Database Reference
In-Depth Information
Table 5.1 Types of folksonomy networks and respective application scenarios
Type of node
Method
Application scenario
U[
9
]
k
-means [
9
],
Social link prediction [
10
]
R[
11
-
13
]
Hierarchical Agglomerative
Clustering [
14
],
User profiling [
12
], Personal content
organization [
12
], Resource
clustering [
13
]
T[
5
,
14
-
17
]
Spectral clustering [
13
],
Community detection
[
5
,
16
,
17
]
Personalized retrieval [
14
], Ontology
evolution [
18
], Search and
navigation [
15
], Tag sense
disambiguation [
5
]
U, R [
19
]
Content recommendation [
13
]
U, T
Fuzzy Biclustering [
19
],
Co-clustering [
20
],
Biclique detection [
21
]
Collaborator Recommendation [
22
],
Content recommendation [
22
]
R, T [
20
]
Tag recommendation [
18
], Resource
(image-page) organization [
20
]
U, R, T [
23
,
24
]
Tripartite clustering [
23
],
Hypergraph clustering [
24
]
User-Resource-Tag recommendation,
Study of Collaborative Tagging
Systems
explained below and further analyzed in the following sections. Table
5.1
presents a
summarized view of the classification.
Type of node
: Depending on the kind of objects that they contain, there is a
distinction between several kinds of communities in Collaborative Tagging Sys-
tems. First, there are three one-mode communities (user based, resource based, and
tag based). In addition, there are three two-mode communities (user-resource,
user-tag, resource-tag). Finally, it is possible to have communities comprising all
three kinds of folksonomy entities (user-resource-tag).
Method
: There is a multitude of methods used to derive folksonomy commu-
nities. Methods range from conventional clustering schemes, such as
k
-means and
Hierarchical Agglomerative Clustering, to state-of-the-art community detection
methods, such as modularity optimization and biclique detection. For two-mode
and three-mode folksonomy networks, more elaborate methods need to be applied,
such as co-clustering and hypergraph clustering.
Application scenario
: The results of community detection can be exploited in the
context of several data analysis tasks. In most cases, community analysis results
are incorporated in recommendation algorithms, for example, friend/collaborator
recommendation, resources (web page/image) recommendation, and tag suggestion.
Additional application scenarios include tag sense disambiguation, personalized
content search and navigation, as well as ontology evolution and population.
Note that the term “community detection” is frequently used to refer to
graph-
based clustering
algorithms [
25
]. In essence, these two fields have a common
objective: the identification of groups of nodes in a network that form
natural
groups
. It is due to their different scientific origin (social network analysis and
statistical physics for community detection and computer science for graph-based
clustering) that these two fields sometimes appear as different despite the fact that
Search WWH ::
Custom Search