The Future of Search Won't Need a Search Box

19 comments

The Yahoo Search Blog has an interesting interview with Yahoo's VP of Emergent Search Technology, Andrei Broder. His basic premise is that search will switch from a pull model to a push model.

Search today is confined to putting in something and getting something back, a pull model. The next stage is for information to come in a context without actively searching, a push model. My favorite example is GPS. Instead of looking up your way on a paper map, you are in your car, and your GPS navigator gives you directions, shows gas stations near you, and so on. A year or two from now perhaps it will show you where those gas stations are, but only when you are low on gas.

Greg Linden notes that what Broder is really talking about is personalized information streams.

Is this really the future of search?

Comments

I think that we have to

I think that we have to believe that we are smarter than machines and know what we want more than machines do. Surely Yahoo! is banking big on the topical stream stuff...but there will always be some need for a pull of some sort.

today

The Lexis LS470 navigation system has gas stations that show up when you are low on gas. Try it for yourself... have POI turned on, and you will not notce any of those little gas pumps at all until you are low on gas. Once the gage says you need gas, those little pumps stand out like campfires. Very handy, but hardly futuristic (this is a 2004 model).

No matter how you slice or dice it

the limits of current technology will not allow this to happen for a long time. However with him talking about push vs. pull it makes me think about the fact that you can't push data unless you know exactly what to push. So Seobook you're absolutely right. To push data (ads, web results, even GPS as Mr. Broder states) you still need information from the end user. A computer is a box. It only does what we tell it or program it to do. It requires human input and interaction to function properly. Sure you can generate data and profile a user but that's clearly not enough information to go on. Unless a user gets very specific and tells you exactly how they want to be targeted.

I won't say that Mr. Broder isn't a genius. He is one of the few out there in my book. Still I think he's way out there on this one and walking a very fine line on the real abilities of today's computing models. This kind of capability is a long time off.

So my question is: Would you give personally specific data on yourself to any company knowing that it can target you perfectly? Would you be more inclined to do so if your name, social security #, and credit card information remained completely non-identifiable?

So...

So my computer will know when I have a migraine and want to look up some information? It'll know when I'm hungry and look for a Chinese restaurant in town? It'll know when I want to order myself a new DVD? It'll know when I want to see some porno?

This sounds like crap I'd see in a Wachowski brothers movie. Perhaps they should work on being able to keep their e-mail up all day before they start trying to read my mind.

Broder missed the boat

The push technology is already here, it's called AdSense, and it's on many web sites suggesting other websites that might meet your search criteria.

Just bind Google's relevancy ad server to the actual search database and you could easily serve up a bunch of relevant document links on ANY PAGE you're looking at as Google already knows it's context and can assume the other documents would also be of interest and present them for you to click on.

Child's play.

The problem you can't easily solve is the starting point, like I'm looking for a wrench, at some point the system has to be seeded with the term wrench to start the process.

I Think It's Possible, Sooner Than You Think

Combine user-provided information, collaborative filtering, persistent cookies, and suddenly you can learn a lot about a user's intent. The key is that the user has to trust that you will use their personal information for the user's benefit and not exclusively for behavioral targeting.

Here's my take (from my blog, last month):

"Consider this scenario. You download a program to your desktop. The program begins by asking you a lot of questions about yourself - from basic demographic questions (age, sex, geographic location), to highly detailed psychographic questions (your personality, your pastimes, your fears, your future goals, your risk quotient).

Moreover, this program follows everything you do on the Internet (and you allow it to do so). It tracks the time you spend on each page, which sites you return to frequently, which sites you leave immediately. You can even add reviews of sites you like and don't like. And finally, if you want, the program compares your behavior to that of millions of other Internet surfers across the universe. It finds surfers similar to you in terms of interests, life stage, or basic demographics.

After a few weeks (or maybe months), the program starts to combine all of this disparate information into a pretty darn accurate profile of your wants and needs. It combines your responses to surveys, your Internet searching habits, your ratings of Web sites, and the activity of similarly-situated Internet users into one big algorithm. And here's the great part. Instead of using this algorithm to show you "search results", it simply takes you to the "right" page.

So now, when you type in "los angeles travel", it takes you directly to the Web site with the best travel deals for you to Los Angeles. Heck, depending on the information you have provided or the system has gleaned from you, it might even know the dates of your travel, your departure location, your preferred method of traveling, frequent flyer numbers, travel companions, credit card information, and whether you need a car, hotel, a kennel for your dog, and some new luggage. All you need to do is review the price of your trip, click submit and voila you're off to LA!"

It ain't gonna happen in the next 12 months - but 3-4 years of Internet time is a looooong time!

Hence my questions:

Would you give personally specific data on yourself to any company knowing that it can target you perfectly? Would you be more inclined to do so if your name, social security #, and credit card information remained completely non-identifiable?

Bill, Adsense is really far from push technology too. Reading the context of a page is not precise and just because I am reading an article on Google's Stock does not necessarialy mean I am remotely interested in investing in stocks. Yet, if I was which stock an I looking to invest in? Adsense can't tell you that or interpret it.

The only way this can be done right now is profiling. Which I go back to my questions. Are you willing to give up that data?

Personally, I might just to see if it would be a better experience. As long as I could be 100% certain that I was personally non-identifiable. I might even give my zip code, area code, and street name if it meant better targeting.

options = $$$

Instead of using this algorithm to show you "search results", it simply takes you to the "right" page.

Amongst choices there are advertising opportunities. If you auto direct and provide no choices some questions will be asked if that relationship is a paid one.

I would question relevancy quite a bit more if it was a paid relationship that was decided as my only option...or some option that was exceptionally hard to unmake.

Where is the legal liability if trademark A sends people to site B without a choice?

So now, when you type in "los angeles travel", it takes you directly to the Web site with the best travel deals for you to Los Angeles.

So do they throw away making profit from that vertical? Or how is the best deal provided?

Many friends have review sites where recommended products are in order of affiliate payout, but those sites are not posing as the default information navigation system of the world.

The Money Doesn't Come From Advertising

Well, there are two ways to monetize personalize without sending someone to SERPs. 1) You charge a user a monthly fee to use your service - there are no ads, there is no preferential rank of sites shown based on the amount they are willing to pay - you simply make an agreement with the user that if they are honest with you with their information and are willing to pay a little bit each month, you will use technology to give them a better searching experience. I equate it to Tivo - Tivo monetizes through a subscription to a service that provides a better user experience than a regular TV. This is the same concept.

The second way to monetize without SERPs is the "AllAdvantage" model, where you pay users (or incentivize users) to provide personal information. The agreement here is: we'll pay you $.50 an hour, in exchange for being able to track you through persistent cookies, user questionnaires, etc. We will then take that information and serve up targeted ads (which we can charge a higher CPC/CPM because of the personalization).

Personally, I like the first idea better than the second, simply because the user is incented to be honest with you (users who pay a monthly fee for Tivo obviously want to provide accurate information, otherwise what's the point?)

The more things change

What's *really* changed in search since the internet first saw search? Nothing. A search box, make money from ads. It was that way years ago, it's still that way now. I'm betting that it'll still be that way in 3-5 years. It's possible it won't be Google/Yahoo/MSN at the top of the search, but whoever it is will have a page centered around a simple text box with a submit button.

possible is different than practical

personalising may be viable but the biggest problem is in the multitasking that most people do and differentiating between the multiple profiles' each user would need. I look at very different groups of sites because 'I' want to and because I need to for clients. It would be great if I could have about 10 different profiles and a way the system could understand what project I was working on at that time but since I might have 10 tabs on 5 different projects open sometimes that would be pretty intelligent programming....

While it might work great for people who work on one PC at the office, another at home, and never work on the weekends, I'd stop using any company which did this without an 'off' switch 'cos I'd rather not be presented with any personalisation than with things that are wrongly chosen because of my conflicting surfing at different points.

One problem that has

One problem that has remained with this concept from the very first time I recall seeing it discussed is that we all tend to misuse terminology and thereby confuse two important terms. Subjective and objective.

We see statements like "the best site" and the "right" answers. The thing is, those are completely subjective terms. The "right" travel page to you is likely the one that saves you the most money, (there could actually be a laundry list of factors that each individual could use to qualify "right" at any given moment), BUT, the "right page" to the service pushing the suggestion would likely be the one that makes the service the most money at any given moment. Now, taken that specific scenario, and assuming we can agree that relelvancy relates even remotely to "the right page", define relevancy.

As to giving them personal information about yourself, THEY already have it. And if THEY don't, just like internet traffic, all THEY have to do is buy it and it's cheap.

That GPS-with-gas-station

That GPS-with-gas-station model is not a "push" model, it is a "pull" model. You have configured some system to deliver something to you when you need it - that is, you ask for it, one way or another.

Pull is the only thing that will ever be successful. Imagine if every coffeshop you passed on your way to work started giving you phonecalls and text messages, just because you might need coffee, wouldn't you be pissed?

You guys need to check out

You guys need to check out Dashboard (http://nat.org/dashboard/) - it's a deskbar which uses desktop search to show you information based on what you're doing - mention someone's name in IM, it pops it up, etc. Read a band's website, it brings up their albums, etc. The project's on hold for a bit pending a re-work of the desktop search framework it's based upon. However, said desktop search supports web APIs...

Written in .NET too so a Windows port would be on the cards.

Edit: http://nat.org/dashboard/textchain.png gives the best example

Very cool

BUT, Dashboard is still input in - data out. Push, not pull.

The only way we will ever be able to determine intent is if the user tells us specifically what they might need at any given moment in time. Leading them through the buying cycle like lost sheep to the herd. You can only get out of the box what you or someone else puts into it.

Still what I don't hear answered is would you be willing to give up enough data on yourself to allow a company to target you in a more accurate fashion? This is the only way I see it working in the forseeable future.

give away data?

yeah we all do it all the time - anyone who's ever completed one of thse surveys in the hopes of winning a mansion in acapulco and as much beer as you can drink gets a trees worth of car insurance renewal quote invites once a year and a voucher for free Felix on the cats birthday.....

generally people are not

generally people are not going to want to pay for what they think they can get fast and free.

alladvantage is going to be skewed toward the poor end of the market.

But TV is Free . . .

So why do people pay for Tivo? Because it is an enhancement to a free service that is worth paying for.

Other examples:

- There is plenty of free content on the web, but people pay for content like WSJ, Salon.com, etc.
- There are plenty of free Internet radio stations, but people pay for LaunchCast radio because it's better

not exactly the same

TiVo enables something more than TV alone. It allows you to get something you would not have gotten for free (because you missed it, won't be able to see it, won't remember to record it, etc). Salon, WSJ etc. provide something you cannot get for free --the content is not the same as other content. Internet radio may be the same... if using a service means you can actually listen (not have to fiddle with readers/players/directories, for example) then it, too is different than simply being an enhancement. I think iTunes straddles the bar... for many it is simple the only way to get a music library playing on your computer. For others, it is an enhanced version of an MP3 downloader/ripper/player/library manager and worth paying for.

Now try and sell a better WinAmp and it's the "why bother when the free one's so good" issue.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.