Semantic Knowledge Representations for Soft Data Fusion - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 24

To compute the extended maximal join of two graphs, we have to find compatible sub-graphs
of the two graphs that are maximally extended in terms of the number of their nodes.
The compatibility of the two subgraphs is processed according to the compatibility of their
concepts and relation nodes. Then, the compatible subgraphs are fused.
Fig. 15. Compatible relations (Extended maximal join)
On the example depicted on figure 15, we try to fuse the graph G 1 and G 2. We can can see
that, according to the initial definition of compatibility between concepts and relation nodes,
the subgraphs of G 1 and G 2 composed of Program and Content concepts, linked through
a content relation on the one hand and Entity and Content concepts, linked through a
content relation on the other hand are compatible. The result of the maximal join is thus the
one depicted on graph G 3.
When looking at the “News” and “The news” titles of the two programs depicted on figure
15 and given the remaining elements of the two descriptions, one would like to fuse the titles.
Indeed, they “obviously” represent the same title and the descriptions are related to the same
TV program. By including domain knowledge thanks to the compatibility testing function, we
obtain as compatible subgraphs the two previous ones, with the titles in addition. The result
of the fusion of G 1 and G 2 using maximal join extended with the fusion strategies defined in
the examples above gives the graph G 3.
6. Experimentations
We describe here experiments that were conducted in order to validate the usefulness of both
the fusion strategies and the association phase within soft data fusion. We used information
acquired within the TV program case study described earlier. The fusion platform that was
developed was also used within biological experimentation and intelligence applications.