| Popis: |
Purpose: The global quarantine measures implemented in response to the coronavirus outbreak have resulted in a surge of psychiatric issues, including but not limited to depressive symptoms, anxiety, and sleep disturbances, which have placed a significant burden on healthcare systems worldwide. The most efficacious prevention strategy for depression is the identification of its early warning symptoms. In recent times, the healthcare sector has increasingly embraced a technology-oriented strategy. The examination of social media content possesses the capability to significantly reduce the incidence of depression and, consequently, the number of suicides. Methods: The present study introduces a computational method to detect social media users who exhibit symptoms of depression. The intelligent framework employs machine learning techniques to identify text in social media posts that pertain to depression. In this study, three distinct machine learning models, namely the bidirectional encoder representations from Transformers (BERT), the long short-term memory (LSTM), and the convolutional neural network (CNN) were subjected to 1 Springer Nature 2021 L A T E X template Article Title pre-processing and pre-training using data sets obtained from the Twitter API and the Reddit database. The ultimate goal of this study is to accurately distinguish between depressive and non-depressive content. Results: According to the empirical analysis, the BERT model demonstrated a 92.4% accuracy rate and achieved the most optimal validation loss during the third epoch. Moreover, in comparison to the CNN and LSTM models, the BERT model exhibited greater efficiency across all performance indicators. Conclusion: The BERT model performed exceptionally well in both text categorization and feature extraction when compared to the other pre-training models utilized in this investigation. The BERT model has been refined by the identification of these unique characteristics, making it better suited to the task of identifying depressive-related information in digital documents and social media posts. |
