Introducing Project Emporia Powered Matchbox Technology: News is Now Yours

When discussing relevance in search, people often cite personalization as the
next frontier in delivering more relevant results. Traditional methods of
personalization have been based on something called ‘collaborative filtering.’
At its simplest, ‘collaborative filtering’ means if user A and user B both like
something, then you can predict user A’s future “likes” based on what user B
“likes.” 

This approach works quite well for things that are static in nature such as
products, restaurants and movies.  For these objects, systems are able to gather
enough rating data to make future predictions because the objects don’t change
all that often.  Contrast that to the frenetic pace of news and real-time
information where a story can materialize in a matter of minutes. How do you
approach personalization in a reality where there will not be enough activity to
enable collaborative filtering systems to kick in?

Today, at SXSW
Interactive
in Austin, TX, we showcased a new approach to
personalization. Rather than looking only at what others like you are
interacting with on the web, the Matchbox technology begins to
‘understand’ the Web more like a human might. This enables us to make
predictions about what you might want to see, not only based on if someone like
you has done something with it but also on what the content says. 

As an example: let’s say over the past several weeks you have been reading
and rating articles on electric cars. Traditional collaborative filtering
techniques would look at other people who have read similar articles and cluster
you with them.  So when a new article comes out about Tesla’s new model, as long
as someone ‘like you’ interacts with it, it might get recommended to you. But
how does that first ‘someone like you’ find it in the first place?  That’s where
Matchbox can help. 

Rather than simply relying on the actions of others, the Matchbox technology
uses the wealth of information about the entities mentioned in an article. In
this case, Matchbox knows that Tesla is an ‘electric car’.  With this additional
information, it can display “Tesla
Model S: 300 miles on 1 charge
” to you without having to receive input from
other like-minded individuals.  For those geeking out, this technique is called
‘feature generalization’ and is a key underpinning of Matchbox.

Project
Emporia
is the first broad implementation of the Matchbox technology that
you can use.  Project Emporia is a new web application and a Windows Phone 7
application that recommends news stories shared on Twitter using the Matchbox
technology. It combines three major pieces of technology:

  • Filter news stories by automatically predicted news categories

  • Deep social integration of Twitter so you can see what stories your friends
    and friends-of-friends are sharing as well as curate your Twitter network

  • Recommend news stories based on your personal preference votes

    This new approach will help users getting on top of the wealth of news coming
    at them every day. 

    Ralf Herbrich

    Principal Dev Manager, Bing Mobile Personalization