Mesurer le désaccord entre vocabulaires d’experts et son impact sur l’analyse de données

A vocabulary expressed through linguistic variables is a way to exploit subjective and expert knowledge to analyze data. When multiple experts are involved, it is crucial to point out possible disagreements between their  respective vocabularies. This paper proposes disagreement measures that take into account discrepancies regarding the structure of their partitions, the linguistic labels of the modalities and their impact on the dataset analysis process. Experiments show that the proposed multi-level disagreement measure highlights complementary and meaningful discrepancies between the expert vocabularies.