Exploring Media Portrayals of People with

Mental Disorders using NLP

Swapna Gottipati

, Mark Chong

, Andrew Lim Wei Kiat

and Benny Haryanto Kawidiredjo

School of Information Systems, Singapore Management University, Singapore

Lee kong Chian Business School, Singapore Management University, Singapore

Keywords: Media Portrayal, Mental Illness, Stigmatization, NLP, Sentiment Analysis, Machine Learning.

Abstract: Media plays an important role in creating an impact in society. Several studies show that news media and

entertainment channels, at times may create overwhelming images of the mental illness that emphasize

criminality and dangerousness. The consequences of such negative impact may impact the audience with

stigma and on the other hand, they impair the self-esteem and help-seeking behavior of the people with mental

disorders. This is the first study to examine the Singapore media’s portrayal of persons with mental disorders

(MDs) using text analytics and natural language processing. To date, most studies on media portrayal of

people with MDs have been conducted in developed Western countries. This study found that media articles

on MDs in Singapore were largely negative in sentiment; even quotes from experts contain aspects of stigma.

In addition, crime-related articles on MDs accounted for a significant portion of the corpus. Our model is also

extended to detect positive health articles that discuss recovery and motivation. We further developed a stigma

classifier based on the machine learning algorithms and text mining techniques. The classifier based on the

XGBoosts performed best with an F1-score around 76%.

1 INTRODUCTION

Mental disorders (MDs) affect a significant segment

of society. According to a survey by the Institute of

Mental Health, one in seven Singaporeans have

experienced an MD in their lifetime (Choo, 2020).

Despite the prevalence of MDs, the afflicted are still

stigmatized: outreach programs have found that

persons with MDs in Singapore felt excluded from

contributing meaningfully to society and did not feel

they could fulfil their personal potential. Media

coverage was cited as one major factor contributing

to this deep-seated stigma (“Understanding the

Quality”, 2020).

This study aims to examine Singapore’s news

portrayal of persons with MDs using text analytics

and natural language processing. To date, the bulk of

studies on media portrayal of people with mental

disorders (MDs) has been conducted in developed

countries such as the U.S., U.K., Canada, and

Australia. No similar study has been conducted in

Singapore, making this the first of its kind. This study

makes two contributions to the literature: First, it

https://orcid.org/0000-0003-3136-3704

provides editors and journalists with important

principles for creating non-stigmatizing stories about

persons with MDs. Second, this study’s approach for

identifying “aspects of stigma” in news articles can

be used by the editorial staff in the newsroom to weed

out stigmatizing messages and thus eliminate their

negative effects.

1.1 Stigma and MDs

Stigma has been defined as “a social construction that

devalues marked members of a community” (Smith,

2007, pp. 235). It involves the categorization of a

person into a group based on a distinguishing

characteristic or mark (Brown et al., 2003; Dovidio et

al., 2003) and results in discrimination, prejudice, or

stereotyping (Smith, 2007). As stigmas are shared

between members of a community, they impact

personal and group interactions (Smith, 2007).

Stigmas and the ensuing negative social attitudes

surrounding those suffering from mental disorders

(MDs) have been key health and social problem

(Corrigan & Penn, 1999). Large portions of

708

Gottipati, S., Chong, M., Kiat, A. and Kawidiredjo, B.

Exploring Media Portrayals of People with Mental Disorders using NLP.

DOI: 10.5220/0010380007080715

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 5: HEALTHINF, pages 708-715

ISBN: 978-989-758-490-9

communities were found to hold highly prejudiced

views and hostile attitudes toward these people

(Bagley et al., 2005; Dahaf, 1997). These negative

attitudes violate the civic rights, self-image, and

family life of those afflicted with MDs, and can

ultimately interfere with their social integration into

the community (Klin & Lemish, 2008).

Stigmatization “often results in the creation of laws

that (a) identify marks (or stigmata), (b) socially

isolate marked groups into geographical locations,

and (c) remove the rights of the marked groups and

other peoples’ obligations to them” (Smith, 2007, p.

235). This process can constrain the stigmatized

person’s access to health care, employment,

education, and housing (Brown et al., 2003; Miller &

Major, 2003], which can directly or indirectly impair

physical health and even lead to death (Smith, 2007).

At its most extreme, the stigmatized may even face

eviction from their communities, have their homes set

on fire (Wiener et al., 2000), or killed (Gilbert, 2010).

1.2 The Media and Stigma

Content analysis studies have shown that the negative

framing of MDs in the media contributes to the

public’s negative attitudes toward the afflicted (Sieff,

2003). Several studies (Granello & Pauley, 2007;

Philo et al., 1994; Thornton & Wahl, 1996; Wahl &

Lefkowits, 1989; McGinty et al., 2014; Corrigan et

al., 2004; Philo, 1998) reported a relationship

between exposure to negative portrayals of those with

MDs and negative attitudes toward them, including

fear of persons with MD. This relationship is true not

only of news stories in newspapers and magazines,

but also of fictional stories in soap operas, film, and

dramas (Philo, 1998).

Significantly, stigma surrounding persons with

MDs has been associated with negative outcomes

such as “poor treatment rates, discriminatory housing

and employment practices, and public opposition to

the expansion of mental health services in local

communities” (McGinty et al., 2018, p. 187). A study

of U.K. television’s depictions of MD argued that the

media imagery fuelled the fear of and hostility toward

MDs, which in turn significantly affected policies of

community care (Rose, 1998).

Nonetheless, longitudinal studies have shown that

campaigns can be effective in targeting the stigma

surrounding mental illness. For example, researchers

found favorable changes in attitudes and challenges

to stigma following the 2009 ‘Time to Change’ anti-

stigma program (Evans-Lacko, 2013). The number of

articles in the UK covering mental illness also saw a

substantial increase from 2008 to 2016, with a small

but significant positive change in newspaper

reporting on mental health topics through an observed

increase in anti-stigmatizing articles (Rhydderch et

al., 2016).

This study aims to examine Singapore’s news

portrayal of persons with MDs using text analytics

and natural language processing. To date, the bulk of

studies on media portrayal of people with mental

disorders (MDs) has been conducted in developed

countries such as the U.S., U.K., Canada, and

Australia. No similar study has been conducted in

Singapore, making this the first of its kind. This study

makes two contributions to the literature: First, it

provides editors and journalists with important

principles/methods for creating non-stigmatizing

stories about persons with MDs. Second, this study’s

approach for identifying “aspects of stigma” in news

articles can be used by the healthcare staff to act upon

when preparing or publishing such articles.

2 MATERIALS & METHODS

We study two text analytics methods to analyse the

articles. Firstly, rule-based approach as the news

articles should be analysed in a heuristic approach to

process human behaviour in discovering stigma in the

sentences. Secondly, we propose a machine learning

based approach for stigma prediction in news articles.

In this section, we present the methodology (data and

analysis) followed by the solution models.

2.1 Data Collection

The data collection process requires three

components; Sources, data crawling, and article

processing.

Data Sources: In consultation with Singapore’s

National Council of Social Services (NCSS), the

scope of the study was narrowed to focus on mental

health-related news articles published by the

Singapore Press Holdings (SPH) and MediaCorp,

Singapore’s largest media organizations.

Specifically, the study focused on The Straits Times

(SPH), The Business Times (SPH) and The New

Paper (SPH), Channel NewsAsia (CNA), and Today.

To commence data collection, a list of mental health-

related terms from a study exploring the portrayal of

mental health in Australian newspapers was used

(Kenez, 2015). These article search keywords were

tweaked to fit the local context (see Table 1).

Crawler: A combination of BeautifulSoup

(Richardson, 2007) and Selenium webdriver

(Narayanan, 2016) was used to crawl for the articles.

Exploring Media Portrayals of People with Mental Disorders using NLP

709

Article Details: As each website had a different

interface, three different programs were required to

retrieve the following article details; title of the

article, body text of the article, author, publisher,

published date, and section that the article belongs to

2.2 Data Preparation

Recall that we propose two different approaches for

the solution design to detect stigma in the articles.

The rule-based approaches are heavily dependent on

the dictionaries. Further, in our preliminary analysis,

we observed that health-related articles also appear

more often with the crime. Therefore, in the data

preparation stage, we consider these two observations

in our data preparation steps.

Lexicon Preparation: Four dictionaries (Table 1)

were prepared to facilitate data cleaning and analysis:

• Stigmatizing phrases: The stigma dictionary

includes phrases such as “crazy”, “lunatic”, “commit

suicide” and “psycho” (Kenez, 2015).

• Crime phrases: The crime dictionary includes

words such as “assault”, “victim”, “threaten”, etc

(“Vocabulary University”, 2020). This list aids to

separating crime and non-crime articles.

• Medical phrases: The medical dictionary

includes words such as “depression”, “depressive”,

“phobia”, “phobic”, etc. This list was created from the

synonyms of the stigma words using Wordnet

(George, 1995).

• Recovery phrases: The recovery dictionary

includes words such as “confidence”, “coping”,

“determination”, etc (Kenez, 2015).

Data Cleaning: The initial results were extensive

but yielded many articles that were not related to

mental health. Thus, several filtering steps were

undertaken to focus on mental health articles.

1. Article Filtering: Preliminary filtering of

articles was done by removing articles found in

sections that are not related to mental health. The

remaining articles were then ranked on the basis of

the number of keyword matches. Articles with fewer

than two mental health keyword matches were found

not to be related to mental health. These articles either

mentioned mental health in passing or were using the

term in a different context. These articles were

dropped from the corpus. Finally, these articles were

manually inspected to ensure that they are related to

mental health. After pruning, a final corpus of 1930

articles remained.

Table 1: Keywords used for the data collection and content

analysis (Examples).

Addiction Disorder, Behavioural Disorder, Bipolar,

Breakdown, Depressed, Depressing, Depression,

Depressive, Depressive Disorder, Despair, Distress,…

Article search ke

words

Bonkers, Commit Suicide, Craziness, Crazy, Deranged,

Freak, Gila, Halfwit, Insane, Loco, Loony, Lunacy, Lunatic,

Mad, Madman, Madness, Maniac, Maniac, Mentally Ill,

Nutcase, Nuts, Psycho, Psychotic, Siao, Unbalanced,..

Stigma dictionary

Acceptance, Actualisation, Adapt, Adapted, Autonomous,

Autonomy, Confidence, Confident, Contentment, Cope,

Coping, Determination, Determined, Efficacy, Esteem,

Fulfilment, Happiness, Happy, Meditate, Meditation, ..

Recovery or wellness dictionary

anxiety, anxious, anxiousness, bipolar, breakdown, clinical,

compulsive, dementia, depress, depressed, depressing,

depression, depressive, disorder, dying, headache, hoard,

hoarder, hurt, injury, manic, mental, nervous, obsessional,..

Medical dictionary

Abuse, Accomplice, Accused, Accuser, Aggravated assault,

Alcohol, Alert, Alias, Alibi, Alienate, Allegation,

Ammunition, Appeal, Armed, Arraignment, Arrest,

Arsenal, Arson, Art forgery, Assailant, Assault, Attack..

Crime dictionary

2. Article Categorization: In this step, the

characteristics of crime versus non-crime articles in

the context of mental health reporting were assessed.

To do this, a crime dictionary (see Table 1) was used

to identify crime articles. A detailed inspection

concluded that articles with fewer than three crime

keyword matches can be confidently classified as

non-crime. However, it was not possible to

confidently classify a document having more than

three crime keyword matches as being related to

crime. Therefore, the rest of the articles had to be

manually determined with the aid of the crime

dictionary.

2.3 Stigma Analysis Methods

In our study on news articles, the analysis is

conducted on both the article-level and quote-level.

1. Article-Level Analysis:

The article-level analysis examined the title,

subtitle as well as the first section of each article.

Based on feedback from NCSS, journalists put the

heaviest effort into crafting these sections of an

article. Similarly, readers tend to focus on the title,

subtitle, and first section, as their attention spans

decline in tandem with the length of the article.

Therefore, these elements are extracted from each

article.

HEALTHINF 2021 - 14th International Conference on Health Informatics

710

2. Quote-Level Analysis:

The quote-level analysis examines only the

quotes in an article. Each quote in an article is

extracted and analyzed on multiple levels. Readers

also tend to read short, highlighted quotes as the

articles might be lengthy. The quotes are usually

placed under the pictures to attract the attention of

readers.

3 SOLUTION DESIGN

3.1 Rule-based Solution Model

Figure 1 shows the rule-based solution model used in

this study to discover the stigmatizing articles.

A. Text Extraction

In this step, for each article, we extract quotes using

python regular expressions such as quotations. The

quotes are used for quote analysis. Similarly, for

article analysis, we extract the title, sub-title, and first

section from the article text. In our preliminary

analysis, the first paragraphs mostly summarise the

information in the news. Moreover, the audience

usually read the first paragraph and the quotations in

the news articles.

B. String Phrase Match

In this step, we perform the textual match of the

keywords using the stigma dictionary to discover the

presence or absence of stigmatizing phrases. We

compiled a list of keywords and their associated

synonym using Wordnet (George, 1995) These

keywords include ‘crazy’, ‘lunatic’, and ‘psycho’.

C. Data Normalization

In sentiment extraction, terms are usually tagged as

negative or positive. In the process of obtaining the

sentiment scores, the medical terms commonly found

in mental health reporting such as depression, panic,

and anxiety are judged to be negative by the sentiment

lexicon. Therefore, such words are filtered through

the use of the medical phrase dictionary.

Figure 1: Rule-based Solution Design.

D. Sentiment scoring

Sentiment analysis computationally identifies the

writer’s feelings and attitude towards a particular

topic – i.e. positive, negative, or neutral. Three types

of lexicon-based sentiment analyses are compared:

Lexicon based UIC Sentiment Analysis (Liu, 2015),

TextBlob Sentiment Analysis (TextBlob, 2020), and

Vader Sentiment Analysis (Hutto and Gilbert, 2014).

a) Lexicon: Liu, 2015 compiled a list of words

for sentiment detection (Liu, 2015). It consists of

around 6,800 positive and negative sentiment words

in the English language. An article’s score is the sum

of the scores of all its words.

b) TextBlob: This sentiment module has a

default implementation, NaiveBayesAnalyzer – an

NLTK classifier trained on movie reviews. TextBlob

returns 2 values, ‘Polarity’ and “Subjectivity”.

c) VADER: This is a rule-based and lexicon

sentiment analysis tool and unlike the typical bag-of-

words model, VADER also implements the

grammatical and syntactical rules. We consider the

compound score to identify the overall sentiment.

E. Data Visualization

Consolidation of the results is achieved by computing

the overall outputs from the stigmatizing phrase

analysis step and the sentiment analysis step. Both

steps help to indicate the presence of stigma in a given

text. The Tableau dashboard is used to visualize and

analyze the results.

3.2 Machine Learning Model

The machine learning approach for this study

attempts to develop a stigma classification model

using supervised learning techniques. Ground truth of

manually labeled mental health articles will be

formulated and be used to train the models. With

input from NCSS, a Google Form survey was

designed to conduct the labeling process. The survey

contains the following questions:

1. Do you think that the title of the article is

stigmatizing?

If yes, please identify the words/phrases in the

title that are stigmatizing

2. Do you think that the main (body) of the

article is stigmatizing?

If yes, please identify sentences in the main article

that are stigmatizing. Then identify a few specific

words/phrases in the main article that are stigmatizing

3. Is this article crime-related?

If yes, does the article imply that the person's

mental health condition caused the crime? If you

answered yes to the above question, what are the

problematic statements/phrases?

Questions 1 and 2 attempt to identify whether the

title or the body of the article is stigmatizing.

Exploring Media Portrayals of People with Mental Disorders using NLP

711

Question 3 is only relevant to crime-related articles

and attempts to find out if a correlation is made

between the mental health condition and crime.

The classifier development process would

involve text preprocessing, model training/tuning,

and model selection. The best performing model

would then be selected based on our selected

performance metric, F1-Score and applying cross-

validation. In this study, we implemented four

classifiers (Christopher, 2006) as described below.

A. Random forest: RF is a classification

algorithm that works by forming multiple decision

trees at training and testing it outputs the class that is

the mode of the classes (classification).

B. Log-regression: Logistic regression (LR)

statistical method is used for analyzing the dataset

and produces a binary outcome. It is a specific

category of regression and it is used in the best way

to predict the binary and categorical output. The LR

is the fast prediction algorithm.

C. Support vector machines: SVM classifier

represents the instances as a set of points of 2 types in

N-dimensional space and generates an (N - 1)

dimensional hyperplane to separate those points into

2 groups. SVM attempts to find a straight line that

separates those points into 2 types and is situated as

far as possible from all those points.

D. XGBoosts: XGBoost is a decision-tree-

based ensemble machine learning algorithm that uses

a gradient boosting framework. It provides a stronger

regularization framework that constrains overfitting

to overcome this shortcoming. Therefore, it has

gained much popularity recently and has become a

state-of-the-art machine learning algorithm.

4 EXPERIMENTS

4.1 Rules-based Evaluations

In this section, we first present the results of the

evaluation of the steps in the rule-based model.

Table 2: Sentiment classification results.

Model Precision Recall F1-Score Accuracy

Vader 0.71 0.77 0.73 0.73

Textblob 0.59 0.61 0.58 0.63

Lexicon 0.7 0.69 0.58 0.61

As depicted in Table 2, Vader sentiment classifier

is the most optimal one in terms of the F1- Score

which is the harmonic mean between the recall and

the precision. Another curious fact is that positive and

neutral labels are effectively separable by Vader

unlike the negative variables. This could be because

extreme negative keywords indicating the stigma and

crime are removed from the data processing stage.

From this analysis, we choose Vader as the best tool

and continued with the sentiment classification for all

the articles.

4.2 Machine Learning Evaluations

We asked the human judges to label 154 articles. Out

of 154 articles being labeled, we found that 22% of

the title to be stigmatizing. We then combined these

titles with the stigmatizing sentence found in the

article body and formed the labeled dataset. This

dataset consists of a total of 218 sentences and titles,

with 99 out of 218 sentences found to be stigmatizing.

For our supervised learning model, we use TFIDF

(Term Frequency – Inverse Document Frequency) as

features. All the classification models were tested and

we used Cross-Validation and F1-Score as

performance metrics to evaluate and select the best

model. Cross-validation with 5 folds was used because

our dataset is small. In this method, the dataset is

divided into five folds. Each run, one-fold will act as a

validation set and the rest will act as the training set.

F1-Score evaluation metric is selected as our

evaluation metric since it considers both the precision

and recall of the test to compute the score. We

perform parameter tuning on all the models using

Random Grid Search with F1 score as our target. The

parameters corresponding to the best F1 score were

then employed to train our model. Table 3 depicts the

model results.

Table 3: Stigma classifier performance.

Model Precision Recall F1-score accuracy

RF 0.824 0.667 0.737 0.773

LR 0.786 0.524 0.629 0.705

SVM 0.833 0.476 0.606 0.705

XGBoost 0.762 0.762 0.762 0.772

XGBoost yielded the best F-Score of 0.762 and it

is selected as our machine learning model. Our

parameters for the model are as follows; 'subsample':

0.8, 'min_child_weight': 1, 'max_depth': 11, 'gamma':

1. Stigma classifier based on XGBoosts is suitable to

predict the stigma on the news articles at the sentence

level.

5 INSIGHTS FROM THE DATA

5.1 Exploratory Analysis of Articles

The article statistics are shown in Figure 2. The

quarter-to-quarter distribution of mental health-

HEALTHINF 2021 - 14th International Conference on Health Informatics

712

related articles remained fairly consistent throughout

the period under study, averaging around 130 articles

each quarter. The number of articles peaked in Q3 of

2016 – this can be attributed to an increase in the

number of editorials on suicide awareness after the

Samaritans of Singapore (SOS) annual report showed

that the number of suicide cases in 2016 was the

highest in recent years.

Figure 2: Distribution of articles over the years.

The Straits Times accounted for nearly half the

total number of articles produced during the period.

The next three publishers, Today Online, Channel

NewsAsia, and The New Paper each published

roughly the same number of articles. The Business

Times had the least number of articles related to

mental health, with just 39 articles across the three-

year period (see Figure 2b).

5.2 Stigma Analysis in Articles

5.2.1 Sentiment Analysis and Crime

Proportion

Articles about crime-related mental health issues

made up about 41% of all articles on mental health

The overall sentiment of mental health-related

articles tends to be negative – negative articles made

up more than half of the corpus (see Figure 3).

Distribution of articles by

sentiment

Distribution of crime

articles b

sentiments

Figure 3: Article statistics by types and sentiments.

Of 41% of crime articles, about 88% of articles

are negatively polarised. This is expected as the

nature of crime words is usually negative. A high

proportion of crime articles could influence the public

to equate MDs with criminality. In the next sub-

sections, we shall study the stigma in the articles.

5.2.2 Stigma Analysis

1. Analysis – Article level

Depression was the most frequently mentioned

MD (see Figure 4). According to the World Health

Organization, depression is the leading cause of

disabilities worldwide and the most common MD in

Singapore (“Depression Vocabulary”, 2020),

affecting one in sixteen Singaporeans. The next two

most common MDs were anxiety and trauma.

Hoarding was the least reported MD.

The color coding represents how positively or

negatively each MD was portrayed; a negative

portrayal is shown in red and a positive portrayal is

shown in green. Personality disorder, paranoia and

brief psychotic disorder were the most negatively

portrayed MD. Hoarding was the only mental

condition with a neutral sentiment. “Mentally ill”,

“psychotic” and “commit suicide” are the top three

frequently used terms to label people. At the article

level, we observed 1.5% of titles are stigmatized,

1.55% of sub-titles are stigmatized and 6.6% of first

sections are stigmatized. Though the numbers are

very low, given the impact of stigma, detecting such

articles during the editorial stage is critical.

Figure 4: Distribution of the type of mental health

condition.

2. Analysis – Quote level

An analysis of the quotes in each article was

conducted to determine if the quotes contain

stigmatizing elements. The result of a sample quote

analysis is shown in Figure 5.

In this example, the word “crazy” in the quote

perpetuates the negative stereotype that people with

MDs are divorced from reality, irrational, or

incapable of making decisions.

Exploring Media Portrayals of People with Mental Disorders using NLP

713

Our model extracted 9914 quotes in total and we

observed only 20 quotes which are 0.002% of quotes

are stigmatized. This indicates that in Singapore's

new articles, the stigma is low in terms of the quoted

content compare to the article title and first sections.

Figure 5: Quote analysis.

5.3 Media Positive Impact Analysis

Articles on wellbeing and recovery help to educate

the public about mental health conditions, decrease

discrimination towards persons with mental

conditions, and encourage help-seeking behavior. To

find these articles, keywords from a recovery

dictionary were used. In total, these articles made up

about 15% of all mental health reports. These articles

are more likely to be positive in sentiment and

frequently appear in the lifestyle and commentary

sections of the media (see Figure 6).

Figure 6: Recovery and wellbeing articles.

6 DISCUSSIONS

This study shows that mental health-related articles in

Singapore were primarily negative in sentiment and

crime-related articles accounted for a significant

portion (40%) of the corpus (see Figure 3). The high

proportion of articles associating MDs with crime is

troubling, as it may influence the public to equate

MDs with danger and violence. To complicate

matters, the quotes – including those given by experts

– contain stigma aspects. As experts have been rated

as the most credible source of information (Edelman,

2020), their words carry disproportionate weight.

As media articles have the potential to promote

mental health and contribute to de-stigmatization

(Klin & Lemish, 2008), editors can play a more

proactive gatekeeping role and counteract the largely

negative portrayal of MDs with articles on wellbeing

and recovery, which accounted for only 15% of the

corpus. In addition, they (and their newsrooms) can

use the solution model in this study to identify and

weed out stigmatizing elements in their articles about

MDs.

There are three limitations to this study. First,

although a keyword-based approach was used to filter

the articles, it is possible that the keyword list is non-

exhaustive. This may have resulted in the unintended

exclusion of some mental health articles from the

study. Second, only the first section of each article

was analyzed for stigma. Future studies can overcome

this limitation by applying the model to the entire

article text. Finally, our medical dictionary is limited

to medical mental illness terms only. There might be

other medical terms in the articles which may affect

the sentiment scores of the mode. In particular, they

may have impacted the negative scoring by the

sentiment model. A machine-learning approach can

overcome this limitation by training the model with

both the stigma and non-stigma datasets.

7 CONCLUSIONS

This study analyzed Singapore media articles to shed

light on media stigmatization of MDs in the country.

It proposes a rule-based solution model based on text

mining and NLP techniques which can automatically

identify aspects of stigma in media articles. In

addition to flagging stigmatizing articles, the model

can identify specific sentences and quotes that are

stigmatizing. Such a model can help editors to remove

stigmatizing elements before the publication of an

article. At the same time, however, it can identify

positive articles (e.g. about recovery help and help-

seeking behavior). The model’s dual advantage can

empower editors to address the negative tone that is

prevalent in media coverage of MDs. Last, this model

can be used to run a course for journalists who cover

MDs. Even though this study focused on news

articles, the methods of analysis and solution model

can be extended to other media, such as social media,

blogs, opinion articles, and expert reviews.

REFERENCES

Bagley, C., King, M.(2005) Exploration of three stigma

scales in 83 users of mental health services:

Implications for campaigns to reduce stigma. J. Ment.

Health. 2005, 14, 343–355.

HEALTHINF 2021 - 14th International Conference on Health Informatics

714

Brown, L.; Macintyre, K.; Trujillo, L. (2003). Interventions

to reduce HIV=AIDS stigma: What have we learned?

AIDS Educ. Prev. 2003, 15, 49–69.

Chen, Tianqi; Guestrin, Carlos (2016). "XGBoost: A

Scalable Tree Boosting System". 22nd ACM SIGKDD

International Conference on Knowledge Discovery and

Data Mining, 2016. ACM.

Choo, C. (2020). More people have experienced a mental

disorder in their lifetimes. Today. Available online:

https://www.todayonline.com/singapore/more-people-

singapore-have-experienced-mental-disorder-their-

lifetime-study-finds (accessed on 8 November 2020).

Christopher M. Bishop. (2006) Pattern Recognition and

Machine Learning.

Corrigan, P. W., Penn, D. L.(1999). Lessons from social

psychology on discrediting psychiatric stigma. Am.

Psychol. 1999, 54, 765–776.

Corrigan, P. W., Watson, A.C., Warpinski, A.C., Gracia, G.

(2004). Stigmatizing attitudes about mental illness and

allocation of resources to mental health services.

Community Ment Health J 2004, 40, 297–307.

Dahaf Survey. (1997). Is Disability Contagious? Findings

of an Opinion Survey. Publisher: Yafo: Dahaf Institute,

Israel, 1997.

Depression Vocabulary. (2020). Available online:

https://www.who.int/news-room/fact-

sheets/detail/depression (accessed on 8 April 2020).

Dovidio, J.; Major, B.; Crocker, J. Stigma. (2003)

Introduction and overview. In The Social Psychology

of Stigma, 1st ed. New York, USA, 2003, pp. 1–28.

Edelman Trust Barometer 2020. Edelman. (2020).

Available online: https://www.edelman.com/sites/g

/files/aatuss191/files/202001/2020%20Edelman%20Tr

ust%20Barometer%20 Global%20Report_LIVE.pdf

(accessed on 12 Novermber 2020).

Evans-Lacko, S. M. (2013) Influence of Time to Change’s

social marketing interventions on stigma in England

2009–2011. Brit J Psychiat 2013, 202, s77-s88.

George A. Miller (1995). WordNet: A Lexical Database for

English. Communications of the ACM Vol. 38.

Gilbert, R. (2010). Strange notions: Treatments of early

modern hermaphrodites. In Madness, Disability, and

Social Exclusion: The Archaeology and Anthropology

of ‘‘Difference”, London, UK, 2010, pp. 168–179.

Granello, D. H., Pauley, P. S. (2000). Television viewing

habits and their relationship to tolerance toward people

with mental illnesses. J Ment. Health Couns. 2000.

Hutto, C.J., Gilbert, E.E. (2014). VADER: A parsimonious

rule-based model for sentiment analysis of social media

text. Eighth International Conference on Weblogs and

Social Media (ICWSM-14). Ann Arbor, MI, June 2014.

Kenez, S. (2015) The portrayal of mental health in

Australian daily newspapers. Aust NZ J Publ Heal

2015, 39, 513-517.

Klin, A., Lemish, D. (2008). Mental disorders stigma in the

media: Review of studies on production, content, and

influences. J. Health. Commun. 2008, 13, 434-49.

Liu, B. (2015). Sentiment Analysis: Mining Opinions,

Sentiments, and Emotions. Cambridge: Cambridge

University Press. doi:10.1017/CBO9781139084789.

McGinty, E. E., Webster, D.W., Jarlenski, M., Barry, C.L.

(2014). News media framing of serious mental illness

and gun violence in the United States, 1997–2012. Am

J Public Health 2014, 104, 406–13.

McGinty, E.E., Goldman, H.H., Pescosolido, B.A., Barry,

C.L. (2018) Communicating about mental illness and

violence: Balancing stigma and increased support for

services. J. Health Politics Policy Law 2018

Miller, C. T., Major, B. (2003). Coping with stigma and

prejudice. In The Social Psychology of Stigma, 1st ed.;

Eds.; Guilford Press: New York, USA, 2003.

Narayanan Palani. (2016). Selenium Webdriver: Software

Automation Testing Secrets Revealed Part 2.

eBooks2go, Schaumburg, IL.

Philo, G. (1998). Media and mental illness. In Message

Received: Glasgow Media Group Research, 1993–

1998, 1st ed., Philo, G., Ed.; Addison Wesley

Longman: New York, USA, 1998, pp. 54–61.

Philo, G., Secker, J., Platt, S., Henderson, L., McLaughlin,

G., Burnside, J. (1994). The impact of the mass media

on public images of mental illness: Media content and

audience belief. Health Educ. J. 1994, 53, 271–281.

Rhydderch, D., Krooupa, A.M., Shefer, G., et al. (2016).

Changes in newspaper coverage of mental illness from

2008 to 2014 in England. Acta Psychiatr Scand 2016.

Richardson, L., (2007). Beautiful soup documentation.

April, 2007.

Rose, D. (1998). Television, madness and community care.

J Community Appl Soc Psychol 1998, 8, 213–228.

Sieff, E. M. (2003) Media frames of mental illnesses: The

potential impact of negative frames. J. Ment. Health.

2003, 12, 259–269.

Smith, R. (2007). Media depictions of health topics:

Challenge and stigma formats. J. Health. Commun.

2007, 12, 223-49.

TextBlob, 2020. Textblob: Simplified Text Processing —

Textblob 0.16.0 Documentation. [online]

Textblob.readthedocs.io. Available at:

https://textblob.readthedocs.io/en/dev/ [Accessed 12

Novermber 2020].

Thornton, J. A., Wahl, O. F. (1996). Impact of a newspaper

article on attitudes toward mental illness. J. Community

Psychol. 1996, 24, 17–25.

Understanding the Quality of Life of Adults with Mental

Health Issues. (2020) Available online:

https://www.ncss.gov.sg/Press-

Room/Publications/Detail-Page?id=Understanding-

the-Quality-of-Life-of-Adults-with-M (accessed on 8

April 2020).

Vocaburary University. (2020). Vocaburary Word List.

Retrieved from MyVocaburary.com:

https://myvocabulary.com/word-list/crime-

vocabulary/. (accessed on 12 Novermber 2020).

Wahl, O. F., Lefkowits, J. Y. (1989). Impact of a television

film on attitudes toward mental illness. Am. J.

Community Psychol. 1989, 17, 521–527.

Wiener, L. S., Battles, H. B., Heilman, N. (2000). Public

disclosure of child’s HIV infection: Impact on children

and families. AIDS Patient Care STDs 2000.

Exploring Media Portrayals of People with Mental Disorders using NLP

715