Train the YouTube algorithm to receive better recommendations on videos is a common practice of many users. Deleting a video from your history, using the “Dislike” button, or requesting that a channel not be recommended would all serve this purpose. Curiously, none of these actions is as effective as it seemsaccording to a Mozilla study.
Does this button work? it is Mozilla’s latest research on the efficiency of YouTube controls to train its recommendation algorithm. The foundation collected data from more than 22,000 users who use various methods to monitor the video recommendations they receive on the platform. The result shows that the “Dislike” button, as well as other options offered by YouTube to improve the experience, are ineffective.
Using more than 500 million recommended videos, the researchers found that YouTube’s algorithm suggested content that the user did not want to see. Among the alternatives that people have to control their experience, the “I don’t like” button and the “I’m not interested” option are the least reliable. Deleting videos from history or requesting YouTube not to recommend the channel are more effective in filtering unwanted videos, however, they are not foolproof.
In our analysis of the data, we determined that YouTube’s user control mechanisms are inadequate as tools to prevent spam. […] In general, a consistent theme is that some of these tools have little effect on improving recommendations, but are inadequate as tools to exert meaningful control.
According to Mozilla, the option “Do not recommend channel” has a positive impact in similar videos on other channels. However, this feature does not prevent the algorithm from suggesting a video from the channel that you explicitly stated that you were not interested in. The researchers monitored behavior for a week and found that unwanted videos leaked into recommendations.
YouTube personalization controls are ineffective
Another interesting finding is that these controls are more effective on the YouTube home page than in the sidebar that appears next to the video being played. Mozilla also discovered that there is a drop in referral rates over time after a video is rejected. The researchers concluded that the tools are confusing and there is no information to “train” the algorithm.
Youtube mentions on their help page that activity on the platform, as well as on Google and Chrome, can influence recommendations.
According to YouTube, if you receive recommendations on topics you are not interested in, you can delete the videos you watched on that topic from your history, or remove them from searches. The site suggests that you use the “I’m not interested” function as a mechanism to personalize recommendations.
Training the algorithm is a myth
After the study was published, Elena Hernández, a YouTube spokeswoman, criticized Mozilla for not considering how customization controls work. The “Not Interested” and “Don’t Recommend Channel” features only target the video and channel in question, not a particular topic. Hernandez told TheVerge that YouTube does not seek to stop recommendations of content related to a topic, opinion or issuer.
The Mozilla study uncovered the rituals that users follow to cure the home page on YouTube. While some watch videos in a private window, others pause playback history, clear cookies, or log out of their account. The truth is that neither has the right effect, since training the algorithm is as effective as closing open applications on the iPhone “to save memory”.