Using Machine Learning to infer the Ideology of Japanese Twitter Users

May 10, 2018 TOPICS

Using Machine Learning to infer the Ideology of Japanese Twitter Users: News Audience Fragmentation in the Japanese Twittersphere

A recent study published by a research team, including Assistant Professor Yuki Ogawa of the College of Information Science and Engineering, has found that, aside from a few exceptions, the ideology-based news audience fragmentation commonly seen among American Twitter users does not appear to exist in Japan.

The study, which used machine learning to estimate the ideology of Twitter users, was led by Associate Professor Tetsuro Kobayashi of the City University of Hong Kong with contributions from Assistant Professor Ogawa as well as Assistant Professor Takahisa Suzuki from Tsuda University and Professor Hitoshi Yamamoto from Rissho University. Their findings were published in the Asian Journal of Communication on April 6.

Twitter and other types of social media make it easier for their users to selectively expose themselves to information that adheres to their pre-existing attitudes, so scholars have pointed out the possibility of news audience fragmentation occurring when conservatives follow conservative media accounts and, likewise, liberals follow liberal media accounts.

If this tendency to only look at information one wants to look at is heightened when accessing news via social media, news audiences can end up fragmenting along ideological lines. This means users lose opportunities to contemplate political issues in consideration of other peoples’ points of view, and public opinion runs the risk of becoming rigidified.

Twitter followers have a tendency to fragment in the US
Scholars have pointed out the possibility of news audience fragmentation occurring on social media - conservatives following conservatives, liberals following liberals etc.

For this study, the researchers recruited Japanese Twitter users who followed at least one Diet member and at least one media outlet as of the fall of 2014. By using machine learning together with survey data and social media data (i.e., the content of Diet members’ and users’ tweets), the researchers were able to predict the users’ political ideologies with more precision than previous methods.

Next, they used this predictive model to estimate the political ideologies of more than 600,000 Twitter users and determine the degree of news audience fragmentation in the Japanese Twittersphere. What they found is that there is no ideological fragmentation among Twitter users who follow NHK and the major national newspapers, and this in turn shows that Japan Twitter audiences are unlike their counterparts in the United States, where fragmentation between conservative and liberal media audiences is more pronounced.

The results did, however, allow the researchers to determine that followers of Sankei Shimbun are more conservative, whereas followers of Tokyo Shimbun are more liberal. Although the numbers of followers with strong ideological slants were low, such users tended not to follow other media outlets - demonstrating, at least for some media outlets, that some small-scale ideological fragmentation does exist.

Project leader, Associate Professor Kobayashi, hopes these findings can help clarify the role that social media plays in the process of public opinion formation in Japan.

To find out more follow the link below to access the full research paper:

News audience fragmentation in the Japanese Twittersphere
(Asian Journal of Communication 2018) Kobayashi, T., Ogawa, Y., Suzuki, T. & Yamamoto, H.


May 01, 2018 TOPICS

The Potential for new Kimono Businesses in Kyoto