Depression Intensity Estimation via Social Me...

Depression Intensity Estimation via Social Media: A Deep Learning Approach

Author Name : Mrs. Asmita Thube, Prof. Anisara Nadaph, Prof. Sneha Tirth

ABSTRACT

Depression is an important component to the world- wide disease burden. Doctors used to diagnose sad folks by referring to clinical depression criteria in person. The COVID-19 pandemic encapsulates, among other things, significant pressures such as unemployment, mortality, and loneliness. Clinicians must distinguish between demoralization and depression when they are called upon. However, more than 70 percent of patients do not seek medical care in the early stages of depression, resulting in further aggravation of their diseases. Meanwhile, as users increasingly rely on social media to express their sentiments and discuss their daily lives, social media has proven to be an efficient tool for diagnosing physical and mental diseases. In this paper, we harness social media (Twitter) data to forecast depressed persons and characterize their depression intensity in order to assist in sounding an alarm. This problem is modelled as a supervised learning task. In a self-supervised approach, we begin by weakly classifying the Twitter data. To represent each user, a rich set of features is extracted, including emotional, topical, behavioral, user level, and depression-related n-gram features. We train a small long short-term memory (LSTM) network using Swish as an activation function to predict depression intensities using these features. Extensive investigations are carried out to illustrate the efficacy of our technology. We beat baseline models for depression intensity estimate by obtaining the lowest mean squared error of 1.42 and by more than 2percent accuracy over the present state-of-the-art binary classification approach. We noticed that depressive users typically use negative terms like stress and melancholy, post late at night, use a lot of personal pronouns, and occasionally describe personal occurrences.

Keywords: Social Media, Machine Learning, Depression detection, Medical.