Geoscience Reference
In-Depth Information
(d R), it is not necessary to calculate the score because the document does not
have to be retrieved (RSV (d)=0).
0
if d R
i=1
p
i
|p
i
>0
·
i=1
p
i
· m
i
(d, c
i
) else
RSV(d)=
[3.2]
1
In summary, we have proposed a new multicriteria IR model that extends those
implementedincriterionaggregation,illustratedin[FAR 08].CMRPpresentsseveral
advantages:
- for each criterion, matching functions (m
i
(d, c
i
)) depend on dimension-
dedicated IRSs;
- for each criterion, expressiveness (i.e. requirement or preference (p
i
)) is
independent of the invoked IRS;
- for each query, the aggregation of the results is partially compensatory: the
RSV (d) score computation takes into account the requirements and the levels of
preference associated with each criterion by the user.
We present hereafter the PIV
3
platform that supports the implementation of
different aggregation models such as CombMNZ [FOX 93], PSM [DA 09, DA 12]
and CMRP [PAL 10a].
3.4.3.3. Multicriteria IR: implementation on the PIV
3
platform
The multicriteria CMRP IR model, described in the previous section, can be
implemented on a single search engine or several search engines federated by a
meta-engine. As recommended by Rasolofo et al. [RAS 03], we have adopted the
second approach in order to develop the PIV
3
multicriteria IR platform.
Figure 3.8 shown a meta-search engine as a broker that splits a query into sub-
queries dedicated to the targeted search engines. This meta-engine also supports a
result lists aggregation process that produces a single final list.
Thus, the PIV
3
meta-engine, detailed in [PAL 10a], implements the
CMRP
model
following the global architecture presented in Figure 3.8. Three “drivers” feature the
PIV
2
_spatial, PIV
2
_temporal and PIV
2
_thematic (based on the
Terrier
[OUN 05] IR
engine) search engines. Spatial, temporal and term-based specific matching functions
are declared in these drivers. PIV
3
thus federates three mono-dimensional IRSs. It
breaks down a multicriteria query into sub-queries and relays each of them to the
corresponding IRS. PIV
3
supports different result aggregation models, including the
CMRPmodelthat, aswehaveseen, allowsustoextendtheexpressivenessassociated
witheachcriterion.Inthiscase,PIV
3
performsthepartiallycompensatoryaggregation
Search WWH ::
Custom Search