Sequence to Backward and Forward Sequences: A Content-Introducing Approach to Generative Short-Text Conversation

Using neural networks to generate replies in human-computer dialogue systems attracting increasing attention over the past few years. However, the is not satisfactory: the neural network tends to generate safe, relevant replies which carry little meaning. In this paper, we a content-introducing approach to neural network-based generative systems. We first use pointwise mutual information (PMI) to predict a as a keyword, reflecting the main gist of the reply. We then propose2BF, a "sequence to backward and forward sequences" model, which generates a containing the given keyword. Experimental results show that our approach outperforms traditional sequence-to-sequence models in terms of evaluation and the entropy measure, and that the predicted keyword can at an appropriate position in the reply.