Geoscience Reference
In-Depth Information
Itmustbenotedthat,inanadvancedqueryingscenario(Figure3.6),theadjustment
oftherequirementsandpreferencesisperformeddirectlybytheuser.However,certain
combinationsareincoherent (e.g. a − requirementanda + preference).It istherefore
necessary to perform controls.
3.4.3.2. Multicriteria IR: formalization of the partially compensatory CMRP
approach
Wethusproposeanewapproach,inspiredfromthoseusedinmulticriteriadecision
support [MAR 99], in order to develop a more expressive IR. The aim is to enable
the user to specify, for each criterion of their query (Figure 3.6), the corresponding
geographic matching operator as well as the modal operators of requirement and of
preference.
The partially compensatory aggregation model of query results which we propose
can thus be modeled as a (C, M, R, P) quadruplet:
- criteria C =(c
1
,...,c
n
),where c
i
is a search criterion of the query;
- matching operators M =(m
1
,...,m
n
),with m
i
: C × D → [0; 1], and m
i
refers to an external matching function used to compare any document d ∈ D with
the criterion c
i
. The score assigned to the document for this criterion corresponds to
a value between 0 and 1. Various operators may be available such as intersection,
equality, inclusion and proximity. These correspond to the functions supported by the
spatial, temporal and thematic IRSs, respectively;
- requirements R =(r
1
,...,r
n
),where r
i
is a modal operator of requirement
applied to the criterion c
i
. It indicates the mandatory (+), neutral (N) or exclusive
(−) character of this criterion;
- preferences P =(p
1
,...,p
n
),withp
i
∈R,and p
i
weighs the search
criterion c
i
according to the corresponding level of importance desired by the user.
Thus, the CMRP model [PAL 12a] allows users to adjust the aggregation of
results according to their expectations. The final score of a document is established
by the partially compensatory aggregation function RSV described in equation [3.2].
This function weighs each intermediate score associated with a document for a given
criterion according to the level of preference initially assigned to the criterion by the
user. Thus, the score of a document d corresponds, for a set of n criteria, to the sum
of the products of the preference level p
i
by the score obtained for each criterion c
i
(the m
i
matching function applied to the document d with c
i
as criterion). This sum
is then normalized by the sum of the positive preference levels. Let us note that, if a
mandatory criterion is not satisfied or, on the contrary, an exclusion criterion is
Search WWH ::
Custom Search