Nothing on the internet quite compares withFacebook Newsfeed, in terms of time spent. That's not just because it happens to be at the centre of the homepage of a site that has one billion users. Newsfeed is actually designed to be a very engaging product, which is why Facebook (FB) users end up spending so much time on it — checking out the status of their friends, clicking the 'like' option, adding their own comments.
When Jocelyn Goldfein joined Facebook asdirector of engineering two-and-a-half years ago, Newsfeed was already the most popular product on the internet. Her first mandate: change it.
"If we don't disrupt ourselves, someone else will disrupt us," she says. "We have to constantly improve. If today's Facebook was the same as the 2007 Facebook, we'd be dead."
Goldfein's is a thankless job. After spending hundreds of thousands of manhours rewriting the algorithm codes for Newsfeed, trying to make it better, she's usually greeted with a chorus of disapproval from users.
"They invariably say 'bring back the old Facebook' everytime we make a change. People respond that way because they have an emotional investment in the product. And it's because we favour high-risk high-reward options that make a big difference."
When Goldfein and her engineers began work their Newsfeed improvement project, users viewed news from their friends in chronological order, with the most recent postings appearing first. The problems with this were many.
Most users had a fairly large number of friends and new users tended to expand their networks rapidly, leading to exponential growth in the number of new feeds. FB itself had no problem handling the volumes, but most users did not have the attention span or the time to go through all the news, even if they logged on everyday.
As a result, news of what FB calls Major Life Events (MLEs) — graduation day, newborn babies — often got buried amid spam, which includes application and games feeds. And then there were postings which generated animated discussion, whose value increased with time, where chronological ordering was counter-productive. "Our goad was to reduce noise and find the best stuff for the Newsfeed" says Goldfein.
"It was about high quality stories versus low quality stories. We also viewed it as an issue of determinism versus magic, fresh versus stale and consumption versus interaction."
In other words, FB needed an algorithm where the top stories would be those that made for interaction in the form of comments and discussion. It wanted fresh stories at the top — but not on the basis of deterministic (read chronological) order.
Instead, it wanted magic — an algorithm that would find the most interesting stories from among hundreds of stories. It wouldn't be an exaggeration to say the task is at the frontiers of computer science, since it uses artificial intelligence derived from the huge amounts of data FB has on user behaviour.
It's also at the core of the company's philosophy as articulated by founder Mark Zuckerberg, who believes the amount of information people share will double with each passing year. That makes for a huge amount of Newsfeed in the future. And though FB has huge amounts of data on how users behave on the site, it still has to mine this information for insights on what makes for higher levels of engagement.
"The product development culture at FB is based on experimentation," says Goldfein. "The Googlealgorithm has one right answer, ours doesn't. The quality of Newsfeed depends on the users and their friends, who create the content. Our way is to experiment and quickly put out prototypes which are used by our own people so we get a sense of what might work. We went through three prototypes for Newsfeed before we launched the new version."
Of clusters & boulders
The first prototype was based on the idea of clusters, which meant grouping together friends talking about the same thing. The most obvious clusters would be related to topics like birthdays and the weather, which weren't interesting enough to appear at the top of the feed.
It was found that interesting clusters don't develop as easily as one might presume. The idea did reduce some of the noise but it failed to create any magic at the top of the Newsfeed.
The second prototype was to model the problem as a newspaper. After all, the product is called a Newsfeed and internally, FB refers to those posting status reports as 'publishers.'
The focus of this prototype was freshness and the magic lay in picking a MLE from the most recent feeds and turning into a veritable headline. Alas, as any newspaper editor might tell you, picking a single headline story is never easy. There are slow news days when nothing seems to be happening and the prototype would end up projecting banal news as headlines stories. And it too often happened that it failed to recognise the best story of the day and project it as the headline.
"We didn't have the process at the algorithm level," says Goldfein. "Maybe we should have brought in a journalist." The third prototype leveraged the newspaper idea but avoided its pitfalls and this is the one being used by the FB Newsfeed today. Called the 'boulder' prototype, it picks out the big stories and assigns them boulder status, with other stories showing up as pebbles.
"Today, if you log on after a week, you get all the important stories since you last logged in. If you hang out on FB, you'll see changed feeds. It has the magic element because you see the best stories first, but it feels deterministic, which is something users like," says Goldfein, who has a degree in computer science from Stanford University.
Even as her team worked on improving the Newsfeed, another team of engineers at FB was working on Ticker, which now occupies space on the upper right side of the home page. The ticker has left less space for advertising on the FB home page and Goldfein uses it as a case in point for Zuckerberg's philosophy.
"It helps to have a leader who is also the company's founder. We're always thinking, what would Mark do in this situation? We know he believes that growth in usage is of prime importance. Anything that is bad for users is bad for advertisers, so there's really no trade-off between users and advertisers. Anything that increases usage will eventually monetize. The culture hasn't changed since the company went public."
With 4,000 employees, FB is still a relatively small operation, given the size of its influence. The company has just opened an engineering office in London, its first outside of the USA, but Goldfein says there are no plans to expand into India, though the company employs lots of Indian engineers in its existing facilities.
"FB has some very talented people, capable of solving some really complicated problems, which is why we have managed to scale up as we have," she says. Goldfein has relatives in India from her husband's side and though she has no plans to set up a software centre in India, she is very interested in the market.
"India has more FB users on mobiles than on desktops, which makes it a special market. Mobile users are different and we are developing products with them in mind. As Mark likes to say, we're only 1% done. There are just so many greenfield oppourtunities in this field, it's going to be good for everybody."
No comments:
Post a Comment