Is it possible to create an app the identifies the top topics in a corpus ranked by frequency/relevancy? We were told by our Watson rep that it is also possible to train the service to fine tune the concepts. Is that also correct?
Thanks!
Answer by Michele Franceschini (636) | May 23, 2016 at 04:18 PM
@SanPietro there is indeed an API call for that:
This will return the top concepts related to the corpus (related in that they do not have to necessarily appear in the corpus. They are divided in concept popularity brackets (using level
) as all other related concepts calls.
The more information is in the corpus the finer is going to be the selection.
Another useful one is the one shown in the dashboard, which lists raw statistics from the corpus itself, in terms of concepts mentioned in the corpus and count of appearances:
Concept insights cannot be "trained": it has it's own world model and it uses it to interpret the documents it's fed. As a corpus gets bigger, however many aspects will improve, mainly because it can build a better statistical model of the conceptual representation of the corpus itself. This is a little bit like training but it's not properly a retraining of the world model (the concept graph).
Using SUM operator in Discovery query language 1 Answer
Difference between graph/annotate_text and document/related_concepts 1 Answer
Concept Insights + AlchemyData News Application 2 Answers
Probable bug in concept graph: Romance_(love) 0 Answers
invalid ids parameter error when calling conceptual search with curl 1 Answer