Unveiling Bias in YouTube Shorts: Analyzing Thumbnail Recommendations and Topic Dynamics

In the dynamic landscape of digital media, YouTube Shorts have emerged as a popular format, captivating users with their brief and engaging content. However, the recommendation algorithms driving these videos often exhibit biases that influence which thumbnails are prominently displayed. This study delves into the biases present in YouTube’s recommendation algorithms, focusing on the thumbnails of YouTube Shorts, which play a crucial role in attracting viewers. Thumbnails, as powerful visual elements, significantly impact user decisions and engagement. By utilizing advanced topic modeling and content generation techniques, we analyzed a substantial dataset of YouTube Shorts’ thumbnails. Our analysis, employing generative AI and BERTopic models, reveals notable shifts in topic distribution across recommendation cycles, it highlights a preference for certain types of content. These biases not only affect content visibility but also steer user engagement towards popular, yet potentially less diverse, topics. The findings of this study enhance the understanding of algorithmic biases in digital platforms and aim to promote more equitable and transparent content recommendation practices.