Multimodal, multiview and multitasking depression detection framework endorsed with auxiliary sentiment polarity and emotion detection

The impact of online social media has aided the users in sharing of knowledge, mood, feelings, and interests to the large volume of audience. The mental health of a person can be easily identified by analysing these expressions consisting of different modalities (text and emojis/emoticons). This research work aims to investigate the mood disorder like depression, low mood and other symptoms using tweets and emoticons. The present work curated the twitter based SentiEmoDD dataset as a benchmark for depression detection, labelled with sentiments analysis, emotions detection and other symptoms important for depression detection. The evolved dataset is equipped with both modalities (text and emojis) of tweets. A novel approach has been proposed based on the multi-view ensemble learning model contemplated to attain the information available in different modalities of a sentence for better depression detection. The proposed approach extracts the results from inter ensemble learning model and intra ensemble learning model. The experimental results clearly indicates that multimodal, multi-view and multitasking proposed framework provides an accuracy of 88.29% for the primary task of depression detection SVM linear kernel function. The stacking technique used here, provides the accuracy of 87.69% to detect depression using the proposed algorithm considering all the expressions of emoji and text combinations.