Generating Natural Language from few seed sentences without human for an intent has been a challenging problem in Chatbot industry. Here is my proposal:
a) Apply a sequence to sequence model with attention on filtered SemEval dataset.
b) Loop through each seed sentence and generate a semantically equivalent sequence to sequence model from step a.
c) Take generated sentences from step b and add it as a seed sentence and recursively apply step b till you get sentences which are not semantically similar as evaluated by say a Siamese Network.
What are your thoughts? Can you please let me know the results, if anyone gets a chance to implement it?
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment