Great work by NextMove, an open, machine-readable, freely-reusable, annotated reaction data set, available for download here https://figshare.com/articles/ChemicalreactionsfromUSpatents1976-Sep2016/5104873

Reactions extracted by text-mining from United States patents published between 1976 and September 2016. The reactions are available as CML or reaction SMILES. Note that the reactions SMILES are derived from the CML.

Reaction SMILES

For convenience the reaction SMILES includes tab delimited columns for: PatentNumber, ParagraphNum, Year, TextMinedYield, CalculatedYield

Now that we have a large initial data set it would be great if others could contribute using the same format.

There is a fabulous detailed review of this invaluable resource on the Depth-First blog http://depth-first.com/articles/2019/01/28/the-nextmove-patent-reaction-dataset/