Thanks for putting this together! Our company ParadigmRPA used #1 to build a dataset for a client that wanted to analyze and categorize content within books.

Manually building the dataset has been, and I can imagine, will be, an ongoing process.