[Search] Enable results highlighting for textual fields

[frascuchon]

Find a way to return highligthting info with search results. We can exploit the elasticsearch highlighting keeping in mind that, for text classification, highlighting for multiple fields is not trivial when no default string searches.

For example, given a dataset with records including inputs.subject and inputs.body the query:

size weight

could match terms in inputs.subject, inputs.body or both.

But, if we select a field in query:

inputs.subject: (size weight)

the highlight results could bring info for both fields subject and body, resulting in highlighting information that is not totally truthful

[dcfidalgo]

I think we could live with this "additional" highlighting info, as long as the highlights of the original query are present.

[frascuchon]

Since the computed highlighting won't be totally accurate and for the sake of performance, we can return only the relevant terms in search that makes record match.

What do you think @dvsrepo @dcfidalgo ???

The client (UI and client) could build the needed highlighting with those keywords.

[dcfidalgo]

Could you maybe provide an example of a query, record, and the returned terms you have in mind? Maybe we can also have a quick call about that.

[frascuchon]

The PR #1201 implements what I propose.

The query list* will match following records text:

- The listener can be implemente....
- Last listen in spotify was .....
- People is listening som.....
- Google is listed as the comp....

And the records keywords suggested in response should be:

- ["listener"]
- ["listen"]
- ["listening"]
- ["listed"]

Ready to talk if you prefer.

[dcfidalgo]

Yes, let's have a discussion about this. Maybe one comment ahead, not sure what the workflow of our WS for token classification will be, but maybe it would be helpful to also return start, end char indexes of the returned keywords. Not sure if this is technically possible.