Multicriteria Information Retrieval in Textual Corpora - Geographical Information Retrieval in Textual Corpora - page 54

framework is destined to evaluate spatial, temporal and thematic IRSs or any
combination of such IRSs. It provides a test collection in French language adapted to
the GIR and, in the medium term, could lead to the proposition of a specific task in
the framework of an evaluation campaign such as G EO C LEF [GEY 05].
The contributions described in this chapter are the result of study conducted on
the framework of the T2I team of the laboratory of LIUPPA and, in particular, of the
CIDRI, PIV 2 and PIV 3 projects. These projects are detailed in [PAL 10d], [PAL 11],
[PAL 12a] and [SAL 11] and in the thesis of Palacio [PAL 10a]. They are also the
result of collaborations with Guillaume Cabanac and Gilles Hubert, researchers at
the IRIT involved in IR activities: together, we have, in particular, set up a GIRS
evaluation framework which we have applied to the multicriteria GIR prototype PIV 3
[PAL 10c, CAB 11].
This chapter is organized into six sections. Following this introduction,
section 3.2 recalls the research questions raised in Chapter 1, which will be studied
here. Section 3.3 presents the propositions found in published literature dealing with
the issues of criterion standardization, on the one hand, and criterion combination, on
the other. Section 3.4 details the multicriteria GIRS PIV 3 that integrates our
propositions for indexing by tiling and the multicriteria IR. Section 3.5 describes the
GIRS evaluation framework we have designed as well as the results of the evaluation
of the GIRS PIV 3 . The summary of our contributions and perspectives are presented
in section 3.6.
3.2. Review of challenges, hypotheses and research objectives
Few research studies are interested in the combination of classic IR services with
specific services dedicated to spatial IR and temporal IR. We can mention Alonso
et al. [ALO 11] who propose the use of the temporal information annotated in texts
in order to constrain a search engine: a query is now composed of a textual part
completedwitha temporal part.Alonso et al. illustratethisproposalwiththeexample
of the query “world war” temporally constrained by the criterion “1939-1945”.
The search thus targets the documents dealing with the Second World War. Alonso
et al. consider that a query combining textual and temporal search criteria has to be
processedinthefollowingway: thehigherthelevelofsatisfactionofthesecriteria,the
higherthecorrespondingscoreofthedocument.They,however,raisethreequestions:
- How should a combined score coming from textual and temporal search criteria
be calculated reasonably (side effects are probable)?
- Doesa documentinwhichthe textual and temporal elementsofcorrespondence
are distant have to be penalized?
- How should a document, which entirely satisfies textual criteria but little or no
temporal criteria, be weighted and vice versa?