Geoscience Reference
In-Depth Information
3.3. Standardization and combination of criteria: literature review
The combination of criteria is an issue that mainly appears in the fields of IR
and decision support systems (DSSs). To avoid a number of possible biases during
the combination phase, a preliminary phase of criteria standardization is generally
proposedintheapproachesrelativetoeachofthesefields.Indeed,theheterogeneityof
themodelsofdatarepresentationandcriterionevaluationinvolvedinthecombination
has to be taken into consideration.
3.3.1. Criterion standardization
We are interested in the approaches to standardization in the fields of IR,
multimedia IR, DSSs and GIR.
3.3.1.1. Normalization and IR
Normalization consists of bounding the relevance score of a document calculated
with respect to the given query. In IR, these scores are mostly bounded between 0
and 1 [KOW 97]. If a query is multicriteria, each of the criteria has to be bounded in
order to be combined later with the others. In a keyword search, a user can specify
several criteria, in other words several possibly constrained words. The IRS will then
perform a search for each criterion and then build a result list composed of ranked
documents. A normalized score (average of the scores, for example) is associated
with each document. In this particular case, given that the data representation model
is unique and the same IRS is used to process each of the criteria, there is no problem
of compatibility between the obtained results.
3.3.1.2. Generalization and multimedia IR
Generalization corresponds here to an approach of simplification aiming at the
reduction of a very large quantity of details. Several approaches allow us to achieve
this simplification. The first is identification: based on multimedia data, the
concepts [WOL 92] (a concept corresponds to a meaningful piece of information, a
category such as car, person and house) present in the document have to be
determined. The second is regrouping/focusing: based on multimedia data, the
information detected in the document has to be regrouped in order to be associated
with more general representation frameworks. For example, the so-called
administrative hierarchy of places (communes, counties, regions, etc.) can be used to
simplify the representation of spatial information [LIE 07, MAR 08b].
The identification approach to generalization is used to determine the concepts
contained in a document. It is necessary beforehand to define the list of suiting
Search WWH ::
Custom Search