=Paper=
{{Paper
|id=Vol-2411/paper4
|storemode=property
|title=The Diffusion of Fake News through the "Middle Media" - Contaminated Online Sphere in Japan
|pdfUrl=https://ceur-ws.org/Vol-2411/paper4.pdf
|volume=Vol-2411
|authors=Hirotaka Kawashima,Hiroyuki Fujishiro
|dblpUrl=https://dblp.org/rec/conf/sigir/KawashimaF19
}}
==The Diffusion of Fake News through the "Middle Media" - Contaminated Online Sphere in Japan==
https://ceur-ws.org/Vol-2411/paper4.pdf
The diffusion of fake news through the ”middle media” -
contaminated online sphere in Japan
Hirotaka Kawashima Hiroyuki Fujishiro
Graduate School of Sociology, Faculty of Social Sciences,
Hosei University Hosei University
2-17-1 Fujimi, Chiyoda-ku, 4342 Aihara-machi, Machida-shi,
Tokyo, Japan Tokyo, Japan
kwsmhr@gmail.com fujisiro@hosei.ac.jp
exposed to 80,000 posts from a Russian propaganda
group during two years at election. Google deposed
Abstract that they had banned 18 YouTube accounts identified
as the Russian propaganda group which had uploaded
The purpose of our research is to determine a total of 1,108 videos. Twitter also deposed that their
how fake news is disseminated on the Japanese statistics on Russia’s organic reach on twitter in two
portion of the internet. We adopted the na- and a half months last year, that is 1.4 million tweets
tional election held in October 2017 as a case. and 288 million impressions by Russian bot accounts.
We found a fake news on an opposition politi- It has become increasingly difficult for the public to
cian was diffused through some intermediate discern facts from fiction.
media, so called ”Middle Media” in Japan, Aside from practical struggle by journalists, aca-
and these media had the key role in hinder- demic approaches have been also reported. [All17],
ing the distribution of correcting information. gathered a database of fake news articles that circu-
Our finding suggests that if middle media have lated in the three months before the U.S. presidential
a large presence in your country, the effect of election in 2016. The authors also conducted an on-
correcting information, currently regarded as line survey to acquire demographics, political affilia-
a solution, will decrease as a result. tion, news consumption and whether participant can
recall some specific news headline including fake news.
1 Introduction Based on regression model, they show some polarized
To elucidate the viral structure of fake news is an emer- beliefs like ”People believe what they want to believe”
gent issue. Triggered by the U.S. presidential elec- and Bayesian model. [Dar17], collected top 50 most
tion, the effects of fake news are being reported in a retweeted tweets of each day from September 1st, 2016
number of countries. Related to the U.S. presidential to November 8th, 2016 (Election Day). They catego-
election in 2016, social media have been mentioned as rized those tweets based on whether the tweet is sup-
a field of intervention from abroad. This maneuver- porting or attacking and that is whether neutral or
ing is named ”Russia gate”. On 2017, October 31th irrelevant to either candidate. They found the Trump
and November 1st, Facebook, Google and Twitter in- campaign was more attacking and Trump supporters
vited by Congress to answer on Russian use of their were more likely to share links from websites which
platform in campaign ([Lap17]). Facebook deposed are of questionable credibility than Clinton support-
that as many as 126 million people have been possibly ers. [Dav16], had developed a web service to evaluate
whether each twitter account is human-controlled or
Copyright c 2019 for the individual papers by the papers’ au- an automated bot.[Sha17] had based on that system.
thors. Copying permitted for private and academic purposes. They focused on the role of social bots in the spread
This volume is published and copyrighted by its editors.
of misinformation and found the active role of social
In: A. Aker, D. Albakour, A. Barrón-Cedeño, S. Dori-Hacohen,
M. Martinez, J. Stray, S. Tippmann (eds.): Proceedings of the
bots in the spread of misinformation. [Zub18] distin-
NewsIR’19 Workshop at SIGIR, Paris, France, 25-July-2019, guished two types of rumors on social media, which
published at http://ceur-ws.org are long-standing rumors and newly emerging rumors.
Based on this demarcation they resolve the solutions claims. RQ2, the role on the context: middle me-
into four component; rumor detection, rumor tracking, dia add or change the context of questionable claims
rumor stance classification, and rumor veracity classi- through the diffusion process.
fication.
Meanwhile, in Japan, little has been reported on 4 Data Collection
the context of fake news though viral structure on
misinformation and disinformation in natural disaster Finally, the dataset in this research is composed of the
situations has been big theme based on Japanese ex- following four types of data. These four data are re-
periences of an earthquake and nuclear accidents at lated to a fake news on an opposition politician, which
2011. The Japanese national election was held in Oc- diffused through mass, middle and personal media in
tober 2017. We targeted this national election as a the campaign.
case and collected questionable claims about candi-
dates and parties on social media during the elections. A. 566 tweets which mention that fake news.
In this research, we empirically show the role of inter-
B. 126 unique web pages (URLs) linked from A.
mediate media between social media and mass media
in the viral structure of fake news though the case
C. 148 tweets which mention the correcting informa-
study on the national election in Japan.
tion.
2 Middle Media Model D. 25 unique web pages (URLs) linked from C.
The structure of media is influenced to no small extent
Here we describe the collecting protocol of data col-
by the country and the society. In Japanese case, there
lection and related mother project. As the first step,
are a lot of intermediates between mass media and
questionable claims related to the election had been
personal media such as news aggregators and content
collected during the campaign, which was adminis-
curators. Here, examples of mass media are TVs or
tered as a temporal project by Japan Center of Ed-
portal sites and examples of personal media are social
ucation for Journalist (JCEJ). We joined this online
media, personal blogs or bulletin boards. [Fuj06] called
verification project. Collaborating journalists from a
these intermediates ”middle media” in the Japanese
total of 19 media companies (newspapers, TV broad-
context. Personal media are assembled and summa-
casts and web media) verified the questionable claims.
rized by middle media. Middle media provide the con-
The project published five debunks related to the cam-
tent to mass media and mass media deliver the news
paign. Of the five debunks, we specifically looked into
made of the contents from middle media to public. Al-
fake news related to an opposition politician. The
though the role of middle media has been recognized
politician is Kiyomi Tsujimoto, an incumbent candi-
and pronounced in the field of Japanese journalism,
date from opposition first party (at the election). She
that has remained at only insight without data from
had already won six national elections (and as a result,
concrete case studies. In this research we adopt this
achieved seventh at this election). She has been elected
middle media model and define them as media that ful-
many times, which indicate her presence and news
fill the existing gap between mass and personal media,
hook. Her twitter account (@tsujimotokiyomi) has
to clarify the viral structure of fake news and try to
stopped tweeting from July 30th, 2015 and the num-
connect this conceptual model with the data from fo-
ber of followers is 19,240 (as of October, 2017), that is
cused case of the Japanese national election. We clas-
relatively small as an active Diet member. Addition-
sify each media related to the focused fake news into
ally, Public Offices Election Act in Japan prohibits all
mass media, middle media or personal media based
candidates to send their message through social me-
on the definition. This classification and tracking fake
dia. In other words, she had difficulty in responding
news during the election gives us factual evidences on
to questionable claims immediately during the elec-
the role of middle media.
tion. The content of the fake news is described be-
low. It was published on September 29th, 2017 by the
3 Research Question middle media named J-CAST News (https://www.j-
Based on these social background and middle media cast.com). The title of the article is ”Kiyomi Tsuji-
model on Japanese context, we fixed a research ques- moto has fallen into ’insanity’, it’s been coming up
tions on the role of middle media. RQ1, the role on the a lot on internet. She remained silent despite ques-
distribution: middle media functions as the interme- tions by reporters. Suddenly, she said something am-
diate not only between the mass media and the public biguous.” After this claim was distributed, J-CAST
(personal media) but also between the personal me- News publish the correcting information at October
dia and the mass media in distributing questionable 4th. Both the article (fake news) and its correcting
information are target of this research to track the vi- shows another information. To see horizontal (time se-
ral paths and features. As the second step of data ries) distribution, once happening the fake news from
collection, we searched related tweets based on the middle media or original tweet which failed the seed of
keyword ”Tsujimoto” and/or words within the article fake news, the news is distributed to other middle me-
from September 29th to November 14th. After col- dia, mass media and personal media in relatively short
lecting related tweets, human coders judged whether term. However, this is different for its correcting infor-
each tweet is about the original fake news or its cor- mation. The diffusion of correcting information is less,
recting information. Here we obtained A. 566 tweets not only in amount but also in velocity. Figure 2 gives
which mention that fake news and C. 148 tweets which us the path of distribution, and quantitative informa-
mention the correcting information. As the third step tion. It indicates that the original fake news content
of data collection, we extracted URLs and its linked diffused through middle media, and once the amplifi-
web pages from those tweets. After deduplications, we cation by middle media reached mass media, middle
also reextracted another URL linked from those web media picked the topic up again, as the article from
pages to another web media. Here we obtained B. 126 mass media (the upper block). On the other hand,
unique web pages (URLs) linked from A and D. 25 middle media did not pick the correcting information
unique web pages (URLs) linked from C. in terms of the amount and velocity (the lower block).
As a result, this asymmetrical property promotes the
5 Results diffusion of fake news, and inhibits the transfer of cor-
recting information. We can summarize two findings
Table 1 shows a breakdown of mass, middle, and per- from the discussion on RQ1. Middle media have an
sonal media in our data. ”Number of sites” refers to active role on the distribution of fake news, mediating
the unique number of sites linked to collected tweets fast and extensively between mass and personal me-
related to the fake news in question. ”Average of num- dia. Also, middle media have a passive role on hinder-
bers of tweets” means the average number of tweets in ing the distribution of correcting information by not
each mentioned media. ”S.D. of numbers of tweets” transmitting it as much as the original fake news.
means the standard deviation of tweets of each men- Second, we review RQ2, the context role: middle
tioned media. media add or change the context of uncertain claims
Figure 1 shows the distributions of each media on through the diffusion process. For RQ2, we show the
Twitter related to the fake news, or its correcting in- transition of the news title in Figure 3. Based on our
formation. Circles represent the media content linked survey, the first seed of this fake news was determined
in tweets, and the size of the circles represent the num- as a TV news video, broadcast on September 28th. In
ber of tweets. The horizontal axis shows the time, and that video, Kiyomi Tsujimoto left the press interview
the vertical axis in each media type demonstrates the with negative comments. When the news program was
order of appearance. aired, a tweet mentioning the interview was posted (A
Figure 2 shows the diffusion of the fake news, and its in Figure 3). The tweet was caught by a tweet cu-
correcting information across three media types. The ration site (seikeidouga.blog.jp, B in Figure 3). At
circle indicates the total number of tweets based on that time, the tweet curation site (middle media) gave
summation, by websites which mentioned the previous the story a headline ”Kiyomi Tsujimoto has gone ’in-
website in that sequence, and circle size indicates the sane’.” Another middle media, J-CAST News, caught
number of tweets. this topic, and published their article with the title,
Figure 3 shows the changes in the fake news head- ”Kiyomi Tsujimoto has gone ’insane’; it’s gone viral.
line through the diffusion process. Horizontal axis in- She remained silent at the reporter’s question. Sud-
dicates the time series, the upper block shows the tran- denly, she said something ambiguous.” (C in Figure
sition of fake news, and lower block shows the transi- 3). The part of the title, ”it’s gone viral” indicates a
tion of correcting information. feature of middle media. Through this path: tweet,
curated tweets, middle media article, the context ”It’s
gone viral on the internet that Kiyomi Tsujimoto has
6 Discussion
gone ‘insane’” was added up. After J-CAST News,
First, we look at RQ1, the distribution role: middle each middle media made their own article, with their
media functions as the intermediate not only between own headlines. The title of the news has repeatedly
the mass media and the public (personal media) but changed (D in Figure 3). J-CAST News is a middle
also between the personal media and the mass media media, which delivers to Yahoo! News. Yahoo! News
in distributing questionable claims. We can see in Ta- is one of a largest portal sites in Japan (i.e. Yahoo!
ble 1 that the correcting information was distributed News is in mass media.) The tipping point of diffu-
less than the original fake news. In addition, Figure 1 sion was when Yahoo! News and the other portal sites
Table 1: Statistics on mass, middle, and personal media websites.
Mass Middle Personal
Fake news Number of sites 3 111 12
Ave. of number of tweets 56.3 3.1 2.5
S.D. of number of tweets 40.1 7.4 3.3
Correcting claims Number of sites 2 17 6
Ave. of number of tweets 3.0 2.6 14.8
S.D. of number of tweets 1.0 4.4 20.1
Figure 1: Temporal distributions of each media related to the fake news in focus, or its correcting information.
Figure 2: Diffusions of the fake news in focus, and its correcting information across three media types.
Figure 3: Changes in the fake news headline through the diffusion process.
published the article from J-CAST News (E in Figure middle media have an active role on the distribution of
3). fake news as the fast and extensive mediator, and func-
On the other hand, the changes in correcting infor- tionally fulfill the existing gap between mass and per-
mation exhibited a different appearance. The point sonal media. Second, middle media have a passive role
of origin was an article on JCEJ blog that posted the on hindering the distribution of correcting information
result of fact-checking. After publication by JCEJ, by not transmitting correcting information, as much as
a mass media website introduced the the debunk by it does original fake news. Third, middle media has a
JCEJ. However, the mass media article (Tokyo Shim- qualitative role in changing the context, by adding or
bun, http://www.tokyo-np.co.jp) was focused more on changing the title of claims of both the original fake
the activity of fact-checking rather than the result of news and its correcting information. Our findings sug-
fact-checking. This article had not been picked up by gest that if middle media have a large presence in your
any middle media. Subsequently, Kiyomi Tsujimoto country, the effect of correcting information, currently
published an official statement on her website. Some regarded as a solution, will decrease as a result.
middle media organizations picked up this statement
(F in Figure 3), but no article on middle media made Acknowledgements
it to mass media. During these transmissions, middle The authors are grateful to Rino Yoshii for sup-
media modified the title of the article. One middle porting data collection and discussion. This work
media organization introduced the correcting claim, was supported by JSPS KAKENHI Grant Number
but the title and contents were focused on the con- JP18K11997.
flict between Kiyomi Tsujimoto and J-CAST News (F
in Figure 3). Here, we found an additional qualita- References
tive role of the middle media in changing the context
by adding or changing the headline of claims of both [Lap17] I. Lapowsky. What Congress
original fake news and its correcting information. should ask tech executives about Russia,
https://www.wired.com/story/what-congress-
7 Limitations and Future Works should-ask-tech-executives-about-russia/ (31
October 2017).
Although our results have consistency with research
questions and their background model, there are three [All17] H. Allcott and M. Gentzkow. Social Media
limitations and corresponding future works. First, the and Fake News in the 2016 Election. Journal of
number of cases is not enough because this research is Economic Perspectives, 31–2: 211-236, 2017.
one of the first attempts to elucidate the viral struc- [Dar17] K. Darwish, W. Magdy and T. Zanouda.
ture of fake news in Japan related to political events Trump vs. Hillary: What Went Viral During the
such as elections. Future research should consider a 2016 US Presidential Election. Proceedings of the
significant sample size of fake news, not just one. The International Conference on Social Informatics,
samples will provide the classification on both the type 143–161, 2017.
of generation status and the type of structural paths.
Additionally, the correlations will progress to clarify [Dav16] C. A. Davis, O. Varol, E. Ferrara and A.
the structure of fake news virality. Second, we do Flammini. BotOrNot: A System to Evaluate
not know who disseminated fake news on social me- Social Bots. Proceedings of the 25th Interna-
dia, because the focus of this research is categorizing tional Conference Companion on World Wide
related media into mass, middle, or personal. In fu- Web, 273–274, 2016
ture works, we should analyze each personal media,
[Sha17] C. Shao, G. L. Ciampaglia, O. Varol, A.
especially Twitter and Facebook accounts. If network
Flammini and F. Menczer. The spread of low-
and public profiles are linked to fake news, clustering
credibility content by social bots. Nature Com-
the layer of transmitters can be clarified. Third, our
munications, 9(4), 2018
model of layered media remains a matter of sophisti-
cation. The interface between mass and middle media, [Zub18] A. Zubiaga, A. Aker, K. Bontcheva, M. Li-
and between personal and middle media are possible akata and R.Procter. Detection and Resolution
feature amount for identifications. of Rumours in Social Media: A Survey. ACM
Computing Surveys, 51(2), 2018
8 Conclusion [Fuj06] H. Fujishiro. ’Middle Media’ and
In this paper, we showed the roles of middle media ’Media Inflation’ [published in Japanese],
in the structure of fake news virality in the Japanese http://d.hatena.ne.jp/gatonews/20061203/1165162065,
media ecosystem. Middle media has three roles. First, (3 December 2006)