Multicriteria Information Retrieval in Textual Corpora - Geographical Information Retrieval in Textual Corpora - page 55

We place our issues (see Chapter 1) in this same framework for reflection. Our
objective is to combine a classic IRS with spatial and temporal IRSs. The
heterogeneity of the corresponding models of information representation and
retrieval does not allow us to directly consider the combination of such systems. For
example, the aggregation of results coming from different models of IR raises
problems. Thus, the first question dealt with in this chapter is: What core models of
information representation and retrieval should be proposed in order to prepare for
the combination? which proposes a preparatory phase in advance of the result
combination phase itself. Consequently, a phase of model standardization will be the
basis of the controlled combination approaches.
Finally, the combination has to be based on aggregation operators adapted to the
context. Let us not forget the need for expressiveness mentioned earlier. This need is
partially covered by the operators proper to each dimension. It nevertheless has to be
completed by a finer formulation of each criterion: is it mandatory, is it associated
with a degree of preference, is it a rejection criterion? The question: What advanced
aggregation operators should be proposed and implemented for a multicriteria IR
combining several IRSs? will thus be dealt with in a second stage.
The retained working hypotheses are the following:
- The standardization of data representation is adapted to spatial and temporal
information.
- Thestandardizationassociatedwithawell-triedIRmodeldoesnotinvolvealoss
of precision with respect to the initially dedicated IRSs.
- A “classic” coupling (such as an arithmetic mean, for example) of spatial,
temporal and thematic IRSs gives better results than a thematic IRS alone or two-
by-two pairings of spatial, temporal and thematic IRSs.
- An “advanced” coupling (offering more power of expression to the user) of
spatial, temporal and thematic IRSs gives better results than a “classic” coupling.
The research objectives corresponding to the announced issues aim mainly at the
design and implementation of:
- a model of standardization of indexed data representation;
- a model of multicriteria IR.
These hypotheses are some of the avenues of study that we havefollowed in order
to implement a GIR fully exploiting the spatial, temporal and thematic dimensions.