Word classification

Question

I have a task to classify the model of a product from its part number using machine learning. Part numbers can be of different lengths and forms and can contain both letters and numbers and also special characters in some cases e.x. 123-456, ABC123, A/B/123.

Can anyone give some ideas on techniques for feature extraction?

UPDATE

To give some more context, this is a typical (synthetic) example data:

Part Number :	Model
MM2 EAB111-444CDE	MM2
HPP6KT4ECCBD8IQI	HI-PROMANIFOLD
5142814	UFG E315 Cable Gland

So in some cases a model is part of the part number, but in other cases the map is not trivial. As another feature I could included the supplier as well.

Answer 1

Based on your (very brief) description of the problem, I understand that you have one model which needs to be classified based on one part number (singular in your description). This sounds like "translate": part -> model.

If so, you could look into sequence-to-sequence models. See "A ten-minute introduction to sequence-to-sequence learning in Keras" for some details.

You would probably need to do the translation based on the character level. You can use sequence-to-sequence models for instance to add numbers (without actually adding them up), so maybe this could work for your problem.

Alternatively, if a model consists of several parts, you could of course build a "bag of words" or use TFIDF (when the number of parts matter) with some multiclass classifier.