I'm trying to convert product shorthands into brand, pack sizes etc. E.g. input: "CG EXHIBITIONIST VERY BLACK WP MASCARA", should produce the output Brand = "Cover Girl", Sub-brand = "COVERGIRL EXHIBITIONIST"
I've got a bunch of existing products that have been coded manually and then some new products. This feels to me like missing values, in that I've got very similar lines with all values present, and I just have to make a best guess.
However I'm a bit stumped on the language side of it. E.g. CG is cover girl here, but there probably isn't much training data on that in the dataset.
A non missing-values approach would be more like parsing. Products are written in a very formulaic way, so you'd think I could extract provided_brand="CG", though there'd be some ambiguity on provided_brand="CG EXHIBITIONIST". Then I'd need a mapping between CG and Cover Girl.
What approach would you take?