The behaviour of consumers and information providers during a pandemic: analysis of information related to Covid-19 and the way consumers in social media reacted to this information

Introduction

Pandemics have a global impact on economic, social, and political issues as well as mortality rates. There is great importance to the vast and varied information that people are exposed to during a pandemic and about it. In this aspect, the information people are exposed to during a pandemic can have a crucial and decisive impact on their behaviour during a pandemic and, consequently, on the pandemic spread rate. Today, people consume information that comes from traditional news providers, articles on the web, posts on social media and more. The concept of social media describes various platforms that enable the creation of social connections and information sharing by their users. Social media has opened new information related possibilities for users. If in the past a person was only exposed to information that was filtered and edited by the traditional media channels (radio, television, written press) and there was control over the flow of information; Nowadays when a person searches for information about a pandemic she can find a vast amount of content that others have written or shared. As mentioned, the social network is a global cultural phenomenon. Despite significant differences in terms of regulation, infrastructure, availability, and access to the internet, culture, and customs in different parts of the world, social networks are expanding in most countries. In most developed countries, over 90% of internet users use social media to some extent. Additional studies found that more than 62% of the world’s population surfs the Internet, and approximately 4.62 billion Internet users currently use social networks. In Israel, 88.6% of Internet users use social media (Kemp 2022; Scribd 2022).

The underlying assumption of this study is that information on social networks may affect people’s behaviour during a pandemic. Thus, the information consumed and shared on social networks can influence the rate of pandemic spread. It is important to emphasize that this work does not directly examine the impact of the information that is disseminated on social networks on the actions and behaviour of people in daily life outside the social network. In the proposed study, we chose to focus on the latest pandemic that broke out at the end of 2019 caused by the SARS-CoV-2 virus, which was defined as a serious global threat and earned the professional nickname Covid-19 By the World Health Organization. During the covid-19 pandemic, we experience the ‘information explosion’ phenomenon—the revelation of large amounts of information in a short time. People are exposed to a wealth of information, at different levels of credibility, and such a flood can affect their behaviour as well as how they perceive information (Tangcharoensathien et al. 2020), hence the importance of this study analysing the covid-19 related information that spreads in social media.

Social media users are not apathetic to the information they encounter with. They share, comment, and use various types of emojis to express their feelings and thoughts on the information they engage with. The covid-19 pandemic affected social media users’ online behaviour. Our study attempts to report and offer a possible explanation for this change. Since the covid-19 pandemic introduced uncertainty to our life on scales that we have never experienced before, we harness the uncertainty reduction theory in an attempt to explain social media users’ online behaviour throughout the pandemic period. We chose to focus and examine the impact of the pandemic on the information circulated in the leading social network in Israel.

In our study, we focus on both the information that providers and users generate on social media. In the context of information produced by the information providers, we examined the volume of pandemic related information on social media.

These issues are of great importance in our era, especially in light of the comments of the Minister of Health of the United Kingdom Sajid Javid: ‘we can't live in a world where the only thing we are thinking about is COVID’ (COVID-19 News, 2021). The proposed research sheds light on the dimensions of discourse and reactions to information regarding covid on social media, with an emphasis on in-depth analysis of the information provided on economic, health and education issues. In addition, we examined the attitude of people on social media to covid-19 related information compared to other information throughout different periods of the pandemic.

We collected information from leading information providers on the popular social network in Israel—Facebook. The information included a variety of news items on general and covid related issues, we characterized the types of news items and examined the public's attitude to those news items.

Literature review

Social networks are a major source of information for many people. People receive and share newsworthy information through these networks (Khuntia et al. 2016; Karnowski et al. 2021). Many studies show that people in different cultures consume their news from social networks. Studies have shown that two-thirds of the US population consumes news from Facebook (Newman et al. 2017).

Information is of great importance in our day-to-day decision-making process, and it affects the way we operate in the world. Studies show that information that is disseminated in the digital space and in particular on social networks, affects people's behaviour in the real world. For example, when the United States faced an outbreak of Ebola, wrong information about the plague was distributed, and in combination with conspiracy theories, it created confusion among the public which led to a lack of trust in the health system (Kilgo et al. 2018). Multiple reports and theories which were reviewed on news sites and social networks, led, de facto, to the lack of cooperation with medical teams. According to health workers, this situation of newsworthy information is biased, sowing seeds of fear and mistrust among the public and thus making their work difficult (Millman 2014; Atlani-Duault et al. 2015). Similar results were observed in a study conducted between 2014-2016 in West Africa regarding the Ebola pandemic. the study examined theories and accusations circulated within the framework of social networks about responsibility for provoking or spreading the pandemic. through data analysis and comments posted on Twitter and Facebook, the researchers showed that the blame landed normally on general entities (bodies or groups). Furthermore, the accusations were most often directed at ‘close’ entities such as governments and municipalities, and less at ‘distant’ entities such as Africans, or global health authorities. The results also indicate that the development of online accusations was by periods: in the early stage of the pandemic the accusations were directed mainly towards the populations first infected, but at the peak outbreak stage, the accusations were gradually directed toward general entities (Roy et al. 2020). In our study, we propose to examine the change in pandemic information and responses to it throughout the pandemic period.

In the context of pandemics, the information that flows on social media is not necessarily negative. Many studies have examined the ability to characterize areas infected with the swine flu (H1N1) around the world using information that has been streamed on Twitter. Previous research proposed various models whose results indicated a high correlation between tweets and the health of the population during the plague, even more than Google searches on the subject (Culotta 2010; Lampos et al. 2010; Aramaki, Maskawa & Morita, 2011; Pawelek, Oeldorf-Hirsch & Rong, 2014; Zhang et al. 2014; Woo et al. 2018; Samaras, García-Barriocanal & Sicilia, 2020). The results of these studies reinforce the importance of the study proposed here which analyses the information on social media during the covid plague.

Past research shows that various parameters affect the resonance of information on social networks. These parameters are of paramount importance in studies that examine the spread of information during pandemics. Since our research deals with these issues there is great importance in analysing the characteristics that contribute to the spread of information on social networks. Many studies suggest a link between text sentiment and social media sharing characteristics. For example, a study that revolved around information on Facebook found a strong connection between the number of words that indicate positive and negative emotions in the post and the number of responses that correspond to it. Furthermore, the results also showed that the emotions expressed in the responses, match the nature of the emotions expressed in the original post (Stieglitz & Dang-Xuan, 2012). Another study which examined records on blogs found that positive sentiment records or negative ones tend to get more responses compared to records with neutral sentiment or emotion involved (Dang-Xuan & Stieglitz, 2012). Continuing with the same line, research that examined the sharing patterns of New York Times articles found that news articles with positive content are more likely to be shared in comparison to negative content (Berger & Milkman, 2012).

The conclusion from this line of research is that the information disseminated on social networks tends to also contain emotions to shape public opinion and attract responses and therefore in our research we will address the emotional aspect emerging from the information. The Uncertainty Reduction Theory (URT) claims that people are uncomfortable with uncertainty and constantly use various strategies to reduce their uncertainty (Berger & Calabrese, 1974). In an attempt to reduce that uncertainty, people tend to seek information that will explain someone’s behaviour during an interaction. This theory was later used to explain online behaviour such as low retweets count for disasters (Son et al., 2020). In our work, we utilize the URT to explain the reactions of users to covid-19 related news throughout the pandemic.

Research goal

The main goal of this work is to investigate peoples’ reactions to covid-19 related news throughout the pandemic.

Methodology

In this study, we examined the characteristics of the covid-related and non-related posts, and for each post, we examined the activity on social networks with an emphasis on the number of shares and the types of responses the post received.

The research method we have chosen is an exploratory data science methodology. To this end, we collected information that includes posts (information/news) from a social network and their sharing features. Using statistical tools, we examined the nature and type of relationships between them.

We chose ‘Facebook’ as the representative social network since it is the most common and is used by all levels of the population, regardless of age, gender, and socioeconomic status. For testing a correlation between the various characteristics and to compare averages (word count, number of responses, etc.) we used T-test and ANOVA tests.

We collected a variety of news items and assembled a broad database containing posts of the leading content providers in Israel. In addition, we collected the characteristics of the various post and their sharing data on Facebook. We chose to focus on content providers in Israel because Israel was a pioneer in vaccinating its entire population and therefore, we were able to focus the research on a period of one year which includes an outbreak of pandemics, closures, vaccinations, and return to routine (after the first wave).

In addition, the authors of the article are familiar with the leading content providers in Israel and therefore the manual analysis was executed in a controlled manner.

Information collection

We used the Python programming language and the CrowdTangle API to retrieve and process the data from Facebook (https://research.fb.com/blog/2020/07/crowdtangle-opens-publicapplication-for-academics/).

The type of data we were able to elicit using the CrowdTangle API included: public posts from the social network Facebook, the type of post (video, photo, text, link), from which page or public account it was published, or to which public group it was published, the posts’ interactions (e.g., likes, comments, reactions, and shares), and a list other pages or public accounts that shared it.

We chose to focus on the top leading pages in Israel in terms of total interactions (sum of all types of responses received by the posts on the page) and extracted all their posts from January 2020 to March 2021 inclusive. The inclusion criteria were pages that contained at least 75000 interactions during the period. All in all, we collected information from 249 leading pages (with at least 75000 interactions).

Categorizing posts related to covid

To know which specific posts are covid related, we first took all posts that contained a variation of one of the words ‘covid’, ‘corona’, ‘Sars’ and ‘pandemic’. Next, we took all the posts from the previous step and joined them into one big text. For each word in that text, we calculated its frequency. We manually checked the most frequent words. We noticed that out of the most frequent words the top 5% words (excluding stop words) were found to be covid19 related. Any post that contains any variation of one of the most frequent words reported above has been marked as related to covid.

Page type classification

Each of the 249 Facebook pages (channels) was manually categorized into one of the following topics: news, entertainment, politics, sports, non-profit organizations, religion, government institutions, health and music.

Post sentiment analysis

For the analysis of the sentiment, we used a pre-trained language model based on BERT architecture which was trained on Hebrew corpus (Chriqui and Yahav 2021). The model calculates the probability of knowing the sentiment and divides it between three types of sentiments - positive, negative, and neutral. From a manual check, we saw that every post that received a score of at least 80% in one of the types of sentiment did indeed manage to classify the post accurately and therefore for sentiment analysis we only examined posts that meet this threshold. In addition, we have a manually verified (a sample) that the posts do correspond to the sentiment they received.

Research periods

The study is divided into five parts of three months each:

1. January - March 2020
2. April - June 2020
3. July - September 2020
4. October - December 2020
5. January - March 2021

We chose to divide the information by annual quarters since we wanted to examine trends at a macro level.

Results

People’s reactions to Covid-19 related posts

We start by reporting the reaction of Facebook users towards information (posts) related and unrelated to covid.

As can be seen from Table 1, covid related posts attract slightly more interactions (7%), and likes (5%) in comparison to non-covid related posts (P <0.01). Covid related posts attract more (~30%) shares, angry and loving reactions in comparison to non-covid related posts (P <0.01). The sad, caring and lol interactions are the most interesting ones, as covid related posts attract significantly more (almost double) sad and caring reactions in comparison to non-covid related posts (P <0.01). On the other hand, covid related posts attract significantly less (almost half) LOL reactions in comparison to non-covid related posts (P <0.01). From the above data, it can be concluded that information about covid triggers different emotions and causes users to react differently than information that is not covid related.

We now turn to report the reaction of Facebook users towards information (posts) when taking into account the post sentiment.

Three metrics that examine people's attitude to the information presented to them: total emotions (lol, sad, wow, etc.,), number of comments, and number of shares.

Table 2 demonstrates the averages of the various interaction indices, broken down into three types of sentiments and separated to covid related and unrelated posts. As can be seen from the table, negative and positive posts related to covid attract more emotions than posts that are not covid related. The gap is particularly high in posts with negative sentiment (about 20%) and is reflected in both comments and shares.

In general, in covid related and unrelated posts, a negative post leads to more comments and shares as well as a positive one which leads to more emotions. This behavioural pattern is more noticeable in covid related posts.

Peoples’ reactions to Covid-19 related posts over different periods of the plague We further examine to what extent the average number of the interactions, responses and shares of covid related shares changed throughout the different pandemic periods. We start by reporting the average monthly total interactions for covid related posts. As can be seen from Figure 1 in March, April, October, 2020 and February 2021 the total average of interactions was high.

Also, it can see that in June 2020, and in November and December 2020 the number of interactions was. The conclusion from these results is that peoples’ attitude toward covid related information has changed throughout the plague and there are sub-periods in which the topic entails more interactions than other sub-periods.

We turn to report the average monthly shares for covid related posts. As can be seen from Figure 2, the average of the shares corresponds to the distribution of the average of the interactions. However, it can be seen that there is a downward trend throughout the pandemic period. The decline is not monotonous, and therefore it suggests that throughout the plague there are times when people feel they need to share more information with their environment.

We conclude by reporting the average monthly comments for covid related posts. As can be seen from Figure 3, the mean total comments correspond to the distribution of the mean interactions. However, it can be seen that there is an upward trend throughout the pandemic period. Here too the trend is not monotonous, and here too a possible explanation is that it indicates that throughout the plague there are periods in which people feel they need to comment more than in other periods.

The popularity of Covid-19 related posts across different information providers

In this section, we report the popularity of covid related posts across the different information providers. The division here is not made based on the post’s topic, rather it is made based on the information channel (content providers) category. To recall, we divided (manually) each of the 249 information channels into 9 pre-defined categories.

Figure 4 shows the percentage of covid related posts out of the total number of posts by the different content providers.

Extensive information related to covid was generated in channels related to health, government, and non-profit organizations. A moderate information related to covid was generated in channels related to news and politics. Substantial information related to covid was generated in channels related to religion, entertainment, music and sports. From the figure, it can be seen that the covid topic was discussed in all types of information channels to such an extent that it has even been discussed in the pages of content providers that are seemingly irrelevant to the topic such as religion and music. This figure is surprising and best illustrates the extensive preoccupation with social media around the covid topic.

Discussion

Peoples’ reactions to Covid-19 related posts

We see that covid related posts drew more interactions (mostly shares, ‘sad’, ‘care’,’ angry’, and ‘love’ emojis) than posts that were not related to covid. From the results, we can see that covid is a highly sensitive topic that touches many people and provokes an emotional storm and therefore they feel the need to vent and share their feelings with others.

The users were not satisfied with just sharing covid related posts but also felt the need for emotional expression and therefore we are seeing more angry, sad, love or care emojis. Interestingly, the use of the LOL emoji was rare. This behaviour teaches us that when people express their feelings online, they carefully consider which emoji to use. As the covid related posts are mostly serious people rarely use the LOL emoji.

In general, a negative post leads to more comments and shares as well as a positive one leads to more interactions both in covid related and unrelated posts. This behavioural pattern is more noticeable in covid related posts. We have also seen that covid related posts with a strong sentiment (positive or negative), have drawn much more interactions. Especially in posts with negative sentiment.

Peoples’ reactions to Covid-19 related posts over different periods of the plague

From the average distribution of interactions over the period, we can see three points of high interactions: March - April 2020, October 2020 and February 2021.

In March - April 2020 the pandemic started in Israel, during this period the first infected and deceased persons were reported, and the authorities announced restrictions (first closure, obligation to wear masks, and social distancing).

The second peak, which occurred in October 2020, can be explained by the fact that on September 25, the tightening of the second closure, which included the closure of all workplaces (except for exceptions), sports and synagogues, took effect, and on September 30, legislation was passed that limited distance demonstrations of one thousand meters from the place of residence.

The third peak, occurred in February 2021 at a time when hospital congestion was high, as was morbidity. Simultaneously with the high dimensions of morbidity, discussions about getting out of the third quarantine started.

On February 7 of that year, some of the restrictions, including travelling distance limits were lifted. Part of the education system opened on 11 February. On February 15 the government decided that on the 21st of the month, a green passport will begin to be issued in culture and leisure, as well as the opening of trading places at the same time as easing the restrictions on gatherings and opening additional parts of the education system. The discussion about the green passport was heated and varied and therefore there was an impact on the reactions on social media.

As expected in June 2020 and November-December 2020, which were the lowest peaks in terms of interactions, there were no significant covid related events that affected the population (closures/restrictions/government decisions).

In terms of the types of interactions, we see that the trend of sharing has been declining over the examined period. When a new event occurred the users thought it was right to share the information further and over time they realized that covid related information is available to all and then reduced the shares.

However, in the peak months, when new important covid related events have occurred, we see a temporary increase in the number of information shared by the users. This behaviour can be explained by the uncertainty reduction theory. When people encounter new events they attempt to reduce their uncertainty by engaging with information sources and spreading the news to others to reduce their uncertainty. When new information does not help people in reducing their uncertainty (throughout time) they do not feel the need to share the information with others.

In terms of the trend of reactions, we see an increase over time. During the pandemic, people consumed a lot of information regarding the pandemic and became experts. When people felt they know enough about the pandemic situation (over time) they felt more comfortable expressing their opinions.

The popularity of Covid-19 related posts across different information providers

From the distribution of the number of posts published, it can be seen that the engagement with the covid subject depends on the type of information provider (page).

During the epidemic, the vast majority of posts on health pages dealt with the covid topic. Because it is a health epidemic, it is natural that the issue will be discussed on health pages. However, the extensive discussion on those pages (almost 90% of the posts were covid related) surprised us.

In our opinion, these results can be explained by the fact that unlike other diseases ,this disease has affected everyone and thus the health pages felt the need to constantly inform people about this disease.

The sheer volume of posts about covid in the health category can be problematic, as it can give people the impression that this is the most acute health problem. As a result of that, it can cause people neglect other health issues.

On government pages and non-profit organizations about half of the posts were covid related. Government pages were required to convey explanations, guidelines and government decisions to create a change in people’s behaviour to eradicate the plague.

As a result of the plague, there was an increase in the number of people who needed help. Non-profit organizations were more involved in helping epidemic victims in various aspects (medical, economic, social, emotional, etc.’) and therefore they posted more covid related information on their pages to attract volunteers and raise funds (Klein, 2017).

In news and politics pages about a third of the posts were covid related. When it comes to a common disease, we would not expect this amount of information on news and politics pages. Since this is an abnormal disease that has reached the dimensions of a global epidemic, political intervention was needed to take control of the infection and curb the epidemic. As a result, there was high engagement on the subject on news pages as well.

In terms of post sentiment, the relationship between positive and negative and neutral depends on the type of page.

In news pages, the reports were more neutral, as they deal with rapidly status updates rather than discussion posts and so unlike other topics, the posts were more neutral. On government pages (e.g., municipalities, security bodies, etc.) the posts were more negative, as these ministries convey the guidelines to the population and express the severity of the situation, to convince the population to carry them out.

The covid related post on non-profit organizations’ pages was more positive. When examining a sample of the posts’ content we were able to understand the reason for this phenomenon. The extensive use of positive posts on non-profit organizations’ pages was to show successes and thus encourage people to donate money or recruit them to volunteer. Pages from the Health category post alerts, information about the pandemic, how to keep ourselves healthy, and behaviour that endangers our health. The amount of covid related posts on these pages was large and diverse (positive messages about patient recovery, relief in guidelines (such as exit from a lockout)), and therefore the ratio of the positive-negative posts was about the same.

As we have seen the subject of the corona has taken over the discourse to such an extent that it has even been extensively discussed in the pages of content providers who are seemingly irrelevant to the subject.

Conclusions

In this work we examined the effect of the covid pandemic on social networks discourse, assuming that this activity has an effect on public opinion and consequently on the culture and behaviour of the population.

To this end, we collected and analysed news from the top 249 pages on Facebook, and examined the types of posts and responses as well as sub-periods, during the covid plague. This examination was conducted by measuring the activity on the social network in accordance with the various characteristics of the posts and the reactions to them. The examination was done on posts related to covid in relation to themselves and in relation to other non-covid related posts.

The data showed that covid related information did indeed take over the discourse on the examined social network, hence we see that the covid topic had a special place in the Israeli atmosphere.

According to the results which emerged from this research, it turned out that information related to covid led to more interactions and reactions. There were more responses to covid related posts, and they were more emotional, the preoccupation with the subject intensified during periods when pandemic-related incidents occurred, or weakened at times when there was a return to a relative routine.

It was further revealed that covid related posts were more common in topics which involve health, politics, economics education and social community needs (volunteering, donation, etc.).

In the age of discourse availability on social networks, every emotion (positive or negative) gets a strong and immediate resonance.

We have seen that interactions, shares and responses correspond with current events and therefore decision-makers can predict an increase in sentimentality, which can cause fear and mistrust, and must demonstrate consistent and empathetic decision-making leadership. Thus, it seems that with proper conduct, leaders can manage crisis times more correctly, and harness social media to convey reliable and useful information which in turn will influence people.

About the authors

Tom Potash is a master’s student at the Information School of Bar-Ilan University, Max and Anna Web, Ramat-Gan, 5290002, Israel. He can be contacted at potas2@gmail.com
Avshalom Elmalech is a Lecturer at the Information School of Bar-Ilan University, Max and Anna Web, Ramat-Gan, 5290002, Israel. He received his Ph.D. from Bar-Ilan University and his research interests are in crowdsourcing, digital humanities and recommender systems. He can be contacted at avshalom.elmalech@biu.ac.il

How to cite this paper