However, in order to do this, it seems like I would need to build some very sophisticated search algorithm into the content of the article and not just the title and description.
Any ideas on how I could start to approach this? Thanks!
I think it depends on how the words are being selected for highlighting.
Are they negative feeling words? or words who’s sentiment can be classified?
If so I would start by looking at sentiment analysis api’s that can rate the whole article first for being negative/positive. Then see if you can get something that goes down to the paragraph / sentence / word level.
This would be a good place to start seeing what’s out there.