Social Loneliness in Youth: A BERT-based Sentiment Analysis of Digital Texts

Abstract

Background: Digital platforms profoundly shape how young individuals form and maintain social connections. While these platforms can enhance social engagement, they may also intensify feelings of social loneliness, particularly among youth navigating identity and peer relationships.

Objective: This study aims to explore the emotional dimensions of young adults' digital experiences by using Natural Language Processing (NLP)-based sentiment analysis, focusing on the manifestation of social loneliness in online discourse.

Methods: Textual data were collected through semi-structured interviews with 300 university students (ages 18–25) in Turkey regarding their digital platform usage. Sentiment classification was conducted using two models: a baseline Naive Bayes algorithm with TF-IDF vectorization and a BERT-based deep learning model (BERTurk). The models were evaluated based on accuracy, precision, recall and F1-score.

Results: Findings revealed that 40% of participants expressed positive sentiment, 35% negative and 25% neutral regarding digital platform use. The BERT model significantly outperformed Naive Bayes in all performance metrics (F1-score ≈ 0.75), particularly in identifying negative emotional content. Qualitative themes highlighted a dual impact of digital media, serving both as a source of connection and emotional detachment.

Conclusion: Context-sensitive deep learning models provide valuable insights into youth's digital social experiences. This research emphasizes the potential of AI-driven sentiment analysis to uncover psychosocial dynamics like social loneliness and offers implications for digital mental health interventions targeting young populations.