At an event in August, Marissa Mayer, Google’s leading VP in search, said social search hasn’t shown much promise, but if it does, Google would be in a good position to incorporate it. I wrote about that here.
In a recent interview with VentureBeat (see Q&A below), she says she considers social search to be an essential part of the leading search engine in coming years.
She hints Gmail may be used to identify your friends, using their search history to influence search results for you and those in your social network. While this network would likely first be built on Gmail contacts, Marissa wouldn’t rule out importing friends from third-party networks down the road. But it’s clear Google wants to have the user explicitly approve them. For now, influencing search results based on information gathered from your MySpace friend connections is a relative non-starter, she says, because such influences haven’t been explicitly approved by MySpace users. She is hopeful users will adopt an annotation model, though — meaning that they write reviews about websites and broadcast them to their friends. Google hasn’t implemented these ideas yet. But I wouldn’t be surprised to see if it’s in the works.
So what about other players? Third parties such as AOL and MySpace already contract with Google to use its search engine. But if they want to enable social search, social networking sites would have to rerank search results. Marissa said Google currently doesn’t let third-party sites do this because Google works hard on ranking. She didn’t say whether or not Google might allow such reranking in the future, but it’s clear that reranking would be essential to social networks that want to fully leverage the social context of their users. (By the way, at a time when it’s a foregone conclusion that open source and open platforms create more value than any individual company could on its own, I find it surprising more people aren’t calling for Google to be more open.)
Marissa has started speaking more about social search lately. Most recently, for example, she said it’s wrong to look at search as all human vs. all algorithm, meaning there could be some combination, and that “expertise” may be more valuable than “trust.” Just because you know someone well doesn’t mean they’ll provide the best answers.
Below is VentureBeat’s Q&A with Marissa. We’ve edited the transcript to remove redundancy. For a few of her answers, we’ve also provided some analysis.
The bottom line is, no one has yet done widespread social search like what Marissa is describing: translating the millions of everyday social searches from the physical realm to the internet realm. Below, we’ll also mention early player Eurekster, and how its early experimentation didn’t get very far. We’ll also talk about how, as a first step, Google is likely to encourage its users to annotate search results. Mahalo and Wikia are two other companies aiming to prove social search works. But until more social search is actually put in action on a wide scale, it’s all speculation about whether it will actually improve search.
VentureBeat: How would you describe social search?
Marissa Mayer: We believe social search is any search aided by a social interaction or a social connection… Social search happens every day. When you ask a friend “what movies are good to go see?” or “where should we go to dinner?”, you are doing a verbal social search. You’re trying to leverage that social connection to try and get a piece of information that would be better than what you’d come up with on your own.
We know that because of the volume of searches like this that happen everyday, that the social component of search is actually very important, and it hasn’t translated well yet to the internet medium.
Social search is hard because the intuitive thing you would do online to mirror normal social networks and other social interactions just aren’t that effective, compelling or even reasonable. So, for example, from the Facebook News Feed analogy, you could just get a social network and broadcast all of everyone’s searches to everyone on their social network, but most people view search as a far more private activity than that. They’re not comfortable letting everyone in their social network know what they searched for, so such a product is clearly not reasonable.
VB: How would you implement social search?
MM: One thing we have tried…is labeling — have users annotate the search results they see and have those annotations be shared with people on their social network or with people of like mind and interest. We did that in the form of Google Co-Op… We’ve seen success there in things like health. There have been a few topical areas that have had a lot of traction, but overall the annotation model needs to evolve.
…Another classic thing to try is “other users like you,” where you build implicit social connections between users who are like each other, much like Amazon does with books. Examples: “Other users like you also searched for” or “other people who did this search also did searches.” It works to give you related queries, but again it doesn’t fundamentally change the social experience or capture that movie or restaurant recommendation request that would be one of the classical examples of verbal social search.
I think there is the possibility of taking a social network and combining some element of annotations and searches done. For example, if I have 400 friends on Facebook, and I knew 10 of them all searched for one topic today, that might interest me. So aggregate statistics might work. In truth, there are a bunch of things you could try. You can take the annotations that people enter through something like Google Co-Op and broadcast the annotations. Presumably when someone is writing a search result, they don’t mind if other people see it and read it.
PageRank itself relies on the link structure of the web to try to find the most authoritative pages. For example, it’s clear that people would attribute more authority to the pages that their friends have visited. So if we took Web History and allowed that data to influence rankings, such that pages that your friends have visited were now bumped up in your search ranking, that that might be a good augmentation to something like personalized search. In essence, it’s a fusion of personalized and social search. In this case, what we would do is say: This Gmail account which maps to Marissa Mayer then maps to these other friends, allow those friends to influence this ranking… But no, we have not done anything like that to date.
VB’s take: She noted that Google has not done anything like influence search results based on friend connections to date, but given reports of a Gmail News Feed, Google might be in the process of implementing these ideas. If Gmail is used to create a social network based on the already existing Google Profile, then Google might complete the transformation to becoming a full-fledged social network. If Facebook releases a social search service, as well as a fully-featured email product and IM, it would increasingly look like Google. Facebook IM and email would likely increase time spent on site, provide more data for monetization, and increase the chance users would conduct web searches on Facebook.
Google or Facebook could create social search products to see how much users are willing to annotate, vote on and share results with friends. Google has tried annotation with Google Notebook, but there aren’t enough users annotating results to make an impact yet. Alerting your friends in a news feed when you annotate a search result might encourage more annotations. A Google News Feed could accomplish this. Social utilities like Facebook could use their existing news feeds. Last month, Google released a new experiment to let users vote on search results explained here, but the experiment page still doesn’t permit users to see it in action.
The question of how Google’s monetization might change if Google succeeds in creating its own network of connections between friends remains open. If social search works, so might social ads, and vice versa.
If it’s relevant to you that people in your network searched a particular term at unusually high frequency today, then you might also be interested in the ads they clicked on. Facebook is already working on this with Social Ads, which identify a friend who is affiliated with a particular brand. In order to create Google’s own version of Social Ads, they’d need to actually get users’ names in the first place. That’s where the leading social networks have an edge. Google starts with tens of millions of accounts, but outside of Orkut, they’re not social network users used to sharing information about their activities with friends. Putting Google users’ names next to ads would likely come as even more of a shock to Google users than Facebook users.
VB: What is the future of search?
MM: If we look at a search engine ten years from now, we know it will be better than Google is today. Google itself gets better every single day because we’re constantly making changes to the relevance… I think one way it will be better is in understanding more about you and understanding more about your social context: Who your friends are, what you like to do, where you are. It’s hard to imagine that the search engine ten years from now isn’t advised by those things.
Given how things have transitioned online – everything from travel planning to purchasing books — my belief is that a lot of those physical questions we ask now – “what movies are good to see?”, “where’s the nearest sushi restaurant that’s good?”, those kinds of questions, will ultimately begin to transition to the online medium and that social context will be leveraged there.
VB: What will navigation look like for social search: search, browse (meaning discovering what you want by perusing pages, such as looking at the most popular list from friends), other type, or does it remain to be seen?
MM: I think it’s too early in the evolution of the space to really know if it will be searches yielding better browse and discovery mechanisms or if it will be manipulation within search itself.
VB: You already have hundreds of millions of users who use Google regularly. Hypothetically, would you leverage that user base to connect people with friends if you were to offer social search for Google?
MM: Yes, I think that’s exactly right. We’re interested in looking at the Google user base and the connections that exist in it. We would use those social connections to influence search.
VB: Does influencing existing Google results with third-party social networks make sense?
MM: It would be a surprise if one day users opened up their browsers, began using Google, and friendships they had formed on MySpace began to rerank their results or in some way influence the page. It would be non-intuitive to them. As a result, we don’t think it would be a great product decision or a great user experience.
VB: Why not let Google users draw on friends from Facebook or MySpace?
MM: We may well. We aren’t likely, without users’ explicit consent or knowledge, to take their connections on MySpace and cause them to influence Google ranking.
You can imagine us forging connections on our own, you can imagine us asking users what connections currently exist that they would like to use. Each of those are viable options.
VB’s take: It sounds like Google will start by forging connections among their own users, then extend from there by cutting deals with big social networks to let their users import their friends lists to Google. If the lasting, sustainable value of a social network is the connections between friends, selling that information to Google for a one-time fee on the cheap is not worth it. If social networks can find a way to still own the networks of friends Google uses to influence searches going forward, then they’ll be better positioned to continually participate in the billions in profits search generates every year.
VB: You license Google search to many websites, including MySpace and CNN. The user experience on some of these partner sites does not include many of the features you get on Google.com. For example, there are no universal search features, no “more results from this site,” and so on. Why don’t you make all of these features available to all search partners? A step further, would you consider letting developers create their own features?
MM: We do license our search and we also do the custom search engine for small businesses. We do actually offer them the full Google search functionality, but obviously we license those for a fee depending on overall volume as well as difficulty of implementation. Features are priced accordingly. In our deals with AOL or with Ask, we do offer them this functionality. Because these features have a cost to us, they do incur a charge to the partner. Our partners then decide what they would like to use.
VB’s take: Yes, “priced accordingly” means factoring in how much those features would cause a given user to return to the partner website to search and force Google to pay a rev-share. That’s a real cost. The fewer features you provide, the less likely the user is to search there. The most profit comes from users going to Google.com because there’s typically no rev-share. This focus on driving all users to Google.com makes business sense now, but too much focus on that might cause Google to miss opportunities to enable innovation in search by people outside Google with data Google doesn’t have. That said, it’s not clear how outsiders like developers or companies have any real chance to improve search, since search is complex and Google holds much of the valuable data.
VB: Are you open to permitting reranking by search partners? Maybe they could test their own approach on social search.
MM: We work very hard on our ranking functions. As a result, our search licensing agreements explicitly prohibit reranking results. A lot of the value our partners pay for is in the ranked list of search results, so we suggest they don’t rerank it. We welcome developers adding things to the UI to make search more useful. Because the ranking remains one of the core aspects of our business, we have as part of our Terms of Service that they not reorder results. The Google brand stands for excellence in relevance and ranking, so we don’t want to provide a co-branded product where it doesn’t reflect our best ranking.
VB’s take: When it comes to reranking, Google cares about a lot more than just branding. Ranking cuts to the core of how Google makes money. If Google lets others rerank results, and they start to do it better than Google, then it puts Google’s entire multi-billion dollar search business at risk. Granted, given the billions invested in search by Yahoo and Microsoft building inferior search products, the idea sounds farfetched. But from Google’s perspective it’s apparently a risk simply not worth taking. A third party that establishes search clout could negotiate a revenue-share much less beneficial to Google than usual. That could cost Google millions of dollars at least. That’s the challenge of dominating the search market: Anything you do that gives others more control of search directly trades off against profitability for Google.
This idea of letting search partners rerank results isn’t a new one. While Google denied early player Eurekster from reranking results in 2003, Yahoo was more open. It let Eurekster and Friendster rerank its results, but both of those companies didn’t get very far. Eurekster tried reranking by looking at what friends and friends of friends click on, an experiment that lasted a year and a half. That deal broke down in part because Eurekster didn’t find the right search augmentation, and social search for Friendster wasn’t a top priority. (Eurekster says, however, it would be open to trying it again with a major social network).
Amazon’s Alexa has made an open search product available for some time, though its unclear whether if Alexa even makes anything available worth building on. Leading social networks like Facebook and Bebo are prime candidates because they have significant traffic to direct into search and have a dataset — including users’ name, age, interests, education and so on — that potentially could improve search. Companies with interesting datasets shouldn’t have to rebuild search from scratch in order to see if they can enhance search. They should be able to plug in the data to see what results they can get. Effectively, innovation in search is being stifled. Yahoo and Microsoft could spur innovation in search by doing something Google has not: actively encourage developers to try to build better search by reranking, using data in new ways and conducting other tests with extensive access to the resources of their search engines.
[Special thanks for contributions to interview questions and the topic: Ada Chen, Chris Campbell, Dan Ackerman-Greenberg, Eugene Shteyn, Greg Sterling, Ido Green, Jack Abraham, Jared Morgenstern, Jessica Alter, Jim Scheinman, John Battelle, Jon Turow, Julian Gutman, Kara Swisher, Noah Lichtenstein, Sachin Rekhi, Satya Patel, Vik Singh and Zach Weinberg.]
[Disclosure: Doug Sherrets owns Facebook shares.]
VB's research team is studying web-personalization... Chime in here, and we’ll share the results.