I am working on a system that compares a source document to a target document and then generate alternative variations of the source document. The goal is to reach a higher cosine similarity between the two documents, say +80%. There's already plenty of articles on how to find the similarity difference, but not many on how to generate suggestions or improvements.
I've never heard of a model that can do this, so how do I build a model that can suggest altered variations of a source document with the ultimate goal of increasing its similarity to a given target document?
Articles on document similarity