figure 5. Prediction performance is based on the logarithmic growth model
measured by the average relative squared error function for (a) Digg and (b) you Tube,
respectively. The shaded areas indicate one standard deviation of the individual
submission errors around the average.
0.5
Relative squared error
0.4
0.3
0.2
0.1
0
0
5
10
15
20
Digg story age (digg hours)
(a)
1
0.8
Relative squared error
0.6
0.4
0.2
0
0
5
10
15
20
25
30
(b)
you Tube video age (days)
these constraints must be considered
to achieve the minimum error possible
allowed by the model.
Social networking
Social networking features in Web 2.0
services are so ubiquitous it is almost
mandatory for a site to offer them to its
users. For example, Digg’s approach to
social networking is to make it possible
for users to be fans of other users, after
which they are able to see what stories
their “idols” submit or digg. This is
essentially a restricted form of collaborative filtering, but users themselves
select the peers they wish to follow. A
similar kind of social network is active
in You Tube, though the feature that allowed users to follow the videos their
friends were watching was nascent in
2008; however, they might have seen
if friends recently uploaded videos.
Due to the limited nature of social-networking options on YouTube in 2008,
we focus on the network of Digg users.
Together with content-popularity data,
we also collected link information using the Digg API. Figure 6 shows a typical snapshot of the Digg social network
in 2007, with about 260 users and 550
links, where a link represents whether
a particular user is a fan of another
user. Users who dugg a particular story
are in red, with no apparent clustering
among them. However, these users are
relatively dense in the neighborhood of
the small social graph in Figure 6, since
the story attracted nearly 15,000 diggs
altogether, considerably more than the
average submission at the time.
It was known that the Digg social
network plays an important role in
making a story visible and popular
when the submission is still in Digg’s
“upcoming” section, with new stories
appearing at the top of the “upcoming”
page on average every ninth second, as
in Figure 2, with about 400 new submissions an hour in 2007. Though all
new submissions are shown in the “
upcoming” section, the list is updated so
quickly that entries left the first page in
about two minutes. The most effective
way to discover new stories should thus
be through the social network, where
recent diggs of a user’s idols are visible
for more time on the user’s personal
page. To what extent then, do diggers
pay attention to what their idols already dugg?
To see how Digg social networking
functioned we took all submissions
for which we had data for at least 12
hours after promotion and measured
the fraction of diggers with at least
one digger among their idols and who
had already dugg the same story. In essence, this measurement is the probability that a new digg is made by users
who may have seen the story through
their social networks. We normalized
the times of diggs with respect to the
promotion time of the individual submissions, so for diggs made before promotion, time is measured backward.
Results are outlined in Figure 7, where
about 20% of diggers have an idol who
dugg the same story before they did,
when it was still in the “upcoming”
phase. However, this figure drops considerably (to 7%) after promotion; most
diggs are cast by users who could not
have seen the submission in their social network before. This falloff in peer
following supports the assumption
that stories are found through the social network in the “upcoming” phase,
but once they are promoted to the front
page and exposed to a diverse audience
for a longer time, the effect of the social network becomes negligible.
While users are about three times
more likely to digg a submission their
idols dugg in the “upcoming” phase
than after it was promoted, the measurement only intuitively suggests that
users pay attention to the activities of
their peers. To determine whether diggers are truly influenced by their social peers, the null hypothesis for user
