Information Technology Reference
In-Depth Information
domain. Thus, the organization should perform a few initial retrieval experiments
and configure empirically this threshold, considering the final results of each re-
trieval effort, which are sorted and presented with their relevance indices.
Class families and properties in an ontology affect the final calculations. In ad-
dition, the number of concepts of a domain could be higher than of others, even
for comparable topic sets, concerning relevance distribution. Nevertheless, even if
the final results are values between 0 and 1, for ontologies representing the same
domain or different ones, they will be specific.
6.2 Discrepant Weights
During the tool assessment, it was possible to observe a false-positive case pro-
duced by the retrieval algorithm that is very interesting. A topic part that has no
relevance at all, considering the ontology, was one of the firsts in the relevance
ranking that was produced.
Analyzing the case, the conclusion was that the problem was due to the pres-
ence of one isolated keyword in the topic that was spelled the same way as a con-
cept that was present in the ontology. Coincidently this concept does not appear
anywhere else. This case caused the associated
idf
to be very high, which influ-
enced the construction of the weight vector for the document equivalent as well as
of the query equivalents vectors. As each concept in the vector represents a coor-
dinate in a multiple space, considering both vectors, the correspondent dimension
was discrepant to the other concept dimensions, in such a way that the cosine of
the angle formed between the correspondent vectors had a very high value.
This discrepant case points out the necessity to include some kind of treatment
in the retrieval algorithm that could avoid highly discrepant weights.
6.3 Distinction between Class Families
In the proposed ontology structure there is no way to specify different weights for
different class families. If implemented, such functionality will become very inter-
esting because it is acceptable that each class family represents an information sub
domain and thus it can be more or less relevant than the other families.
The inclusion of class weights could aggregate a refinement to the information
retrieval mechanism that is used by the implemented tool.
7 Conclusions
In this chapter it is presented an approach to perform semantic information re-
trieval upon wikis. The idea was to provide a tool to follow up news or participa-
tion on consumer discussions. The wiki should contain articles and discussions
that are inserted continuously during a time frame, but its ideas can be ported to
other social media, such as blogs and discussion lists.
The proposed tool can be used in several other scenarios where information re-
trieval is necessary or can be used for improvements. The main differential
to other similar tools and mechanisms is manifold: the semantic nature of the
