawing on these findings, the study develops the Expectancy-Regulation-Amplification (ERA) Model, which conceptualizes academic stress as a dynamic process shaped by the tension between external expectations and perceived capabilities, limitations in self-regulatory resources, and the cumulative amplification of stress across sociocultural and digital environments. By mapping how academic pressure is linguistically reproduced and sentimentally intensified in algorithmic settings, the ERA model provides an interpretive framework for understanding the semantics of student vulnerability and contributes new insights to targeted interventions in educational and mental health contexts.