Netflix has made it a goal to cater to streaming subscribers with content recommendations. But how does the company arrive at its conclusions? According to Todd Yellin, Netflix vice president of product innovation, it isn't magic. And it isn't perfect, either.
Speaking to The Financial Post, Yellin outlined the recommendation process employed by the streaming giant to suggest films and TV shows to its 20 million+ subscribers.
Each piece of content in the Netflix library is marked with more than 100 tags. The company employs a team of about 25 freelancers to watch and tag video. Most live in L.A. and do script coverages -- a type of pre-reading -- for studios. Others are moms who want to work part time tagging children's films.
Netflix tried employing similar experts to watch movies and intuitively assign related titles. But mathematics beat out human expertise.
Yellin explained that such a method "fail[ed] miserably," leaving the company to instead rely on "metadata and clustering." Part of that info is based on answers new subscribers provide to 22 questions they're asked when they create an account. Other data plugged into the company's recommendation algorithm is determined based on the content they actually watch. When combined, the expectation is that Netflix recommends apt titles for each individual customers -- whether they love artsy-fartsy foreign films or mindless shoot 'em ups. (via Hacking Netflix)
Has your experience with Netflix's recommendations been helpful or problematic? Let us know in the comment section.
