contributed articles
Doi: 10.1145/1787234.1787254
Early patterns of Digg diggs and YouTube
views reflect long-term user interest.
By GaBoR SzaBo anD BeRnaRDo a. huBeRman
Predicting
the Popularity
of online
content
the eAse oF producing online content highlights
the problem of predicting how much attention any
of it will ultimately receive. Research shows that user
attention9 is allocated in a rather asymmetric way, with
most content getting only some views and downloads,
whereas a few receive the most attention. While it
is possible to predict the distribution of attention
over many items, it is notably difficult to predict the
amount that will be devoted over time to any given
item. We solve this problem here, illustrating our
approach with data collected from the portals Digg
( http://digg.com) and You Tube ( http://youtube.com),
two well-known examples of popular content-sharing-and-filtering services.
The ubiquity of Web 2.0 services has transformed
the landscape of online content consumption. With
the Web, content producers can reach an audience
in numbers inconceivable through conventional
channels. Examples of services that
have made the exchange between producer and consumer possible on a
global scale include video, photo, and
music sharing, blogs, wikis, social
bookmarking, collaborative portals,
and news aggregators, whereby content is submitted, perused, rated, and
discussed by the user community.
Portals often rank and categorize
content based on past popularity and
user appeal, especially for aggregators,
where the “wisdom of the crowd” provides collaborative filtering to select
submissions favored by as many visitors as possible. Digg is an example,
with users submitting links to and
short descriptions of content they have
found on the Web and others voting on
them if they find them interesting. The
articles attracting the most votes are exhibited on the site’s premiere sections
under headings like “recently popular
submissions” and “most popular of
the day.” This placement results in a
positive feedback mechanism leading
to rich-get-richer vote accrual for the
very popular items, though the pattern
pertains to only a small fraction of the
submissions that rise to the top.
Besides Digg, anyone with Internet
access can watch You Tube videos, reply
to them through their own videos, and
leave comments. The way the online
ecosystem has developed around YouTube videos is impressive by any standard, and videos that draw millions of
viewers are prominently displayed on
the site, like stories on Digg.
Content providers, Web hosts, and
advertisers all would like to be able to
predict how many views and downloads
individual items might generate on a
given Web site. For example in advertis-
key insights
Site administrators, advertisers, and
providers would all find it useful to be
able to predict content popularity.
Prediction is possible due to the extreme
regularity with which user attention
focuses on content.
early patterns of access indicate long-
term popularity of content.
