<![CDATA[In the digital era, social media has become a primary means for individuals to express emotions, including symptoms of depression. Posts reflecting feelings of despair and loneliness are increasingly common, particularly on platforms like Reddit. This phenomenon underscores the importance of automatically detecting depressive emotions at an early stage through technology-based approaches, to mitigate negative impacts on mental health. This study employs three linguistic approaches—Lexical Base, WordNet, and GLUE—to enrich semantic understanding and enhance model performance in multilabel classification of depressive emotions. A total of 6,037 text data points were used and split into training, validation, and test sets with a ratio of 70%:15%:15%, following initial processing and linguistic preprocessing stages. Evaluation was conducted using precision, recall, and F1-score metrics on both macro and micro averages. Overall, the study indicates that while linguistic approaches such as Lexical Base, WordNet, and GLUE can enrich text representation, their performance does not always surpass BERT without preprocessing. This suggests that the effectiveness of integrating linguistic information is highly dependent on data context, and further research could explore combining it with multimodal data or advanced mechanisms such as attention to improve depressive emotion classification performance. These findings have potential applications in AI-based mental health monitoring systems, such as chatbots or early detection platforms, to assist in automatically identifying depression symptoms in social media users.]]>
Copyrights © 2025
