Recommendation Trends and Echo Chambers in the Korean COVID-19 YouTube Video Network

Abstract

During the initial spread of COVID, there was a frenzied release of conflicting information about the virus quickly spread by news sources and social media, including on YouTube. YouTube has been a main source of information sharing amongst Koreans with regard to the pandemic, but studies have found a substantial amount of misinformation on the platform. At the same time, social media platforms, along with YouTube, have been faulted for creating echo chambers as their algorithms feed into user confirmation bias. The risk of misinformation is heightened when shared in echo chambers as users are less discerning of information quality within echo chambers. There have been academic studies on Covid misinformation spread on social media, but most have focused on other platforms like Twitter or Facebook.

In my thesis, conduct network analysis to study the impact of YouTube’s recommendation system on the Korean Covid-19 video network as well as the existence of echo chambers in the network. I constructed the video network by querying the Covid-19 keyword for videos and then conducting a two-step crawl to query the related videos. I found that the largest strongly connected component contains videos relevant to the study. I then studied the interaction patterns between clusters of videos with the same channel types. I found that news and independent health channels were most prevalent in the SCC. However, the recommendation system seemed to favor the independent health channels through frequent recommendations. Looking at engagement showed that view and comment count of each channel type correlated with the size, and like count was more aligned with recommendation patterns. I also identified 50 clusters in the network using the page rank clustering algorithm. Of these 50 clusters, I identified 4 potential echo chambers. Of these four, one cluster had frequent commenters that did not engage in other topics and uniformity in the topic.