Sur le consensus en catégorisation libre

摘要

Starting from individual judgments given as categories (i.e., a profile of partitions on an item set X), we attempt to establish a collective partitioning of the items. For that task, we compare two combinatorial approaches. The first one allows us to calculate a consensus partition, namely the median partition of the profile, which is the partition of X whose sum of distances to the individual partitions is minimal. Then, the collective classes are the classes of this partition. The second one consists in calculating, first, a distance D on X based on the profile and then in building an X-tree associated to D. The collective classes are then some of its subtrees. We compare these two approaches and more specifically study the extent to which they produce the same decision as a set of collective classes.