Geoscience Reference
In-Depth Information
is presented in [PAL 09], [PAL 10a] and [PAL 10d], as well as in [SAL 12]. These
studies answer the following questions:
- Does the approach of standardization have an interest? Does it induce a too
significant loss of precision capable of very strongly degrading the quality of results
retrieved by a spatial or temporal IRS?
- Does a tiling range or a weighting formula differentiate itself in the particular
context of our corpus?
- Does a different weighting formula have to be used for the type of relationship
(spatial/temporal adjacency, inclusion, orientation, etc.) processed?
- Does the use of an index adapted to the spatial or temporal scale of the query
improve the results?
3.5.2.1. Evaluation of the spatial IR
Thisanalysisconsistsofevaluatingtheinterestofspatialstandardization.Forthis,
we have compared our standardized spatial PIV
2
IRS with the initial spatial PIV IRS:
40 topics containing ASFs and RSFs of different ranges were tested.
Figure 3.9. Comparison of the PIV and PIV
2
spatial IRSs
AswecanseeinFigure3.9,PIV
2
givesresultssuperiorinprecisiontothoseofPIV
by 13%. The loss of precision induced by tiling is balanced by the use of frequency
calculations.
3.5.2.2. Evaluation of the temporal IR
Thisanalysisconsistsofevaluatingthetemporalstandardization.Asforthespatial
information, we have compared our PIV
2
standardized temporal IR system with the
initial PIV temporal IR system: 35 topics containing ATFs and RTFs of different
ranges have been tested.
A summary of the results presented in Figure 3.10 shows that the temporal
standardization associated with PIV
2
gives similar results to those of PIV (−0.5%).
Contrary to spatial, temporal information is less in number and paragraphs
containing several temporal references are rare. This indeed explains that the use of
Search WWH ::
Custom Search