A View from Rachel Metz
Live Updates From Facebook's Graph Search Announcement
Mark Zuckerberg demonstrated a search service that taps into the information shared by a Facebook user’s friends.
Here are the takeaways from today’s announcement:
- Facebook’s Graph Search will provide a way to sort through all the information shared by those in your social network using natural language queries.
- Based on company demos, Graph Search looks like a promising way to find products or services you might like, and to identify local businesses you might want to visit.
- Facebook has partnered with Microsoft’s Bing to deliver results from the Web when Graph Search can’t find something.
- Zuckerberg says he’d love to work with Google but Microsoft is more willing to work on its terms.
- The service will be rolled out to a few hundred thousand people initially.
- Facebook plans to broaden it to non-English languages and to other platforms such as mobile, but did not provide a timeline.
2:14 PM With that we’re done with the formal presentation, and I’m off to check out Graph Search at the demo stations! Thanks for reading!
2:13 PM: Sixteenth question: how will businesses be able to use Graph Search?
Tom says most business page owners focus on distribution through News Feed. This is a new context where their information can exist. So it’s more important to make sure the info on your business page is up to date. Your tag is the right type for your restaurant, etc.
Seventeenth question: will Instagram data be in there?
Mark says “That should be on the list of things we will hopefully one day get to.” He reiterates that this is a beta product and they know they didn’t get to everything. That’s because it’s really hard and takes years and years.
2:10 PM Fourteenth question: have you tested this with people outside of Facebook?
Tom says they bring users in and let them try it out, and that’s what led to a lot of the changes they made over time. Mark says user tests were “very positive”. They didn’t want to come up with a specific grammar people had to learn. You can just type things like “friends New York” to get friends who live in New York. Someone notes that it doesn’t have to be as structured as the way they phrased things during the demos.
Fifteenth question: in sorting places, are you relying on “likes”?
Lars says when you see a set of results in Graph Search it’s sorted by a number of features–who has liked, checked in to a place, etc. all the info they have about a thing is fed into their ranking algorithm.
2:00 PM Eighth question: was there any consideration of working with Google?
Mark says, “I would love to work with google. When we did our Bing Web search integration we were very public about the fact that this wasn’t a thing we were trying to do with Bing–we want to make search social in general.”
Ninth question: Why even have Bing in there if people aren’t coming to Facebook to do Web searches?
Mark says there is a lot of content you can find with Graph Search, but a lot you can’t find with it, too. It’s better to show good Web results than nothing when people ask certain things, he says.
Tenth question: search is very resource-intensive–what did you have to do to get this all together?
Lars says building out the software for this took more than a year, lots of engineering work.
Eleventh question: any plans to couple voice search with Graph Search?
Mark says not right now.
Twelfth question: What do you want Google to do that they’re refusing to do?
Mark is smiling and not sure what to say. He says when people share something on Facebook they want to give them the ability to change things later–take things down quickly, for example. Google has a system that works really well for them about how they treat information, but he thinks Facebook is different. Hmm, okay.
Thirteenth question: you’re talking about creating a great user experience, and that might be when you see restaurants your friends like to show more info–prices, menus, etc–any plans to add this stuff?
Mark says they have lots of teams at Facebook. They spend most of their time on helping people share content with other people. As that info comes online in the coming years, they definitely want to add it to Graph Search, he says.
1:55 PM Ah, now they’re doing Q&A with Tom, Lars, and Mark.
First question: how will you make sure people understand the difference between hiding from the timeline and hiding from search?
Tom says that at the end of last year they sent a message to users encouraging them to take a look at their settings, and rolled out easier privacy controls. And it’s also about pointing out the differences between hiding things from timeline and from graph search.
Second question: can this be built into products, apps, etc?
Mark says they would love to. “This is really a beta of a V1,” he says, and there are years of work ahead. Eventually he thinks it will be useful for developer apps, so that sounds like that coming.
Third question: where do you see advertising, sponsored placement fitting in?
Mark says this could be a business opportunity for them over time, but for now they want to make it something that is high-quality that does what users want. They have had sponsored search results for a while, he reminds us.
Fourth question: how soon will this be available to how many people, and how soon on mobile?
Mark says it’s going to be available to a “very small audience” today–hundreds or thousands of people. They don’t know what people will search for, so it’s hard for them to estimate rollout. Mobile is the next thing they want to take on, but no “concrete estimates” for how long.
Fifth question: how will what I search for influence the ads I see? And who else can see my searches?
Lars says nobody else can see what you search for.
Sixth question: how do you know this is what people want from Facebook?
Mark says they try to understand what people are trying to get out of Facebook and synthesize that into products and services anyone can use. People have been asking for a long time about place search, he says. “We’re pretty confident from the feedback we’ve gotten that this will be useful for a lot of folks,” he said.
Seventh question: did Microsoft have any involvement in Graph Search?
Lars reminds us that if Facebook can’t answer your question very well with graph search, it will show you results from Bing. They worked closely with the Bing team to make this happen, he says.
Mark says they have “great partnership” with Microsoft and that what they do highlights the differences between Graph Search and Web search.
1:53 PM Mark says that there are other Graph Search things they want to get to–mobile, for example, and other languages beyond English. They also want to index all posts and content on Facebook
Eventually they want to get to Open Graph, too, meaning all the other content fed into Facebook. “It’s really an honor to be able to build this service and offer it to the world,’ he says. The service will be in limited beta starting today at facebook.com/graphsearch. They’re slowly rolling it out, he says
Oh, now we’re going to watch a quick video about graph search. Lights down! Thoughtful music swelling! The video is showing people doing various searches and results a commercial for graph search. “photos of my friends” for an ex; “photos before 1995”, that sort of thing. These people look like they’re having lots of fun. The end. Clapping, not me, but others. I feel no need to clap the ad.
“Graph Search is the kind of product we love to build at Facebook. It’s a big technology problem, and it’s also a big social problem,” Zuckerberg says. “This is who we are–we love building things like this,” he says.
1:49 PM Mark is back on the mic. He says Facebook built a few tools to demonstrate all the things people will be able to see with Graph Search. He says the company wants to get these tools in front of people before everyone in the world gets access to Graph Search.
So there will be an “encouragement” on everyone’s home screen that lets them see what will be happening. And you can click things you don’t want showing up in search.
Another thing: when you can’t find what you’re looking for, Facebook will show some search results from the wider Web through a partnership with Microsoft’s Bing. It’s Bing-powered web search on Facebook. Uh-oh, Google.
So if you’re searching for an artist’s new album or the weather in Menlo Park, this would show you results drawn from Bing but on Facebook. Mark doesn’t expect people will come to Facebook to do general Search queries, but he thinks it’s useful for Facebook to have it.
1:44 PM Tom is telling us one last thing: they realize people will care about what shows up about them in search, and they have some “great tools” to help with that. Is Facebook a little touchy about privacy, perhaps?
The existing privacy shortcuts are available on the upper right side of screen if you click the little lock icon, Tom says. You can see your activity log in there–things you’ve liked/posted/commented on/etc. And on the left are filters. You can click on one–say the photos one–and see the set of photos that has been uploaded and tagged with you in them. You can see which ones are publicly shared, shared with friends of friends, hidden from timeline, etc. You can then send a note to the owner of those photos (assuming they’re not yours) asking them to take them down if you hate them–rather than just untagging yourself from embarrassing photos.
1:38PM Okay, Lars is back on the scene to show us Graph Searches involving places. He also wants to remind us you can still search for friends by name, pages, groups, apps, all by name. He talks about how he had a toothache and needed a dentist, so he typed in “dentists liked by my friends”. The second result, of course, was Mark Zuckerberg’s dad, who is a dentist. But he chose a closer one (Zuck’s dad is on the East Coast, of course).
He then searches for “restaurants in San Francisco, California”. Lots of restaurants show up–they show what his friends like, what’s popular. To get more specific he types “restaurants in San Francisco, California liked by my friends from India” and it shows him some Indian, Thai, and Mexican places.
This could be very useful–I often want to know what kinds of restaurants and shops my friends dig. And this should be making Yelp very nervous.
Then he searches for “restaurants in San Francisco, California liked by culinary institute of America graduates”. Veerrrry clever, Lars! It gives a list of a lot of expensive restaurants in SF, but also some not-so-pricy ones.
1:33 PM Now Tom is back, showing us a search for “movies my friends like”, and that will show you a list of movies your friends like, as well as movies people who liked those movies liked.
This could make it easier for me to figure out what to watch–an endless struggle with Netflix. There’s also an option to see video clips of TV shows that your friends like. “Videos by TV shows liked by my friends” is the search for this. Now he’s searching for “TV shows liked by software engineers”.
Heh, “Big Bang Theory” is top of the list. But they also like “Game of Thrones” and “Friends”… hmm. Next query: “TV shows liked by doctors”. Ha, it turns out doctors like Grey’s Anatomy and House–those are the first and second ones on the list.
Wow, here’s an interesting query “music liked by people who like Mitt Romney”. This shows Johnny Cash, Pink Floyd, Metallica. “Music liked by people who like Barack Obama” shows Michael Jackson and Lady Gaga.
You can also search for “languages my friends speak”. And you can see just the friends who speak a particular language–useful if you need something translated, perhaps.
1:28 PM Another example: recruiting. Let’s say you want to recruit people from NASA—and who doesn’t? Stocky searches “NASA Ames Research Center employees who are friends of Facebook employees”, and a list of these people is the result. Aand you can see who the Facebook employees are that are friends with these NASA people. Very neat.
Another search: “People who have been product managers and who have been founders”. Now he’s seeing all the people he can access who have this information in their Facebook profiles.
Lars is taking the reins to show us how you can use Graph Search to find photos. Finding the best photos of your friends? Easy, he says. “Photos of my friends” is the search. Graph Search picks out the best photos based on likes, comments, he says. “For graph search, this is a walk in the park. It’s one of the simplest things it does, but i don’t think it’s been available before,” he says. A search for “photos of my friends taken in Paris, France”, “photos of my friends taken in national parks”. These yield all sorts of fun outdoors photos. Whoa, a llama! At Machu Picchu. Very cute. Lars, by the way, is wearing a sweet red t-shirt that says “hackathon” on it.
Another neat example of image search: “Photos of Berlin, Germany in 1989” (the year the Berlin Wall came down). It turns out plenty of people have shared some cool photos of this. You can hover over the photos and see who shared them and with what audience–these are a bunch of publicly-shared photos. It s a very neat way to see history.
They stress that in Graph Search you can only see content you could already see before. This is just making it way easier to find that content.
You can also search for “photos i like” to see all the photos you like in one place. That’s actually very cool.
1:25 PM You really have to see Graph Search for yourself to get a feel for it, Zuck says. So Tom Stocky and Lars Rasmussen, who worked on it, are invited up to show it off live. Now we’re going to walk through different search scenarios…
Ruh-roh, a tiny technical glitch here with the background screens. Ah, ok - back to normal.
So there’s a search box at the top, and a number of suggestions come down below it that suggest things to search for. Someone types “friends who like star wars and harry potter” and gets back a set of friends, in this guy’s case 16 of them.
Basically you’re typing in the view of how you want to see Facebook content, and the content fills in. These results are entirely unique to you, they say–if anyone else does this same search they’d get different results, because they’re ordered by how close you are to people.
Another case: you meet someone in real life and want to friend them on Facebook. So if you meet someone at a party named Chris who is a friend of Lars, who went to Stanford, you can search for this. Stocky does this and gets one person–Chris Cox–and can reasonably determine this is the right person.
Another thing we can do with graph search is find dates! Alright. Let’s say you want to set someone up with somebody else. So he searches for “my friends of friends who are single men” and makes it specific to San Francisco and that the woman in question should be from India, because the person looking for a date is an Indian woman, hypothetically.
1:18PM What they came up with looks like a real-time search. You can type “photos of my friends in Arizona” or whatnot, and it will try to come up with the rest of the query you’re looking for as you type. It’s pretty neat-looking, I must say.
Let’s talk about what graph search can do, says Zuck. Okay, let’s.
For the beta version, it’s focused on: people, photos, places, interests. Zuck explains a few months ago “Game of Thrones” was going to be on and he wanted to invite people so he typed in “my friends who live in Palo Alto, California, and like game of thrones”. People are first sorted by their connection to you, so the first result is his sister.
Another use case: what photo to use on a holiday greeting card? “Photos of me and priscilla chan” was the query, and the best photos show up first, he says. (I think he means ones with most likes/comments but I’m not sure).
Another situation: going places. So you can say “Mexican restaurants in Palo Alto, California my friends have been to.” That way you’ll see which friends have been where, which ones liked it, and which entries they liked for each restaurant.
1:16 PM So what’s the difference between web search and graph search? Zuck is glad you asked!
Web search “is great” he says, but very different from graph search. If you do a web search for “hip hop”, for example, Web search will show you links with possible answers to questions you might have about the subject. In general, web search meant to take any open-ended query with links that have answers to questions you might want to ask. Graph search is meant to take a precise query and return an answer. So you can ask “who are my friends who live in San Francisco?” And it should show you the answers.
A big design problem they had to solve was: how can they make it so people can ask these very precise questios in a natural and precise way? How does graph search work? Filters, he says! “That’s my joke for you today,” he says. Huh.
Most structured search products out there rely heavily on filters. But Facebook realized in product development that this wouldn’t scale to all the connections they want to support. So they came up with something more natural.
1:14 PM They wanted to build privacy-aware search. So with graph every piece of content has its own audience. Most content is not public and you can only search for content that has been shared with you. (Mark is going talking very fast.)
Already today, Zuck says, about 10 percent of Facebook’s computational capacity is spent computing privacy checks. “That just gives you a sense of how seriously we take letting people share anything they want with any audience out there.”
1:11 PM This is called Graph Search. It is NOT web search, zuck says “We’re not indexing the web here; we’re indexing our map of the graph.” The graph is big ad constantly changing: 1 billion people, 240 billion photos, 1 trillion connections, and this is always growing; billions and billions of new connections each day, he says. And there are lots of different kinds of connections–likes, comments, tags, etc.
Indexing all this content and making it so you can retrieve it quickly is a very hard technical problem they’ve been working on for a long time, he says.
1:09 PM Facebook is like a big database, and you should be able to query it, Zuckerberg says. He mentions a few major use cases: First, “what’s going on with people around me”? (That’s news feed). “Who is this person? Tell me their story.” That’s timeline. Today, we’re going to talk about the third! What is it, Zuck asks? Is it things my friends like? is it photos of my friends doing things? What about places my friends have been or like?
“The reality is what’s more interesting than any of these things individually is giving people the power and the tools to take any cut of the graph and make any query they want.”
1:05 PM Oh, here’s Mark Zuckerberg. This is the first big product announcement at this campus, he says. Facebook’s mission is to make the world more open and connected. The way we do this is by giving pepoplle tools to map out their connections, he says.
Most people think of Facebook as a tool for getting connected with people they already know. But before newsfeed people used Facebook to browse and discover things in their network, he says, and we’re on a trip down memory lane to see how the social graph helps people make new connections.
1:02 PM (Eastern Time) We’re going to start in a few minutes it looks like. We’re in a packed room with two big screens in the front with the Facebook logo and blue background on them. It’s reporters at tables and a crowd of Facebook employees in the back behind us.
I will be posting live updates from today’s invite-only Facebook event from the social network’s headquarters in Menlo Park, California, starting at around 10 a.m. local time.
Facebook sent out an invitation last week that coyly read “Come and see what we’re building.” Predictions about just what the company is building include a phone (which founder Mark Zuckerberg has denied in the past), a new advertising product, or perhaps a search engine.
Stay tuned for details on the company’s mysterious announcement, as well as observations about the company’s headquarters and executives’ outfits.
AI is here. Will you lead or follow?
Join us at EmTech Digital 2019.