Can we use machine learning to generate a text output based on the input strings

Question

Problem : Generate a text output based on input strings which will be combined using a number of rules.

Example :

    Feature1               Feature2                    O/P

Rule 1  Enum_Domain           Priority          /Enum_Domain/Priority

Rule 2 Enum_Domain.EnumData   Name              /Enum_Domain/EnumData/Name

Rule 1  Trunkgroup            Gateway            /Trunkgroup/Gateway

Rule 2  GatewayGrp.Gateway    IP                /GatewayGrp/Gateway/IP

This is a simple programming problem, but is there any machine learning algorithm that can learn these rules and generate the output based on the two inputs.

Answer 1

Yes, sequence 2 sequence models attempt to do this. This can be used in a number of domains, from typo fixing to machine translation. They are encoder -> decoder based, which means you have a part that encodes your input and then a decoder that generates a new sequence based on this encoding (and usually some attention). In this case your encoder would likely be two recurrent neural networks of which the output would be concatenated and then a decoder that takes this concatenated output and turns this into a new sequence. If you want to use attention you need to adapt the standard attention a bit because you have two textual inputs, but if you understand how it works this would not be too difficult to adapt.