Abstract
This study explores the role of psychological support in moderating the relationship between emotion regulation and teacher-student interaction, specifically addressing its impact on loneliness and social isolation among university students in online learning environments. By utilizing a mixed-methods approach, the research examines how emotional regulation strategies like cognitive reappraisal can mitigate feelings of loneliness, while poor emotional regulation, such as emotional suppression, exacerbates isolation. The findings underscore the importance of fostering strong teacher-student interactions to alleviate the psychological and emotional toll of online learning, offering strategies to combat social isolation and enhance well-being. Using a mixed-methods approach combining educational data mining (EDM) and neural network analysis, we analyzed behavioral logs and self-reported data from 1,200 Chinese university students. Results revealed that cognitive reappraisal strategies significantly enhanced psychological fitness, while emotional suppression correlated with adverse outcomes. Teacher-student interaction quality mediated these effects, particularly in online environments. Our findings highlight the need for personalized psychological interventions and data-driven models to support students’ mental health in digital education. This study contributes to educational psychology by integrating machine learning with traditional surveys, offering scalable solutions for cross-disciplinary mental health challenges.
The Digital Isolation Paradox
The rapid shift to online learning has created a significant challenge: while students are more connected to information than ever, they are increasingly isolated from their peers and instructors. This "digital isolation" often leads to decreased psychological fitness, higher burnout rates, and poor academic outcomes. Traditional psychological interventions typically rely on self-reported surveys, which are "lagging indicators"—by the time a student fills out a survey, they may already be in crisis. This research seeks to bridge this gap by using behavioral data (logs of how students interact with online platforms) to predict psychological states and identify the specific mechanisms, such as emotion regulation strategies, that can prevent loneliness.
Methodology: Integrating EDM and Neural Networks
The researchers employed a mixed-methods approach that integrates traditional psychological theory with modern machine learning. The core of this framework is Educational Data Mining (EDM), a four-step process involving data collection from diverse sources, preprocessing to handle complex formats, data mining (classification and clustering), and interpretation for pedagogical decision-making.
To improve predictive accuracy, the study utilized an Extended Neural Network (E.NN) based on a three-layer Backpropagation (BP) architecture optimized with Genetic Algorithms.
The model was further optimized using the Levenberg-Marquardt (LM) algorithm, which proved significantly more efficient than standard gradient descent, achieving convergence in just 88 iterations. Additionally, Natural Language Processing (NLP) was used to perform sentiment analysis on teacher feedback, providing a quantitative measure of interaction quality.
Key Insights: The Mechanics of Emotion Regulation
The study highlights that how students manage their emotions is critical to their well-being in digital spaces. Two primary strategies were identified:
- Cognitive Reappraisal: Reinterpreting a situation to change its emotional impact was a strong predictor of psychological fitness ($\beta = 0.58$).
- Emotional Suppression: Inhibiting the outward expression of emotion was directly linked to burnout ($\beta = -0.32$).
A crucial finding is that teacher-student interaction quality mediated 40% of the relationship between a student's emotional regulation and their psychological fitness. This suggests that supportive instructors can significantly offset the negative effects of poor self-regulation. Furthermore, behavioral patterns on digital platforms (e.g., video engagement, forum participation) were found to be viable proxies for internal psychological states.
Results and Performance
The integrated model, combining network, performance, and behavioral data, achieved a Precision of 0.72, a Recall of 0.75, and an F1-Measure of 0.71. The LM-optimized neural network achieved a convergence precision of $3.15 \times 10^{-13}$, vastly outperforming standard training functions. Sensitivity analysis revealed that the model's robustness lies primarily in feature engineering rather than hyperparameter tuning, as changes in the number of convolution kernels had a negligible effect on results.
Conclusion and Future Work
While the study provides a powerful framework for monitoring student well-being, it is currently limited by its demographic focus on Chinese university students. Cultural differences in emotional expression mean these results require further calibration for global application. Additionally, the authors acknowledge the ethical challenges of monitoring student behavior logs for mental health purposes.
Future research should focus on developing real-time, scalable intervention tools embedded directly into Learning Management Systems (LMS). These tools could provide "emotional scaffolding" the moment a student's behavioral patterns indicate rising isolation, fostering a "people-first" approach in educational technology.
Summary updated