For years people have talked about personalized search as the next
evolution of our amazing technology.  Over the years, the biggest obstacle facing search engineers is the simple fact that human behavior is not predictable. Don’t
get us wrong, people can be creatures of habit and we can build
functions that enable us to display different results based on logical
assumptions made in the aggregate.  For example, ‘traffic’ at 6pm on a Friday
likely refers to road conditions, not the movie or the band. So we can generally
use some math magic to make really good predictions about what you mean when you
type ‘traffic’ at 6pm tonight, but what if we’re wrong?  It’s easy to see how
that could happen especially if you, say, walk to work. 

In that case a more personal search would benefit you; having more
detailed information about the person doing the search can make results more
effective.

We think one of the challenges with delivering results which are truly
individualized is that, to date, personalized search “can’t see the forest for
the trees”. In other words everyone is collecting everything and trying to
figure out the foibles of human behavior from a mass of digital bits. To an
extent, we’ve all been looking at the wrong inputs which in turn haven’t given us
the output we want.

We’ve found something interesting: a person’s history or profile often does
not necessarily deliver better results across all types of queries. In
other words, depending on the type of search a person is performing (are they
exploring? Looking for a particular site? Trying to find something they’ve seen
before?), there are different personalization techniques, or even combinations
of techniques, that yield better personal search results. In real life, you
take into account many things when making a decision: for example, where you
are, what time it is, and what your friends think. We asked – shouldn’t your
search engine do the same thing?

Even as we continue to develop more relevant search through smart
personalization, we are very focused on maintaining an industry-leading privacy
stance. For more information, see here.

We’re
currently ‘flighting’ (or “testing”, for non search-geeks) a raft of experiments
to see which techniques deliver the best results for a given user behavior, but
today we want to talk about two we’ve recently put out there for you all! First, something relatively simple: automatically tailoring search
results based on your physical location

As 76% of people use search engines to plan trips, events or social
gatherings, Bing Local has always provided you with maps to nearby business
listings, authoritative reviews and areas of interest.  Starting today, we’re
going a step further with new improvements that take into account where you are
and serve locally relevant information directly in the body of the results
page. 

How does it work?

In the past, if you were looking for information related to where you are,
you would need to include the location as part of the search.  For example, if I
wanted the websites of Seattle pizza joints I might try “Pizza in Seattle.” 
Today, I can simply type “Pizza” and Bing will immediately recognize where I am,
in this case Seattle, and provide a link to a popular local pizza place directly
in the results.

Before: searching for “pizza”

After: Searching for “pizza” with tailored results based on physical
location

Here’s another example.  We’re in San Francisco for work and looking for things
to do in the city over the weekend.   Where before we had to specifically
highlight my location, now all I have to do is type “things to do.”   With
today’s improvements, Bing recognizes our location and conveniently serves “Top
10 things To Do in San Francisco” as the second link.  

Second, we’re introducing a feature that helps Bing present the most relevant
website based on an individual’s previous searches. Here is where we
really begin to see our theory come to life to show that different user
behaviors benefit from different thinking around personalization.

We know about 30% of queries are “navigational” queries – meaning a user is
trying to find a particular website like Facebook, Expedia, or the New York
Hilton. Now, if a user issues a query such as {acs} the most relevant
result for that user is not necessarily the same as that for the majority of
people in the U.S. To numerous users with an interest in pursuing a career in
chemistry, the most relevant result may be the American Chemical Society,
but to someone interested in how they can get involved in the fight against
cancer, the most relevant result is more likely to be the American Cancer
Society.

Suppose, in this latter case, the chemistry fan selects American Chemical
Society . Our research shows that users commonly re-issue such navigational
queries and the intent of that user rarely changes. This new personal search
feature uses this human behavior as its core premise – if Bing
thinks a user is trying to “re-find” a site, the relevant result is promoted to
the top position on the page:

The beauty of thinking differently about personalized searching is that it
enables us to construct elegant solutions that require a minimal amount of
personal information and, frankly, often exhibit better results than a more
computationally complex predictive model alone.

There is much more to come, but take Bing for spin and tell us what you
think!

– Aidan Crook & Sanaz Ahari, Bing Search