Paun 2018 - Comparing Bayesian Models of Annotation

Routing Summary

Full capture of Paun et al. (TACL 2018): six Bayesian annotation models (MULTINOM, Dawid-Skene, MACE, HIERD&S, ITEMDIFF, LOGRNDEFF), four datasets, evaluation against gold, predictive density, annotator characterization and difficulty. 5 notes.

Concept Map

ConceptNoteTypeDepends OnKey Result
MULTINOM, D&S, MACEPaun 2018 - Pooled and Unpooled Models Multinomial Dawid-Skene MACEdefinitionOverviewconfusion matrices vs credibility
HIERD&S, ITEMDIFF, LOGRNDEFFPaun 2018 - Partially Pooled Models Hierarchical D&S ItemDiff LogRndEffdefinitionD&Spartial pooling avoids overfit
Evaluation resultsPaun 2018 - Evaluation Protocol and Results Tablesexamplemodelsstructure beats majority vote
Label switching, non-centeringPaun 2018 - Implementation Priors Label Switching and Guidelinesconceptresultsrelabel chains with gold sample

Notes

Sources

Paun et al 2018 - Comparing Bayesian Models of Annotation

See Also

Hierarchical Models · Monsters and Mixtures