An hybrid deep learning approach for depression prediction from user tweets using feature-rich CNN and bi-directional LSTM

Depression has become one of the most widespread mental health disorders across the globe. Depression is a state of mind which affects how we think, feel, and act. The number of suicides caused by depression has been on the rise for the last several years. This issue needs to be addressed. Considering the rapid growth of various social media platforms and their effect on society and the psychological context of a being, it’s becoming a platform for depressed people to convey feelings and emotions, and to study their behavior by mining their social activity through social media posts. The key objective of our study is to explore the possibility of predicting a user’s mental condition by classifying the depressive from non-depressive ones using Twitter data. Using textual content of the user’s tweet, semantic context in the textual narratives is analyzed by utilizing deep learning models. The proposed model, however, is a hybrid of two deep learning architectures, Convolutional Neural Network (CNN) and bi-directional Long Short-Term Memory (biLSTM) that after optimization obtains an accuracy of 94.28% on benchmark depression dataset containing tweets. CNN-biLSTM model is compared with Recurrent Neural Network (RNN) and CNN model and also with the baseline approaches. Experimental results based on various performance metrics indicate that our model helps to improve predictive performance. To examine the problem more deeply, statistical techniques and visualization approaches were used to show the profound difference between the linguistic representation of depressive and non-depressive content.