Geoscience Reference
In-Depth Information
The ad hoc spatial and temporal models that we propose are close to the
SpatialML and TimeML models. The transition of one representation to another is
made possible by model transformation, without the possibility, however, of
associating numeric representations with the RSFs and RTFs in SpatialML and
TimeML. We have made the choice of working with ad hoc models since, on the one
hand, we conduct a simplified annotation in the framework of our process flows and,
on the other, we do not have the need to exchange representations. The spatial and
temporal extraction, interpretation and indexing processes, as well as the spatial and
temporal IR models, have been implemented in an empirical way on textual samples
extracted from our corpus. These indexing process flows and the corresponding IRSs
have been tested. Like the work described in [PUR 07], our experiments
[SAL 07a, LEP 07] show that the specialized PIV IRS gives better results than a
classic thematic IRS for an IR containing only spatial or temporal criteria.
In [SAL 07a], we also show that a “rough” coupling of spatial and thematic IRSs
gives better results than a unique thematic IRS for an IR containing spatial and
thematic criteria.
This work enabled us to develop different prototype versions each corresponding
to contributions to the projects described in Chapter 1:
- Web services of spatial and temporal processing of textual documents: for the
annotation of archival documents in the GEOTOPIA (see Figures 2.10 and 2.11)
project and for the annotation of hiking reports in the GEOCIME project;
- process flows aiming at the extraction of geographic terms candidate for the
enrichment of ontologies in the GEONTO project;
- specialized spatial and temporal IRSs dedicated to information indexing and
retrievalintraveloguesforthePIVprojectandinhikingitinerariesfortheGEOCIME
project.
Letusalsomentiontwootherprototypes.Ontheonehand,afirstextensionofPIV
is dedicated to spatial indexing by patterns [LES 07]. The corresponding prototype
implements a method of indexing by classification guided by the evaluation of spatial
characteristics specific to the SFs of indexed textual units. These characteristics allow
the grouping of SFs and the association of itinerary, local description, point of view,
and comparison of places patterns with documentary units with as diverse ranges as
the paragraph, the section, the chapter, etc. The IR scenarios can now be different and
the retrieved results no longer contain only paragraphs but also document fragments
of different structure and size. On the other hand, the second prototype aims at the
automatic interpretation of itineraries [LOU 08a]. This prototype, called PIIR , takes
as the input a rough text related to a travel account and gives an interpretation in
XMLformatofthestepscorrespondingtotheitinerarydescribedintheaccountasthe
output.
Search WWH ::




Custom Search