The recommendations of different platforms such as Spotify, Netflix, or even Amazon have a little something magical: the platform knows you so well that rather than embarrassing you with content that you would certainly not like, you are entitled to a preselection of content… that will keep you captive on the site.
These recommendations work more or less well – and require more or less conscious and deliberate user intervention. YouTube, on the other hand, seems to be based on both your video search history, your “Likes” and the presses of the “Dislike” and “Not interested” buttons.
Mozilla study criticizes YouTube’s recommendation algorithm
However, as a recent study by Mozilla shows, the buttons “I don’t like”, “Not interested”, and even “Do not recommend the channel” seem to have a strangely limited effect on the following recommendations that will be offered to the Internet user. .
Based on data from 20,000 users, the foundation’s researchers explain that in the best of cases, the algorithm continues to present the user with recommendations similar to channels and content that they are not interested in – in more than half of the recommendations.
In detail, pressing “dislike” reduces unwanted recommendations by 12% while “not interested” only reduces the percentage by 11%. “Do not recommend the channel” is by comparison much more effective, since it makes it possible to reduce on average 43% of bad recommendations.
Removing a result from the history reduces the number of unwanted recommendations by 29%. According to the researchers, the remedies offered by YouTube to improve the recommendations remain insufficient at this stage:
“Youtuve should respect the feedback users give about their experience, and treat them as meaningful signals about how people want to spend their time on the platform”can we read in the study.
Apparently this is not a bug
Questioned by The Verge YouTube explains that this behavior is intentional and criticizes the conclusions of the researchers:
“Our controls do not completely block entire topics or viewpoints as this could negatively affect users. We support academic research on our platform. That’s why we recently enhanced our data access API through our YouTube Researcher Program. Mozilla’s study doesn’t take into account how our systems work, so it’s hard for us to get a lot of insights from it.”explains Elena Hernandez, spokesperson for Google.
Mozilla indeed draws its figures from a source independent of YouTube. The foundation asked volunteers (20,000 in total) to use the Mozilla RegretsReporter extension, which superimposes a general “Do not recommend” button on videos viewed by participants.
The sample was also, behind the scenes, categorized into groups based on their use of the ‘dislike’, ‘not interested’, ‘don’t recommend channel’, ‘remove from history’ buttons. The data of the participants who used these options the most was compared to that of a control group who used youtube without going through this type of interaction.