YouTube users go to the website for videos of all kind (e.g., music, education, etc).
YouTube targets everyone globally with access to the internet and an interest in a video.
The goal is to enable a user to 1) post a video and/or 2) watch a video. Ideally, the process of finding a desired video will be seamless, and similar videos will be recommended to keep the user on the website (this helps increase YouTube’s business profitability).
YouTube helps its users accomplish their goals by 1) allowing a user to subscribe to a particular publisher/channel to get notifications of new uploaded videos; 2) recommending videos that might be relevant based on the latest video seen; 3) combining similar music videos together into a playlist to avoid having to manually select the next song/video
It makes perfect sense to have scenario design twice given that a user can be either 1) someone looking to publish a video, or 2) a user looking for a specific video.
Google open-sourced the technology behind YouTube’s recommendation system, which is based on TensorFlow using neural networks. At a simplistic level, there are two neural networks - one for videos uploaded/searched/viewed and another for ranking (a user can rank a video as good or bad). While complex, the process has aspects common to other recommendation systems such as, normalizing data, weighting features, choosing a classification or regression model, etc.
(source: http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45530.pdf)
YouTube currently doesn’t distinguish between a video officially uploaded by an organization and a video from another independent user. For example, sometimes music videos appear in search results where the video is a recording from an individual’s cell phone. Another example is an educational video from John/Jane Doe versus one from MIT/Stanford/etc. It would be nice if YouTube gave more weight in the ranking of “official” videos uploaded by organizations.