Netflix Recommendations and Page Rank
Netflix is one of the most popular tv and movies streaming services and has one of the highest active users. Over the past decade, Netflix has curated its platform to constantly provide quality content, and more importantly tailored recommendations to each of their users. In fact, “80% of stream time is achieved through Netflix’s recommender system” as they seek to increase their user retention rate.
The way Netflix currently recommends content is through a “row-based ranking system” where each row is categorized by theme, and the strongest recommendations will be at the top, and left of the screen. They use a couple different algorithms to determine the content provided in each of these rows. The first is Personalized Video Ranking which filters the catalog by a specific genre. The second is a Top-N Video Ranker which only looks at the head of the catalog rankings. A Trending Now Ranker that is based on seasonal trends like holidays, or one time events like a pandemic that could lead to a rise in documentaries about a certain topic. Another is the Continue Watching Ranker which looks at content that the user has watched but not finished and ranks the probability that the user will finish the item based on various factors. The last one is the Video-Video Similarity Ranker which recommends similar videos based on the user’s history.
Netflix uses a “machine learning approach” where they try to create a scoring function by training an algorithm using past information from their users about homepages that they have created and how users interacted with those pages. Netflix asks new members to fill out a survey so that they can start with a baseline for recommendations. If this step is skipped, then Netflix displays a broad selection of popular items.
In class we learned about Page Rank, and how those rankings are determined based on links between hubs and authorities. However, Netflix’s ranking algorithm seems to be more complex than that because movies or series do not have direct links to other content, so they use multiple processes to develop robust recommendation systems. They need to analyze a multitude of factors rather than just clicks like watch time, genre, current events, and user history. It is interesting to note how based on the service that is being provided to users, the recommendation systems and algorithms associated with those models are different. The models that Netflix has developed has allowed them to become the go-to streaming service for most people worldwide. They continue to deliver quality content for their users, and now that they also fund content of their own, known as “Netflix Originals,” they are using all the vast data that they collect to influence the topics and genres that those items are based on.
Sources:
https://towardsdatascience.com/deep-dive-into-netflixs-recommender-system-341806ae3b48