Re: intent-driven search

13 Jun 2005 - 3:27pm
9 years ago
2 replies
606 reads
Doug Anderson
2004

Hi Ben,

Isn't what you're talking about a "recommender system"?

This category of applications holds great promise for asynchronous social collaboration about the sorts of things you're talking about.

I've had the opportunity to hear Joseph Konstan of the University of Minnesota & CHI talk about his adventures in this market segment, but I haven't yet read his book, "Word of Mouse."

There's a lot of programming, yes, and there also needs to be a substantial amount of data, yours and others'. The possibilities go well beyond "People who looked at this book also looked at..." I think the intersection of computer-based "intelligent" software agents featuring user-situation awareness, location-awareness, and with data about user inclinations / preferences will bear some interesting fruit in the next several years.

Peace,
Doug Anderson

Original message:
Date: Mon, 13 Jun 2005 13:35:56 +0100
From: Ben Hunt <ben at scratchmedia.co.uk>
Subject: Re: [ID Discuss] intent -driven search
To: vimal sharma <vimalsharmaa at yahoo.co.in>
Cc: discuss at ixdg.org
Message-ID: <42AD7DAC.8070403 at scratchmedia.co.uk>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed

Wow, what an exciting area! Thanks Vimal.

<snip>

3) Google Local has added a third dimension: space! Now you can search
for a hotel near a certain postal code, which is really great! I love
Google Local! Wow, it can show me all the hotels near the event in a way
that I can quickly interpret, and choose which to contact, in the right
order!

But this isn't enough.. Whenever I sense that something isn't quite
enough, I look at the real world and at how existing systems solve
problems. In the real world, when I want a babysitter, I would ask my
friends: people like me whom I trust, if they can recommend someone. I'd
want someone within travelling distance of my house. I'd want someone
with similar values to mine (i.e. won't sit the kids in front of the TV
all night and feed them chocolate, won't use intimidation or violence...)

There's no search engine that can do this yet, but I've designed a
concept to do this. It adds the following dimensions, in addition to
keywords and locality:

* Preferences:
In real life, we all have preferences and prejudices that
influence our buying choices. Do you prefer to use liberal or
conservative businesses, vegetarians, start-ups or established
businesses, businesses that recycle or have ethical policies,
Muslim or Christian or Jewish or ethnic-run? Why not click a few
boxes to tell the search engine?
* Cross-referenced ratings:
What other people think really matters. If a computerized system
is going to do as good a job as asking people who know, it has to
use all the information available. My design would allow people to
rate any service from any provider. These would have to be
filtered carefully to prevent abuse, but this isn't too hard. The
trick will be to use ratings *in context*. i.e. Ratings of people
like me, or who live near me, should carry more weight.
Preferences, locality, and any other socio-economic indicators
should be used to predict the services/products I'd be most likely
to choose myself.

That's it, really. Conceptually, this would be a simple system. Sure,
there would be a lot of programming involved.

Google don't want to go down this route, because they have a strategic
intention to collect as little personal information as they can. I'd
argue that they will *need* to allow people to supply more information
in order to get better search results. They actually suggested that I
have conversations with Yahoo! or eBay, to see if there's more synergy
there.

Anyone from Yahoo! product development on this list?

- Ben

Comments

13 Jun 2005 - 4:26pm
Ben Hunt
2004

Douglas, yes, I think this is a "recommender system": a kind of Yellow
Pages, plus recommendations along the lines of Ciao, and using a web of
trust such as eBay's.

The cleverness comes in its knowing which variables to use when. For
example, some things are local-only (window cleaners, home help),
whereas others may be global (programmers, astrologers).. In other
areas, personal preferences or price may be paramount..

All that information may be garnered with a bit of intelligence and
care. Just learn by watching which providers got clicked on (for contact
details) by which kinds of people in what contexts. I bet there are AI
programs out there today, or on MSc students' workbenches, that could do
just this kind of thing.

My argument with Google is: why not?! Why not ask me what newspaper I
read, how many kids I've got, whether I'd rather do business with a
small local company or a global chain, if it'll help you predict what I
would most likely choose given all the information?

I think that Google's #1 job has to be "not to become the next
Microsoft", i.e. ubiquitous but disliked, "the devil you know".. They're
a trustmark right now (not a lovemark, because their service is too
mundane), and if they are going to keep that trust, we all must continue
to think they're on our side. That must be the motivation for not
wanting to ask for, or store, too much personal information. But I want
the choice to be able to provide personal information, if it'll save me
time and money.

The risk for Google is that another competitor could take their crown in
exactly the same way they took Altavista's in 2000(/01?), by simply
doing it better, by giving me slightly better search results. I think
that this personal information will be the glass ceiling on Google's
performance. (But Google *do* hold personal information about us, see:
http://www.cnn.com/2005/TECH/internet/06/03/google.privacy.reut/).

Who in the search world today is bold enough to ask me up-front for the
information they need to know to get the search job done better?

Peace,

Ben

Anderson, Douglas W. wrote:

>[Please voluntarily trim replies to include only relevant quoted material.]
>
>Hi Ben,
>
>Isn't what you're talking about a "recommender system"?
>
>This category of applications holds great promise for asynchronous social collaboration about the sorts of things you're talking about.
>
>I've had the opportunity to hear Joseph Konstan of the University of Minnesota & CHI talk about his adventures in this market segment, but I haven't yet read his book, "Word of Mouse."
>
>There's a lot of programming, yes, and there also needs to be a substantial amount of data, yours and others'. The possibilities go well beyond "People who looked at this book also looked at..." I think the intersection of computer-based "intelligent" software agents featuring user-situation awareness, location-awareness, and with data about user inclinations / preferences will bear some interesting fruit in the next several years.
>
>Peace,
>Doug Anderson
>
>
>

13 Jun 2005 - 8:16pm
George Olsen
2004

Discerning intent _is_ the great difficulty of search. After all intent
really determines relevance.

Unfortunately, people are unwilling to put much effort helping search
engines figure this out. The average query length has stayed around 2.5
words -- and actually it's effectively less because it includes a lot of
multi-word terms -- e.g. "Brittany Spears" -- that are really a single
concept. And there are a lot of reasons _why_ someone might be searching for
Brittany-related results.

While search geeks will fill out preferences or assign a rating to a result,
ordinary people won't make the effort. Search is a place where everyone goes
but no one wants to be there. (They'd rather be at their destination.)

Another problem with preferences is they're contextual. Whether you prefer
research-related or shopping-related results depends query-to-query on what
you're looking for. There's a similar problem with ratings. It's one thing
to rate a concrete thing like a book or a camera, but if I rate a site
highly you'd probably want to know whether I was looking for the same thing
as you are before you'd take my rating into a account. (Plus there's lots of
room there for gaming the system and black-hat SEOs already do lots of
automated trickery to boost sites.)

You can try to infer intent from observed behavior, but that's tricky. If
someone clicks on a link, how do you _know_ whether they found it relevant.
There are clues from click tracking, but they're not conclusive.

Yahoo's My Web beta <http://myweb.yahoo.com/> -- which I helped design --
does to do some interesting things by letting users save sites as bookmarks
and block other sites from ever showing up in the results again. I can't say
more about it, but you draw your own conclusions about how that feedback
could be used.

George
___________________________________________________
George Olsen Principal, Interaction by Design
650 329 1728 george at interactionbydesign.com

Syndicate content Get the feed