Deep Learning for Depression Detection of Twitter Users

Abstract

Mental illness detection in social media can be considered a complex task, mainly due to the complicated nature of mental disorders. In recent years, this research area has started to evolve with the continuous increase in popularity of social media platforms that became an integral part of people's life. This close relationship between social media platforms and their users has made these platforms to reflect the users' personal life on many levels. In such an environment, researchers are presented with a wealth of information regarding one's life. In addition to the level of complexity in identifying mental illnesses through social media platforms, adopting supervised machine learning approaches such as deep neural networks have not been widely accepted due to the difficulties in obtaining sufficient amounts of annotated training data. Due to these reasons, we try to identify the most effective deep neural network architecture among a few of selected architectures that were successfully used in natural language processing tasks. The chosen architectures are used to detect users with signs of mental illnesses (depression in our case) given limited unstructured text data extracted from the Twitter social media platform.

The Influence of COVID-19 on the Well-Being of People: Big Data Methods for Capturing the Well-Being of Working Adults and Protective Factors Nationwide

“…There are primarily two approaches to taking social media data, as psychological data and coding them into psychological variables. One such approach is machine learning, which includes methods such as regression and Latent Dirichlet Allocation ( Schwartz et al, 2013 ; Park et al, 2015 ), support vector machine ( Hutto and Gilbert, 2015 ), long short-term memory, convolutional neural network ( Wang et al, 2017 ; Husseini Orabi et al, 2018 ), cross autoencoder ( Lin et al, 2014 ), and the more recently introduced transformer-based pre-trained language models ( Zhang et al, 2020 ). The other is rule-based modeling, which includes the widely used Linguistic Inquiry Word Count (LIWC) and the sentiment analysis tool VADER (i.e., Valence Aware Dictionary and sEntiment Reasoner).…”

Detection of Suicidality Among Opioid Users on Reddit: Machine Learning–Based Approach

“…This is especially valuable in this context because drug-related words are often slang, and expressions related to getting high or intoxicated may be falsely categorized as negative when they are not due to the inability of sentiment lexicons to learn domain specificity [ 40 ]. Furthermore, there is no need for exhaustive feature engineering because weights can be learned [ 41 ].…”

“…Kim Yoon [ 42 ] demonstrated the ability of a simple convolutional neural network (CNN) with 1 layer of convolution in sentence classification for multiple data sets, finding that the performance of a simple CNN is comparable with that of traditional methods [ 42 ]. Orabi et al [ 43 ] used CNN and recurrent neural networks (RNNs) to predict depression for Twitter data, showing that CNN performs better than RNN. Johnson and Zhang [ 44 ] successfully used word sequences to classify documents with CNN, whereas Kim and Orabi only classified short, sentence length texts [ 44 ].…”

Do Models of Mental Health Based on Social Media Data Generalize?

“…For example, feature analyses have highlighted higher amounts of negative affect and increased personal pronoun prevalence amongst depressed individuals (Park et al, 2012;De Choudhury et al, 2013). Given these consistencies, the field has largely turned its attention toward optimizing predictive power via state of the art models (Orabi et al, 2018;Song et al, 2018).…”