Multicriteria Information Retrieval in Textual Corpora - Geographical Information Retrieval in Textual Corpora - page 81

tile frequency for the results score (RSV) computation does not contribute to the
improvement of their precision.
Figure 3.10. Comparison of the PIV and PIV 2 temporal IRSs
3.5.2.3. Summary of the evaluation of PIV 2
Spatial standardization coupled with the usage of classic IR formulas in PIV 2
improvestheresultsofthespatialPIVIRapproach.Temporalstandardizationcoupled
with the usage of classic IR formulas in PIV 2 has not shown significant changes with
respect to the temporal PIV IR approach. To answer the three other questions, the
experiments described in [PAL 10a] have shown that:
- communaltiling(respectivelymonthly)coupledwiththeTF p weightingformula
gives the best results;
- no matter the spatial (respectively temporal) relationship involved in the
information retrieval, the TF p weighting formula gives the best results;
- theuseofanindexadaptedtothespatialortemporalscaleofthequeryimproves
the results.
3.5.3. Evaluation of the multicriteria IR applied to tiling
We have compiled a test collection MIDR_2010 3 containing a corpus, topics, the
corresponding qrels and geographic resources [PAL 10c]. We have based our study
on this repository to evaluate the multicriteria IR according to the CombMNZ model
(foratotallycompensatoryaggregation),ontheonehand,andaccordingtotheCMRP
model (for a partially compensatory aggregation), on the other.
3.5.3.1. Evaluation of the totally compensatory multicriteria IR
This experiment has been carried out on the MIDR_2010 test collection for 44
topics containing spatial, temporal and/or thematic criteria. We have evaluated the
PIV 3 GIRS implementing the CombMNZ totally compensatory aggregation model:
3 This is a French test collection accessible at http://t2i.univ-pau.fr/MIDR.