Prediction using NLP/ML/word2vec/tensorflow. I have 1000's of records from service request utility. I have list of sevice requests & corresponding action taken by executive for each service request which is in natural english langauge(text could be different for same type of request from different user).
Sample record-
request: addition to xxx mail alias for the user id yyy
action: ask user to get approval from manager and then manually add the user id to mail alias group.
My understanding is that this data can be used as training data for my model(NLP/ML/word2vec), in such a way that if any new request of same type given as input to model, model should be able to suggest the required actions that executive should take. This is what I want to achieve but not sure how to and from where to start. I have executed samples from
https://github.com/tmikolov/word2vec
https://github.com/tensorflow/tensorflow/tree/r0.12/tensorflow/models/embedding
Any pointers.
I think you can approach this as a translation problem. Just instead of translating between languages you translate from request to action. The TensorFlow page has a translation tutorial you could try: https://www.tensorflow.org/tutorials/seq2seq/
For an even simpler go at it you could also just train a character-prediction RNN on the whole text file that contains the requests and answers. This is described well with example code at http://karpathy.github.io/2015/05/21/rnn-effectiveness/.
The RNN will then be able to generate or auto-complete requests as well :).