Mitigating social polarization in video sharing platform using unbiased recommendation system: A case study of South Korea political youtube channels

In a digital age marked by surging polarization and the rise of echo chambers on video-sharing platforms, the imperative for unbiased, diverse content recommendation systems has never been more apparent. This study aims to propose a novel approach for building an unbiased recommendation system on a video-sharing platform to address the issue of escalating polarization and biased content. Our novel system targets the heart of social polarization, aiming to provide users with balanced exposure to a spectrum of perspectives on contentious issues. By integrating advanced techniques for polarity measurement, harnessing user activity, and leveraging video characteristics, we have constructed an innovative unbiased recommendation system. Commencing with seed channels, we collect sample data to establish user activity models and gauge the bias of each channel. The comprehensive feature set we utilize includes video titles, hashtags, posting times, comment distribution, and active users, each meticulously embedded to ensure the utmost precision in addressing user preferences. The output of the model is designed to provide users seeking relevant videos for social events without polarization bias. We use political polarization as a representative case, amassing data from Korean Political YouTube channels via our TubePlunger system. Beginning with six seed channels and approximately one million users, we analyzed more than six million comments on 24,000 videos to create an expansive dataset, eventually encompassing over 50 other relevant channels. More than 80% of survey participants agree that our system offers a means to achieve information balance across the political spectrum, in contrast to YouTube recommendations. They view TubePlunger as a valuable solution for addressing media-related social polarization issues. This proposed recommendation system approach has the potential to encourage a more equitable and informed political discourse, curbing polarization and enhancing democratic engagement in the online realm. This research signifies a pivotal stride in the realm of unbiased recommendation systems, transcending politics to extend its influence across diverse fields, providing balanced perspectives and bridging divides in a polarized digital world.