Estimating Media Effects Using an Ideologically Diverse News Curation Platform

• In Event: Information Diet and Its Effects

Abstract

In this paper, we collaborated with Gunosy, a major news curation app in Japan with more than 10 million domestic downloads. We examine how ideologically diverse readers encounter ideologically diverse news content and how they react, utilizing machine learning and text mining techniques and a vast amount of data. We argue that the app’s unique features and the sheer size of data allow us to estimate the truly cognitive portion of the effect of selective exposure. We also touch upon the implications of the persuasive effect and other behavioral patterns of readers.

News curation apps gather news from various media and display them in parallel on the same screen. They are becoming a major source of traffic to publisher sites in many countries, including Japan. In contrast to traditional media, news curation apps allow readers to see ideologically diverse content at a glance on a smartphone or PC screen. In the case of Gunosy, readers can choose which article to read by clicking on the link, and they can also hit the “iine” (“like” in Japanese) button when they agree with the views expressed in the article. News curation apps thus provide a novel and real setting to rigidly estimate particular media effects.

We analyzed the behavior of more than 60 thousand readers of Gunosy’s political news articles using machine learning techniques. We first estimated the articles’ ideological position: Of the nearly 200,000 articles published between 2018 and 2020, we focused on ten salient issues marked by distinct ideological conflict. By utilizing existing measures for the left-right slant of Japanese media sources and left-right dictionaries, we assigned ideological scores for each article. We then estimated each reader’s ideological position by examining which items she clicked to read and which she also clicked the “iine” button for five of the ten issues. We classified readers who made ideologically consistent choices to either the “left” or “right” and others to the “center.” We also assigned ideological scores to each of the “left” and “right” readers.

Next, we analyzed how readers encountered and reacted to ideologically diverse news content, focusing on articles related to the remaining five contentious issues. As expected, we found that ideological readers were more likely to click on articles ideologically consistent with their own views. They were also more likely to hit the “iine” button on those articles. These findings are in line with selective exposure theories. More surprising, though, was that ideological readers also tended to click on articles written from an ideologically opposing viewpoint. The ratio of clicks for ideologically opposite articles to ideologically consistent articles ranged from 28%–41%—higher than suggested by recent research on selective exposure. Furthermore, perhaps due to the high ratio of clicks for ideologically opposite articles, about 2%–4% of ideological readers hit those articles’ “iine” button, which implies some sort of a persuasive effect. We also found, analogous to some previous studies, asymmetrical behavioral patterns between the “left” and “right.”

We suspect that features unique to Gunosy may have led to the high clicking ratio for ideologically opposite articles and a smaller-than-expected selective exposure effect. A contemporary media environment where cable TV, the internet, and social media are widely available offers people more media choices. Such an environment is usually regarded as giving rise to selective exposure because one can focus on the media that is ideologically consistent with one’s own. News curation apps such as Gunosy, however, display ideologically diverse news on a single screen. Readers can access articles that diverge from or are even counter to their personal beliefs at very low cost. These features of news curation apps, we believe, allow researchers to sort out the truly cognitive portion of selective exposure because one can access views on both sides of a contentious issue at equally low cost. If we can precisely estimate the cognitive portion of selective exposure through analysis of news curation apps, we can also more precisely estimate increased selective exposure due to other factors, such as the contemporary media environment. There are certainly other possible reasons for the high clicking rate. For example, since Gunosy readers of political news are likely to be more interested in politics than the standard reader, they are more likely to be motivated to learn the opposing viewpoint.

This paper is an initial attempt to analyze readers’ behavior on a news curation platform. Despite limitations, we believe the novel and realistic setting of a news curation app and the vast amount of data made available by Gunosy allow us to add genuine insights to previous studies of media effects.

Authors

• Sota Kato, International University of Japan

• Takafumi Nakanishi, Musashino University

• Yoshifumi Seki, Gunosy Inc

• Budrul Ahsan, Philips Japan

• Shun Ibaragi, The Tokyo Foundation for Policy Research