-
Notifications
You must be signed in to change notification settings - Fork 24
Extraction flow
Zheng Tang edited this page Aug 21, 2021
·
3 revisions
The main entrypoint to the EidosSystem for extraction is the extractFromDoc
method (there is a comparable one for extracting from text that annotated the text and then calls this method). In this method, you will see the creation of Refiner
s which are used in the extraction pipeline. If you follow the method path, you'll see that in the overloaded method, the flow is as follows:
- The
DocumentRefiners
filter or adjust the document prior to extraction (e.g., filtering long sentences or those which are likely mis-parsed tables from a pdf). - Odin is run over the document by successive application of enabled
Finders
- The Odin Mentions are post-processed with any enabled
odinRefiners
(e.g., handling of hedging, negation, etc.) - An
AnnotatedDocument
is created with the resulting Odin Mentions. During this process, each Odin mention is mapped to anEidosMention
, which is a wrapper class that holds additional metadata. - The EidosMentions are post-processed with enabled
EidosRefiners
(e.g., ontology grounding, grounding of gradable adjectives/Quantifiers).