Our data set is a lot of ship movements. A captain usually enters destination in form of a short code like: "CHN", "Singapore", "XIAO HU" etc. These keywords are pretty short, can be somewhat misspelled and can be a code for a port, name of the city or just random gibberish depending on user's mood :)
We have a large dictionary of destination words and want to try and match the code into a real destination (Like Singapore Port X).
So, basically it's a word => words predictor, depending on earlier history (ie. same vessels usually goes on same route).
Is there a good NLP model to use to implement these kind of predictions? Right now we're using regular text/pattern matching and it WORKS, but thinking this could be solved better by ML.
Since we also have the actual trip data, we do know the actual destination afterwards, so it's possible to have a decent training data set. Unfortunately the captain might be different for same vessel on different trips, so the "style" of codes might not be consistent.