Retrieving / annotating enrichment metadata about a doc/blob

Prerequisite: Data returned from the prediction service is stored in the main datastore using the defined enrichment format.

These annotations are visible on the document to the user, and a reduced version from the metadata stored after AI service calls. They can be created either automatically from a call to an external API, or manually by the user.

Possible kinds of annotations :

• Classification : List of String (labels), representing classes this document belongs to
• Tagging :

• Object Recognition in Image with a label and a box associated
• Entity Recognition in Text with a label and word index associated

• Transcription : String with a transcription of an audio stream or an image

When needed, we should create simple interfaces for these annotations. For classification and transcription this should require minimal interface, Tagging might need a more complex interface with annotation in the image or the text file.

Requirements:

• Existence of a flag that defines if the tag was created by an human or automatically from an API call

• In case of an annotation with multiple labels, changing one value will make it manual, even if other values where not changed by a human

• An annotation is related to a blob (text, image, etc). If the original document changes, the annotation value is void and should be recalculated/reannotated

Nice to have:

• Maybe a reference to the metadata used to create it, if this was automatic
• An historic, who changed and what changed

Decision point:

• Text classification is to be done on extracted text. Annotation should be done on the same text, but strange !
• Image classification is potentially done on a small version of the original. Annotation to be done on the image to be used for classifciation