This study aims to detect depression-related content in social media posts automatically. The Sentiment140 dataset was used as the data source, with the text data preprocessed using natural language processing techniques. The processed texts were then converted into numerical vector representations using the GloVe word embedding method and analyzed with various classification algorithms. Deep learning models (CNN, LSTM, BiLSTM) and traditional machine learning algorithms (Naive Bayes, Random Forest, Decision Tree, SVM, Logistic Regression) were evaluated and compared for performance. Evaluation metrics included accuracy, precision, recall, and the F1 score. The experimental results show that deep learning models, particularly the CNN architecture, outperformed traditional classification methods in detecting depression-prone content. Overall, the study presents a practical and applicable approach for conducting mental health analysis based on social media data.
Güneyli et al. (Sun,) studied this question.