by Jukka Corander
Research Report 1997:12
Department of Statistics, Stockholm University, S-106 91 Stockholm, Sweden
Abstract
Subjective classifications of objects where the categories are not specified in advance are considered. Several classifications of the same object set are assumed to be available. A probability model is introduced to capture consensus and variability between the classifications. In its simplest form the model distributes the classifications in accordance with a unimodal distribution. A generalization based on mixture distributions is used to describe partial consensus among the calssifications. Inference concerning consensus classifications and variability parameters is made possible by applying simulation algorithms and a modified likelihood approach.
Key words: Consensus of Classifications, Markov Chain Monte Carlo Methods, Random Graphs, Subjective Classifications.
Last update: 1997-12-16 / KH