Community Detection in Collaborative Tagging Systems - Community-Built Databases: Research and Development - page 112

they address the same problem. In the rest of this chapter, we will use the term
community detection for consistency, but in most cases it is legitimate to read it as
graph-based clustering. Similarly, a community can be thought of as a cluster.

5.2.3 Types of Communities in Collaborative Tagging Systems

There can be several kinds of communities in a Collaborative Tagging System
depending on the kind of objects that constitute them. One-mode communities, i.e.,
communities comprising objects of one kind, are by far the most commonly studied
structures within folksonomies. These communities are implicitly defined upon the
assumption that their members are similar to each other. The employed measure of
similarity is usually empirically selected based on the achieved results, as well as on
computational efficiency considerations.
Among the possible three kinds of one-mode communities within a Collaborative
Tagging System, tag communities are the ones attracting the most research interest
[ 5 , 14 - 18 ]. The problem of tag community identification is attractive mainly for two
reasons: (a) tag communities usually correspond to semantically related concepts or
topics, which make them suitable for a series of applications (see Sect. 5.2.5 ), (b) the
number of unique tags is limited, 5 which makes the problem of tag community
detection easier from a computational point of view. It is also for these reasons that
tag communities constitute the main focus of this chapter.
Before the advent of Collaborative Tagging Systems, the problem of resource
(web page) community identification was tackled by means of web graph [ 6 , 7 ] and
content-based [ 26 ] analysis techniques. In view of the wealth of user input that is
now possible in the context of Collaborative Tagging Systems, the discovery of
resource communities attracts once again significant research interest [ 11 - 13 ].
Collaborative tagging has been particularly valuable for resources with no textual
component, for example, images and videos.
Detecting user communities in tagging systems is probably the least studied kind
of folksonomy communities. This is mainly due to two reasons: (a) users are typi-
cally characterized by multiple interests; therefore, it is hard to group them into
communities, unless overlapping community detection methods are devised;
(b) even when communities of users are identified within a tagging system, there
is no straightforward means of evaluating the quality or utility of the identified com-
munities. Recently, some interesting work has appeared [ 9 ] that attempted to cluster
users of a Social Tagging System based both on the tags that they use and on the
temporal patterns of their tag usage. Such analysis could be potentially interesting
for the administrators of a tagging application, for example, for improving the effec-
tiveness of online advertising.

5

Even though tags are allowed to have any form, there are practically a finite number of distinct
tags and when tag filtering and normalization techniques come into play, this number is even
lower.