August 9, 2013
Netflix devotes significant resources in order to develop its rating and recommendation systems, which is a key component of the service. The company employs 800 engineers to operate the service. Netflix estimates that 75 percent of viewing activity is now recommendation driven. The company uses several types of customer data in order to create the personalized recommendations, and uses behavior of similar users to suggest preferences.
In March, Netflix shipped four billion DVDs, while during the first quarter of 2013, it streamed more than four billion hours, reports Wired. This summer, Netflix released a profile feature that allows family members to have separate preferences with individual queues.
Netflix’s recommendation system examines metadata and finds similarities among shows, such as ratings, year produced and director, said Carlos Gomez-Uribe, VP of product innovation and personalization algorithms. Netflix also uses customer behavior, such as browsing, playing and searching.
The company has more than 40 staff members that hand-tag TV shows and movies. Most are freelancers and are trained to be objective as they catalog video content.
Ratings have less impact in streaming since users can view a program and stop it if they choose, according to Xavier Amatriain, engineering director. However, DVD relies more on ratings, since there is more of a time investment by the customer.
Netflix monitors what customers play, search for and rate — as well as when and on what device, explains Amatriain. “We even track user interactions such as browsing or scrolling behavior. All that data is fed into several algorithms, each optimized for a different purpose.”
The company has been working on introducing context into the recommendation system, explains Amatriain. “We have data that suggests there is different viewing behavior depending on the day of the week, the time of day, the device, and sometimes even the location. But implementing contextual recommendations has practical challenges that we are currently working on.”
Placement of title selections also matter. “The closer to the first position on a row a title is, the more likely it will get played,” suggests Gomez-Uribe. “The higher up on the page a row is, the more likely it is to generate a play.”