A Systematic Analysis of Depression-Related Discourse Within

Facebook: A Comparison Between Brazilian and American Communities

Silas Lima Filho

1,2 a

, M

onica Ferreira da Silva

2 b

and Jonice Oliveira

1,2 c

Instituto de Computac¸

ao (IC), Universidade Federal do Rio de Janeiro (UFRJ), Rio de Janeiro, Brazil

Programa de P

os-Graduac¸

ao em Inform

atica (PPGI), Universidade Federal do Rio de Janeiro (UFRJ),

Rio de Janeiro, Brazil

ﬁ

Keywords:

Depression, Social Media, Social Networks, Textual Analysis.

Abstract:

Identifying the symptoms of a depressive disorder can help potential sufferers seek professional help, increas-

ing their chances of recovery. This article presents the operationalization of systems and tools to systematize

the analysis process using data from depression-related communities within Facebook. We discuss how we

can utilize the data to understand details about depression and the discourse surrounding the disorder through

textual analysis using LIWC. The results show a low correlation between textual analysis and the features of

social media interaction. This study, through a systematic use of data collection and analysis tools, aims to

provide explanatory insights into messages discussing the topic of depression.

1 INTRODUCTION

Computer science has studied textual language,

speciﬁcally in the natural language processing do-

main. However, this concern is also shared in the

ﬁeld of health studies. In Castro’s work, Lacan’s con-

tributions are discussed and compared with other ap-

proaches regarding the importance of language and

its structuring in the unconscious. As mentioned in

the paragraph above, the author discusses various lan-

guage interpretations. Language is recognized as a

manifestation of already formed thoughts or the up-

shot of an unconscious process (De Castro, 2009).

Social media are potential tools for monitoring

populations in epidemic control, information dissem-

ination, and combating misinformation about certain

diseases (Skaik and Inkpen, 2021). Public health en-

tities can understand patterns in speciﬁc groups or

populations through systems that integrate social me-

dia information to identify public opinion on services

that may not be as good as they should be, iden-

tify individuals at risk, and communicate potential ur-

gent diseases (Horvitz and Mulligan, 2015). The use

of technology directly supports institutions and pro-

https://orcid.org/0000-0002-5611-7297

https://orcid.org/0000-0003-0951-6612

https://orcid.org/0000-0002-2495-1463

fessionals, helping to raise awareness of certain dis-

eases. During the COVID-19 pandemic, many pa-

tients sought information about the disease on social

media (Chen and Wang, 2021).

According to the World Health Organization

(WHO), approximately 300 million people of differ-

ent ages suffer from some level of depression

. De-

pressive disorders are the fourth leading cause of dis-

ability and have progressed steadily over the years

(Brody et al., 2018) and (James et al., 2018). Ma-

jor depressive disorder is classiﬁed as such when the

patient presents a set of predeﬁned symptoms (Asso-

ciation. and Association., 2013), e.g., daily depressed

mood, loss of interest in regular activities, weight

loss, and insomnia.

Several factors can complicate the diagnosis of de-

pressive disorders, intrinsic to the clinical approach,

e.g., costs, longer patient follow-up time by the pro-

fessional, and the number of patients each profes-

sional can assist (Li et al., 2020). In Brazil, there

is unequal access to healthcare professionals and fa-

cilities, which can make identifying mental disorders

challenging. Populations with lower income do not

always have easy access to social media

www.who.int/en/news-room/fact-

sheets/detail/mental-disorders [accessed 01-03-2023]

http://revistapesquisa.fapesp.br/tempos-de-incerteza/

466

Filho, S., Ferreira da Silva, M. and Oliveira, J.

A Systematic Analysis of Depression-Related Discourse Within Facebook: A Comparison Between Brazilian and American Communities.

DOI: 10.5220/0012381100003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 2, pages 466-473

ISBN: 978-989-758-688-0; ISSN: 2184-4305

Given the described scenario, identifying symp-

toms related to depressive disorders and promptly and

discreetly assisting someone who may be a potentially

depressive patient can be very useful, both for the pa-

tient and the professional. This identiﬁcation process

involves challenges such as correlating the same signs

and symptoms of depressive disorders from the clini-

cal environment with the often abundant data from so-

cial media. We present a study on the operationaliza-

tion of tools for the problem of identifying depressive

symptoms and an analysis of the correlation between

different obtained features. We apply textual analy-

sis through psychological variables to understand and

identify possible patterns within the text used by com-

munities discussing depression phenomena. The re-

sults show a low correlation between textual analysis

and the features of social media interaction.

Next, Section 2 presents related works. Section 3

deﬁnes the concepts and supporting tools. Section 4

presents the methodology and research method. Sec-

tion 5 presents the results obtained. Section 6 dis-

cusses the results. Section 7 concludes the work.

2 RELATED WORK

Studies that address the identiﬁcation of depressive

users on social media employ algorithms and pattern

recognition techniques, leveraging Natural Language

Processing (NLP) to perform a systemic analysis of

text in social media posts. They delve into techni-

cal aspects of computational pragmatics, incorporat-

ing information and knowledge from the Health do-

main, such as Psychology and Medicine, and utilize

psychometric questionnaires (Giuntini et al., 2020).

(De Choudhury et al., 2013) collect data from so-

cial media, among the many articles and research on

identifying depression in the population using social

media information. They employ psychometric ques-

tionnaires representing the theory and technique of

measuring mental processes as applied in Psychol-

ogy and Education. In this work, they extracted data

from Twitter from individuals with a clinical diagno-

sis of depression through crowdsourcing. They create

a corpus and develop a probabilistic model to detect

whether a post indicates depression. (Tsugawa et al.,

2015) apply the same methodology but on a group of

Japanese users, analyzing the replicability of results

from previous studies.

Using a different social media platform, (Park

et al., 2015) demonstrates how activities on Facebook

are associated with users’ depressive states. Observ-

ing an increase in the suicide rate among students,

[accessed 01-03-2023]

they aim to raise awareness of depression issues at the

university where the study was conducted. (Andal-

ibi et al., 2017) explore self-disclosure posts on Insta-

gram marked with the #depression tag to understand

which disclosures are highly sensitive in this social

media. (Li et al., 2016) take a qualitative approach

to understand the behavior and describe the Chinese

population’s understanding of depression. It differs

from the previous studies in that it explores post-

disclosure without creating a classiﬁcation model and

does not rely on a quantitative epistemology.

An observational study was conducted to under-

stand the interactions between clinically depressed

users and their network connections compared to

a group of users without depression (Vedula and

Parthasarathy, 2017). The authors identify relevant

linguistic and emotional signals in social media ex-

changes to detect symptomatic signs of depression.

3 CONCEPTS

This work adopts a textual analysis approach, consis-

tent with the literature, to ﬁnd helpful information that

can aid in identifying symptoms of depressive disor-

ders. Tools for data capture from social media, text

analysis, and statistical analysis were utilized. The

Crowdtangle tool was used for data capture, prepara-

tion, and preprocessing, which were carried out using

Python routines, and the captured message data was

selected for textual analysis.

Crowdtangle is a research tool that collects data

from public proﬁles and groups for link veriﬁcation

and post monitoring. It monitors over 7 million ver-

iﬁed Facebook public pages, groups, and proﬁles,

2 million public Instagram accounts, and 20k sub-

reddits. Two post-monitoring approaches are avail-

able: deﬁning sources or searching for popular posts.

Crowdtangle returns the number of reactions to a post

(likes, reactions, and shares) and calculates the inter-

action rate of a post compared to the interaction his-

tory of a page or community

The Linguistic Inquiry and Word Count (LIWC)

system is used in related works. It allows for text

processing and analysis using a lexicon predeﬁned by

dictionaries, where words are categorized into differ-

ent domains. It was developed for exploratory text

analysis, with the premise that daily words carry psy-

chological characteristics, e.g., emotions, beliefs, and

habits (Boyd et al., 2022). It provides processing and

analysis modules for various purposes. Its primary

www.crowdtangle.com/resources [accessed 01-03-

2023]

A Systematic Analysis of Depression-Related Discourse Within Facebook: A Comparison Between Brazilian and American Communities

467

analysis module uses dictionaries where each ana-

lyzed word is compared with those already deﬁned

in the dictionaries, with their pre-established values

in different domains. One of the advantages is the en-

hancement of the original dictionary by including new

terms. The Brazilian Portuguese version is limited to

an older version of the original dictionary, dating back

to 2015.

The primary analysis module allows for the quan-

tiﬁcation of four aspects. Analytical thinking cap-

tures the extent to which people use words that sug-

gest formal, logical, and hierarchical thinking pat-

terns. People with low levels of analytical thinking

tend to write and think using more intuitive and per-

sonal language. High language scores in Analytical

Thinking are rewarded in academic environments and

correlate with grades and reasoning skills. Language

with low scores in analytical thinking is often seen as

less formal and rigid and more friendly and personal.

The second aspect is Clout, which refers to the

relative social status, trust, or leadership that people

display through their writing or speaking. The third

aspect, Authenticity, deals with when people reveal

themselves as ”authentic” or honest. They tend to

speak more spontaneously and do not self-regulate or

ﬁlter their words. Examples of texts with low authen-

ticity scores include prepared texts (i.e., pre-written

speeches) and texts in which a person is socially cau-

tious. Examples of texts with high authenticity scores

tend to be spontaneous conversations between close

friends or political leaders with little or no social in-

hibition.

Finally, although LIWC-22 includes positive and

negative tone dimensions, Emotional Tone combines

these two dimensions into a single summary variable.

The algorithm is constructed so that the higher the

number, the more positive the tone. Numbers below

50 suggest a more negative emotional tone.

4 METHODOLOGY

This work presents an empirical post-positivist ap-

proach (Wohlin et al., 2012), an explanatory perspec-

tive on depressive disorders. Data analysis extracts in-

formation and patterns that explain the phenomenon

and problem addressed without exhausting all pos-

sibilities and subject to improvement (Creswell and

Creswell, 2018). While the primary goal of related

works is the development of artifacts and constructs

to identify depressive users on social media, there is

room for those that aim to investigate and analyze

these phenomena and behaviors (Recker, 2012).

It is challenging to develop computational systems

considering the sociotechnical aspect as a relevant

factor, not just the technical factor (Boscarioli et al.,

2017), considering subjective and humanistic aspects.

The sociotechnical factor contributes to the construc-

tion of solutions in society (Cafezeiro et al., 2017).

Previous research was conducted among healthcare

professionals to explore which information is most

relevant to them in therapy or a similar process for

identifying someone with depression. They assess the

relevance of a computational and technological arti-

fact in patient care and how such an artifact could be

helpful as support for verifying patients’ social media

data (Lima Filho et al., 2022).

This research uses the concepts of the Design Sci-

ence Research (DSR) approach (Wieringa, 2014) as a

reference. Given that the literature deﬁnes that an ar-

tifact should be relevant to domain experts who deal

directly with the problem at hand (Pimentel et al.,

2020). In this research context, the artifact created

aims to be useful for healthcare professionals, such as

psychologists or psychiatrists.

Grounded in the Design Science Research (DSR)

paradigm with a focus on supporting healthcare pro-

fessionals, this research adopts a mixed-method ap-

proach, aligning with the post-positivist epistemo-

logical perspective described by Creswell (Creswell

and Creswell, 2018). This paradigm acknowledges

the traditional reductionist approach in scientiﬁc re-

search, where data acquisition and analysis aim to

comprehend the complexity of ”reality,” emphasizing

the absence of absolute truth and the inherent suscep-

tibility of evidence to imperfections. In adhering to

this methodological stance, the study employs corre-

lation analyses that integrate qualitative and quanti-

tative elements, exploring metrics and variables from

analysis tools and scrutinizing speciﬁcities in online

communities. The analyses conducted herein lay the

groundwork for future research focused on addressing

the overarching question: ”How to identify symptoms

of psychological diseases through social media?”

4.1 Method

The research method is divided into two stages: the

data collection stage and the analysis stage. Both

stages and their implementation details are described

below.

4.1.1 Data Collection

The data were collected from open Facebook commu-

nities that directly discuss depression or related top-

ics, obtained through Crowdtangle. Communities dis-

cussing depression-related issues in both English and

Brazilian Portuguese were sought in September 2022,

HEALTHINF 2024 - 17th International Conference on Health Informatics

468

a period when media actions usually promote mental

health care.

Crowdtangle allows the ﬁltering of the type of

posts to be selected, such as video posts, images, and

text. Textual content posts of the “status” type, where

the user generates textual content, were selected. The

search was limited to open groups only. In this type

of search, the tool excludes paid advertisements, ver-

iﬁed proﬁles, and Facebook pages, which may repre-

sent companies.

Initially, depression-related terms were deﬁned

based on works that used the same approach after an

extensive systematic literature review. The terms de-

ﬁned in Brazilian Portuguese and English were, re-

spectively, in Brazilian Portuguese: “quero morrer (I

want to die)”, depress

ao (depression), deprimid (de-

press), depressiv, angustia (anguish); in English: de-

pression, depressed, depressive, anguish, distress.

To reﬁne the search, the focus was on groups’

status that directly discussed depression. There

are a total of four English communities that objec-

tively discuss depression, totaling 835 posts. Seven

Portuguese-language communities, including 1945

posts, are divided into four communities that directly

discuss depression, two on psychological treatment,

and one on mourning.

4.1.2 Data Analysis

Relevant attributes from the data collection were se-

lected as continuous: “Post Views,” “Total Views,”

“Post Created,” “Post Created Date”, “Post Created

Time”, “Total Interactions”, “Likes”, “Comments”,

“Shares”, “Love”, “Wow”, “Haha”, “Sad”, “Angry”,

“Care”, and “OverperformingScore”; or as categor-

ical: “Type”, “Group Name”, “User Name”, “Page

Category”, “Facebook Id”, “Message”, and “URL”.

The categorical attribute that identiﬁes the post au-

thor is null; therefore, it does not allow system users

to access such sensitive data. Only the “Message” at-

tribute was selected to analyze the content of these

posts. An approach adopted in this method was the

translation of community messages from Portuguese

to English. Thus, the exact version of the lexical was

applied to both communities. Given that the most

recent version of the Portuguese dictionary is from

2015. This would make using the metrics mentioned

in Section 3 impossible. After proper selection and

export to a .csv ﬁle, the data set with only the in-

dex and message text was analyzed in the LIWC tool.

Classiﬁcations of the most commonly used words in

each community were generated, and the attributes of

the main analysis module of the tool were also ana-

lyzed. The step for processing textual data, i.e., pre-

processing, was performed in the LIWC system.

With the values of the attributes described in Sec-

tion 4.1.2, Pearson and Spearman correlation mea-

sures were applied between the continuous variables

obtained by Crowdtangle and the attributes obtained

by LIWC. The Pearson correlation measure is a para-

metric measure, while Spearman is non-parametric.

Both generate values between -1 and +1. A value

of -1 indicates a negative correlation, meaning that

when one variable increases, the correlated variable

decreases. A positive correlation, or a correlation

value of +1, indicates that changes in one variable

affect the behavior of the other. A value closer to 0

implies a weak or nonexistent correlation. Pearson’s

correlation is used for linear data, while Spearman is

used for non-linear data.

4.2 Ethical Aspects

The Certiﬁcate of Presentation and Ethical Apprecia-

tion (CAAE) assigned by the Research Ethics Com-

mittee (CEP) is 54865821.5.0000.5263. The data do

not identify users since the collection system provides

data sets without user identiﬁcation, and no additional

sensitive user data was collected.

5 RESULTS

The ﬁrst data analysis involved quantifying the fre-

quency of word usage to discover the most commonly

used words in each community. Figure 1 presents the

most repeated words in Brazilian Portuguese commu-

nities. The highlighted words in Figure 1 are irrele-

vant since no preprocessing was applied to the Por-

tuguese words, as the LIWC system does not have

data processing for this language.

Figure 2 represents the same community but with

an automatic translation approach using the Python

library googleTrans

. Meanwhile, Figure 3 shows the

most used words in English-language communities.

Table 1 provides detailed frequency occurrences,

showing the most used words in English and Brazilian

Portuguese communities. It is worth noting that the

top four words in both tables are the same, only dif-

fering in their positions. Therefore, the words “anxi-

ety, depression, people,” and “life” are the most com-

monly used in both English and Portuguese commu-

nities. Some words further down the ranking, even

though their positions may vary, are repeated, such as

“pain, day,” and “time.”

Metrics values for “authenticity,” “inﬂuence,”

“emotional tone,” and “analytical thinking” are also

github.com/ssut/py-googletrans [accessed on 29-05-

2023]

A Systematic Analysis of Depression-Related Discourse Within Facebook: A Comparison Between Brazilian and American Communities

469

Figure 1: Wordcloud of communities in Portuguese.

Figure 2: Wordcloud of words translated into English from

communities in Portuguese.

obtained for both data sets. Figure 4 shows the dis-

tribution of these values for messages in English and

Brazilian Portuguese communities. The metrics mea-

sured for these attributes range from 0 to 100. Despite

differences between the communities, the nuances in

data distribution have some similarities. It’s impor-

tant to remember that the number of messages differs

for each community, with 835 messages for English

and 1945 for Brazilian Portuguese communities.

In the “Analytical Thinking” attribute, the highest

concentration of messages is found with low scores (0

to 30), which applies to both communities. There is

a more signiﬁcant distribution of higher scores in the

Portuguese community. This dynamic also applies to

the ”Inﬂuence” attribute, with a higher concentration

for low scores (between 0 and 10). However, for both

communities, there is a slight increase for a score of

Figure 3: Wordcloud of communities in English.

40 and between 90 and 100. The ”Authenticity” at-

tribute has a higher concentration of messages with

scores between 90 and 100, but it also has a concen-

tration of messages with low scores between 0 and

20. The score for the ”Emotional Tone” attribute has

three main concentrations. The most signiﬁcant con-

centration is for lower scores, between 0 and 10. The

second-largest concentration is around 20. The last

concentration is for higher values, between 90 and

100.

Table 3 provides values representing statistical

measures for reactions, likes, shares, and the num-

ber of comments on posts from each community

analyzed. In English communities, speciﬁc reac-

tions are more commonly used than others. The

most frequently used reactions fall into the cate-

gories of “Love,” “Sad,” and “Care,” while responses

like “Wow,” “Haha,” and “Angry” are less commonly

used. The reactions most widely used are those of

“Sad,” followed by “Care,” and then “Love.” These

reactions also have a higher dispersion in their sum

for each post, while reactions with lower averages

have less dispersion and, therefore, more uniformity

among the posts. The average number of likes and

shares corresponds to the amount of “Sad” reactions.

In Brazilian Portuguese communities, the se-

quence of average reactions is similar to the previous

case. There is a more minor difference in the average

between “Sad” and “Care” reactions and a more sig-

niﬁcant difference between “Care” and “Love” reac-

tions. Therefore, the ﬁrst two reactions are used more

frequently in Portuguese communities. The standard

deviation for the most commonly used reactions in-

dicates greater dispersion in the most used reactions,

similar to the English community, and less dispersion

in the less used reactions. However, the standard de-

viation value for the “Sad” reaction (the most used in

both communities) is considerably lower in the Por-

tuguese language groups.

Table 1: Ranking of the most used words in Portuguese-

speaking communities.

Word

Language

Com.

Freq.

Posts

Percent.

anxiety Pt-Br 373 317 23.105

depression Pt-Br 371 297 21.6472

people Pt-Br 328 240 17.4927

life Pt-Br 292 217 15.8163

pain Pt-Br 250 180 13.1195

god

Pt-Br 250 167 12.172

day Pt-Br 228 177 12.9009

time Pt-Br 177 135 9.8397

today Pt-Br 164 138 10.0583

live Pt-Br 162 129 9.4023

Figure 5 shows the heat map of the non-parametric

Spearman measure for the English-speaking commu-

HEALTHINF 2024 - 17th International Conference on Health Informatics

470

Figure 4: Distribution of values for both sets of communities.

Table 2: Ranking of the most used words in English-

speaking communities.

Word

Language

Com.

Freq.

Posts

Percent.

life Eng 111 88 15.5203

anxiety Eng 99 83 14.6384

depression Eng 97 91 16.0494

people Eng 77 61 10.7584

talk Eng 75 64 11.2875

time Eng 64 55 9.7002

pain Eng 58 51 8.9947

day Eng 51 44 7.7601

feeling Eng 49 39 6.8783

hard Eng 48 40 7.0547

nity, displaying relevant scores only between the total

number of interactions and the number of likes. In

this representation, the goal is to discover which at-

tributes are related. With these correlation measures,

we aim to ﬁnd out if there is an association between

the emotional measures obtained by LIWC and how

the post was received by the community it was posted.

6 DISCUSSION

Not all analyzed messages relate to mental health or

directly discuss the disorder. Many texts have a reli-

gious theme, contain motivational messages, or share

personal experiences. This dynamic occurred for

searches with terms in both English and Portuguese.

While applications and translations are available

for the LIWC dictionaries in Portuguese, dated 2015,

the current version of the dictionary, dated 2022, still

needs a Portuguese version. Some recent works, such

as (Carvalho et al., 2019) and (Nascimento et al.,

2020), use the Brazilian Portuguese dictionary. In the

LIWC 22 manual, it is suggested to translate the text

using automated methods to English rather than trans-

lating the system’s base dictionary (Boyd et al., 2022),

i.e., translating the text to be analyzed and not the dic-

tionary itself.

Analytical thinking and authenticity metrics re-

ﬂect how spontaneous and informal a user’s discourse

is (Boyd et al., 2022). A low value for analytical

thinking reﬂects that the person is not using formal

words and is being more intuitive. The highest values

for analytical thinking are found in the Brazilian com-

munities, which include therapy discussion groups.

As for Authenticity, the higher the metric, the greater

the spontaneity. This is reﬂected in the image. How-

ever, messages with low authenticity also have con-

siderable value. The emotional tone metric reﬂects

the type of discussion in the communities. Both for

Brazilian and non-Brazilian communities, most mes-

sages have a negative tone, with values below ﬁfty.

Using correlation measures, strong correlations

between the continuous variables obtained by Crowd-

tangle and LIWC were expected to be found. How-

ever, neither correlation measure obtained relevant

correlation values, whether positive or negative. The

values for correlations between post-reactions and

textual analysis metrics do not show a relevant degree

of correlation. The highest correlation values were

concentrated in attributes from the same tool. For

example, the highest Spearman correlation value was

A Systematic Analysis of Depression-Related Discourse Within Facebook: A Comparison Between Brazilian and American Communities

471

Table 3: Statistical Description of Community Message Reactions.

Language Com. Love Wow Haha Sad Angry Care Likes Comments Shares

Avg. Eng. 2.617 0.041 0.062 8.380 0.013 3.076 8.058683 15.549701 8.427545

Std.Dev. Eng. 4.957 0.276 0.392 12.288 0.133 4.191 7.362661 16.863349 19.510323

min Eng. 0 0 0 0 0 0 0 0 0

25% Eng. 0 0 0 0 0 0.0 4 4 0

50% Eng. 1 0 0 4 0 1 6 12 3

75% Eng. 3 0 0 11 0 4 10 22.5 8

max Eng. 60 6 7 102 2 27 69 194 220

Avg. Pt-Br 0.676 0.097 0.099 3.945 0.035 2.560 10.513 28.335 0.762

Std.Dev. Pt-Br 4.488 0.469 0.461 5.466 0.294 3.638 9.409 20.160 2.181

min Pt-Br 0 0 0 0 0 0 0 0 0

25% Pt-Br 0 0 0 0 0 0 5 16 0

50% Pt-Br 0 0 0 3 0 2 8 25 0

75% Pt-Br 0 0 0 6 0 4 13 36 1

max Pt-Br 164 13 8 111 6 76 131 213 42

Figure 5: Heatmap for attributes referring to communities in English.

between the number of shares and the number of sad

reactions. The second-highest positive value was be-

tween Crowdtangle attributes, total interactions, and

the number of likes. The most relevant negative cor-

relation was between the attributes Authenticity and

Inﬂuence, both obtained in the LIWC tool.

7 CONCLUSION

The present work introduced an approach to social

media analysis focused on communities discussing

the topic of depression. It is worth noting that this

study, through a systematic use of data collection and

analysis tools, aims to provide explanatory insights

into messages discussing depressive disorder. Despite

the results presented and discussed here, there is room

for further investigation of the correlations between

the values obtained from these tools.

Identifying more communities within Facebook

for data extraction and obtaining data from other so-

cial media platforms are areas to explore in the fu-

ture. Creating a dataset with greater variety and diver-

sity would be possible with data from various sources.

This would allow for the development of more robust

and reliable classiﬁcation models, which healthcare

professionals could use in potential analysis tools.

ACKNOWLEDGEMENTS

This work was carried out with the support of CAPES

- Notice nº 09/2020 - Prevention and Combating Out-

breaks, Endemics, Epidemics, and Pandemics. Proc.

#223038.014313/2020-19, and partially supported by

the Oracle for Research program (award number

CPQ-2160239).

REFERENCES

Andalibi, N., Ozturk, P., and Forte, A. (2017). Sensi-

tive Self-disclosures, Responses, and Social Support

on Instagram: The Case of #Depression. In Pro-

ceedings of the 2017 ACM Conference on Computer

Supported Cooperative Work and Social Computing

- CSCW ’17, pages 1485–1500, Portland, Oregon,

USA. ACM Press.

Association., A. P. and Association., A. P. (2013). Diagnos-

HEALTHINF 2024 - 17th International Conference on Health Informatics

472

tic and statistical manual of mental disorders : DSM-

5. American Psychiatric Association Arlington, VA,

American Psychological Association. 750 First Street

NE, Washington, DC 20002-4242, 5th ed. edition.

Boscarioli, C., Araujo, R., and S. Maciel, R. (2017). I

GranDSI-BR Grand Research Challenges in Informa-

tion Systems in Brazil 2016-2026. SBC, Porto Alegre.

Boyd, R. L., Ashokkumar, A., Seraj, S., and Pennebaker,

J. W. (2022). The development and psychometric

properties of liwc-22. Austin, TX: University of Texas

at Austin, pages 1–47.

Brody, D. J., Pratt, L. A., and Hughes, J. P. (2018). Preva-

lence of depression among adults aged 20 and over:

United States, 2013-2016. US Department of Health

and Human Services, Centers for Disease Control

and . . . , 1600 Clifton Rd., Atlanta, GA 30329.

Cafezeiro, I., Viterbo, J., Costa, L., Salgado, L., Rocha,

M., and Monteiro, R. (2017). Strengthening of the

Sociotechnical Approach in Information Systems Re-

search, pages 133–147. SBC-Sociedade Brasileira

de Computac¸

ao, Av. Bento Gonc¸alves, 9500, Bairro

Agronomia, 91501-970.

Carvalho, F., Rodrigues, R., Santos, G., Cruz, P., Ferrari, L.,

and Guedes, G. (2019). Avaliac¸

ao da vers

ao em por-

tugu

es do liwc lexicon 2015 com an

alise de sentimen-

tos em redes sociais. In Anais do VIII Brazilian Work-

shop on Social Network Analysis and Mining, pages

24–34, Porto Alegre, RS, Brasil. SBC.

Chen, J. and Wang, Y. (2021). Social media use for health

purposes: Systematic review. J Med Internet Res,

23(5):e17917.

Creswell, J. and Creswell, J. (2018). Research Design:

Qualitative, Quantitative, and Mixed Methods Ap-

proaches. SAGE Publications, 1 Oliver’s Yard, Lon-

don, EC1Y 1SP, England.

De Castro, J. C. L. (2009). O inconsciente como linguagem:

de Freud a Lacan. CASA: Cadernos de Semi

otica Apli-

cada, 7(1):12–42.

De Choudhury, M., Counts, S., and Horvitz, E. (2013). So-

cial Media As a Measurement Tool of Depression in

Populations. In Proceedings of the 5th Annual ACM

Web Science Conference, WebSci ’13, pages 47–56,

New York, NY, USA. ACM.

Giuntini, F. T., Cazzolato, M. T., de Jesus Dutra dos Reis,

M., Campbell, A. T., Traina, A. J. M., and Ueyama,

J. (2020). A review on recognizing depression in so-

cial networks: challenges and opportunities. Journal

of Ambient Intelligence and Humanized Computing,

11:1–17.

Horvitz, E. and Mulligan, D. (2015). Data, privacy, and the

greater good. Science, 349(6245):253–255.

James, S. L., Abate, D., Abate, K. H., Abay, S. M., Ab-

bafati, C., Abbasi, N., Abbastabar, H., Abd-Allah, F.,

Abdela, J., Abdelalim, A., Abdollahpour, I., Abdulka-

der, R. S., Abebe, Z., Abera, S. F., Abil, O. Z., Abraha,

H. N., Abu-Raddad, L. J., Abu-Rmeileh, N. M. E.,

Accrombessi, M. M. K., and Acharya, D. (2018).

Global, regional, and national incidence, prevalence,

and years lived with disability for 354 diseases and

injuries for 195 countries and territories, 1990–2017:

a systematic analysis for the global burden of disease

study 2017. The Lancet, 392(10159):1789–1858.

Li, G., Li, B., Huang, L., Hou, S., et al. (2020). Automatic

construction of a depression-domain lexicon based on

microblogs: text mining study. JMIR medical infor-

matics, 8(6):e17650.

Li, G., Zhou, X., Lu, T., Yang, J., and Gu, N. (2016). SunFo-

rum: Understanding Depression in a Chinese Online

Community. In Proceedings of the 19th ACM Con-

ference on Computer-Supported Cooperative Work &

Social Computing - CSCW 16, pages 514–525, New

York, New York, USA. ACM Press.

Lima Filho, S. P., Ferreira da Silva, M., Oliveira Sam-

paio, J., and Couto Ruback Rodrigues, L. (2022).

A study about gathering features in depression de-

tection’ problem with health professionals commu-

nity. iSys - Brazilian Journal of Information Systems,

15(1):10:1–10:25.

Nascimento, R., Nascimento, G., Carvalho, F., and Guedes,

G. (2020). Minerac¸

ao de opini

oes com liwc: abor-

dagem pr

atica sobre sistemas judiciais eletr

onicos

brasileiros. In Anais do IX Brazilian Workshop on

Social Network Analysis and Mining, pages 132–141,

Porto Alegre, RS, Brasil. SBC.

Park, S., Kim, I., Lee, S. W., Yoo, J., Jeong, B., and Cha, M.

(2015). Manifestation of Depression and Loneliness

on Social Networks: A Case Study of Young Adults

on Facebook. In Proceedings of the 18th ACM Con-

ference on Computer Supported Cooperative Work &

Social Computing, CSCW ’15, pages 557–570, New

York, NY, USA. ACM.

Pimentel, M., Filippo, D., and Santos Marcondes, T. (2020).

Design Science Research: pesquisa cient

ıﬁca atrelada

ao design de artefatos. RE@D - Revista de Educac¸

a Dist

ancia e eLearning, 3(1):37–61.

Recker, J. (2012). Scientiﬁc Research in Information Sys-

tems: A Beginner’s Guide. Springer Publishing Com-

pany, Incorporated, Springer Publishing Company, In-

corporated.

Skaik, R. and Inkpen, D. (2021). Using Social Media for

Mental Health Surveillance. ACM Computing Sur-

veys, 53(6):1–31.

Tsugawa, S., Kikuchi, Y., Kishino, F., Nakajima, K., Itoh,

Y., and Ohsaki, H. (2015). Recognizing depression

from twitter activity. In Proceedings of the 33rd An-

nual ACM Conference on Human Factors in Comput-

ing Systems, CHI ’15, page 3187–3196, New York,

NY, USA. Association for Computing Machinery.

Vedula, N. and Parthasarathy, S. (2017). Emotional and

linguistic cues of depression from social media. In

Proceedings of the 2017 International Conference on

Digital Health, DH ’17, page 127–136, New York,

NY, USA. Association for Computing Machinery.

Wieringa, R. J. (2014). Design Science Methodology

for Information Systems and Software Engineering.

Springer Berlin Heidelberg, Berlin, Heidelberg.

Wohlin, C., Runeson, P., H

ost, M., Ohlsson, M. C., Reg-

nell, B., and Wessl

en, A. (2012). Experimentation in

software engineering, volume 9783642290. Springer

Berlin Heidelberg, Heidelberger Platz 3, 14197 -

Berlin.

A Systematic Analysis of Depression-Related Discourse Within Facebook: A Comparison Between Brazilian and American Communities

473