Multicriteria Information Retrieval in Textual Corpora - Geographical Information Retrieval in Textual Corpora - page 85

Figure 3.13. Illustration of the multicriteria IR supported by PIV 3 prototype -
“hikes, (+) to the south of Lourdes, (ˆ0.8) in the 1890s”
These studies have allowed us to develop different prototype versions mainly
contributing to the CIDRI project, described in Chapter 1 of this topic:
- a model of generalization of indexed data representation for multicriteria IR;
- amodelofmulticriteriaIRfortakingintoaccountgeographicinformationneeds
with a strong power of expression;
- corresponding process flows for the implementation and testing of PIV 2 and
PIV 3 IRSs.
An initial experiment based on geographic queries and the MIDR_2010 test
collection shows that, in terms of precision, the PIV 3 GIRS associated with the
CMRP model surpasses the configuration integrating the CombMNZ [FOX 93]
model by +54.16%. A final experiment based only on thematic queries of TREC-8
shows that PIV 3 associated with the CMRP model improves the precision of search
results in a significant way (+26.96%) in a context that implements expressive topics.
These studies are based on a first set of results presented in Chapter 2 and are the
fruit of numerous collaborations having mainly involved Guillaume Cabanac and
Gilles Hubert, academic researchers at the IRIT, and Damien Palacio, a doctoral
student at the LIUPPA.