Netflix’s New ‘My Profile’ Feature Means Fewer Outrageously Bad Recommendations

Netflix’s New ‘My Profile’ Feature Means Fewer Outrageously Bad Recommendations

By Aaron Pressman | The Exchange – 14 hours ago

Netflix (NFLX) subscribers get annoyed when their kids flip on the TV and see a suggested list of raunchy comedies alongside rows of more-appropriate fare. Or when the suggestions mix PBS’s pre-school cartoon “Caillou” with AMC’s drug-dealer saga “Breaking Bad.” But don’t worry – it drives Todd Yellin, Netflix’s vice president of product innovation, crazy, too. The mix-ups happen when Netflix’s automated algorithms attempt to analyze the viewing patterns of a whole household of people watching different kinds of shows through one account. Mom and Dad’s late-night “Scandal” binge gets thrown in with a tween’s “Cheetah Girls” fixation. The online programming service’s new profiles feature should help end the confusion, Yellin says. Starting this month, U.S. subscribers can set up different profiles for each viewer. Remember to hit the icon for the proper person before selecting a show, and the suggestions algorithms will suddenly seem much smarter.Better personalization

“It’s going to make it a lot easier for use to personalize recommendations,” Yellin says. “People do have different tastes but probably not as extreme as being huge fans of ‘Caillou’ and ‘Breaking Bad.’”

Bad suggestions aren’t just annoying – they also affect Netflix’s customer retention and subscriber growth rates, the company has found. Netflix’s stock price tripled this year in large part because of fast growth in its online streaming service. The company counted 38 million members at the end of June, up from 28 million a year earlier. Improving suggestions should help keep up that pace of growth.

That’s why a large team among the 800 engineers working at Netflix focuses on improving the service’s personalization features. Back in 2007, the company even paid a team of outside programmers a $1 million prize for coming up with an improved algorithm. With a few tweaks, the formulas developed by the outsiders were incorporated into the Netflix service, where they remain today.

Useful suggestions are also increasingly important as the service morphs from its DVD mailer roots to more of a premium cable channel model similar to HBO or Showtime.

That’s because Netflix is spending less of its $2 billion annual content budget to maintain a huge catalog of classic, well-known movies and TV shows – the kind of programs people know they want to watch before they turn on the television. Increasingly, Netflix is focusing on acquiring exclusive and original fare, shows such as “Orange is the New Black,” which viewers may not even have heard of previously.

The suggestions engine, which spits out recommended shows in milliseconds, works from two different sets of data. On one side is historical data about which shows people who watched one program also watched. That’s similar to the method popularized for shoppers on Amazon.com: People who bought this also bought that.

Another stream of data comes from a team of human reviewers who tags every show and movie to highlight different characteristics. Reviewers fill out a lengthy questionnaire about each show, choosing, for example, whether it has an upbeat or downbeat ending, revolves around a main character or an ensemble cast or takes place in the past, present or future.

So if a viewer watches “Breaking Bad,” the traditional suggestion formula might offer that other people who liked that show also liked “Mad Men.” But the tag-based formula might notice that the show’s violent crimes and dark style are similar to the Netflix original show “Hemlock Grove.”

The best blend

When Netflix pitted the two methods against each other in a test with subscribers, the tag-based suggestions provided viewers with more shows which they ending up watching. But a blending of the two methods worked best of all.

Netflix is also experimenting with new ways to offer suggestions. In one recent test, a talking virtual personality, dubbed Max, quizzes viewers about what kind of show they’d like to watch next and how they’d rate a handful of related choices. The “Max” experience acts sort of like a playful game show host as it prompts viewers to rate and select new shows.

So far the voice interface only works one way – viewers still enter their choices onscreen with a remote or other controller. But, eventually, as voice recognition technology becomes more common, a future version of Max could respond to answers shouted back by viewers.

Netflix subscriber Anthony Berry was one of the customers included in the “Max” test via his PlayStation 3. He was interested at first but ultimately felt the feature might be more appealing to kids. “It ultimately wasn’t something I could see myself using regularly,” Berry says. “I usually just go on word of mouth recommendations.”

A new feature that has graduated from the testing phase applies the suggestion formulas in a new way, setting the order of shows that a subscriber has added to their instant viewing queue. Netflix now calls the user-selected queue “My List.”

The metaphor of a queue made sense when Netflix had to mail DVDs to subscribers, Yellin explained. But with online streaming, where every show is just a click away, the ordering can be more flexible.

Users who prefer to remain in charge of the order of their lists can switch back to manual under settings.

“You can override it to make it your own order,” says Yellin. “ But we’ve found that users will want to go with our order.”

And that also suites Netflix’s goals right now.