Task is to analyze an existing recommender system by
-Perform a Scenario Design analysis -Attempt to reverse engineer what you can about the site, from the site interface and any available information that you can find on the Internet or elsewhere -Include specific recommendations about how to improve the site’s recommendation capabilities going forward
YouTube is the world’s largest platform for creating, sharing and watching
videos.YouTube’s recommender systems are run by Google Brain that was later opensourced by Google as TensorFlow. They approached the problem using deep learning, where the idea is that non-linear combinations of features may bring a better prediction than a traditional matrix factorization system can.
Beyond gender, YouTube has an age divide. It is mainstream enough to have a broad demographic reach. Even the older people watch video, although not to the degree of millennials and GenX. Overall, YouTube is the second most clicked webpage in US.
To find a videos of their interest at the moment like favorite song, movie clip or TV show, also to find a tutorial videos for any given problem they experiencing
YouTube can help their users accomplish their goals by: - creating an easy to navigate interface where a customer can search and view videos easly and without downloading the content - following users search history in order to recreate and add additional videos of similar content and high approval rating from likes clicks - asking for user experience approval after the search or rather tracking it what did they like or did not like.
YouTube is using a Deep Learning approach towards Recommender Systems focused on two networks: - the candidate generation network which is using the viewers’ watching history and find some applicable videos to the user
As I mentioned above, the ranking network doesn’t always shows relevant recommendations of the videos, which can be sometimes frustrating especially when there is no history of the user available to the recommender system, then basically the recommendations are a lost to candidates, who is not interested in the high approval rating videos rather finding the most helpful video.
I would rather focus on fining relevant and the highest ranked videos to the user whose history is not relevant and not existing to the current search. Some recommendation can be really off and left user without any great choice.