A Multimodal Deep Learning Approach for Identification of Severity of Reflective Depression

Abstract

Social media consumes a greate time of our dialy times that generate a significant amount of information through expressing feeling and activities, sharing admiral contents, viewing, and more. This information mostly contains valuable discoveries. Despite many attempts to mining such produced data, it is still unexploited in certain issues and attracts many research areas. In this paper, we use the data extracted from social media from female’s pages to detect possibility of depression. A new deep learning model based on the psycholinguistic vocabulary to create the embedding words is developed. First, we extract the features from the data before and after the preprocessing phase. Second, the Convolutional Neural Network (CNN) is used to label the data for extracting the remaining features. Based on the previouse two phases; the developed model succeeded to predict the depression possibilty. For evaluation and comparative analysis purpose, three datasets extracted from twitter are used: these are (DB1: contains 700 samples from different countries; DB2: includes 80 samples from KSA and DB3: it is a benchmark CLPsych shared task 2015). The proposed indicator model proved promising results in predicting depression.