General Conclusion - Geographical Information Retrieval in Textual Corpora - page 90

- A “classic” coupling of spatial, temporal and thematic IRSs supported by the
CombMNZ model gives better results than a single thematic IRS or a two-by-two
pairing of spatial, temporal and thematic IRSs.
- An “advanced” coupling of spatial, temporal and thematic IRSs, supported by
our partially compensatory aggregation model called CMRP, gives better results than
a “classic” coupling. It offers, in addition, more power of expression to the user.
4.2. Perspectives
The central element of our concerns is thus the geographic information illustrated
at the beginning of this topic by Figure 1.2 in section 1.2.2. We consider geographic
informationasamoleculenotonlycontainingaspatialcomponent,butalsoatemporal
component and a thematic, or “phenomenon” component [USE 96, GAI 01].
Let us recall that our work's focus is accessing the content of non-structured
geographic textual corpora. The experiments relative to the different versions of our
prototypes, carried out on such corpora, have allowed us to observe information
making and reference to places spatial indications, as well as descriptions of
landscapes, temporal indicators and dates, implying a pronounced interest of these
documents for the geographic aspect. We have mainly worked on travelogues (see an
excerpt in Figure 1.1 in section 1.2.2) in which the authors frequently use an identical
logical structure: the text is divided into paragraphs describing a part of their travel.
Each part can consist of a description of an itinerary, a stage, a point of view, an
observation and/or an event, etc. Although the retrieval and availability of
information from travelogues remains a current issue [HAO 10], our results can,
nevertheless, be applied to other types of corpora such as tourist guides, hiking
guides as well as travel description blogs whose structure of discourse also presents a
certain homogeneity.
Numerousapproachesarefocusedonthe thematic dimensioninIRtoday,whereas
others, also on the increase, associate the spatial, temporal and thematic dimensions
in geographic information retrieval (GIR). Inspired by successive GIR conferences
[PUR 11], here is a non-exhaustive list of open research problems in these fields:
1) Specification of data models taking into account the representation of relations
between the spatial, temporal and thematic dimensions [PER 07, EZZ 11, MAT 11,
HOF 12].
2) Emergence of domain ontologies for indexing documents, contextualizing
queriesorguidingmatchings[LIN 06,BAZ 07,DUD 10,MEI 10,POO 10,DAO 11,
FER 11, BAS 12, LAT 12, OHT 12].
3) Detection of geographic references in the form of place names or calendar
features, and their corresponding qualifiers in natural language within texts [PUR 11,
LEI 11].