Depression detection model based on social media post using sentiment analysis in Bengali language

Sadness is a basic human emotion. In our interview with four experts in psychology, they agreed that a person feeling sad for 14 days or longer can be said to be suffering from chronic sadness. Chronic sadness is a major symptom of depression. Numerous research studies have previously been conducted on the detection of depression. The technique adopted for analysis in those studies is Lexicon-based (text) Analysis, which is a popular type of Sentiment Analysis. Previous research on depression did not address symptom identification. If we were to map text polarity in those studies with depressive symptoms based on expert comments, then the symptoms identified in those research studies would be mapped with sadness. However, there are several other potential symptoms of depression that need to be considered, such as insomnia. To identify this symptom, we propose the new DepSympdetect model. The technique used to develop our model is Lexicon-based Analysis with a linguistic dictionary. Our model is focused on the Bengali language. The main reason for choosing Bengali is that it is one of the most widely spoken languages in the world. Another reason is that no such study has ever been conducted in Bengali to identify depressive symptoms using a linguistic dictionary approach. We enhanced the existing linguistic dictionary that was compiled through an opinion review from a past research study in the Bengali language, resulting in the addition of more than 300 new negative sentiment words and 45-degree modifiers. This expanded linguistic dictionary will enable us to identify more negative words for depression analysis. For data collection, we first obtained the consent of ten participants. Then, we manually crawled their Facebook profiles and groups for posts. In our improved DepSympdetect model, we used a number of new features such as text polarity, statement polarity and time of post. With the help of these features, we successfully identified two depressive symptoms, which are chronic sadness and insomnia. We also sought experts to verify the model and the dataset. The experts provided help by validating the newly added negative sentiment words and the degree modifiers. We measured the accuracy of DepSymdetect model that was 93.33%.

Text 113136.pdf Download (914kB)

Actions (login required)

View Item