An Analysis of Twitter Communities Related to the 2022 War in Ukraine

Karolina Sliwa

1 a

, Ema Ku

sen

2 b

and Mark Strembeck

1,3,4 c

Vienna University of Economics and Business (WU), Vienna, Austria

University of Vienna, Faculty of Informatics, Austria

Secure Business Austria (SBA), Austria

Complexity Science Hub (CSH), Austria

Keywords:

Community Detection, Emotion Analysis, InfoMap, LIWC, Twitter, Ukraine War.

Abstract:

In this paper, we analyze a dataset including more than 189 million tweets related to the ﬁrst month of the

2022 war in Ukraine. Our analysis especially focuses on communities of Twitter users and their collective be-

havior. In particular, we applied the InfoMap community detection algorithm and found on average 44079.63

communities of Twitter users per day. Our behavioral analysis especially focuses on the ﬁve largest daily

communities (i.e. the communities that have been detected for each day during the ﬁrst month of the war).

We found that: 1) hashtags played an essential role in framing conversations, 2) communities often publicly

called on international organizations or ofﬁces such as @potus, @NATO, or @UN to aid in conﬂict resolution,

3) anger was the dominant emotion in all communities and 4) negative tweets spread wider than the positive

ones.

1 INTRODUCTION

People use online social media platforms to dissem-

inate messages throughout various natural and man-

made disasters. Data collected during such events

have the potential to reveal large-scale opinions and

stances towards the crisis (Ku

sen and Strembeck,

2019; Stieglitz et al., 2017).

Social media users often form communities or

groups who interact with each other more frequently

than with the out-group members. As documented in

(Bedi and Sharma, 2016), community members will

often exhibit similar opinions, attitudes, and prefer-

ences. From the sociological point of view, such a

similarity stems from various behavioral, sociodemo-

graphic, and intra-personal characteristics. In gen-

eral, people tend to connect with the others who

are perceived as similar to themselves (“similarity

breads connection”), thereby forming homogeneous

networks (McPherson et al., 2001). A subgraph C is

said to be a community if each node of the subgraph

has more connections within its own community than

with the rest of the nodes in the corresponding net-

https://orcid.org/0009-0006-3271-0199

https://orcid.org/0000-0003-1145-6778

https://orcid.org/0000-0003-1680-9296

work (Flake et al., 2002).

Community detection has revealed hidden net-

work structures (Javed et al., 2018) and brought valu-

able insights into human behavior – from characteriz-

ing follower-followee structures in online social net-

works (Bedi and Sharma, 2016) to detecting social

botnets on Twitter based on the behavioral similarity

(Lingam et al., 2020a).

This paper presents an analysis of the tweets sent

during the ﬁrst month of the 2022 War in Ukraine. In

particular, we aim to detect communities of interact-

ing users and characterize their behavior with respect

to the topics they predominantly talked about, asso-

ciated sentiments, and analyze their inﬂuence on the

Twitter discourse. Our ﬁndings indicate that in the

early weeks of the war, Twitter helped to frame the

conﬂict as an international crisis, negative emotions

were tweeted more frequently, with anger being the

most prominent emotion, and communities frequently

contacted foreign organizations for assistance in con-

ﬂict resolution.

The remainder of this paper is structured as fol-

lows. In Section 2, we give an overview of com-

munity detection methods and emotions during man-

made crises. We then describe the research procedure

in Section 3.2. In Section 4, we report and discuss our

ﬁndings, before we conclude our paper in Section 5.

Sliwa, K., Kušen, E. and Strembeck, M.

An Analysis of Twitter Communities Related to the 2022 War in Ukraine.

DOI: 10.5220/0011980700003485

In Proceedings of the 8th International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2023), pages 27-36

ISBN: 978-989-758-644-6; ISSN: 2184-5034

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

2 RELATED WORK

Community detection in large-scale online social net-

works is a prominent social network research area.

Given the increasing availability of large-scale net-

work data-sets, it has received much scientiﬁc atten-

tion, allowing us to discover latent communities and

their structures (Schaeffer, 2007). A social network’s

communities of users can be identiﬁed using graph

clustering algorithms. Commonly, divisions are made

to reduce connections across clusters while maximiz-

ing the number of connections (edges) within a cluster

(Schaeffer, 2007).

Recently, community detection has shown to

be quite beneﬁcial in understanding the spread of

COVID-19 (Xueting Liao, 2021). Moreover, by mon-

itoring online social networks and virtual communi-

ties, it is possible to better understand and foresee

potential threats from extremist organizations (R

ıos

and Mu

noz, 2012). Another use of community de-

tection might be found in identifying social botnet

communities on Twitter based on behavioral similar-

ity scores associated with the respective user accounts

(Lingam et al., 2020b). In addition, community de-

tection may also be utilized to enhance Bitcoin’s au-

ditability through de-anonymization (Xueshuo et al.,

2021).

Some of the most commonly used community de-

tection algorithms include the Louvain method, In-

fomap, Girvan-Newman algorithm, Label Propaga-

tion Algorithm, and HAC (Hierarchical Agglomera-

tive Clustering). The Louvain method was ﬁrst pro-

posed by (Blondel et al., 2008). For example, it has

been used to identify communities in the political

sphere, such as in (S

anchez et al., 2016) where the

authors applied the algorithm on a sample of tweets

and users to identify similar political preferences. In

(Fani et al., 2016), identiﬁed users with comparable

temporal tendencies in their topics of interest. The

Girvan-Newman algorithm was ﬁrst proposed by Gir-

van and Newman (Girvan and Newman, 2002). For

example, it was applied to identify and analyze com-

munities from a set of users who posted messages

on Twitter during three signiﬁcant crisis events in

2011 (Gupta et al., 2012). The Label Propagation

Algorithm is based on the idea of propagating la-

bels through the network and grouping nodes with the

same label into communities (Raghavan et al., 2007).

HAC (Hierarchical Agglomerative Clustering) is an

unsupervised method that starts with each node as a

separate cluster and then groups them based on simi-

larity. The Infomap algorithm was proposed by (Ros-

vall and Bergstrom, 2008) and has also been applied

to detect Twitter communities, e.g. in a study of opin-

ions about human papillomavirus (HPV) vaccines.

3 RESEARCH APPROACH

The goal of this paper is to explore the communities

that emerged on Twitter during the early stage of the

2022 war in Ukraine. In particular, we analyze the

ﬁrst month of the war (24 February 2022 until 25

March 2022).

3.1 Research Questions

Our analysis is guided by the following research ques-

tions.

RQ 1: Which Communities of Users Emerged in the

First Month of the War?

By applying a community detection algorithm

(see Section 3.2), we identify Twitter communities on

a daily basis for the ﬁrst month of the war.

RQ 2: Which Narratives Are Dominant for Each

Community Over Time?

For each community, we explore the dominant

narratives. To detect narratives, we use the hash-

tags posted in each community. Moreover, we use

the count of original tweets posted by the community

members as a measure of the intensity of the contri-

bution to the Twitter discourse.

RQ 3: Which Emotions Dominate in Each Commu-

nity Over Time?

We explore the dominant emotions (anger, anx-

iety, sadness, as well as positive emotions) in each

community over time.

RQ 4: How Inﬂuential Are the Communities Over

Time?

For the purposes of this paper, we deﬁne inﬂuence

by the number of likes and retweets a tweet receives.

Both likes and retweets have the potential to boost the

visibility (and reach) of a tweet on the Twitter net-

work.

3.2 Procedure

Our research procedure is organized into the follow-

ing phases.

Phase 1: Data Extraction. To extract the data re-

lated to the war, we used the following list of hash-

tags and key terms

that were selected after monitor-

ing the discourse about the war on Twitter. Our data

#sanctionsrussia, #westandwithukraine, #clos-

ethesky, #closetheskyukraine, ”slavaukraini”, #Rus-

siaUkraineConﬂict, #StopWarRussia, #UkraineUn-

COMPLEXIS 2023 - 8th International Conference on Complexity, Future Information Systems and Risk

extraction resulted in 193,948,858 tweets in the En-

glish language.

Phase 2: Data Pre-Processing.

Next, we removed duplicate entries, resulting in a

dataset consisting of 189,854,201 unique tweets. We

then processed all unique tweets using the Linguis-

tic Inquiry and Word Count tool (LIWC) to detect

the presence of three emotions (sadness, anxiety, and

anger), as well as the intensity of positive and nega-

tive tones in each tweet.

Phase 3: Derivation of the Communication Net-

work. To derive the communication network, we

followed the @-mentioning traces in tweets and

recorded the following information: source (screen-

name of a user who authored a tweet that contains

an @mention), target (screenname of a user being

mentioned), dominant emotion, dominant emotional

tone, and time stamp. Our network is directed and

weighted, where the weight represents the number of

messages exchanged between a pair of nodes. In to-

tal, our network for the ﬁrst month of the war consists

of 4,333,571 nodes and 50,544,405 edges.

Phase 4: Community Detection. To detect com-

munities, we used the Python implementation of In-

fomap

. Infomap (Rosvall and Bergstrom, 2008) is a

clustering algorithm that is based on the map equation

(Rosvall et al., 2009).

Infomap was applied to our daily communica-

tion networks. In each iteration of the algorithm, we

recorded a community identiﬁer (a numeric label),

a list of nodes belonging to each community, and a

tweet ID corresponding to the author node assigned

to each community.

4 RESULTS

The daily volume of English-language tweets in the

ﬁrst month of the war is shown in Figure 1, with an av-

erage of about 560K tweets per day. We assume that

users will be less engaged and the number of tweets

will decline the longer the war continues. Our ﬁnd-

ings support this assumption.

derAttack, #UkraineCrisis, #RusyaUkrayna, #Russi-

aUkraine, #ukraine russia, #PrayForUkraine, Ukraine,

putin, @KremlinRussia E, #standwithukraine, @Ze-

lenskyyUa, Ukrainian, #russianinvasion, #StopRus-

sianAggression, #StopRussia, #PrayingForUkraine, Kyiv,

#stopputinnow, #ukrainerussianwar , #putinswar, zelenskiy,

#ukrainerefugees, #ukraineinvasion, #ﬁghtforukraine,

#ukrainewillresist, #supportrussia, #proxywar, #Russian-

Army, #ukrainazi, #istandwithrussia, #NoWarWithUkraine,

#WarinUkraine, #UkraineRussiaWar, #UkraineWar

https://mapequation.github.io/infomap/python/

The ﬁrst peak in the volume of tweets is shown

at the beginning of the war between February 24 and

February 26 where the average number exceeds 1 mil-

lion tweets per day. A second peak around March

1 (again reaching over 1 million tweets) corresponds

to requests to exclude Russia from the UN Security

Council and mobilize military and humanitarian aid

for Ukraine. A Russian attack on the Maternity Hos-

pital in Mariupol on March 10 coincides with the third

peak. Since March 17, the daily average number of

tweets stabilized at 286K (see Figure 1).

Figure 1: Frequency of English-language tweets.

Similar trends are also observable in a volume of

@mentioning tweets. Figure 2 shows the daily num-

ber of nodes (including tweet authors and those being

mentioned in a tweet) as well as edges (number of

messages sent).

28-02 07-03 14-03 21-03

DATE

NODES

EDGES

(in million)

750K

500K

250K

Figure 2: Nodes and Edges in daily networks.

Given that the users are exchanging informa-

tion/opinion and expressing support, a direct engage-

ment is anticipated (see, e.g., the attack on the West-

gate mall in Kenya in 2013 (Simon et al., 2014) or the

2011 Norway terrorist attack(Steensen, 2018)).

An Analysis of Twitter Communities Related to the 2022 War in Ukraine

Figure 3: Hashtag cloud of the most frequently used hash-

tags among all communities based on their occurrence.

4.1 Emergence of Communities Over

the 4-Week Period (RQ1)

In total, we identiﬁed 1,313,143 Twitter communities

over the entire 4-week observation period, with on av-

erage µ=43771.43, sd=21386.3 communities emerg-

ing per day. Figure 4 shows the daily community

count and membership size. On initial inspection, it

is evident that the number of communities decreases

over time, while the average membership size per

community does not follow this trend. For exam-

ple, on the ﬁrst day of the war, there were on aver-

age 8.93 (sd=70.9, median=3) members per commu-

nity and 107,364 unique communities. The member-

ship size increased to 9.60 (sd=68.3, median=3) and

10.6 (sd=89.9, median=3) on the second and third day

of the war, while the community count dropped to

88,408 and 74,910, respectively.

An increase in the membership size (yet, a de-

crease in the number of communities) is observable

for the entire 14-day period since the start of the

war. In the remaining observation period, commu-

nity membership only slightly decreased to an av-

erage of 8.59 (sd=43.21, median=3) on the 25

h of

March 2022.

In addition, we also observed an increase in the

membership size of the largest communities over the

ﬁrst two weeks (e.g., max(members)

day1

= 12,286;

max(members)

day3

= 17,882, max(members)

day9

16,198; max(members)

day14

= 19,260), while the

remaining two weeks showed communities with a

smaller membership size.

The Twitter communities quickly adopted an in-

ternational perspective, framing the issue as one of

global signiﬁcance. Even though the war is between

Ukraine and Russia, the dominant hashtags were writ-

ten in English. The word cloud in Figure 3 was

created using tweets from all communities. Table 1

refers to the hierarchy of the most used hashtags. The

most prominent hashtags include #StopPutin, #Sto-

pRussia, and #Putin. The image also shows the pres-

ence of hashtags oriented towards geographical loca-

tion such as #Ukraine with 8,9% occurrence and a

a a a

Figure 4: Daily community count and membership size.

hashtag showing social support #StandWithUkraine

with 3,11%.

Table 1: The most frequently used hashtags among all com-

munities based on their percentage occurrence.

Hashtags (#) # count %

#Ukraine 4 483 479 8.87

#StopPutin 4 125 594 8.25

#StopRussia 3 182 749 6.36

#Putin 2 348 161 4.69

#StandWithUkraine 1 559 353 3.11

4.2 Dominant Narratives of

Communities (RQ2)

In this section, we focus our analysis on the ﬁve

largest communities (based on their membership size)

on each day during our observation period (5 x 28

days = 140 communities). It is important to note

that community labels do not match across days, e.g.,

community 1 on day 1 is not the same as community

1 on day 2. This is due to the community detection

procedure that was ran separately for each observa-

tion day. We address this issues in Section 4.5.

To gain a better understanding of what each of the

140 communities tweeted about, we compiled a list

of the most frequently used hashtags for each com-

munity. Figure 6 depicts the most prevalent hash-

tag within each group, scaled by the frequency at

which they occur. When we look at the most com-

monly used hashtags in tweets during the ﬁrst four

weeks of the war, we can observe that #Ukraine is

the most popular. Next, we can identify a group

of hashtags related to supporting Ukraine, such as

#IStandWithUkraine, #SolidarityWithUkraine, along

with #HelpUkrainianRefugees at the beginning of the

war, and #YouTubeKyiv, that urged the Ukrainian

branch of YouTube to relocate its headquarters from

COMPLEXIS 2023 - 8th International Conference on Complexity, Future Information Systems and Risk

24-02

Figure 5: Top Target Mentions per Community.

Moscow to another city.

Another set of hashtags with similar narratives

relates to several anti-war and anti-Putin actions:

#StopPutin, #Putin,#StopRussia, #PutinIsaWarCrim-

inal, #StopWar, #EmbargoOnrussianOil.

Another useful piece of information we may ex-

tract from the data-set are the tagged users in tweets

(via @username). The accounts who are mentioned

most frequently by each community are displayed in

Figure 5. The majority of references concern organi-

zations and ofﬁces including the POTUS (President of

the United States), NATO, UN, Ursula von der Leyen

(@vonderleyen), and Boris Johnson (@borisjohn-

son). The most tagged account during the ﬁrst 4

weeks of the war is @potus (see Table 2).

Messages referring to European or US politicians

typically stress the Ukraine’s need for weapons and

humanitarian aid, as in the following tweet: Ukraine

needs weapons and humanitarian assistance to de-

fend against #Putin. Stop innocent civilian deaths.

@POTUS , provide #SafeAirliftUkraine #StopPutin.

Tagging @rtcom (Russia Today, a Russian state-

afﬁliated media outlet) is mainly associated with re-

stricting ”Russia Today” accounts in major European

nations. Accounts mentioning @zelenskyyua fre-

quently express their respect, support, and prayers for

the Ukrainian president. For instance, @ZelenskyyUa

i admire your bravery your honor President Zelenzky

you stand still and never surrender nappy salute to

you or @ZelenskyyUa The world is with you. Russia

is a terrorist state..

Table 2: Top ﬁve most mentioned targets.

Date Target Mentions Hashtag (#) # count

04-03 @potus 319,043 #Putin 659,464

28-02 @nato 174,087 #StopPutin 646,721

03-03 @potus 102,935 #StopPutin 206,164

01-03 @un 91,318 #StopPutin 449,382

06-03 @potus 80,777 #Putin 159,874

4.3 Dominant Emotions in

Communities Over Time (RQ3)

In this section, we investigate the emotional tone of

the tweets in the top ﬁve communities using the Lin-

guistic Inquiry and a Word Count tool (LIWC) (Pen-

nebaker et al., 2007). Figure 7 represents a heatmap of

the average tone of each community. LIWC emotion

scores less than 50 indicate a more negative emotional

tone. For comparison, reference values for each mea-

sure can be accessed on LIWC’s website ((Cohn et al.,

An Analysis of Twitter Communities Related to the 2022 War in Ukraine

Figure 6: Hashtag mentions per community in the course of 4 weeks of the war.

2004), (Pennebaker et al., 2003)). It is evident that

throughout the ﬁrst four weeks of the conﬂict, there

was a noticeable presence of negative emotions. The

average emotional tone over all communities is 34.94.

The most negatively-inclined communities

use hashtags and mentions afﬁliated with Rus-

sia (e.g. #StopRussia), #Putin, and #StopPutin,

e.g.,@UN@POTUS@NATO the more you wait, the

more of our people die. Everyday. SO CLOSE

THE SKY OVER UKRAINE! WE NEED ACTIONS

NOW! #ClosetheSkyoverUkraine #ClosetheSky

#StandWithUkraine #StopRussia #NoFlyZoneUA

#StopRussianAggression #StopBelarusianAggres-

sion. In contrast, there are only a few communities

that showed a slightly positive tone during the ﬁrst

four weeks. An example can be seen on February

27, when the most frequently mentioned target and

hashtag were @potus and #Ukraine, respectively.

Other positive communities are mostly associated

with @potus, e.g., @POTUS @USDoDGov It’s sim-

ple. Get them planes and rockets, and Ukraine will

handle the rest. They have pilots. #PlanesforUkraine

#RocketsForUkraine #StandwithUkraine.

Following the dominant negative tone in the

largest communities, we further examined how nega-

tive emotions such as anxiety, anger, and sadness were

distributed throughout selected communities over the

ﬁrst four weeks of the war. Table 3 displays three neg-

ative emotions as well as their mean value. The mean

value of each emotion was derived by taking into ac-

count all of the expressed sentiment scores for each

emotion. The consolidated mean reveals that anger

was the emotion expressed most often over the entire

observation period. Anger had a consolidated mean

of 1.9945%, which was more than 6 times higher as

compared to any other emotion. This suggests that

during the start of the Ukraine war, communities dis-

played more anger on social media than anxiety or

sadness. This observation is in line with other events

of extreme violence, such as the 9/11 terrorist attack

(Lee et al., 2015) or the Boston Marathon bombing

(Back et al., 2010).

COMPLEXIS 2023 - 8th International Conference on Complexity, Future Information Systems and Risk

17-03 18-03 19-03 20-03 21-03 22-03 23-03

WEEK 1

WEEK 3

WEEK 2

WEEK 4

24-02 25-02 26-02 27-02 28-02 01-03 02-03

08-0303-03 04-03 05-03 06-03 07-03 09-03

10-03 11-03 12-03 13-03 14-03 15-03 16-03

Figure 7: Heatmap of tone per community in the course of 4 weeks of the war.

Table 3: Mean results of selected communities in 4 weeks.

Week Anxiety Anger Sadness

W1 0.2815 1.8178 0.2896

W2 0.2863 2.6099 0.2527

W3 0.3079 1.7724 0.2917

W4 0.2863 1.7782 0.2885

4.4 Inﬂuence of Communities Over

Time (RQ4)

To investigate the impact of the communities, we as-

sessed the total number of likes and retweets to the

number of messages posted by community members.

Figure 8 shows the results of this analysis. In most

communities, the number of likes outnumbers the

number of retweets. Few communities, though, dis-

play a signiﬁcant disproportion in the number of likes

and retweets. The ﬁrst example is from February 28,

when tweets within a community received fewer likes

but more retweets. A similar pattern can be observed

when Russian Today is the target of a tweet. In addi-

tion, the most frequently retweeted content was as-

sociated with a negative tone. Table 4 shows the

emotional tone and number of retweets for the most

retweeted messages.

For example, more than 8 million people have

retweeted a message asking European countries to

block Russia Today. On February 26, the community

that received the most likes was community number 1.

The most inﬂuential member within this community

was @zelenskyyua, who had a 6,390,407 followers

and received around 250K likes for a message which

read: Had a phone conversation with @BorisJohn-

son. Grateful to the British Prime Minister for his

position, new decisions to enhance the combat ca-

pabilities of the Ukrainian army. Agreed on further

joint steps to counter the aggressor.. This message

expresses gratitude towards the British Prime Min-

ister for his support of Ukraine and mentions new

decisions to boost the military capabilities of the

Ukrainian army.

Table 4: Most retweeted messages.

Date Target Tone Retweets Likes

23-03 @rt com 25,80 8,654,223 41

09-03 @rt com 25,87 6,455,412 183

16-03 @rt com 25,81 5,485,776 65

10-03 @rt com 25,85 5,478,833 123

The link between the emotional tone of a tweet’s

content and its chance of getting retweeted on Twitter

An Analysis of Twitter Communities Related to the 2022 War in Ukraine

WEEK 1

Total number of likes

Total number of retweets

Date Date

WEEK 3

Total number of likes

Total number of retweets

Date

Total number of likes

Total number of retweets

Date

Total number of likes

Total number of retweets

WEEK 2

WEEK 4

Figure 8: Likes and retweet count per community.

has been the subject of several studies. For example,

a recent study by (Sch

one et al., 2021) explored the

spread of emotions on Twitter in the aftermath of pos-

itive and negative political situations and discovered

that negative emotions were more likely to be shared

and disseminated among users, supporting our ﬁnd-

ings.

Table 5 shows that the sum of retweets for nega-

tive messages is substantially higher than for positive.

Table 5: Number of positive and negative messages (excl.

neutral messages) and the sum of retweets for both.

Week Positive Retweets

W1 5,753,639 6,038,874

W2 4,026,927 3,401,901

W3 2,069,306 1,882,124

W4 1,869,300 1,733,706

Week Negative Retweets

W1 6,125,831 23,530,321

W2 6,064,396 18,372,479

W3 2,632,016 9,041,539

W4 2,455,817 9,467,937

4.5 Limitations

Although community detection in Twitter data is an

important approach for analyzing the structure and

behavior in online social networks, it does have cer-

tain limitations. One of our study’s limitations is that

the data is restricted to English-language tweets. This

makes detecting underrepresented communities (ev-

eryone who does not send English language tweets)

in the data challenging to impossible. Furthermore,

hashtag sampling might reveal insightful perspectives

into certain cultural and sociopolitical discussions.

However, it introduces its own set of biases (Tufekci,

2014) and is a key constraint of our study.

5 CONCLUSION

In this paper, we analyzed communities that emerged

during the ﬁrst month of the 2022 war in Ukraine. The

ﬁndings underscore the importance of Twitter as a

medium for communication among communities dur-

ing the early stages of the war in Ukraine. The use

of hashtags was pivotal in structuring conversations,

with generic hashtags such as #StopPutin and #IS-

tandwithUkraine effectively portraying the conﬂict as

COMPLEXIS 2023 - 8th International Conference on Complexity, Future Information Systems and Risk

a global crisis. We found that communities using

hashtags in relation to anti-war and anti-Putin senti-

ments tend to exhibit a more negative tone than those

communities associated with expressions of support

for Ukraine. Additionally, the study revealed that

there was a concentration of communities around spe-

ciﬁc targets. International organizations and ofﬁces

such as @potus, @NATO, and @UN were frequently

mentioned by users and were typically addressed as

potential facilitators of a potential conﬂict resolution.

Moreover, our ﬁndings indicate that the reac-

tions within the top 5 communities were predomi-

nantly characterized by negative emotions, particu-

larly anger, and tend to spread more quickly and more

widely on Twitter than positive emotions. In our fu-

ture work, we plan to extend this study by applying

a temporal community detection algorithm to identify

the dynamics evolution of network communities and

also provide a more ﬁne-grained analysis of related

user behavior.

REFERENCES

Back, M. D., K

ufner, A. C., and Egloff, B. (2010). The emo-

tional timeline of september 11, 2001. Psychological

Science, 21(10):1417–1419.

Bedi, P. and Sharma, C. (2016). Community detection in

social networks. WIREs Data Mining and Knowledge

Discovery, 6(3):115–135.

Blondel, V. D., Guillaume, J.-L., Lambiotte, R., and Lefeb-

vre, E. (2008). Fast unfolding of communities in large

networks. Journal of statistical mechanics: theory

and experiment, 2008(10):P10008.

Cohn, M. A., Mehl, M. R., and Pennebaker, J. W. (2004).

Linguistic markers of psychological change surround-

ing september 11, 2001. Psychological science,

15(10):687–693.

Fani, H., Zarrinkalam, F., Bagheri, E., and Du, W. (2016).

Time-sensitive topic-based communities on twitter. In

Canadian conference on artiﬁcial intelligence, pages

192–204. Springer.

Flake, G. W., Lawrence, S., Giles, C. L., and Coetzee, F. M.

(2002). Self-organization and identiﬁcation of web

communities. Computer, 35(3):66–70.

Girvan, M. and Newman, M. E. (2002). Community struc-

ture in social and biological networks. Proceedings of

the national academy of sciences, 99(12):7821–7826.

Gupta, A., Joshi, A., and Kumaraguru, P. (2012). Identi-

fying and characterizing user communities on twitter

during crisis events. In Proceedings of the 2012 work-

shop on Data-driven user behavioral modelling and

mining from social media, pages 23–26.

Javed, M. A., Younis, M. S., Latif, S., Qadir, J., and Baig,

A. (2018). Community detection in networks: A mul-

tidisciplinary review. Journal of Network and Com-

puter Applications, 108:87–111.

sen, E. and Strembeck, M. (2019). An analysis of

emotion-exchange motifs in multiplex networks dur-

ing emergency events. Applied Network Science, 4.

Lee, J., Rehman, B. A., Agrawal, M., and Rao, H. R. (2015).

Sentiment analysis of twitter users over time: the case

of the boston bombing tragedy. In Workshop on E-

Business, pages 1–14. Springer.

Lingam, G., Rout, R. R., Somayajulu, D., and Das, S. K.

(2020a). Social botnet community detection: A novel

approach based on behavioral similarity in twitter net-

work using deep learning. In Proceedings of the 15th

ACM Asia Conference on Computer and Communica-

tions Security, ASIA CCS ’20, page 708–718, New

York, NY, USA. Association for Computing Machin-

ery.

Lingam, G., Rout, R. R., Somayajulu, D., and Das, S. K.

(2020b). Social botnet community detection: A novel

approach based on behavioral similarity in twitter net-

work using deep learning. In Proceedings of the 15th

ACM Asia Conference on Computer and Communica-

tions Security, ASIA CCS ’20, page 708–718, New

York, NY, USA. Association for Computing Machin-

ery.

McPherson, M., Smith-Lovin, L., and Cook, J. M. (2001).

Birds of a feather: Homophily in social networks. An-

nual review of sociology, pages 415–444.

Pennebaker, J. W., Booth, R. J., and Francis, M. E. (2007).

Linguistic inquiry and word count: Liwc [computer

software]. Austin, TX: liwc. net, 135.

Pennebaker, J. W., Mehl, M. R., and Niederhoffer, K. G.

(2003). Psychological aspects of natural language use:

Our words, our selves. Annual review of psychology,

54(1):547–577.

Raghavan, U. N., Albert, R., and Kumara, S. (2007).

Near linear time algorithm to detect community struc-

tures in large-scale networks. Physical review E,

76(3):036106.

ıos, S. A. and Mu

noz, R. (2012). Dark web portal overlap-

ping community detection based on topic models. In

Proceedings of the ACM SIGKDD Workshop on Intel-

ligence and Security Informatics, ISI-KDD ’12, New

York, NY, USA. Association for Computing Machin-

ery.

Rosvall, M., Axelsson, D., and Bergstrom, C. (2009). The

map equation. Eur. Phys. J. Spec. Top., 178:13–23.

Rosvall, M. and Bergstrom, C. T. (2008). Maps of random

walks on complex networks reveal community struc-

ture. Proceedings of the national academy of sciences,

105(4):1118–1123.

anchez, D. L., Revuelta, J., Prieta, F. D. l., Gil-Gonz

alez,

A. B., and Dang, C. (2016). Twitter user clustering

based on their preferences and the louvain algorithm.

In International Conference on Practical Applications

of Agents and Multi-Agent Systems, pages 349–356.

Springer.

Schaeffer, S. E. (2007). Graph clustering. Computer Sci-

ence Review, 1(1):27–64.

Sch

one, J. P., Parkinson, B., and Goldenberg, A. (2021).

Negativity spreads more than positivity on twitter af-

An Analysis of Twitter Communities Related to the 2022 War in Ukraine

ter both positive and negative political situations. Af-

fective Science, 2(4):379–390.

Simon, T., Goldberg, A., Aharonson-Daniel, L., Leykin, D.,

and Adini, B. (2014). Twitter in the cross ﬁre—the

use of social media in the westgate mall terror attack

in kenya. PloS one, 9(8):e104136.

Steensen, S. (2018). Tweeting terror: An analysis of the

norwegian twitter-sphere during and in the aftermath

of the 22 july 2011 terrorist attack. In Social media use

in crisis and risk communication. Emerald Publishing

Limited.

Stieglitz, S., Bunker, D., Mirbabaie, M., and Ehnis, C.

(2017). Sense-making in social media during extreme

events. Journal of Contingencies and Crisis Manage-

ment, 26.

Tufekci, Z. (2014). Big questions for social media big data:

Representativeness, validity and other methodological

pitfalls. In Eighth international AAAI conference on

weblogs and social media.

Xueshuo, X., Jiming, W., Junyi, Y., Yaozheng, F., Ye, L.,

Tao, L., and Guiling, W. (2021). Awap: Adaptive

weighted attribute propagation enhanced community

detection model for bitcoin de-anonymization. Ap-

plied Soft Computing, 109:107507.

Xueting Liao, Danyang Zheng, X. C. (2021). Coronavirus

pandemic analysis through tripartite graph clustering

in online social networks. Big Data Mining and Ana-

lytics, 4(4):10.

COMPLEXIS 2023 - 8th International Conference on Complexity, Future Information Systems and Risk