of images and a soundtrack). Criteria aggregation, on the other hand, is used to help
a user make a decision corresponding to the best compromise (all the criteria can
rarely be satisfied). Concerning the GIR, the modes of combination are generally of
filtering type: they consist of successively querying the IRSs dedicated to each of the
dimensions followed by performing the intersection of the results. Finally, the
multicriteria IR integrates the processes of aggregation which, depending on the
case, allow a criterion to compensate for one or more other non-represented criteria.
126.96.36.199. Fusion and multimedia IR
Generalization, as explained in the previous section, enables rough information
a video, is illustrated in [PAL 10a]. In a preliminary generalization step, the images
(colors, textures, forms, etc.) and the audio track (sounds, speech, etc.) are analyzed
and higher level concepts are extracted (e.g. “car” and “sport”). Fusion then consists
of gathering information of the same nature. For a video, the concepts extracted from
the text it is associated with [PHA 07]. It has to be noted that this fusion is done in an
different studies mentioned, none weigh the different types of data used. This fusion
of abstraction qualified as semantic or conceptual.
188.8.131.52. Aggregation and decision support systems
DSSs have been used for approximately 40 years to facilitate decision-making
[POW 97]. Reality is multi-dimensional [ZEL 82]; consequently, the criteria (also
called constraints) to take into account for decision-making are numerous [BOU 93].
The objective of a multicriteria decision support is to propose to its user scenarios
best satisfying these previously defined criteria in the search of a solution. It has to be
noted that the expressed criteria cannot all necessarily be respected in the proposed
scenarios. Thus, criterion aggregation is used in order to find the best compromises.
These approaches are based on a strong interaction with the user; he or she has to
determine the importance of each criterion by assigning a weight to it and modulate
the desired compensation between criteria.
More precisely, every multicriteria decision problem can be formulated by the
model “A, A/F, E” [MAR 99]. A is the set of alternatives, in other words the possible
choices. A/F is the set of attributes or features (Attribute/Feature), in other words the
constraints expressed by the user. E is the set of the evaluations of the alternatives'
performance according to each criterion; the extent to which the alternative fits the