“YouTube's recommendation algorithm does not 'push' specific content. It 'pulls' videos that best match the viewer's interests.”
Todd Beaupré, Senior Director leading algorithm optimization at YouTube, explained how the recommendation algorithm operates. He noted that YouTube does not prioritize showing certain types of content. Instead, it uses over 80 billion data points, such as views, watch time, 'likes,' and 'dislikes,' to provide personalized videos for viewers. Regarding concerns that YouTube's recommendation algorithm could create a 'filter bubble' by repeatedly showing specific information, he stated that it does not filter or reinforce specific political viewpoints. He added that in the case of news or information content, it emphasizes videos from reliable sources, helping viewers access balanced information.
Director Beaupré has led the development of recommendation algorithm technology across numerous corporations in Silicon Valley for 40 years, making his career closely intertwined with the history of this technology.
How does YouTube's recommendation algorithm work?
“YouTube's recommendation system aims to find videos that viewers like and find valuable. Think of YouTube as a vast video repository. Here, the algorithm analyzes billions of videos rapidly and recommends the optimal videos for each viewer. This process starts from the fact that every individual's viewing habits are unique. In other words, YouTube's recommendation system does not rely on a fixed 'recipe book.' It utilizes artificial intelligence (AI) and deep learning technology to learn over 80 billion 'signals' every day and evolves. It comprehensively analyzes user reactions such as views, watch time, 'likes,' 'dislikes,' and 'shares' to gauge viewer satisfaction. Recently, it has been further refined by integrating large language models (LLM) and AI technology. Through this, the algorithm is optimized to better understand viewer interests and recommend the latest videos at more appropriate times. YouTube thus provides a 'valuable' viewing experience, helping viewers find content they will be satisfied with in the long run.”
Recently in Korea, concerns about filter bubbles are growing due to the potential for algorithms to reinforce political bias.
“First, YouTube's community guidelines apply equally to all viewers. They are consistently enforced regardless of political views. Additionally, the recommendation system is designed not to recommend or filter content based on specific political viewpoints. In news and information content, where the quality and context of information are critical, YouTube emphasizes content from reliable sources. As a result, a variety of news content spanning the political spectrum is provided to viewers from established news outlets to independent news creators. In other words, YouTube's recommendation system helps viewers access news content from trustworthy sources and provides information that reflects various scales and perspectives.”
Is there a possibility that YouTube's recommendation algorithm will prioritize certain types of content?
“YouTube's recommendation system does not push specific content. It operates by pulling videos that match the viewer's interests. That is, it reflects the preferences of others whose previous viewing history and tastes are similar to the viewer's. YouTube continuously improves its algorithm to ensure it reaches the appropriate audience, regardless of content format, length, or topic. The goal is to help creators (video producers) meet more viewers.”
There is also a possibility that the videos recommended by the algorithm may provide false information. It is important to ensure content reliability.
“For topics like breaking news or complex scientific research, quality and context of information are of utmost importance. Therefore, YouTube emphasizes information from trustworthy sources to help people find more accurate and useful information. For example, accurate facts are also important regarding health-related knowledge. YouTube has introduced a so-called 'health information panel' to show the sources of videos, helping identify trustworthy sources for health topics. Additionally, a 'health content section' was introduced to highlight these reliable sources' videos when searching for specific health topics.”
Since the algorithm refers to viewer data, privacy protection must also be important.
“Of course. The parent company, Google, understands that users provide information based on trust and is deeply aware of the responsibility that comes with it. Therefore, it is doing its best to enhance privacy protection and user control over their information. Google utilizes collected information to provide personalized recommendations and search results but does not sell personal information to anyone. Additionally, YouTube supports users in managing how their data is stored and used via the 'Your Data' page on the platform.”
If someone wants to prevent their video viewing history from being stored.
“YouTube has improved the 'Home feed' environment for users who wish to delete their viewing history or do not want it saved at all. For instance, if users stop using viewing history or have no significant previous viewing history, the YouTube Home feed feature that recommends videos based on viewing history will be disabled. Thus, when accessing YouTube, the search bar and left navigation menu remain unchanged, but recommended videos are not displayed. This way, users who prefer searching rather than exploring recommended content can use YouTube in a more simplified environment.”
Are there protective measures in place for young viewers?
“Absolutely. YouTube has implemented various protective measures to improve the algorithmic experience for young viewers. Experts note that if young people are repeatedly exposed to messages that ideally depict certain standards, they are likely to form more negative self-perceptions than adults. Reflecting this, YouTube is strengthening safety measures while supporting young people in freely exploring topics they are interested in.
Additionally, in collaboration with experts, YouTube is identifying categories of content that may not present issues when viewed as a single video, but could have negative effects on young people when viewed repeatedly. For instance, content that idealizes certain physical conditions or promotes social aggression falls into this category. YouTube is training its system based on data to limit the repetition of such content recommendations for young viewers. Recently, it has expanded the categories of content to which these limitations apply.
Plus point
How to reset the algorithm when the recommended videos are not to your liking
If the videos recommended by YouTube do not match your interests, you can adjust the algorithm to receive new personalized content. The simplest way is to delete your history. To reduce the frequency of recommendations on a specific topic, you can delete the viewing or search history for that topic. You can selectively delete records only after a specific date. It is also possible to temporarily stop using your viewing and search history. For example, if a student repeatedly searches for an uninteresting topic for school assignments, they can adjust this to prevent unnecessary influence on the recommendation algorithm.
To reduce unwanted videos in the Home feed, you need to provide feedback on individual videos. By clicking 'More' (the three-dot icon) on the video and selecting 'Not interested,' the recommendation frequency for that type of content will decrease. If you do not want recommendations from a specific channel, you can select 'Don't recommend channel.' Additionally, since YouTube recommends personalized videos based on your activity history within the Google app, you can more effectively reset the recommendation system by going to the 'Your Data on YouTube' tab in the account settings and turning off web and app activity.
Beaupré director explained, “YouTube allows you to manage your viewing and search history at any time.”