Multicriteria Information Retrieval in Textual Corpora - Geographical Information Retrieval in Textual Corpora - page 79

dimensionsofgeographicinformation.Thisisthereasonwhyweproposethecreation
of a collection including the following:
1) Topics coveringallorapartofthethreedimensions.Forexample,a topic could
be entitled “Transhumance in the Pyrénées in the 19 th Century” with the narrative
“Will be considered relevant the documents invoking transhumance or related events
(daily life of a shepherd in the pasture) in the Pyrénées mountains between the years
1800 and 1899”.
2) A corpus dealing with the three dimensions: the conventionally considered
thematic aspect is completed by spatial and temporal elements.
3) Qrels by dimension for which the assessor evaluates the consistency between
each of the three dimensions (thematic, spatial and temporal) and the document. The
mere presence of the three dimensions in the document is not sufficient to deduce its
relevancy for the query. Let us consider, for example, the case of a document dealing
with the spa industry, then mentioning “Gavarnie” as the birth place of the narrator.
Although spatially relevant, it does not match the query “spa industry in Gavarnie”.
Such a subtlety requires the assessment of the global match between the query and
the document. In judging a given document, the assessor first evaluates its adequacy
according to each of the three dimensions. Not to overwhelm assessors, we opted for
a per dimension binary judgment: a document is either relevant or non-relevant to the
considered query and dimension. This rationale is akin to the conclusions of Bucher
et al. [BUC 05] who underline the fact that the dimensionally gradual assessments
are “unnecessarily cumbersome”. Finally, considering the three per-dimension binary
judgments, as well as the aforementioned global binary judgment, we compute the
document relevance value v ∈ {0; 1; 2; 3; 4}. This both represents the number of
satisfied dimensions (from 1 to 3) and global relevance (4). No assumption was made
regarding the relative importance of dimensions: they were equitably considered.
4) Geographic resources necessary, on the one hand, for the geo-referencing of
the SFs and the interpretation of TFs contained in the corpus, on the other.
Itisonthisbasisthat,inthefollowingsection,wedescribethetestcollectionsand
experimental protocols implemented to evaluate the PIV 2 and PIV 3 prototypes based
on the approach of indexing and IR by tiling.
3.5.2. Evaluation of the spatial and temporal IR applied to tiling
Tovalidateourdifferentpropositions,wehavesetupseveralexperiments.Wehave
evaluated the interest of our standardization approach mainly concerning the loss of
precision induced by tiling. For this, we have compared our spatial PIV IRS with
the PIV 2 IRS integrating the process of spatial standardization. Following the same
principle, we have compared the temporal PIV IRS with the PIV 2 IRS integrating
the process of temporal standardization. A detailed description of these experiments