Community Detection in Collaborative Tagging Systems - Community-Built Databases: Research and Development - page 110

Fig. 5.1 Tag co-occurrence network formation example. In case of (i) tag co-occurrence is
considered in the context of resources, while in (ii) it is considered in the context of both resource
and user

are variations in how tag co-occurrence is computed. For instance, in [ 5 ],
tag co-occurrence is also defined in the context of both a user and a resource,
i.e., c ij ¼ jf9

. These two different tag co-
occurrence definitions are exemplified in Fig. 5.1 . Alternative tag similarity
measures are also possible to use for building a tag graph, for example, tag context
similarity [ 5 ] and FolkRank [ 2 ]. Independent of the measure used to build the tag
graph, the tag neighborhood operator is defined around an input tag t 0 , N ( t 0 )

u 2 U ; r 2 R jð u ; r ; t i Þ; ð u ; r ; t j Þ2 Y gj

¼ { t i ∈

T j ( t 0 , t i ) ∈ E T }. This returns all tags that co-occur with t 0 on G .

5.2.2 Related Work

The concept of community in online systems is extremely wide. In the past, it has
been commonly used to denote communities of web pages [ 6 , 7 ] and web users [ 8 ]. In
these works, communities are defined with reference to an underlying network (e.g.,
the Web graph [ 6 , 7 ]) and correspond to the result of community detection ,i.e.,a
process that discovers groups of nodes in the network that are more densely connected
to each other than to the rest of the network. Naturally, since Collaborative Tagging
Systems comprise three types of entities ( U , R , T ), one should expect that even in
the specific context of folksonomies, the notion of community is still wide in scope.
To this end, we present a classification of pertinent works along three dimensions:
(a) type of node, (b) application scenario, and (c) method. These dimensions are briefly