Analysis of Content Popularity in Social Bookmarking Systems - Evolving Application Domains of Data Warehousing and Mining - page 237

temporal analysis of the user posting and com-
menting activity in Slashdot, a social bookmark-
ing and public discussion application focused on
technology news, revealed that the time intervals
between a post and its comments closely follow a
log-normal distribution with periodic oscillatory
patterns with daily and weekly periods (Kalten-
brunner et al., 2007a). Subsequently, the same
authors managed to predict the future Slashdot
user activity based on their past behavior by creat-
ing prototype activity profiles (Kaltenbrunner et
al., 2007b). Another related study is provided by
Cha et al. (2007) where the authors analyze the
temporal video viewing patterns in YouTube‎ 4 and
Daum‎ 5 . In line with these studies, we devote part
of this chapter to analyzing the story popularity
evolution in Digg as well as the temporal activity
profiles of its users.
Another aspect of content popularity pertains
to the correlation between the popularity of book-
marked online items (as quantified by number
of votes or hits) and their semantics, which are
conveyed by means of their textual features‎ 6 . Con-
siderable work has been carried out with the goal
of separating between different classes of content
based on machine learning methods that make
use of features extracted from their text (Yang &
Pedersen, 1997). For instance, automatic methods
based on machine learning have been devised for
differentiating between positive and negative online
product reviews (Dave et al., 2003; Pang et al., 2002;
Turney, 2002). Further text classification problems
involve the automatic classification of textual items
based on their utility (Zhang & Varadarajan, 2006)
or their quality (Agichtein et al., 2008). In part of
the case study presented in this chapter, we examine
the potential of automatically predicting whether a
given bookmarked item will become popular or not
based on its textual content. Although this problem
is very complex to tackle by means of the machine
learning paradigm adopted in the aforementioned
studies, we can establish significant correlations
between the popularity of content items and their
textual features.

Finally, the study of social network effects
on the behavior of users constitutes another
analysis perspective for content popularity in
social bookmarking applications. In (Richardson
& Domingos, 2002), evidence is provided sup-
porting the significance of network effects on a
customer's online purchase behavior. In an effort
to exploit such effects, Song et al. (2007) propose
an information flow model in order to exploit the
different information diffusion rates in a network
for improving on recommendation and ranking.
On the other hand, an empirical study by Leskovec
et al. (2007) based on an online recommendation
network for online products (e.g. books, music,
movies) indicated that there is only limited im-
pact of a user's social environment on his/her
purchasing behavior. Finally, the study by Lerman
(2007) concludes that the users of Digg tend to
prefer stories that their online friends have also
found interesting. Here, we define two measures
of social influence on (a) content popularity and
(b) users' voting behavior, conceptually similar
to the ones introduced by Anagnostopoulos et al.
(2008). Then, we carry out a set of experiments to
quantify the extent of the social influence effects
on bookmarked content popularity and consump-
tion in Digg.

SBS AnAlySIS frAMeWork

This section introduces the Diggsonomy frame-
work, which aims at facilitating the study and the
description of the phenomena arising in social
bookmarking applications. This framework was
originally presented in (Papadopoulos et al., 2008);
here, we repeat the definition of the framework.
The framework considers an SBS and the finite
sets U , R , T , S , D , which stand for the sets of us-
ers, resources, timestamps, social relations and
votes on resources respectively. Note that T is
an ordered set.
Definition 1 (Diggsonomy): Given an SBS, its
derived Diggsonomy B is defined as the tuple B