Every Saturday we present you a company and its way of using data. After having done an article on two professional experts of recommendation systems: Tik Tok and Netflix, let's come back today on the second search engine in the world and which is not left behind on recommendation systems: YouTube

Quick overview of YouTube

YouTube is a website, then an application, which allows to create and view videos. The company was created in 2005, before being bought a little more than a year later, in 2006, by Google. 

In 2020, YouTube passed the 2 billion monthly user mark, meaning that more than a quarter of the world's population uses YouTube every month! 

Youtube allows content creators to get closer to users and monetize their videos. On the users' side, they can discover quality content in all sectors: cooking, entertainment, DIY, ... The possibilities are endless. 

Data usage at YouTube

The different types of data collected by YouTube

YouTube collects 2 main types of data: 

• Videos and their metadata: title, description, tag and thumbnail, which will allow us to understand the video and use it in the recommendation algorithms. By looking at the diversity of colors in a video, for example, we can calculate indicators of video quality. Of course the different interactions (like, share, comments, ...) will play on the virality of a video.
• User behavior data: this is what YouTube uses to build its various recommendation algorithms 

The different recommendation algorithms of YouTube

YouTube offers different levels of recommendation algorithms for its users: 

• On the home page, with different blocks, mixing videos related to channels you are subscribed to and videos affinities with you. Since more recently, the platform will also offer you shorts
• In a video, on the right if you are on a computer and below if you are on mobile, YouTube will mix affinity content with the video and affinity content with you 
• At the end of the video, YouTube will also offer you videos
• In the search results of its search engine, YouTube will offer neutral and personalized results. 
• YouTube also allows its creators to organize videos in different ways, including through playlists

The keys to a successful video

YouTube takes into account various signals to decide whether or not to highlight videos: 

• The average percentage for which the video is watched
• Likes and dislikes 
• Geographic location: YouTube will suggest videos that appeal to a group of people based on the location - e.g. country - of the people who liked them
• Previous videos seen by Internet users
• Channel activity: YouTube will recommend active channels
• Video engagement: likes, comments and shares, compared to the number of views

Here, for the data experts, is the algorithm that YouTube presented in 2016 and which has surely evolved a lot since then: 

YouTube said in 2021 that most views came primarily from its home page and recommendations on the side or bottom depending on the device used. 

YouTube is the second largest search engine in the world

When we talk about search engines, we often think of Google, Bing, Yahoo, Ecosia or counterparts used abroad like Yandex and Baidu. And yet, YouTube is the second most used search engine. A significant portion of Internet users use YouTube's powerful search engine to get information. Rather than reading, some Internet users prefer to watch a video to learn about a subject and therefore go directly to the king of video, YouTube, to find this information. 

YouTube is therefore a video platform, leader in its field, which uses data to categorize its videos and offer them in the most relevant way possible to its users. 

‍