Semantic Knowledge Representations for Soft Data Fusion - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 14

5.3 Soft data association
In this section, we focus on the soft observation association phase described earlier. The aim is to
compare and analyze the different observations and decide which ones should be fused. More
precisely, we detail here the comparison of two observations, taking into account the domain
knowledge previously modeled as well as the values of the elements (i.e. nodes and edges)
that make them up. This allows us checking their fusability.
We first describe here our proposition for a similarity measure between two conceptual
graphs. Then we show how to use this measure in order to test the compatibility of two
graphs in the association phase.
All the measures that we propose within this section are normalized. Extended proofs of this
property are available in Laudy (2010)
5.3.1 Comparison of concepts
To measure the similarity between two concepts, we propose to compare their conceptual
types, their values as well as their immediate neighborhood. The study of the neighborhood
gives clue about the context in which a concept is used.
5.3.1.1 Comparison of conceptual types: diss type
We first describe how to compare two concepts, regarding their difference, through
dissimilarity processing. The dissimilarity between conceptual types is used to measure how
much two situations are different. We adapt the distance between types proposed by Gandon
et al. (2008), in order to obtain a normalized dissimilarity measure.
Fig. 8. Constraints on the dissimilarity over conceptual types
The main idea, is that the difference between two concepts is processed according to the
number of edges that separate them from their nearest common parent. Furthermore, the
deepest this common parent is in the lattice of types, the smallest the difference is between the
two compared types.
The difference between two types with a nearest common parent of a depth d in the type
lattice is always smaller than the difference between two types with a nearest parent of depth
of d-1 , whatever the number of edges between the types and their parents is.
As an illustration, looking at the figure 8, we want to have the following inequalities: