Sentiment Analysis from Depression-Related User-Generated Contents from Social Media

In this paper, we try to detect the sentiment levels such as positive, negative and neutral sentiments from depression related posts and comments generated in social media platforms. Social media platforms such as Facebook, Twitter are not only used for communication or building networks among connections, but also are getting useful for supporting needy peoples who are on special need or care in terms of mental support. In Facebook, there are several depression support groups, which are very much effective to provide mental support to the victims. In this paper, we try to formalize the depression-related posts and comments into a concise lexicon database and detect the sentiment levels form each instance. We have segmented the total work into two parts: sentiment detection and applying machine learning algorithms to analyze the ability to detect sentiment from such special category of texts. We have utilized python textblob package to detect the sentiment levels and applied traditional machine learning algorithms such as Naïve Bayes (NB), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Sequential Minimal Optimization (SMO), Logistic Regression (LR), Adaboost (AB), Bagging (Bg), Stacking (St) and Multilayer Perceptron (MP) on the linguistic features. We have determined the precision, recall, F-measure, accuracy, ROC values for each of the classifiers. Among the classifiers Random Forest has outperformed others showing 60.54% correctly classified instance. We believe such sentiment analysis on special category of texts may lead to further investigation in natural language understandings.