Sentiment-aware bidirectional long short-term memory model for depression detection from social media texts

Abstract

The number of people living with depression is growing at an alarming rate. Depression impairs normal life, ranging from loss of performance in studies or work to being the cause for other health issues either due to lack of care or self-harm and in rare cases, even suicide. Early-detection is crucial for a better course of treatment. Younger generations usually tend to avoid seeking medical help due to the fear of being “judged” but on the other hand share their feelings on social media. There have been various studies in the past to detect depression from social media posts but detection accuracy is limited due to a number of factors such as informal language. Modeling emotional states from text continues to be a core challenge in human-centered artificial intelligence (AI). The sentiment-aware composite attention bidirectional long short-term memory (SCAB-Net) model aids in screening regular social media posts from the ones indicative of depression. The results show that the proposed model outperforms nine baseline models- support vector machine (SVM), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM),convolutional neural network (CNN), CNN LSTM, CNN BiLSTM, CNN with attention, BiLSTM with attention, CNN bidirectional gated recurrent unit (BiGRU) achieving an accuracy of 89.85%.