Composite search scenarios

[scossu]

I am considering using Elucidate to store image analysis annotations (not IIIF related), e.g. from color palette extraction processes.

A couple of limitations I have come across so far are related to how I would search for these annotations once stored (or before they are stored, to know if they exist already).

Given the following sample annotation that defines two palette sets extracted from a single image:

{
        "@context": "http://www.w3.org/ns/anno.jsonld",
        "id": "https://dummy.getty.edu/anno/my-palette-01",
        "type": "Annotation",
        "body": [
             {
                 "type": "TextualBody",
                 "format": "application/json",
                 "value": "<JSON data>",
                 "purpose": "https://data.getty.edu/local/thesaurus/purposes/low_contrast_palette"
            },
            {
                 "type": "TextualBody",
                 "format": "application/json",
                 "value": "<JSON data>",
                 "purpose": "https://data.getty.edu/local/thesaurus/purposes/high_contrast_palette"
            }
        ],
        "target": "https://media.getty.edu/iiif/image/77902e8c-f844-4956-8f56-186aa2f59e05",
        "motivation": "https://data.getty.edu/local/thesaurus/motivations/image_analysis#histogram"
    }

I would like to:

• Know in advance if an annotation with a target of https://media.getty.edu/iiif/image/77902e8c-f844-4956-8f56-186aa2f59e05 and a motivation of https://data.getty.edu/local/thesaurus/motivations/image_analysis#histogram already exists, and if it does, obtain its URL (so that I can let Elucidate generate the id of the annotation);
• Find all annotations with a body with a purpose of https://data.getty.edu/local/thesaurus/purposes/low_contrast_palette.

From what I read form the documentation, I can only search by either target or motivation, not by both. For my specific purpose I got around this limitation by crafting the URI myself based on these two parameters and using PUT to create or replace the resource; but that wouldn't work if I wanted to back out from overwriting.

Mathmos seems to be more oriented at IIIF Search API compatibility, which is focused on free-text content search, while Elucidate seems to have more semantic/structural querying capabilities. Is this something that either one can support (from this latest assumption, I would guess Elucidate would be the more appropriate place)?

[danielgrant]

Hi @scossu,

In terms of what you'd like to do, Elucidate already supports searching by the id or source value of a target field, however you are correct that it doesn't support searching against motivation, and doesn't support searching for purpose within the body field.

So this would involve:

• Adding extraction of the purpose field of a body field and persisting this in the DB.
• Adding extraction of the motivation field and persisting this in the DB.
• Adding support for querying for annotations that contain a given purpose.
• Adding support for querying for annotations that contain both a given target and motivation.

Does that sound right? If so, would you like us to put together an estimate for implementing these enhancements?

Cheers, Daniel.

[scossu]

Thanks for the breakdown @danielgrant . I would generalize the fourth point as: "adding support for querying annotations using multiple criteria in any combination". I understand that might be more cumbersome and possibly even require changing the API request structure, but it would be greatly more comprehensive.

[danielgrant]

Hi @scossu, I appreciate what you're saying r.e. the fourth point. My concern here is less about changing the API request structure, but more around the implications for the underlying code and, in particular, the database interaction. This part of Elucidate is unfortunately quite clunky, so enhancing point four further to allow for any arbitrary combination of supported search fields to be applied to a single search would likely require more effort, if not a complete rethink of how we approach this currently.

If desired I can take this away and get a feel for the effort to implement all four points, with separate estimations for effort for a targeted target and motivation search, and searching using multiple criteria in any combination.

Let me know what you think.

[scossu]

That would be terrific, thanks. Unfortunately I have no Java expertise and cannot help with hands-on coding, but any other way I can help, e.g. with DB design I'd be glad to.