A Neural
Conversational Model
What is this paper about?
 It proposed a neural conversational model which generates an answer to a question using
sequence to sequence framework.
What’s better than previous paper?
 This model can be trained end-to-end and thus requires much fewer hand-craft rules.
 This model can generate simple conversations given a large conversational training dataset.
 This model has its simplicity and generality and thus can be applied to machine translation,
question/answering, and conversations without major changes in the architecture.
What’re important parts of
technic and method?
 This model adopted sequence to sequence framework and be done by backpropagation in
learning to maximize the cross entropy of the correct sequence.
 During inference, this model used a greedy inference approach.
 A less greedy approach would be to use beam search.
How did they verify it?
 They Asked four different humans to rate this model versus CleverBot with picked 200 questions
from http://ai.stanford.edu/~quocle/QAresults.pdf.
Is there a discussion?
 It can generate simple and basic conversations, extract knowledge and answer to many types of
questions without any rules.
Is there a paper to read next?
 Neural machine translation
 RNNs
 LSTM
 Sequence to sequence framework
Paper Information
 https://arxiv.org/abs/1506.05869

A neural conversational_model

  • 1.
  • 2.
    What is thispaper about?  It proposed a neural conversational model which generates an answer to a question using sequence to sequence framework.
  • 3.
    What’s better thanprevious paper?  This model can be trained end-to-end and thus requires much fewer hand-craft rules.  This model can generate simple conversations given a large conversational training dataset.  This model has its simplicity and generality and thus can be applied to machine translation, question/answering, and conversations without major changes in the architecture.
  • 4.
    What’re important partsof technic and method?  This model adopted sequence to sequence framework and be done by backpropagation in learning to maximize the cross entropy of the correct sequence.  During inference, this model used a greedy inference approach.  A less greedy approach would be to use beam search.
  • 5.
    How did theyverify it?  They Asked four different humans to rate this model versus CleverBot with picked 200 questions from http://ai.stanford.edu/~quocle/QAresults.pdf.
  • 6.
    Is there adiscussion?  It can generate simple and basic conversations, extract knowledge and answer to many types of questions without any rules.
  • 7.
    Is there apaper to read next?  Neural machine translation  RNNs  LSTM  Sequence to sequence framework
  • 8.