General Conclusion - Geographical Information Retrieval in Textual Corpora - page 98

Figure 4.7. Example of a search targeting documents invoking “the inn of
Platane in the commune of Pagolle and the musical instruments” - “txistu” is
a flute defined in the corresponding musical domain ontology
This a posteriori analysis is justified for several reasons:
- The in-depth analysis, a priori, of the document repository requires too much
significant resources; the GIR system that we have proposed is based on targeted and
lightnaturallanguageprocessing(NLP)toolsaswellasIRtoolstocreateafirstresult
list.
- The sentences to be analyzed as well as the relevant spatial, temporal and
thematic features within these sentences are only known after the execution of the
query; this results in a new targeted analysis of the features and of their relationships
performed a posteriori.
The information to be analyzed can be found in the same sentence or in different
sentences (Figure 4.7 cases 1.a, 3.a and 3.c as well as cases 1.b, 2.a, 3.b and 3.d,
respectively). The tools used in the analysis will consequently have to be adapted
to the analysis of elements located within a single sentence or a series of sentences,
which we call the discourse.
The working hypothesis relative to the two following sections is therefore that
the existence of a semantic relation between the features involved in the selection
of the document reinforces its relevance. The score of a document now reflects its
global relevance with respect to the query rather than the weighted sum of the scores
corresponding to each of the search criteria evaluated independently.