The document contains the name of the disease only once (in the main headline), and it has subheadlines with sometimes relevant mentions.
Since the document is quite long I broke it down into chunks with Watson Document Conversion. Now, in order to have enough relevant mentions that I can annotate, I added the main document headline and the subheadlines to the chunks and imported them to WKS. The result would then look something like this (the 3 parts separated with dashes):
Arthrosis — Green tea — Green tea can be a good help because it is anti-inflammatory.
In the annotation process I would then annotate “Arthrosis” as disease, and both mentions of green tea as treatment. Addiitonally, I would twice annotate a relation “isCureFor” between both mentions of green tea and arthrosis.
My question is: does this approach make sense, or is it a “hack” that harms the machine learning process? What could be an alternative in cases where relations are hard to annotate because the necessary entities can seldom be found in one sentence?
Thanks in advance!