Sequence to Backward and Forward Sequences: A Content-Introducing Approach to Generative Short-Text Conversation
This addresses the issue of bland responses in human-computer dialogue systems, though it is an incremental improvement over existing sequence-to-sequence methods.
The paper tackled the problem of neural networks generating safe, meaningless replies in dialogue systems by proposing a content-introducing approach that predicts a keyword and uses a seq2BF model to generate replies containing it, resulting in significant improvements in human evaluation and entropy measures.
Using neural networks to generate replies in human-computer dialogue systems is attracting increasing attention over the past few years. However, the performance is not satisfactory: the neural network tends to generate safe, universally relevant replies which carry little meaning. In this paper, we propose a content-introducing approach to neural network-based generative dialogue systems. We first use pointwise mutual information (PMI) to predict a noun as a keyword, reflecting the main gist of the reply. We then propose seq2BF, a "sequence to backward and forward sequences" model, which generates a reply containing the given keyword. Experimental results show that our approach significantly outperforms traditional sequence-to-sequence models in terms of human evaluation and the entropy measure, and that the predicted keyword can appear at an appropriate position in the reply.