Aspect-Level Sentiment Analysis of Filipino Tweets

During the COVID-19 Pandemic

John Paul S. Guzman

1 a

, Charibeth K. Cheng

2 b

, Jan Michael Alexandre C. Bernadas

3 c

and Angelyn R. Lao

1 d

Department of Mathematics and Statistics, De La Salle University, Manila, Philippines

Department of Software Technology, De La Salle University, Manila, Philippines

Department of Communication, De La Salle University, Manila, Philippines

guzmanjps@gmail.com, {charibeth.cheng, jan.bernadas, angelyn.lao}@dlsu.edu.ph

Keywords:

Aspect-Based Sentiment Analysis, Twitter, COVID-19.

Abstract:

During the COVID-19 pandemic, X (formerly known as Twitter) was teeming with rich discussions as people

shared their experiences and concerns. Understanding the sentiments in these tweets could aid in gauging

public reactions and enhancing public health communication. While some studies analyze public health sen-

timents, few speciﬁcally focus on aspect-level sentiments in the Global South. In this study, we examine

tweets published in the Philippines during the COVID-19 pandemic and aspects relevant to the pandemic. The

sentiment polarities of tweet-aspect pairs are annotated. We analyze these pairs to understand the sentiments

expressed during this period. These insights can improve health communication in the Philippines by assessing

public receptiveness to policies, monitoring events that inﬂuence sentiment, and identifying communication

gaps. Notably, we observed disproportionately high amounts of negative sentiment toward the Sinopharm and

Sinovac vaccines. This sentiment indicates distrust and racial bias against Chinese brands. Moreover, the

consistent negative sentiment toward face shields over an extended period highlights shortcomings in health

communication about their effectiveness.

1 INTRODUCTION

The Internet has become an essential tool for day-to-

day communication. Social media platforms like X

facilitate the widespread sharing of people’s thoughts,

opinions, and knowledge. During the COVID-19 pan-

demic, X was teeming with rich discussions as people

shared their experiences and concerns (Pastor, 2020;

Wang and Chen, 2022). The vast amounts of data gen-

erated through online discourse can be processed to

yield valuable insights (Ghani et al., 2019).

A common method for extracting information

from text data is sentiment analysis. Sentiment analy-

sis, or opinion mining, refers to the process of extract-

ing opinions or sentiments from bodies of texts (Pang

et al., 2008). It has a wide variety of applications,

such as monitoring social media, processing customer

reviews, and deciding political campaign strategies

https://orcid.org/0009-0007-3050-1825

https://orcid.org/0000-0002-2993-6211

https://orcid.org/0000-0002-3828-0314

https://orcid.org/0000-0003-3028-9610

(DeNardis and Hackl, 2015; Ramteke et al., 2016;

Salinca, 2015).

One particularly promising application of senti-

ment analysis lies in health communication. Health

communication is the study of communication strate-

gies designed to inﬂuence individuals to adopt behav-

iors beneﬁcial to their health (Schiavo, 2013). Sen-

timent analysis can be used to assess the public’s

risk perception and attitudes toward infection control

strategies (Alhajji et al., 2020). In addition, it can

measure the public’s concern about diseases (Cabling

et al., 2018; Himelboim et al., 2020; Ji et al., 2015).

However, much of the existing research is situated in

the Global North (Wang and Chen, 2022), which can

have vastly different experiences compared to coun-

tries in the Global South (Balfour et al., 2022). This

study uses sentiment analysis to examine public reac-

tions and identify communication gaps in the Philip-

pines during the COVID-19 pandemic.

Sentiment analysis is mainly done at three lev-

els of granularity (Liu, 2012). The ﬁrst level is

document-level, which extracts a sentiment from an

entire document (Cabling et al., 2018). This is use-

Guzman, J. P. S., Cheng, C. K., Bernadas, J. M. A. C. and Lao, A. R.

Aspect-Level Sentiment Analysis of Filipino Tweets During the COVID-19 Pandemic.

DOI: 10.5220/0013053100003911

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

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 343-350

ISBN: 978-989-758-731-3; ISSN: 2184-4305

343

ful in cases where the documents were selected to ex-

press opinions on a ﬁxed entity, but it is not applica-

ble when documents have multiple entities. The sec-

ond level is sentence-level, which aims to assign sen-

timents to each sentence in a document (Himelboim

et al., 2020; Wang and Chen, 2022). This assumes

that each sentence expresses an opinion on one entity

but will fail when a sentence contains multiple enti-

ties. The third level is aspect-level or aspect-based

sentiment analysis (ABSA), which extracts a senti-

ment towards a given aspect or entity in a sentence.

Knowledge about the speciﬁc target of the opinion al-

lows for a more ﬁne-grained analysis of sentiments.

For instance, smartphone reviews may contain sen-

timents about various aspects of the phone, such as

battery life, camera, and screen quality. Smartphone

manufacturers may use the insights from aspect-level

sentiment analysis to make targeted improvements.

Jang et al. (2021) conducted an ABSA study an-

alyzing Twitter data during the COVID-19 pandemic

in North America. Their research uncovered various

negative topics linked to the pandemic, including anti-

Asian racism and the spread of misinformation. They

employed topic modeling to identify topics and as-

pects. They then used a combination of English lexi-

cons, part-of-speech tagging, and dependency parsing

for sentiment polarity classiﬁcation.

A signiﬁcant challenge in sentiment analysis for

the Global South is the lack of tools for processing

multilingual text. One approach, employed by Math-

ayomchan et al. (2023), involves ﬁltering out non-

English tweets and then applying existing tools to the

remaining English tweets. Unfortunately, this method

excludes tweets written in local languages or those

code-mixed with English.

In this study, we examine tweets published in the

Philippines during the COVID-19 pandemic and as-

pects relevant to the pandemic. This is done to gain a

deeper understanding of public sentiment and the fac-

tors that inﬂuence it during this critical period. We

manually identiﬁed the topics and aspects and then

classiﬁed sentiment polarity. We then study the senti-

ments by aggregating aspect-level sentiments in vari-

ous ways. These insights can improve health commu-

nication in the Philippines by assessing public recep-

tiveness to policies, monitoring events that inﬂuence

sentiment, and identifying communication gaps.

2 METHODOLOGY

The data used in this study originate from the tweets

collected by Chan et al. (2022). These tweets

were posted between December 4, 2020, and June 4,

2021. The tweets were ﬁltered to ensure they orig-

inate from the Philippines based on geolocation and

contain at least one of the speciﬁed health-related key-

words: “covid-19,” “covid,” “coronavirus,” “corona,”

“tb,” “tuberculosis,” “WorldTBDay,” or “TBFreePh”

(Chan et al., 2022). Aspect terms were selected

based on topics frequently discussed during the pan-

demic, including but not limited to different vaccine

brands, travel bans, community quarantine, the use

of masks/face shields, hospital capacity, government

agencies, and isolation facilities. Aspects related to

tuberculosis, such as “cough,” “ubo,” and “fatigue,”

were excluded to avoid conﬂating tuberculosis senti-

ments with those of COVID-19. In total, 590 aspects

were identiﬁed (Guzman, 2024).

Two human annotators were hired to label the

sentiment polarity towards an aspect term in a given

tweet. These annotators are native Filipino speakers

ﬂuent in English and have backgrounds in jobs requir-

ing communication skills in both languages, such as

teaching. Before starting the annotation process, the

annotators underwent a brief training session. They

were instructed to classify each tweet-aspect pair into

one of four categories: positive when the sentiment

towards the aspect is positive, negative when the sen-

timent towards the aspect is negative, neutral when

the sentiment towards the aspect is neither positive

nor negative, and no relation/conﬂicting when there

is no sentiment expressed towards the aspect or when

the sentiment is conﬂicting. The annotators were also

provided with examples to illustrate each category.

To reduce the noise in the labeling, tweet-aspect

pairs labeled as no relation/conﬂicting were dis-

carded. The pairs where the annotations do not match

were also removed, such as when one annotator labels

a pair as positive while another labels it as negative.

A total of 6,600 pairs remained after the removals.

Among these, 2,840 are labeled as negative, 2,705 as

neutral, and 1,055 as positive (Guzman, 2024).

We evaluated the inter-annotator agreement be-

tween the two hired annotators using Cohen’s Kappa

statistic κ before discarding annotations (Cohen,

1960). The calculated κ value was 0.5116, suggest-

ing a degree of agreement but at a relatively low level

(McHugh, 2012).

3 RESULTS

Aspects are categorized into different topics, which

can be further divided into subtopics. This approach

enables us to aggregate sentiments at various lev-

els of generality. The aspects are categorized into

ﬁve main topics: Facilities, Government, Responses,

HEALTHINF 2025 - 18th International Conference on Health Informatics

344

Figure 1: Categorization of aspects into selected topics and subtopics. Aspects related to facilities, government organizations

or ofﬁcials, pandemic responses or policies, testing, and vaccines are assigned to speciﬁc topics under COVID-19. Within

the Vaccines topic, particular vaccine brands are classiﬁed into subtopics. Similarly, frequently discussed pandemic responses

within the Responses topic are organized into subtopics. Finally, important pandemic facilities within the Facilities topic are

further categorized into subtopics.

Testing, and Vaccines. Aspects related to facilities,

government organizations or ofﬁcials, pandemic re-

sponses or policies, testing, and vaccines are assigned

to these topics. Within the Vaccines topic, speciﬁc

vaccine brands are classiﬁed into the subtopics: As-

traZeneca, Moderna, Pﬁzer, Sinopharm, and Sinovac.

Frequently discussed pandemic responses within the

Responses topic are categorized into the subtopics:

Assistance, Border control, Contact tracing, Face

shield, Isolation, Lockdown, Mask, Social distanc-

ing, and Vaccine deployment. Finally, important pan-

demic facilities within the Facilities topic are classi-

ﬁed into the subtopics: Hospitals, Isolation facilities,

Testing facilities, and Vaccination facilities. Any as-

pect within the ﬁve main topics that do not ﬁt the

aforementioned subtopics is not further categorized

(Guzman, 2024). A diagram of the selected topics

and subtopics is shown in Figure 1.

Figure 2 shows the distribution of sentiments

across various topics. Each topic contains a signif-

icant portion of neutral sentiments. Notably, a con-

siderable amount of negative sentiment is expressed

towards Government, Responses, and Facilities. In

contrast, Vaccines elicits more positive sentiments.

Figure 3 displays the sentiment distribution of

subtopics of Vaccines. Each subtopic shows a sig-

niﬁcant proportion of neutral sentiments. Notably,

AstraZeneca and Pﬁzer exhibit a moderate amount

of positive sentiments. Meanwhile, Sinopharm and

Sinovac have larger amounts of negative sentiments.

Figure 4 depicts the distribution of sentiments

within the Responses subtopics. Most of them show

low amounts of positive sentiments. The exception is

Mask, which is predominantly positive. Conversely,

Face shield and Lockdown mainly exhibit negative

sentiments. Similarly, Figure 5 shows that nearly all

Facilities subtopics have a low proportion of positive

sentiments. The only exception is Vaccination facili-

ties, which predominantly shows positive sentiments.

Figure 2: Distribution of sentiments across various topics.

Each topic exhibits a substantial proportion of neutral senti-

ments. There are signiﬁcant amounts of negative sentiment

toward Government, Responses, and Facilities. Meanwhile,

Vaccines elicits more positive sentiments.

Figure 3: Distribution of sentiments among subtopics of

Vaccines. Each subtopic has a large portion of neutral sen-

timents. AstraZeneca and Pﬁzer exhibit moderate amounts

of positive sentiments. On the other hand, Sinopharm and

Sinovac have large proportions of negative sentiments.

Aspect-Level Sentiment Analysis of Filipino Tweets During the COVID-19 Pandemic

345

Figure 4: Distribution of sentiments among subtopics of Re-

sponses. Most subtopics show low amounts of positive sen-

timents, except for Mask, which is primarily positive. In

contrast, Face shield and Lockdown predominantly exhibit

negative sentiments.

Figure 5: Distribution of sentiments among subtopics of Fa-

cilities. Almost all subtopics show low levels of positive

sentiment. The exception is Vaccination facilities, which is

predominantly positive.

Next, we visualize trends in selected topics and

subtopics by separating tweets by month. In Figure

6, we see that the sentiment for Sinovac is consis-

tently negative and neutral. However, there has been

an increasing trend in positive sentiment since March.

A similar upward trend in positive sentiments is ob-

served for AstraZeneca.

In Figure 7, we observe an increase in positive

sentiment regarding Vaccine deployment in March

and April. However, there is a spike in negative

sentiment in May. Discussions about Assistance

rose in April, though it is rarely discussed in the

other months. Furthermore, Face shield consistently

evokes negative sentiment. Notably, Lockdown re-

ceived a substantial increase in negative sentiment in

March. During March and April, we also observed

a surge in negative sentiments directed towards the

Government and Testing, as illustrated in Figure 8.

4 DISCUSSION

The hierarchical nature of this approach facilitates a

better understanding of sentiments. We notice pre-

dominantly positive and neutral sentiments when ex-

amining the distribution of sentiments related to Vac-

cines, as shown in Figure 2. However, delving into

the subtopics in Figure 3 reveals a disproportion-

ately high amount of negative sentiment targeting

Sinopharm and Sinovac. This discrepancy suggests a

lack of trust in the Sinopharm and Sinovac vaccines.

The overall sentiment towards the Responses

and Facilities topics is mostly negative and neu-

tral. Nonetheless, the Mask and Vaccination facili-

ties subtopics stand out with a large amount of posi-

tive sentiment, as depicted in Figures 4 and 5. This

reﬂects public support for wearing masks during the

pandemic and satisfaction with vaccination sites.

Beyond analyzing sentiment distribution, we

delve into the contents of the tweets to identify com-

mon reasons for negative sentiments. Our goal is to

understand these causes in order to ﬁnd potential so-

lutions. Firstly, we consider some tweets from the

Facilities topic expressing negative sentiments:

• “there are no more hospital beds in ncr. the sooner

the government acknowledges this, the quicker we

can get any intervention.”

• “only 27k people tested for two days in a row. al-

most the same number two weeks ago, before the

start of ecq kala ko ba increase testing capacity?”

A common concern is the potential of hospital or ICU

beds reaching full capacity. This worry is highlighted

in a news article by Tomacruz (2021a), which re-

ports an alarming occupancy of ICU beds in Metro

Manila. Another common concern observed in the

tweets is insufﬁcient testing capacity, which was also

acknowledged by government ofﬁcials (Luna, 2020).

The emphasis on facilities highlights the efforts of Fil-

ipinos on social media to frame the COVID-19 prob-

lem as structural and societal rather than solely indi-

vidual. This emphasis exposes the existing ﬂaws in

the Philippine healthcare system and problematic le-

gal approaches to a public health crisis (Bernadas and

Ilagan, 2020; Navera and Bernadas, 2022).

We turn our attention to the tweets from the Face

shield and Lockdown subtopics under Responses, as

they predominantly exhibit negative sentiments:

• “2020 mecq....gcq...ecq 2021 granular lockdown,

hard/soft lockdown.... tapos heightened gcq... ano

pa naiisip nyong forms ng quarantine?”

• “i still don’t get the rationale for the use of face

shields when in fact, it did not even lower the

HEALTHINF 2025 - 18th International Conference on Health Informatics

346

Figure 6: Trends in sentiments on Vaccines subtopics. The trendlines were calculated using 2-month moving averages. The

sentiment for Sinovac is consistently negative and neutral. Nonetheless, there has been a notable increase in positive sentiment

since March. A similar rise in positive sentiments is observed for AstraZeneca.

Figure 7: Trends in sentiments on Responses subtopics. The trendlines were calculated using 2-month moving averages. We

observe an increase in positive sentiment regarding Vaccine deployment in March and April. However, there was a notable rise

in negative sentiment in May. Furthermore, Face shield consistently evokes negative sentiment. Discussions about Assistance

rose in April, though it is rarely discussed in the other months. Finally, there was a substantial rise in negative sentiment

concerning Lockdown in March.

Figure 8: Trends in sentiments on the Government and Testing topics. The trendlines were calculated using 2-month moving

averages. The positive sentiments are consistently low towards Government and Testing. In March and April, there was a

surge in negative sentiments directed at these topics.

Aspect-Level Sentiment Analysis of Filipino Tweets During the COVID-19 Pandemic

347

covid cases? it just adds plastic trash. so, can

someone shed some light? fuck rules!”

• “after a year of covid-19 pandemic in the country,

we have the following: world’s longest lockdown

last to legally vaccinate in southeast asia best”

The negative sentiments toward face shields stem

from skepticism about their effectiveness, leading to

online debates and controversy (Pamintuan, 2021).

Health agencies can combat this by designing mes-

sages focused on the positive outcomes of preven-

tive measures, known as gain-framing. Meanwhile,

negative sentiment towards lockdowns stems from the

prolonged duration, notably being one of the world’s

longest lockdowns (Patag, 2021; See, 2021). Ad-

ditionally, people expressed dissatisfaction with the

complicated and seemingly erratic types of commu-

nity quarantine (Gotinga, 2020).

We examine the tweets related to the Sinovac

and Sinopharm subtopics under Vaccines, which had

a disproportionately high amount of negative senti-

ment. During this analysis, we identiﬁed some inter-

esting words frequently brought up in these tweets.

Figure 9 shows the most frequent words. Certain

words, like vaccine names, are expected to have

high frequencies. However, some unexpected terms,

namely “chinese” and “efﬁcacy,” surfaced with no-

table frequency. This discovery prompts a more fo-

cused examination of these speciﬁc terms.

Figure 9: Frequently occurring words from negative tweets

in the Sinovac and Sinopharm subtopics. Certain words,

like vaccine names, are expected to have high frequencies.

However, some unexpected terms, namely “chinese” and

“efﬁcacy,” surfaced with notable frequency.

We investigate how the words “chinese” and “efﬁ-

cacy” relate to negative sentiments by ﬁltering tweets

that contain either of these keywords:

• “may history kasi ang sinovac ng bribery sa mga

chinese ofﬁcial dati.”

• “sinovac ang vaccine dito sa qc uhuhuhuhuh i

dont trust chinese product fuuuuuuk”

• “in chinatown/chinese warehouses/chinese owned

condos located from batanes to jolo millions upon

millions of sinovac/sinopharm vaccines are ”la-

bel switching” into pﬁzer/biontech, astrazeneca,

moderna, j&j, covavax!!! very, very alarming!!!”

• “people are hesitant not on the vaccines in gen-

eral but on the efﬁcacy and safety proﬁle of your

favorite chinese brand. people want pﬁzer, az, and

moderna. throw sinovac in the trash.”

Negative sentiment may be linked to Sinovac

Biotech’s past involvement in bribery cases with Chi-

nese drug regulators (Bergonia, 2020). Additionally,

rumors about fake vaccines contribute to the negative

perception of these vaccines (The Food and Drug Ad-

ministration of the Philippines, 2021; The Philippine

Star, 2021). A signiﬁcant point of contention about

the Sinovac vaccine is its low efﬁcacy. Critics, includ-

ing lawmakers and former advisers, question the gov-

ernment’s decision to procure the Sinovac vaccine.

They argue that other vaccines have higher efﬁcacy

rates at lower costs and express concerns about the

lack of transparency in Sinovac’s clinical trials (Bon-

doc, 2021; Jalea, 2020; Romero, 2020). Moreover,

the frequent occurrence of the term “chinese” in a

negative context reinforces the racialization of health

issues and reﬂects the precarity of Sino-Philippines

relations (Caba

nes and Santiago, 2023).

Lastly, we analyze the trends in certain topics and

subtopics. The rise of positive sentiment towards

Sinovac, depicted in Figure 6, coincides with the start

of the COVID-19 vaccine rollout in the Philippines

on March 1, 2021. This event likely contributed to

the public’s favorable perception of the vaccine brand.

Similarly, the vaccine rollout likely inﬂuenced the in-

crease in positive sentiment toward Vaccine deploy-

ment, as illustrated in Figure 7.

It was announced near the end of March that low-

income Filipinos were set to receive supplemental aid

(Rey, 2021). This development reﬂects a rise in dis-

cussions about Assistance in April, as seen in Figure

7. Furthermore, we observed that Face shield consis-

tently evoked negative sentiment. This persistent neg-

ativity suggests that health communication about their

beneﬁts has been inadequate for an extended period.

Public health campaigns should promote the beneﬁts

of using face shields more assertively to bridge this

communication gap. Lastly, the substantial increase

in negative sentiment toward Lockdown in March

2021 coincides with the one-year anniversary of the

initial lockdown implemented in March 2020 (Medi-

aldea, 2020). By this time, many expressed dissatis-

faction with the lockdown, which had become one of

the world’s longest (Patag, 2021; See, 2021).

In March and April, there was a surge in nega-

tive sentiments towards the Government and Testing,

as illustrated in Figure 8. Much of this negativity

HEALTHINF 2025 - 18th International Conference on Health Informatics

348

stemmed from the inadequacy of the pandemic re-

sponse despite the prolonged lockdown. Criticisms

directed at the government included a lack of ac-

knowledgment or downplaying of problems and mis-

trust in reported numbers. As for testing, the main

issue was the absence of mass testing (Cabico, 2021).

Despite the insights gained from this study, we

have encountered several limitations. The reliance

on manually annotated data posed several challenges.

First, the manual labeling process is labor-intensive

and difﬁcult to scale. Second, sentiment analysis has

inherent subjectivity, as individual biases can inﬂu-

ence annotations. Consequently, the resulting dataset

was relatively small, consisting of only 6,600 tweet-

aspect pairs. This limited sample may not fully repre-

sent the broader population’s sentiments.

5 CONCLUSIONS

In this paper, we studied the tweets collected by Chan

et al. (2022) and the tweet-aspect sentiment polarity

annotations of Guzman (2024). We analyzed this data

using three approaches.

First, we examined the distribution of sentiments

at different levels of granularity. We have found a

disproportionately high amount of negative sentiment

speciﬁcally targeting Sinopharm and Sinovac. This

sentiment suggests a lack of trust in the Sinopharm

and Sinovac vaccines. Equally important, this study

reinforces health as racialized and reﬂects the com-

plexity and precarity of the Sino-Philippines relation-

ship (Caba

nes and Santiago, 2023), even before the

pandemic. On the other hand, the sentiment towards

the Mask and Vaccination facilities subtopics stand

out with predominantly positive sentiments. This re-

ﬂects public support for wearing masks during the

pandemic and satisfaction with vaccination sites.

Second, we explored the content of tweets to pin-

point common reasons for negative sentiments. For

the Facilities topic, we have found that a common

theme is the possibility of hospitals reaching full ca-

pacity and insufﬁcient testing capacity. In the Face

shield and Lockdown subtopics, negative sentiments

revolve around skepticism regarding the effectiveness

of face shields, prolonged lockdown duration, and

complicated types of community quarantine. Con-

cerning the Sinovac and Sinopharm subtopics, com-

mon issues include Sinovac Biotech’s past involve-

ment in bribery cases with Chinese drug regulators,

rumors about fake vaccines, the low efﬁcacy of Sino-

vac with its relatively high price, and lack of trans-

parency in clinical trials.

Finally, we examined trends in the sentiments to

identify events that might have inﬂuenced changes

in sentiment. In March, people showed more posi-

tive sentiment toward Sinovac and AstraZeneca likely

the rollout of COVID-19 vaccines in the Philippines.

During the same period, negative sentiment towards

Lockdown increased as it marks the one-year anniver-

sary of the initial lockdown. This anniversary likely

contributed to the rise in negative sentiment toward

Government and Testing. The negative sentiment was

driven by criticisms of the inadequacy of the pan-

demic response and the lack of mass testing despite

the prolonged lockdown. Furthermore, sentiment to-

wards Face shield remained consistently negative,

indicating poor public health communication about

their effectiveness over an extended period. To ad-

dress skepticism and misinformation regarding face

shields, health ofﬁcials should adopt a more proactive

and assertive approach to emphasizing their beneﬁts.

Future studies should explore ways to overcome

these limitations. For manual annotations, we rec-

ommend providing more detailed guidelines and ex-

amples to increase inter-annotator agreement. Addi-

tionally, future research could investigate integrating

machine learning techniques to complement manual

annotation, thereby facilitating data collection on a

larger scale. Moreover, collecting data from different

periods and regions could yield further insights.

We recommend exploring alternative topics re-

lated to the pandemic. Our categorization of topics

was selected based on what we deemed relevant dur-

ing the pandemic. However, other applications may

yield topics and hierarchies different from those in

Figure 1. For instance, one may want to track top-

ics and subtopics related to the spread of misinforma-

tion or mental health. A different categorization could

help gain insights into new topics and subtopics.

REFERENCES

Alhajji, M., Al Khalifah, A., Aljubran, M., and Alkhali-

fah, M. (2020). Sentiment analysis of tweets in Saudi

Arabia regarding governmental preventive measures

to contain COVID-19.

Balfour, R., Bomassi, L., and Martinelli, M. (2022). Coro-

navirus and the Widening Global North-South Gap.

Bergonia, T. (2020). Bribery past hounds Chinese ﬁrm

eyeing PH market for coronavirus vaccines. IN-

QUIRER.net. Accessed: January 6, 2025.

Bernadas, J. M. A. C. and Ilagan, K. (2020). Journalism,

public health, and COVID-19: some preliminary in-

sights from the Philippines. Media International Aus-

tralia, 177(1):132–138.

Bondoc, J. (2021). ‘No pickiness,’ just accept Sinovac with

50% efﬁcacy. The Philippine Star.

Aspect-Level Sentiment Analysis of Filipino Tweets During the COVID-19 Pandemic

349

Caba

nes, J. V. and Santiago, F. (2023). Consuming Digi-

tal Disinformation: How Filipinos Engage with Racist

and Historically Distorted Online Political Content.

ISEAS - Yusof Ishak Institute.

Cabico, G. K. (2021). DOH: No ’mass testing’, just ’risk-

based testing’ to get more people tested. Rappler. Ac-

cessed: January 6, 2025.

Cabling, M., Turner, J., Hurtado-de Mendoza, A., Zhang,

Y., Jiang, X., Drago, F., and Sheppard, V. (2018).

Sentiment Analysis of an Online Breast Cancer Sup-

port Group: Communicating about Tamoxifen. Health

communication, 33(9):1158–1165.

Chan, E. Y., Chan, M. R. C., Ching, S. J. S., Sie, S. L.,

Lao, A. R., Bernadas, J. M. A. C., and Cheng, C. K.

(2022). How Health Information Spreads in Twit-

ter: The Whos and Whats of Philippine TB-data. In

HEALTHINF, pages 421–429.

Cohen, J. (1960). A coefﬁcient of agreement for nominal

scales. Educational and psychological measurement,

20(1):37–46.

DeNardis, L. and Hackl, A. M. (2015). Internet governance

by social media platforms. Telecommunications Pol-

icy, 39(9):761–770.

Ghani, N. A., Hamid, S., Hashem, I. A. T., and Ahmed, E.

(2019). Social media big data analytics: A survey.

Computers in Human Behavior, 101:417–428.

Gotinga, J. (2020). ECQ, MECQ, GCQ, MGCQ: Who can

go where? Rappler. Accessed: January 6, 2025.

Guzman, J. P. (2024). Aspect-based sentiment

analysis of Filipino COVID-19 tweets with

memory networks. Master’s thesis, De La

Salle University Manila. Retrieved from

https://animorepository.dlsu.edu.ph/etdm math/10.

Himelboim, I., Xiao, X., Lee, D. K. L., Wang, M. Y., and

Borah, P. (2020). A Social Networks Approach to Un-

derstanding Vaccine Conversations on Twitter: Net-

work Clusters, Sentiment, and Certainty in HPV So-

cial Networks. Health Communication, 35(5):607–

615. PMID: 31199698.

Jalea, G. (2020). DOST: Despite 50% effectivity, PH still

can’t use Sinovac sans full data. CNN Philippines.

Accessed: January 6, 2025.

Jang, H., Rempel, E., Roth, D., Carenini, G., and Jan-

jua, N. Z. (2021). Tracking COVID-19 discourse on

twitter in North America: Infodemiology study using

topic modeling and aspect-based sentiment analysis.

Journal of medical Internet research, 23(2):e25431.

Ji, X., Chun, S. A., Wei, Z., and Geller, J. (2015). Twitter

sentiment classiﬁcation for measuring public health

concerns. Social Network Analysis & Mining, 5:1–25.

Liu, B. (2012). Sentiment analysis and opinion mining.

Synthesis lectures on human language technologies,

5(1):1–167.

Luna, F. (2020). Testing capacity for COVID-19 still ’not

enough’ — Palace. The Philippine Star.

Mathayomchan, B., Taecharungroj, V., and Wat-

tanacharoensil, W. (2023). Evolution of COVID-19

tweets about Southeast Asian Countries: topic mod-

elling and sentiment analyses. Place Branding and

Public Diplomacy, 19:317–334.

McHugh, M. L. (2012). Interrater reliability: the kappa

statistic. Biochemia medica, 22(3):276–282.

Medialdea, S. (2020). Memorandum from the Executive

Secretary On Community Quarantine Over the Entire

Luzon and Further Guidelines for the Management of

the Coronavirus Disease 2019 (COVID-19) Situation.

Ofﬁcial Gazette. Accessed: January 6, 2025.

Navera, G. S. and Bernadas, J. M. A. C. (2022). ‘Shoot them

dead’: rhetorical constructions of the COVID-19 pan-

demic in Philippine presidential addresses. Journal of

Communication in Healthcare, 16(1):93–102.

Pamintuan, A. M. (2021). Face shield debate. The Philip-

pine Star. Accessed: January 6, 2025.

Pang, B., Lee, L., et al. (2008). Opinion Mining and Senti-

ment Analysis. Foundations and Trends® in informa-

tion retrieval, 2(1–2):1–135.

Pastor, C. K. (2020). Sentiment Analysis of Filipinos and

Effects of Extreme Community Quarantine Due to

Coronavirus (COVID-19) Pandemic. SSRN Electronic

Journal.

Patag, K. J. (2021). In world’s longest lockdown, process-

ing grief is a lonely and lingering struggle. The Philip-

pine Star. Accessed: January 6, 2025.

Ramteke, J., Shah, S., Godhia, D., and Shaikh, A. (2016).

Election result prediction using Twitter sentiment

analysis. In 2016 international conference on in-

ventive computation technologies (ICICT), volume 1,

pages 1–5. IEEE.

Rey, A. (2021). 2021 ECQ ‘ayuda’: What we know so far.

Rappler. Accessed: January 6, 2025.

Romero, P. (2020). Senators: 50% Sinovac efﬁcacy unac-

ceptable. The Philippine Star.

Salinca, A. (2015). Business reviews classiﬁcation using

sentiment analysis. In 2015 17th International Sym-

posium on Symbolic and Numeric Algorithms for Sci-

entiﬁc Computing (SYNASC), pages 247–250. IEEE.

Schiavo, R. (2013). Health communication: From theory to

practice, volume 217. John Wiley & Sons.

See, A. B. (2021). Rodrigo Duterte Is Using One

of the World’s Longest COVID-19 Lockdowns to

Strengthen His Grip on the Philippines. Time Mag-

azine. Accessed: January 6, 2025.

The Food and Drug Administration of the Philippines

(2021). Public Health Warning on Fake COVID-19

Vaccines. Accessed: January 6, 2025.

The Philippine Star (2021). EDITORIAL - Fake vaccines.

The Philippine Star. Accessed: January 6, 2025.

Tomacruz, S. (2021). Metro Manila’s ICU bed capacity at

‘alarming’ 64.5% as COVID-19 cases soar. Rappler.

Wang, Y. and Chen, Y. (2022). Characterizing discourses

about COVID-19 vaccines on Twitter: a topic model-

ing and sentiment analysis approach. Journal of Com-

munication in Healthcare, 16(1):103–112.

HEALTHINF 2025 - 18th International Conference on Health Informatics

350