The Benefits of a Model of Annotation

This Research paper providers another full [[ Bayesian ]] model for inferring when rater decisions can be combined to produce reliable labels.

Key benefits of this paper:

  • It uses EM to train the model.
  • They provide a Mutual Information based method for confirming label confidence.
  • They have nice graph plate diagrams.

This is a thing to build.

Backlinks

No backlinks yet


The whole shebang

UntitledAdvocacy Federation HubArizmendi BakeryBlog NotesSurface Data Collective: First StepsSurface Data Collective - Second StepsBuddhismBuddhist GeeksCelo ChainColony DAOCommonsCentral HubCulture HubFinance HubFranchise HubInnovation HubMission Lock HubCooperative PatternShared Services HubData Commons GovernanceDemocratic User JourneysDesjardins Federation of Credit UnionsEnspiralEthereumEvergreen CooperativePolygon ChainThe Benefits of a Model of AnnotationCLARA: Confidence of labels and ratersThe Power of Scale for Parameter-Efficient Prompt...Examining Scaling and Transfer of Language Model...Self-supervised and Supervised Joint Training for...Towards Universality in Multilingual Text...Whose Language Counts as High Quality? Measuring...ResearchShared Capital CooperativeStreamrSurface Data CommonsThing to buildUnited States Federation of Worker CooperativesWeb3xDai Chain