Databases Reference
In-Depth Information
Figure 2.8: Average semantic distance from the original tag to the most closely related one [Cattuto
et al., 2008].
al. find out that the tags obtained via tag context or resource context are synonyms or siblings of the
original tag, while FolkRank and the tag co-occurrence measure rather provide more general tags.
In [Passant, 2007], Alexandre Passant addresses the limitations of free tagging systems and uses
Semantic Web technologies to solve some of these problems. In particular the author identifies the
following limitations of free tagging systems:
•
Tags variation:
If syntactically different tags describe the same concept, it is problematic to create
the semantic connection between these tags. For example, it is dicult for a tagging system to
discover the connection between the tags “high definition” and “HD”. Variations can also be caused
by simple typo errors.
•
Tags ambiguity:
If one tag can describe different concepts, the system will again not able to make
any difference. The tag “apple”, for example, can stand for the fruit or the company with the same
name.
• Flat organization of the tags:
In contrast to ontologies, for example, tags do not form any hierarchy.
Again, the semantic relations between tags have to be discovered first, e.g., by using data mining
algorithms.
In order to tackle these problems Passant proposes to mix Social Web folksonomies and Semantic
Web ontologies. The idea is to link tags to ontology concepts to enhance an information retrieval engine
for blog-posts.
In [Mika, 2007], a unified model is presented which covers both social networks as well as semantics.
The idea is to extract lightweight ontologies from a folksonomy to better model the concepts of a particular
community. For this reason, the traditional bipartite model of an ontology is extended by the user
dimension, leading to a tripartite model of actors (users), concepts (tags), and instances (items) which
basically corresponds to the tripartite model of a folksonomy. The author shows how two lightweight
ontologies based on overlapping communities (
O
ac
) and overlapping sets of items (
O
ci
) can be extracted
from the unified model. The network of associations
O
ac
, for example, is built by only considering the
associations between actors and concepts. The
O
ci
network, on the other hand, focuses on the associations
between concepts and instances.
