Home' Technology Review : March April 2010 Contents TR 57
WWW. TECHNOLOGYREVIEW. COM
such data, for a fee, is clearly something Twitter could start doing,
Feld says. He adds that the company could eventually sell keyword-
based ads as well.
Though Twitter aspires to be the planet's pulse, its own physiology
is a bit weak in some areas. The company has no apparent owner-
ship rights to the basic technology for microblogging; its only real
assets are its brand and user base. And while Twitter data in theory
should be salable to companies, the marketplace is still skeptical.
A recent survey by Kognito, a business intelligence firm, found
that only 14 percent of market research companies surveyed had
immediate plans to mine social-networking data at all.
Twitter needs to win more eyeballs, motivate users to tweet more,
and make sure the most useful tweets reach people who might ben-
efit from them. The company understands this. "We are coming out
of a year, 2009, which was really about scaling up," Williams says.
"We've gotten really good about getting [employees] in here and
showing that there is interesting scale to work with here. Just like
people who are thinking about using Twitter, people coming to work
at Twitter thought it was some trivial thing---'What are you having
for lunch?'---not a global real-time way of finding out what is hap-
pening right now. As people learn that it fills a role we actually need
filled, we will attract more engineering talent and more users."
The deals with Google and Bing made Twitter profitable. But
they are also a means to another end: Twitter skeptics might well
be won over if their Web searches start turning up useful tweets.
By the same token, a tweet that lands high in Web search results
will encourage the person who sent it to keep making timely and
useful observations, says Dan Weld, a computer scientist at the
University of Washington, Seattle. "When people are able to search
tweets more e ectively, it will change the content," he says. "Peo-
ple's behavior will be a ected. But this will require real-time dis-
tribution and highly e ective search."
Making real-time search e ective is tricky, though. It's not
enough just to give searchers the newest tweets that happen to con-
tain a requested keyword. The reputation of the twitterer means
something; if you want fresh information about the Haiti earth-
quake, for example, you'd like to hear it from responsible sources,
not just from anyone who happened to include the word Haiti in a
tweet. As a first step, Google evaluates tweets in part with a tech-
nological analogue to its PageRank technology, which analyzes
the link structure of Web pages to judge their relevance. Generally
speaking, the more links to a page---and the more pages linked to
the linkers---the more relevant Google's search engine considers
it. Similarly, Google concludes that the more people who "follow"
a Twitter user---and the more people who follow those people---the
more credible and relevant his or her tweets probably are.
But such e orts are only the beginning. Consider a real-time
search for "iPod." An engineer doing such a search might be seek-
ing insights into its software, a high-school student might be most
interested in friends' opinions or whether the gadget's retail price
has dropped, and a music executive might want to look for trends
in what kinds of music people are downloading. Finding out what
specific people want might require some analysis of their social
networks, their past tweets, and the tweets of the people they fol-
low, says Eugene Agichtein, a computer scientist at Emory Uni-
versity who is doing research on social search.
The location from which a tweet was issued can be an enor-
mous help. Messages from mobile devices with GPS receivers
can include location information. Twitter began allowing such
information to be attached to tweets last summer, and Google
and others are exploring ways to use this data to provide more rel-
evant real-time results. "If you follow me and know that I work in
Mountain View or live in Menlo Park, you just assume that when
I publish something---'I saw five fire engines'---you know it's in the
general Bay Area," says Dylan Casey, Google's product manager
for real-time search. "But that comment would become even more
powerful if you knew the exact geolocation."
Twitter itself recently refined its home page's "Trending Topics"
feature---a compendium of the most common phrases appearing
in messages---to allow twitterers to see what subjects are being dis-
cussed where they live. (Twitter determines the location partly on
the basis of twitterers' IP addresses or their reported home cities.)
The new feature, called "Local Trends," is the next logical step in
making real-time search more relevant and interesting. "Search isn't
just a box and a button; it's about serendipity," Williams says. "It's
about hopefully, and ideally, surfacing information to you that you,
as a twitterer, didn't ask for, but wanted---and right at that moment.
In the world of real-time search, the Holy Grail is that you anticipate
what the user wanted. Our opportunity and challenge is solving that
problem---thinking of how users behave in the ecosystem. It gets
you much quicker to realizing the value of Twitter."
There is no one magical algorithm that provides serendipity
without spam, relevance without rubbish, to all people in all
places. But Twitter and other companies see tremendous oppor-
tunities in the ocean of tweets---and, more generally, in mining
the social Web for information that people want. "What's really
important here is the notion of the social Web and using your
social network---people that you trust---for information discovery,"
says Feld. "We've got a long way to go here, which is really excit-
ing for any entrepreneur playing in this domain." Even if the
future business model of Twitter is not certain, the fast-changing
nature of the Web itself provides evidence enough that one will
emerge---and might even represent the next seismic shift in the
DAVID TALBOT IS TECHNOLOGY REVIEW'S CHIEF CORRESPONDENT.
Links Archive January February 2010 May June 2010 Navigation Previous Page Next Page