The COVID-19 pandemic has been characterized by many controversies regarding the illness itself as well as the vaccination process. Social media platforms play a major role in spreading both valuable and scarce information. This paper aims to conduct a social network analysis of Twitter posts mentioning one of the three major vaccine producers. Twitter was chosen because of its user base, open API access and the vast amount of information spread. Data were collected daily from Twitter API 1.1 over a period of nine months from November 2021 to July 2022. Graph metrics, groups and node metrics were calculated using SNAP. The analysis is focused on the most important nodes in the network ranked by betweenness centrality. For the highest ranked user, the content of his posts and the amount of engagements was analyzed. The results show that the highest ranked users are usually (not always) non-professionals who mainly post misinformation, fake-news or only negative true information about vaccines. Another interesting fact revealed is that some users are present in diferent datasets and their posts get engagements from users not speaking the same language. A more thorough analysis of this data using other techniques such as calculating the path of the information flow may reveal further valuable information.
The internet era we live in has massively contributed
sult is the attraction of users to fake news. Fake news is
defined to be “fabricated information that mimics news
media content in form but not in organizational process
or intent” [
COVID-19 pandemic and is defined by WHO as false information related to a disease [6]. After a period of nonpharmaceutical interventions such as lockdowns, restrictions of movement and social gatherings, face masking, physical distancing [7] the development and widespread of diferent kinds of COVID-19 vaccines have brought great hope for the world. Very soon anti-vaccination nEvelop-O LGOBE personel/msc-arber-ceni (A. Ceni); https://fshn.edu.al/departments/ departamenti-i-informatikes/personel/dr-denada-collaku (D. Çollaku)
0000-0003-1373-1438 (A. Ceni); 0000-0002-3598-7285 (A. Kika); 0000-0002-6398-5156 (D. Çollaku) mation [9].
Social media platforms have a major role in the spread of (mis/dis) information, fake news or anti-vaccine movements that spread rumors about the alleged dangers of vaccines [10]. In this paper, Twitter tweets associated with COVID-19 vaccines from a period from November 2021 until July 2022 are analyzed using graphs for each month and type of vaccine. For each graph that is created betweenness centrality is used to determine the top 10 users. Betweenness centrality is the number of shortest paths passing through a node, showing its influence over the flow of information in the network [ 11, 12] and [13]. Betweenness centrality is very important in the analysis of social networks. It can be used to measure the influence of the vertex or the person in a social network.
The paper consists of four further sections: Section 2 reviews recent studies on social network analysis related to vaccination for COVID-19. The data collection, strucmethodology that is used to create graphs, is presented in Section 3. In Section 4, the network analysis metrics and results are discussed. The conclusion and a concise summary conclude this paper.
through social media networks. Social network analysis is used to understand the structure and dynamics of these movements and vaccination hesitancy arose among sev- ture and storage that is used in this study, as well as the
Term Number of posts in order to identify all posts (tweets) that contain the moderna 3009645 term. The Twitter API 1.1 was used to search Twitter pfizer 3311945 and download the data [17]. The API had diferent access AstraZeneca 1224705 levels ranging from free to Premium 1. The free access vaccine 3138370 level allowed to search for posts created in the last 7 coronavirus 3138830 days up to a maximum of around 18000 posts per search. COVID-19 2903138 To gather as much data as possible without being rate covid 3567153 limited, we decided to perform data collection for each vaksina 4170 of the terms once per day every day. For this purpose, TOTAL 20 297 956 we built a console application which would authenticate with Twitter API using OAuth 1.0a [18], search for tweets containing the specified search term, get information for networks and how misinformation spreads within them. the users involved in the conversation and store every
Carvalho et.al [14], applied topic discovery to tweets thing into a database. We setup a task in Task Scheduler with geo information extracted from the COVID-19 vacci- to run this application for each of the terms, so in total nation theme. They used the Latent Dirichlet Allocation we configured eight tasks equally distanced through the for topic modeling, and clustering for visualization, with 24 hour period. the t-SNE, enabling a more detailed view of the distribution of topics and polarities, according to the number 3.1.2. Data structure of tweets, in time and Brazilian geographic space. The analytical process provided a framework containing a set Twitter returns the queried data using the JSON format. of tools to deliver information that can help authorities For each of the posts that contain the search term, the to understand the evolution of public opinion on vaccina- API returns a JSON object, whose fields were used to tion and identify cities with significant numbers of posts build the graph. Below we show some of the fields used. according to the extracted topics [14]. For a complete list of fields and their detailed description,
A study by Olszowski et al [15], conducted a so- please refer to [19]. cial network analysis on Twitter discussions regarding mandatory COVID-19 vaccination in the polish speaking community by using NodeXL to generate betweenness centrality and network clusters, and Clauset–Newman–Moore algorithm to identify two important groups of users. The results of this study revealed a substantial degree of polarization, a high intensity of the discussion, and a high degree of involvement of Twitter users [15].
3.1. Data Motivated by the large amount of social media discussion that the COVID-19 pandemic and vaccines have produced [15, 16], we focused our research on terms related to these topics. We identified eight terms to be of interest (Table 1). • created_at - Date/time when the post was created. • entities - Hashtags, users and URLs contained in the post. • id – Unique identifier of the post. • in_reply_to_screen_name - The user which this posts replies to. • in_reply_to_status_id - The post ID which this post replies to. • is_quote_status - This post is a quote of another post. • lang - Language of the post. • retweeted_status – The original post that this post is a retweet of. • text - The post content. • user - A nested JSON object containing information about the author of the post.
For the user, the main fields used during graph creation were:
• id – Unique identifier of the user.
access level. At the time this paper was written, Twitter introduced the new access level system and revoked the free access level.
3.2.1. Relationships By analyzing the JSON object of a post, we identified five diferent kinds of interactions or relationships between users.
1. Mention – A “mention” relationship will exist between User A and User B if User A has mentioned User B’s username in the post that he has authored. Mentioning in Twitter is done by writing another user’s username (i.e. @userb) in the posts’ text. 2. ReplyTo – A “replyto” relationship will exist between User A and User B if User A has replied to a post authored by User B. A user can reply to another user by clicking the “Reply” button in a post or by preceding his post with a mention (i.e @userB...). 3. Retweet – A “retweet” relationship will exist between User A and User B if User A has retweeted User B’s post by using the Retweet option in Twitter. 4. MentionInRetweet – A “mentioninretweet” relationship is a special case of the “retweet” relationship. Using the current classification technique, in the case where User A has retweeted User B and the post contains a mention of User C, then a “mention” relationship will be created between User A and User C. This is not entirely correct as User A has not directly mentioned User C, although there is a mention. To fix this type of problems, we introduced the “mentioninretweet” relationship which will have as source User A and as destination User C. 5. Tweet – A “tweet” relationship will exist between User A and himself if User A has authored a post which doesn’t contain a “mention”, is not a “replyto”, “retweet” or “mentioninretweet”. This relationship is considered as self-loop, where the source and the destination is the same. 3.1.3. Data storage
better-built queries. The second solution is more flexible regarding changes done by the API in the JSON structure.
Given the fact that the database would be used mainly as a storing mechanism, most of the queries would use either the date or the search term, and the introduction of Twitter API 2.0 which in fact changed the JSON structure, we chose to apply solution number 2. MySQL was chosen as a database management system for its flexibility, price and the ability to partition tables. The structure of the database tables is shown in Figure 1.
The amount of collected data and the fact that the whole JSON object is stored in a single text column, has a negative efect on the performance of the queries. To compensate for this performance loss, we partitioned An important thing to notice here is that a single post the table using HASH partitioning [21] on the “Hashed- may generate multiple relationships. CollectionSearchTerm” column since most of the search queries would be performed on this column. A subparti- 3.2.2. The Graph tioning by date would further benefit the performance since “DateUtc” is the second most used column in the Each of the relationships described above constitutes an search queries, but HASH partitions cannot be subparti- edge with equal weight in our graph. The nodes of our tioned in MySQL [22]. graph are the users. Beside the required information to build a graph (node for nodes and source and destination Table 2 for edges), other important information is included as Graphs size for ”Pfizer” well.
The data were collected from November 2021 until July Month Nodes Edges 20222 and the number of posts collected for each of the November 2021 367294 731875 terms are reported in Table 1. Given the vast amount of December 2021 278266 543490 data collected, we decided to split them in a “by term” and January 2022 198258 451691 “by month” basis. Building a huge graph consisting of February 2022 229197 673992 all the collected posts would not allow us to run metrics MAaprrcilh22002222 224043119638 761078161447 and analyze the graph. The graph generation was done May 2022 210783 626985 on an Intel Xeon Silver 4110 with 16GB of RAM. June 2022 172249 526720
A number of possible graph file formats were consid- July 2022 40060 75511
ered to be used for storing the generated graph. In
particular we investigated GEXF [23], GDF [
The general idea behind the graph generation algo- matter of hours compared to never completing. rithm is to get from the database all the records matching the search term provided by the user. From those records extract the authors of the posts and store them in a dictio- 3.2.3. Results nary in memory. This is done to prevent duplicate users To further study the user polarization and (mis)inforbecause there cannot be duplicate nodes in a graph. For mation spread, we selected three terms: Pfizer , Modeach of the records, extract users that have been men- erna and AstraZeneca. For these terms and for the nine tioned or replied to and add them to the dictionary as well months period under consideration (November 2021 – if they are not present. Last but not least, for each record July 2022), we constructed 27 graphs in total. Table 2, 3 determine if the there is a “ReplyTo”, “Retweet”, “Men- and 4 show the size of the constructed graphs. tion”, “Tweet” (self-loop) or it is a “MentionInRetweet” and create the appropriate edges.
Some challenges were encountered while executing 4. Analysis
this algorithm. A straightforward select query from the
database resulted in a huge response time because of the In this section and the following subsections we present
RawStatusJson field which is set to be of type TEXT. This diferent metrics regarding the generated graphs. Each
also posed a second challenge: it required more RAM subsection contains an analysis of the graphs for the
memory than the machine had available just to load the terms Pfizer, Moderna and AstraZeneca respectively. All
data. Working with the data to create the graph would calculations were done using SNAP [
the data analyzed.
Table 3 of 1106 engagements. As we can see the engagements Graphs size for ”Moderna” in this case are more evenly spread in regards to the diferent types of engagement than in other datasets.
Month Nodes Edges This might be because the user (reuters) is a news agency
November 2021 358701 602031 and is considered trustworthy. It would be interesting to
December 2021 267581 430523 perform a deeper analysis for paths [
March 2022 93629 176314 In the February 2022 dataset the user with the second April 2022 66175 201440 highest betweenness centrality is pokrath. He is a doctor May 2022 79209 222855 and has a Twitter verified account. He has tweeted 9 June 2022 56726 130264 times and the content of those tweets is about vaccine July 2022 23721 39747 eficacy for the Omicron variant. 156 other users have retweeted these tweets and 1 has replied to. Also in this case we can notice that a small amount of engagement 4.1. Pfizer has produced a high betweenness. In March 2022 user lakovosjustice is ranked second in In Table 5 some overall graph metrics are presented. In the list. The account is now suspended, but at the time of Table 6, 7 and 8 the top 10 users (nodes) ranked by be- the import it had 28166 followers, was created in October tweenness centrality are shown. A node with high be- 2021 and had authored more than 3000 tweets in such tweenness centrality means that it is central to the flow of a short time. In the dataset, this user authored 9 tweets information in this network. The oficial pfizer account and all of them contain anti-vaccine claims with most not is expected to be in this list as everyone is talking about having any source of information. These tweets generthem so it gets mentioned or replied to very often. It is ated 4792 engagements, with 4747 retweets, 27 reply-to, interesting to analyze what makes the other nodes so 16 mentions and 3 mention-in-retweet. important. In April 2022 we notice user jakeshieldsajj being sec
In the November 2021 dataset user processic has the ond in the list. He is a form MMA/UFC world champion second highest betweenness centrality. As we will see in and he has a verified account with more than 300k folfuture sections, this user is part of other datasets as well. lowers. During this period he has authored 3 tweets comDuring this period the user has created 7 posts where plaining about Twitter suspending user accounts that 1 is a tweet, 5 reply-to and 1 retweet. The user posts created negative posts about the Pfizer vaccine. These useful information about vaccination centers and their tweets got retweeted 4928 times, mentioned 2 times and location in his country. The user is from Thailand, has replied to 16 times. joined Twitter in 2008 and has 44000 followers. His posts In May 2022 user kwagular has the second highest becreated 19047 interactions, where 19041 were retweets, 1 tweenness centrality. The user has made 2 posts stating mention and 5 reply-to. This means that nearly half of that a new document has emerged from Pfizer that sugthe user’s followers engage with him. gests to not breastfeed after vaccination and baby formula
In December 2022 user reuters is second in the list. is running out. A simple fact check in fact checking sites3 The user posted 52 times during this period and their reveals that these claims are not true. However, these content were vaccine related news. These posts were retweeted 792 times, replied-to 60 times, mentioned 79 times and mentioned-in-retweet 170 times for a total
box.google.com/factcheck/explorer/ second dose of the Pfizer vaccine. These posts got 5755 engagements, of which 5590 were retweets, 73 reply-to, Feb 2022 March 2022 April 2022 81 mentions and 11 mention-in-retweet. Given the low pfizer pfizer pfizer number of followers for this user, it is interesting to know pokrath lakovosjustice jakeshieldsajj how the other users engaged with this tweet. Although suddhi2 techarp f_philippot there is no way to determine if the user’s claim is true bokuwa_kumaa manopsi techarp or false, the fact that the account does not exist anymore f_philippot somorangi_e blemontd make these claims suspicious. drsimonegold f_philippot drpacomoreno1 In July 2022 user tuckercarlsson has the second highest lakovosjustice afshineemrani merissahansen17 betweenness centrality. The user is a well-known jourlereveildatlas follforfight manopsi nalist. He has authored a single post where he mentions kdoiastchloegseretvat vbeuribtyb_lefsraanii_ce mcaoztargsamno_ft Pfizer, but is not related to the vaccine but to medications in general. This tweet has produced 739 engagements with 686 being retweets, 27 mentions, 11 reply-to and 15 mention-in-retweet. posts got 5792 engagements, of which 5698 are retweets, As we can see from this analysis, betweenness cen43 reply-to, 49 mentions and 4 mention-in-retweet. The trality is a good indicator of importance in a network. user has less than 1500 followers and has joined Twitter Users with high betweenness centrality mean that other since 2013, so it is interesting to know who the users that users are engaging with them many times. We can notice produced so many engagements are? that among these users are professionals who responsibly
In the June 2022 dataset, user bluewoodhomes is sec- share true data, but there are also other users who like to ond in the list. It is interesting to notice that the account share and amplify fake-news or disinformation. Another is not active anymore, but at the time of import the user interesting fact is the presence of user drjohnb2 in the had 996 followers and joined Twitter in 2011. The user top 10 users list for January 2022. We will see this user has authored 3 tweets in the dataset and is claiming that being present in other datasets as well. his son had serious heart-related adverse events after the mRNA vaccines. These tweets have produced 22196 reactions, of which 22144 are retweets, 12 reply-to, 3 mention,
Feb 2022 March 2022 April 2022 and 5 mention-in-retweet. As we can see once again the moderna_tx moderna_tx moderna_tx main type of engagement in this case is retweet. jordanschachtel louietraub merissahansen17 In December 2021 another user from Thailand, plobjai, orwells_ghost_ follforfight yuzawn is second in the list of highest betweennes centrality. This craig_a_spencer sensanders elonmusk user has authored 6 tweets mainly giving information p_mcculloughmd manopsi ryan_wigand about how the vaccination process works in Thailand and perpetualmaniac reuters tomtsec notifying his followers that he is going to get vaccinated.
sbancel faesq3639 manopsi These tweets got 16655 reactions of which, 16422 are donaldjtrumpjr theirberge disclosetv retweets, 226 reply-to and 1 mention. In this case the mpafnizoeprsi npyftiizmeers zimermpfaiznerricardo reply-to number is higher than for user processic in the previous dataset.
In January 2022 user nomoretimecafe has the second highest betweenness centrality. The interesting fact in
this case is that this user has only 374 followers and has Some overall metrics for the constructed graphs for Mod- authored 3 posts. In one of his tweets the user asks for erna are shown in Table 9. We will now consider the people who have extra Moderna vaccines to donate them. top 10 users ranked by betweenness centrality for the This tweet got 11027 engagements, all of them retweets. generated graphs for the “Moderna” term displayed in For the February 2022 dataset we will consider the Table 10, 11 and 12. second user in the list given the fact that the first user in
In the November 2021 dataset, we once again have user the list is moderna_tx and it is expected to be so since processic ranked second, same as in the Pfizer dataset the conversation is around them and many users engage for the same period. During this period this user has with them. User jordanschachtel has authored 21 posts authored 39 posts mainly sharing information regarding during this period. He has a verified account with 255000 followers and mainly posts against big pharma companies Moderna vaccine that claims to provide better and longerand vaccines. This user claims that the approved vaccines lasting immunity response. This single tweet has proare not the same as the emergency approved ones and duced 9145 engagements, double the number of us_fda, no citizen has access to the approved vaccines. His posts but still in the list user unrulycat2511 is positioned below produced 9303 engagements of which 9105 retweets, 107 us_fda. This is because betweenness centrality is not a reply-to, 31 mentions and 61 mention-in-retweet. mere representation of the amount of connections in a
In March 2022 user louietraub has the second high- network, but rather a representation of a node’s imporest betweenness centrality. He describes himself as an tance in the whole graph structure. advocate for vaccines injury and during this period he User ratchakorn is ranked second in the top users list authored 22 posts talking about his own injuries after for July 2022. He has authored 2 posts where he shows the second dose of the Moderna vaccine. His Twitter concern about some expired Moderna vaccines being account is shadow-banned (not all his posts are visible) used. On his second tweet he shares a document which and his Facebook account is restricted. He has a total states that the expiration date of the vaccines has been of 16000 followers and 8338 people engaged with his 22 extended by 2 months. These tweets have produced 11825 posts. 8196 were retweets, 123 reply-to, 18 mentions and engagements where 11823 are retweets and 2 are reply-to, 1 mention-in-retweet. despite the fact that the user has only 1300 followers.
In the April 2022 dataset user merissahansen17 is second in the list of users with highest betwenness centrality. 4.3. AstraZeneca During this period the user has authored 2 tweets. The ifrst tweet is news about the CFOs of Pfizer and Mod- Table 13 shows overall metrics about each constructed erna both resigning within 72 hours over vaccine safety. graph. In Table 14, 15 and 16 are shown the top 10 users Performing a fact check on this claim we can notice that (nodes) ranked by betweenness centrality. A node with this news is false4. However, these tweets got retweeted high betweenness centrality means that it is central to 6461 times, replied-to 77 times, mentioned 15 times and the flow of information in this network. The oficial asmentioned in a retweet 16 times. trazeneca account is expected to be in this list as everyone
In May 2022 there are two users with betweenness is talking about them so it gets mentioned or replied to
centrality higher than moderna_tx. The first user is in- very often. It is interesting to analyze what makes the
conforme75. He authored 2 posts where the second one other nodes so important.
is a retweet of his first tweet. This post reports an accu- User processic has authored 38 posts in the November
sation of Russia towards high USA oficials, Pfizer and 2021 dataset, 37 of which are simple posts and one is a
Moderna regarding bioweapons in Ukraine. A simple reply to another user. These posts have produced 26522
search on Google about this topic yields results from interactions where 26520 are retweets and only 2 are
trustworthy media that this is fake-news5. This post reply-to. As we can see the interaction is a form of a
got 4474 engagements, of which 4397 are retweets, 55 broadcast [
The June 2022 dataset captures a nice feature of the to the newest (at that time) COVID variant, Omicron. betweenness centrality. In this dataset we can notice user Users interacted with these posts 6302 times, 6167 of us_fda being second in the list and user unrulycat2511 which were retweets. 64 times users replied to, 40 times being in third place. Us_fda is the oficial FDA Twit- mentioned the user and 16 times retweeted his tweets ter account and has tweeted about a committee meeting while mentioning other users. regarding emergency authorization for the Moderna vac- In January 2021 the first user in the list is drjohnb2. cine. These tweets have produced 4430 engagements. He is part of the top 10 users list in other datasets as well. User unrulycat2511 has tweeted about the new improved For this dataset it is interesting the fact that this user has a higher betweenness centrality than astrazeneca. He authored 25 posts, of which 22 were tweets and 3 were 4https://www.newsweek.com/fact-check-pfizer-moderna-cfosquit-within-72-hours-over-vaccine-safety-1701912
5https://www.npr.org/2022/03/25/1087910880/biologicalweapons-far-right-russia-ukraine
translate his posts. the list. During this time 34 posts were authored where 28 were simple tweets and 6 were retweets (shares) of
Feb 2022 March 2022 April 2022 his own earlier tweets. All of the tweets contain adverse astrazeneca astrazeneca astrazeneca events following vaccination and a reference to a pubdrjohnb2 mateo85966574 marianoalbert2 lished paper explaining the case. A total of 25 papers tsererak gaditanasinmor1 funesta were cited, 12 of them contained reports of single cases. rhodofansie m_ebrard santacarmelac Twelve papers were referring to thrombotic events and k4ats arwen0506 vaxreports1 one was not peer-reviewed. These posts produced 4341 jefyindy robertkennedyjr ex_infirmier interactions. 4287 of them are retweets. 2 are reply-to nathrnetn tonyhinton2016 ake2306 (one being in French), 14 mentions (one in Spanish and drpaulofaria22 drjohnb2 julesserkin 12 from the same user) and 8 mention-in-retweet. spfoonllgfeobrfoigbhcattz drhonehnsd1e0r0kkamp stoerzireasloefviannjutry In March 2022 user mateo85966574 has the second highest betweennes centrality. During this period he has authored a single tweet in which he claims that the doctors have diagnosed a tumor in his body right after retweets of his previous tweets. All of the tweets contain vaccination and he has retweeted a tweet from an account adverse events following vaccination and a reference to a which is suspended. A total of 2239 users engaged with published paper explaining the case. A total of 22 papers these 2 posts, of which 2209 were retweets, 29 replywere cited, 14 of them were single report cases, 7 referred to and 1 mention. Another interesting fact is that this to thrombotic events7. There were 3106 engagements user account was created on December 2021 and the user with these tweets and all of them were retweets. The describes himself as anti-COVID vaccination. creation date of his account is November 2021. In the April 2022 dataset user marianoalbert2 is second In February 2022 user drjohnb2 is again at the top of in the list. He has authored a single tweet commenting his COVID-19 symptoms after three doses of the As7By that time a number of countries had stopped the use of traZeneca vaccine. A total of 1261 users reacted to this AstraZeneca COVID-19 vaccines. tweet, 1245 of which retweeted it and 15 replied to. The posts related to the three main vaccine producers. From account creation date is November 2018 and the user has this collection of data we built 27 graphs by analyzing posted tweets related to other subjects as well. user engagements such as retweet, reply-to, mention and
In May 2022 user adversereports can be spotted be- mention-in-retweet. For each of the constructed graphs ing second in the list. This user account was created in we calculated overall metrics as well as betweenness April 2022 and mainly reports about adverse events with centrality for each of the nodes in a graph. Betweenness the vaccines. During this time the user has authored centrality was used to rank users from the most to the 200 posts, usually within 5 min from each-other. These least important in regards to his position in the network. tweets produced 2719 engagements, 2486 of which are An analysis of the shared content by the top ranked user retweets, 132 reply-to, 11 mentions and 90 mention-in- has been conducted. retweet. From this analysis we conclude that betweenness cen
In the June 2022 dataset user buckyouhorses is second trality is a good metric to distinguish the most central in the list of highest ranking users. This user has pro- users in a network as it does not rely on the amount of duced 10 posts with 86 interactions. The user account engagements alone, but captures the network structure is now suspended, but the account creation date was as well. The main form of engagement in Twitter was 2009. In the description the user states that is a widow found to be retweet. Among the highest ranked users we because her partner deceased after an AstraZeneca vac- found professionals who responsibly share trustworthy cination. These posts created 4134 engagements, 4017 information, common individuals who try to help others were mention-in-retweet (because the original tweet con- as well as suspicious accounts who share fake-news, mistained mentions), 11 were reply-to and 102 were men- information or real news noticing only the negative efect tions. Unfortunately it is impossible to know the reason of the vaccines. From the later group of users we found of an account suspension. that they are present in diferent datasets (i.e. drjohnb2).
In July 2022 we notice user bfntvv has the second While most of the accounts that share fake-news are now highest betweenness centrality. The user has tweeted one suspended, the accounts that do share real but negative time stating that a BBC radio presenter has been deceased news are still active. Given the fact that they share real due to vaccination and a retweet of this tweet. This tweet news they should not be suspended, but the information has produced 1074 user engagements, of which 1065 are they share is not all the information there is. There is retweets, 5 reply-to and 4 mentions. A curious fact about also a positive side of the story. Can deliberately sharing this user account is that it is suspended, but the reported negative only information around a topic be classified as creation date in the dataset is April 2022. an information disorder? Should there be another term
As we can see most of these accounts had a recent for this behavior? Should there be eforts to stop this creation date which makes one think that they were de- behavior? liberately created to spread information for a certain Suspending an account is a reactive response, meaning purpose. Some of the accounts are already suspended that those accounts got a lot of engagement at first and and some are not. We noticed that the accounts not sus- then got suspended. A proactive response would be more pended are indeed sharing true information with proper desirable. Being able to predict if a post is going to attract referencing, but those cases are not the whole picture a lot of engagement and checking if that information is and while it is not dis/mis-information, it seems like the true or false so we can stop it before it spreads is a powpurpose of sharing only that information is to stop peo- erful feature to have. We aim to study the possibility of ple from getting the vaccine. In the analyzed datasets achieving this by using graph neural networks in future there is a noticeable diference between the information work. shared by verified users (i.e. drericding) and other users (i.e. drjohnb2, adversereports, mateo85966574 etc.) even though both information can be regarded as “negative” Acknowledgments with regard to vaccines. The same analysis can be further extended to include other users (nodes) in the top 10 list and check the information that those users share.
Kërkimit Shkencor dhe Inovacionit (National Agency for Scientific Research and Innovation). Its content is the author’s responsibility and the opinions here expressed are not necessarily the opinions of NASRI.
G. Pennycook, D. Rothschild, M. Schudson, S. A. fpubh.2021.700213.
Sloman, C. R. Sunstein, E. A. Thorson, D. J. Watts, [9] S. A. Nossier, Vaccine hesitancy: The greatest J. L. Zittrain, The science of fake news, Science threat to covid-19 vaccination programs, Journal of 359 (2018) 1094–1096. URL: https://www.science. the Egyptian Public Health Association 96 (2021). org/doi/abs/10.1126/science.aao2998. doi:10.1126/ doi:10.1186/s42506- 021- 00081- 2. science.aao2998. [10] K. Gunaratne, E. A. Coomes, H. Haghbayan, [2] I. J. Borges do Nascimento, A. Beatriz Pizarro, Temporal trends in anti-vaccine discourse J. Almeida, N. Azzopardi-Muscat, M. An- on twitter, Vaccine 37 (2019) 4867–4871. dré Gonçalves, M. Björklund, D. Novillo-Ortiz, URL: https://www.sciencedirect.com/science/ Infodemics and health misinformation: A sys- article/pii/S0264410X1930876X. doi:https: tematic review of reviews, Bulletin of the //doi.org/10.1016/j.vaccine.2019.06.086. World Health Organization 100 (2022) 544–561. [11] M. Needham, A. E. Hodler, Graph algorithms: Pracdoi:10.2471/blt.21.287654. tical examples in Apache Spark and neo4j, O’Reilly, [3] S. C. Briand, M. Cinelli, T. Nguyen, R. Lewis, D. Pry- 2021.
bylski, C. M. Valensise, V. Colizza, A. E. Tozzi, [12] A.-L. Barabási, M. Pósfai, Network science, CamN. Perra, A. Baronchelli, et al., Infodemics: A bridge University Press, 2017. new challenge for public health, Cell 184 (2021) [13] L. C. Freeman, A set of measures of centrality based 6010–6014. doi:10.1016/j.cell.2021.10.031. on betweenness, Sociometry 40 (1977) 35. doi:10. [4] A. Chiolero, How infodemic intoxicates pub- 2307/3033543.
lic health surveillance: From a big to a slow [14] V. D. de Carvalho, T. C. Nepomuceno, T. Poleto, data culture, Journal of Epidemiology and Com- A. P. Costa, The covid-19 infodemic on twitter: munity Health 76 (2022) 623–625. doi:10.1136/ A space and time topic analysis of the brazilian jech- 2021- 216584. immunization program and public trust, Tropical [5] E. Wilhelm, I. Ballalai, M.-E. Belanger, P. Benjamin, Medicine and Infectious Disease 7 (2022) 425. doi:10.
C. Bertrand-Ferrandis, S. Bezbaruah, S. Briand, 3390/tropicalmed7120425.
I. Brooks, R. Bruns, L. M. Bucci, N. Calleja, H. Chiou, [15] R. Olszowski, M. Zabdyr-Jamróz, S. Baran, P. Pięta, A. Devaria, L. Dini, H. D’Souza, A. G. Dunn, W. Ahmed, A social network analysis of tweets J. C. Eichstaedt, S. M. A. A. Evers, N. Gobat, related to mandatory covid-19 vaccination in M. Gissler, I. C. Gonzales, A. Gruzd, S. Hess, poland, Vaccines 10 (2022) 750. doi:10.3390/ A. Ishizumi, O. John, A. Joshi, B. Kaluza, N. Khamis, vaccines10050750.
M. Kosinska, S. Kulkarni, D. Lingri, R. Ludolph, [16] W. Ahmed, J. Vidal-Alaball, J. Vilaseca, A social T. Mackey, S. Mandić-Rajčević, F. Menczer, V. Mu- network analysis of twitter data related to blood daliar, S. Murthy, S. Nazakat, T. Nguyen, J. Nilsen, clots and vaccines, International Journal of EnviE. Pallari, N. Pasternak Taschner, E. Petelos, M. J. ronmental Research and Public Health 19 (2022) Prinstein, J. Roozenbeek, A. Schneider, V. Srini- 4584. doi:10.3390/ijerph19084584. vasan, A. Stevanović, B. Strahwald, S. Syed Abdul, [17] Standard v1.1 | docs | twitter developer platform, S. Varaidzo Machiri, S. van der Linden, C. Voegeli, 2023. URL: https://developer.twitter.com/en/docs/ C. Wardle, O. Wegwarth, B. K. White, E. Willie, twitter-api/v1.
B. Yau, T. D. Purnat, Measuring the burden of [18] Oauth 1.0a | docs | twitter developer platform, infodemics: Summary of the methods and re- 2023. URL: https://developer.twitter.com/en/docs/ sults of the fith who infodemic management con- authentication/oauth-1-0a. ference, JMIR Infodemiology 3 (2023) e44207. [19] Tweet object | docs | twitter developer platform, URL: https://infodemiology.jmir.org/2023/1/e44207. 2023. URL: https://developer.twitter.com/en/docs/ doi:10.2196/44207. twitter-api/v1/data-dictionary/object-model/tweet. [6] Infodemic, 2023. URL: https://www.who.int/ [20] User object | docs | twitter developer platform, health-topics/infodemic#tab=tab_1. 2023. URL: https://developer.twitter.com/en/docs/ [7] N. Perra, Non-pharmaceutical interventions dur- twitter-api/v1/data-dictionary/object-model/user. ing the covid-19 pandemic: A review, Physics [21] Mysql 5.7 reference manual :: 22.2.4 hash partitionReports 913 (2021) 1–52. doi:10.1016/j.physrep. ing, 2023. URL: https://dev.mysql.com/doc/refman/ 2021.02.001. 5.7/en/partitioning-hash.html. [8] O. V. Ebrahimi, M. S. Johnson, S. Ebling, O. M. [22] Mysql 5.7 reference manual :: 22.6 restricAmundsen, Ø. Halsøy, A. Hofart, N. Skjerdingstad, tions and limitations on partitioning, 2023. S. U. Johnson, Risk, trust, and flawed assumptions: URL: https://dev.mysql.com/doc/refman/5.7/en/ Vaccine hesitancy during the covid-19 pandemic, partitioning-limitations.html.
Frontiers in Public Health 9 (2021). doi:10.3389/ [23] Gexf file format, 2023. URL: https://gexf.net/.