Introducing semantic knowledge in high level information fusion
A major stake of future decision support systems is to automate the processing of pieces of information coming from different sources in order to ease their understanding. The aim of our work is to recognize specific predefined situations from the observations gathered on different sources. Within the global information fusion process for situation recognition, we focus on three steps.
First, the situations that must be recognized are represented using the conceptual graphs formalism. Using conceptual graphs allow us to take advantage of the theoretical studies that were achieved on this model. Then, compatible observation identification allows checking whether two observations relate to the same situation before any attempt to fuse them. The identification step relies on the use of domain adaptable similarity measures between conceptual graphs.Finally, the information fusion process proposed relies on the use of the maximal join operation on conceptual graphs. In order to fuse observations that are not exactly identical, we introduce domain knowledge inside the maximal join operation and relax the constraint of strict equality between the values of the concepts nodes of the graphs
The validation of our work emphasizes on two aspects. First, we validate the validity of our approach and the usefulness of introducing domain knowledge inside the fusion process. To do so, the fusion platform developed during this thesis was used within a TV program recommendation system. Then, we validated the genericity of our approach and the adaptability of the fusion platform to new application domains by using the fusion platform within four other case studies.