A total of 202 news articles were analyzed. A total of 68 news sites from the US, Canada, UK, Ukraine, Russia, and the Czech Republic were analyzed. 169 articles were written in English, 31 in Russian, 1 in Ukrainian, and 1 in Czech. Most articles were sourced from sites like Newspunch.com (38 articles), Neonnettle.com (27 articles), Worldnewsdailyreport.com (20 articles), Huzlers.com (11 articles). The Figure 21 provides a list of sites that have posted an article that has been flagged by fact-checking resources as untrue or misleading:
56
Figure 21: List of sites with fake news (source: author)
In the process of analyzing the data obtained, the main topics of fake news were identified.
Most of the articles were devoted to the topic of politics - 37.1% (75 articles). The top three
57
also included violence or crime (17,8% or 36 articles) and health (15,3% or 31 articles). A large gap between the political topic in fake news and the backward ones can also be traced in the scientific work of Vosoughi, Roy & Aral (2018). The figure below represents the categorization of articles by topic:
Figure 22: Categorization of fake news by category (source: author)
Of the 202 news articles analyzed, 63 articles were related to websites that position themselves as satirical publications or publications that publish news for entertainment purposes only. The diagram below shows the ratio of articles related to satirical publications and the number of articles from publications that position themselves as news sites:
58
Figure 23: The ratio of satirical articles to "real" ones (source: author)
In addition, most of the fake news articles were written about ordinary people, who may not have existed but were invented exclusively for news. The diagram below represents the ratios of articles about famous people and ordinary (possibly fictional) ones.
Figure 24: The ratio of articles about famous people to ordinary people (source: author)
Donald Trump is the absolute favorite among the articles that have been written about famous people. 23 out of 77 articles about famous people are written about Trump, which is 29.87%.
59
During the analysis of 202 fake news articles, 137 of them had no source citation, while only 65 articles had a source, which is 32.18%, as shown in the diagram below.
Figure 25: The ratio of articles with links to a source of articles without a source (source: author)
Most of the articles without links to the source refer to satirical publications that do not have any links to the original source due to their complete fiction. However, those fake news articles that have links to sources cannot be called reliable. So, for example, the satirical news site The Burrard Street Journal always refers to its own articles in its news. Other fake news sites often link to sources that are not trusted, or simply link to the home page of reputable publications such as The Washington Post or The Guardian. Obviously, news cannot be found on the pages of these resources. This information allows us to conclude that fake news, due to its origin, do not have links to sources, or the sources are not reliable;
In the process of studying fake news, their readability was studied using Flesch – Kincaid Readability Testing. The table below shows the average results for the four most frequent sites:
Website URL Flesch – Kincaid
Readability Testing score
Average Readability
worldnewsdailyreport.com 57,77 8,05
newspunch.com 58,46 7,66
60
neonnettle.com 64,73 6,44
huzlers.com 65,07 6,72
Table 3: Readability score of the 4 most frequent fake news sites
At the same time, the average level of readability of the text in 202 articles was 62.78% on the Flesch – Kincaid scale, which corresponds to the 7th level of readability. Grade level of about 7 should be easily understood by children aged 12-13. This level roughly corresponds to the reading level of the book "Harry Potter and the Order of the Phoenix" in English, which has 65.88 points. The result was compared with the readability analysis of 21 reputable news sources such as Washington Post, Buzzfeed, The Guardian, BBC, and CNN made by Tauberg (2019). The figure below depicts the average readability for news sources:
Figure 26: News Publication by Average Readability (Tauberg, 2019)
Based on the data obtained, we can conclude that the average level of readability of fake news is much lower than that of reliable sources. For example, the level of readability of the MSNBC website corresponds to the level of a 19-year-old person, and the level of readability of the BBC website corresponds to that of a 16-year-old student. This conclusion confirms the
61
hypothesis that fake news is written in the most accessible way, in order to be understood by the widest possible audience.
In total, Facebook users showed 8,488,676 reactions to posts with links to fake news articles.
In addition, users left 3,345,971 comments under posts linking to fake news from the dataset.
At the same time, users shared posts with links to fake news 3,636,136 times. On average, there were 2.54 user reactions per 1 comment and 2.33 user reactions per 1 share.
Using the BuzzSumo.com tool, it became clear that most users used the emoticon 😂 (Haha):
among all reactions, the share of this emoticon is 44.25%. The next most popular was the 😯 (Wow) emoticon with 20.58%. Next are the 😠 (Angry) emoticon with 19,57%, ❤ (Love) with 9,34%, and 😢 (Sad) with 6,26%. At the same time, the general statistics of user reactions on Facebook to posts show that the most popular reaction is ❤ (Love) emoticon, while the most rarely used emoticon is 😠 (Angry). The prevalence of the emoticon 😂 (Haha) in our study can be explained by a large number of user reactions to publications with links to satirical publications. For example, the largest number of reactions have posts linking to the World News Daily Report: almost all of their posts are entertaining. The second most popular are posts about politics or incidents, to which users most often reacted with emoticons 😯 (Wow) or 😠 (Angry).
Figure 27: Using reactions to posts with a link to fake news (source: author)
62
Figure 28: Reaction usage in top shared posts on Facebook in September 2017 (Statista, 2019)
Using the CORREL function in Excel, a correlation was found between the number of reactions and user comments to the news on Facebook. The correlation coefficient was 0.866, which indicates a strong relationship between the two data sets. The graph below depicts the correlation between the number of user reactions and comments under news posts on Facebook:
Figure 29: Correlation between the number of user reactions and comments under news posts on Facebook (source: author)
This correlation tells us that with an increase in the number of user reactions to a post with a news, the number of comments under this post also increases. The same correlation was found between the number of user reactions and the number of shares. The correlation coefficient for these datasets was 0.854. The graph below shows the correlation between the number of user reactions and shares of posts on Facebook:
63
Figure 30: correlation between the number of user reactions and shares of posts on Facebook (source: author)
In addition to the relationship between the number of user reactions, the number of comments, and the number of reposts, a connection was also found in the qualitative characteristics of fake news articles.
1) A relationship was found between the criteria "Does the site pose as a real news portal?" and
"Are there any links to the" source "?". The association coefficient for these criteria was 0.65, and the contingent coefficient was 0.37. Analysis of this connection showed that out of 63 satirical articles (from sites that are not positioned as real news), 59 articles do not have any links to the source, which is 93.65%. To paraphrase, it turns out that the vast majority of satirical articles do not have links to sources.
2) A relationship was found between the criteria "Is the date when the news was published?"
and "Are there any links to the "source?". The association coefficient for these criteria was 0.67, and the contingent coefficient was 0.39. Analysis of this connection showed that among 59 fake news articles that do not have a publication date, 56 do not have a link to the source, which is 94.92%. To paraphrase, it turns out that the vast majority of articles without a publication date also do not have links to sources.
3) A relationship was found between the criteria "Are there photos? Quantity?" and "If there are some images, do they have a source?". The association coefficient for these criteria was 0.94, and the contingent coefficient was 0.68. The analysis of this connection showed that among 19 articles where photos have a source, 16 articles have only 1 or 2 photos, which is
64
84.21%. To paraphrase, it turns out that most articles in which photos have a source have only 1 or 2 photos.
4) A relationship was found between the criteria "Does the title match the content?" and "Can the news be perceived as real?". The association coefficient for these criteria was 0.78, and the contingent coefficient was 0.42. The analysis of this connection showed that among the 101 fake articles that look like real news, 97 have a headline that matches the article, which is 96.04%. To paraphrase, it turns out that in the vast majority of articles whose title matches the content, the article also looks real.
In the process of analyzing news articles, I came across almost all types of fake news that were presented in the theoretical part. The most common techniques used by fake news authors were manipulation of facts, the omission of important information, clickbait headlines.
For example, Newspunch.com published a news item on February 16, 2018, with the headline
"Obama Calls For Nationwide Gun Ban After Florida Shooting." Obama did not call for a nationwide gun ban but rather stressed the importance of gun laws. This news is a good example of clickbait where the reader responds to the headline, while the article is written about something else.
Another interesting example of the spread of fake news is an article in the satirical publication Burrardstreetjournal.com, which talks about the upcoming purchase of the famous Chilean football player Alexis Sanchez by Manchester United. This news is completely fictional, however, it was posted in the fan community of Manchester United with nearly 50,000 members.
While analyzing news articles, I came across an interesting article about the arrest of Milia Obama, the daughter of former US President Barack Obama. The article did not have a publication date, which is a common phenomenon for fictional news. The news, published by the satirical news agency Daily World Update, claimed that Malia was detained for discovering
"a white, powdery substance" at her house (Eagleton, 2019). The article also does not mention the date when the incident occurred. The absence of the publication date and the date of the incident made it possible for the Facebook communities to publish the news multiple times, as shown in the picture below:
65
Figure 31: Posting the same news at different times on Facebook (source: author)
As shown in the picture, a Facebook community “America's Last Line Of Defense”, with an audience of over 220,000 users, posted the news three times: July 19, 2019, October 23, 2019, and March 10, 2020. This example clearly shows the importance of specifying the publication date, since news without a date and time can easily confuse the reader.
Despite my best efforts to find fake news about coronavirus, I have not been able to find at least a few such news. At the time of this writing, Google's algorithms were extremely quick to remove all unverified news regarding COVID-19. Even when writing in search queries like
"Coronavirus is a myth" or "Coronavirus was created on purpose", you only get links to articles that expose the myths. The same thing happens when using Bing, Yandex, Seznam, or Ukr Net.
Appendix B shows search engine responses to the query "coronavirus is a myth".
4.4 Summing up
In total, 202 fake news articles circulating on the network from 2018 to 2020 were investigated.
In order to obtain more detailed information, the web tool CrowdTangle and BuzzSumo.com were used. Most of the fake news was written about politics (37.1%) and crime/violence (17.8%). The average reading time for one news article was 3 minutes, which corresponds to approximately 600 words in the text. The hypothesis was confirmed, according to which fake
66
news is written as accessible as possible in order to be understood by a wide audience. The average reading level was 62.78 on the Flesch Kincaid Reading Ease, which corresponds to a 13-year-old student.
In addition, it was found how Facebook users react to fake news. In total, users have reacted to publications with links to fake news 8,488,676 times and also left 3,345,971 comments. In addition, users made 3,636,136 reposts of publications with fake news. Facebook users mostly reacted to the posts using the 😂 (Haha) emoticon, since the satirical posts were the most popular fake news. The least commonly used emoticons were ❤ (Love) and 😢 (Sad).
During the analysis of fake news, the hypothesis was also confirmed that fake news, due to its origin, do not have links to sources, or the sources are not reliable. Also, correlations were found between the number of user reactions to posts and the number of comments and reposts.
The links between the qualitative characteristics of fake news were discovered and described.
Moreover, empirically, key characteristics of fake news have been discovered, such as the use of clickbait headlines or manipulation with facts.
During the analysis of fake news, it was noticed that search engines such as Google, Bing, Seznam, etc. are blocking all misleading news regarding Covid-19. When trying to search for information such as "coronavirus is a myth", search engines returned only reliable articles about the virus or articles that exposed myths about the virus, but not articles with conspiracy theories.