Tweeting the Blues: Leveraging NLP and Classification Models for Depression Detection

Depression is a global mental health problem while Twitter is a valuable medium on which content indicating emotional affect is generated. Our research draws upon NLP to analyze tweets by the application of Naive Bayes, Support Vector Machines, and Logistic Regression classifiers. The first objective will be the tweets' classification into positive and negative sentiment categories with a secondary goal to explore the possibility of the detection of community characteristics including gender, age, and location. We investigate the tweet content by using NLP in an attempt to pinpoint the sensing patterns and the department that can reliably be used to detect depression. Results of the experiments proved comprehensive model accuracy with Support Vector Machines taking a better position on the matter of correctly classifying depression tweets. Especially, Support Vector Machines gave an accuracy of 85%, and values for precision, recall, and f1 were 0.89, 0.63, and 0.63, respectively. Therefore, the remaining ones for attaining higher accuracy and recall are difficult to achieve in models that are depressed but there is still the necessity for more data generation. This research thus clears the way for NLP in the diagnosis of mental health by providing vital assessment data that can be used by psychologists, researchers, health authorities, and policymakers to develop targeted intervention strategies and advancement of public health goals.