Page 135 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

124 Chapter 5 Depression discovery in cancer communities using deep learning

developmental processes underlying mental functioning” [2]. Ac-
cording to a 2014 report by the World Health Organization [3],
depression is the most common ailment globally, and almost
20% of kids and youngsters have suffered from mental ailments
at some point in their lives. Roughly half of these disorders begin
before the age of 14 years. Furthermore, approximately 23% of
deaths globally are the result of mental and substance use disor-
ders. If the current depression rates endure, it is assessed that
by 2020 the affliction of depression will upsurge to 5.7% of aggre-
gate affliction of all the ailments [4]. This calls the need for smart
systems to investigate people's emotions and discover data sets
for precise and opportune depression detection.
Depression detection on social media is an intricate task.
However, with the continuous surge in the number of online por-
tals and users in the past few years, research in this field has
blossomed in full swing. Social media platforms have now
become a fundamental part of people's reality, reflecting their
private lives within certain precincts. These platforms have
amassed a fortune of detail related to their users and hence serve
as a rich means for detection of mental illnesses.
However, due to the inherent complication of mental disor-
ders and difficulty in detecting mental disorders using social me-
dia platforms, depression detection from social media platforms
continues to remain a challenging task. Further complicating
this task is the lack of sufficient amount of annotated training
data, thereby reducing the applicability of supervised machine
learning (ML) approaches such as deep neural networks. In this
chapter, we find the most effective deep neural network architec-
ture for depression detection in cancer communities chosen out
of some of the deep neural network architectures that have been
successfully used in NLP tasks in the past.

2. Related work

In this section, we present a step-by-step detailed analysis of
related literature. First we will discuss about the field of senti-
ment analysis (SA) on social media. According to a survey carried
out in 2014 [5], largely three SA approaches have evolved over the
years. These three approaches are lexicon based, ML based, and a
hybrid of these two. At the onset, we first present a detailed inves-
tigation of all these approaches and carry out their comparative
analysis with each other with respect to their strengths and weak-
nesses. Next, we list out some of the other approaches prevalent
for SA apart from these three approaches. Then, we carry out a
comprehensive study of the use of SA for depression detection
in online communities.