Event banner
Event details
Pre-annotation possibilties:
For domain experts (who we need to label our examples) it is often easier to start with pre-annotated documents. That way they have something to (dis)agree with.
One way we have been using in the past, is a simple keyword (list) match to perform machine annotation, followed by the manual activity of domain experts.
And about these domainexperts: sometimes it depends which person annotates the content, in other environments there is the detection of 'disagreement' between annotators. Is that still needed today ?
IanStory
Microsoft
MicrosoftHi Michel!
One of the great benefits of Syntex is we provide a set of prebuilt models in addition to allow you to create your own models. I would say in the case of prebuilt models, this isn't needed today, you just take the model (like for invoices or receipts - more coming soon) and apply it and let it do its magic. In the case of building your own models using unstructured document processing, structured document processing, or freeform document processing, it is absolutely helpful to have examples to (dis)agree with, and so I think that's still useful as a concept. However, you don't have to "pre-annotate" them, more just have a small training set ready that whoever is building the model is familiar with (and I get that perhaps you'd have a separate set of documents that a domain expert had, perhaps even graphically annotated, to help you if you're building the model but not the expert yourself). One of the grand things we're trying to do with Syntex though is make it so that the domain expert can build the model themselves, and not need to split this into two separate roles!