Dataset Creation and Curation - Christiaan Swart

We talked about:

• Christiaan’s background


• Usual ways of collecting and curating data


• Getting the buy-in from experts and executives


• Starting an annotation booklet


• Pre-labeling


• Dataset collection


• Human level baseline and feedback


• Using the annotation booklet to boost annotation productivity


• Putting yourself in the shoes of annotators (and measuring performance)


• Active learning


• Distance supervision


• Weak labeling


• Dataset collection in career positioning and project portfolios


• IPython widgets


• GDPR compliance and non-English NLP


• Finding Christiaan online

Links:

• My personal blog: https://useml.net/


• Comtura, my …

curation dataset