Deep neural modeling of emotion recognition and subjective well-being assessment for college students based on social media text streams

With the rapid growth of social media platforms such as Reddit, users – particularly college students – actively express their emotions online, providing meaningful and context-rich insights into their mental health conditions. Traditional subjective well-being (SWB) assessment methods, including retrospective questionnaires, are limited because they capture real-time emotional variations effectively. To address this limitation, this study proposes a novel deep learning-based framework that integrates BERT-based emotion recognition with Pearson correlation analysis for real-time SWB assessment. The model is trained and evaluated using the GoEmotions dataset, which consists of 58,000 English-language Reddit comments annotated with 27 distinct emotion categories. BERT is fine-tuned to classify emotions accurately from textual data. The identified emotions are then analyzed using Pearson correlation to examine their relationships with SWB indicators such as happiness and life satisfaction. The proposed approach achieves strong performance, with 98.03% accuracy, 98.04% precision, 98.01% recall, and a 98.02% F1-score. Correlation analysis reveals a significant negative relationship between Positive and Negative Affect (−0.422), as well as weak negative correlations between Positive and Ambiguous Affect (−0.285) and between Negative and Ambiguous Affect (−0.188). The dataset was divided into 80% training and 20% testing, with four training epochs and evaluation conducted on 9955 samples for SWB analysis.