javalstmdl4j

deeplearning4j generate response to input


I have recently been trying to learn DL4J but have run into some issues. They have an example of a neural network generating Shakespeare-like text based off and input character but I can't seem to find anything that wold indicate a possible way of creating a response to an input statement.

I would like to use an input string such as "Hello" and have it be able to generate a response of varying length depended on the input. I would like to know if this is possible using LSTM and have a point in the right direction as I have no idea where to even start.


Solution

  • We have plenty of documentation this actually. This gives you a layout of what an RNN looks like: http://deeplearning4j.org/usingrnns

    The model you would be looking at is character level, in general what you want is question answering though. You may want to look at an architecture like this: https://cs.umd.edu/~miyyer/pubs/2014_qb_rnn.pdf

    If you are completely new to NLP, I would look at this class: https://www.youtube.com/playlist?list=PLhVhwi0Pz282aSA2uZX4jR3SkF3BKyMOK

    It covers question answering as well.