Multi-Hop Memory Network with Graph Neural Networks Encoding for Proactive Dialogue

Abstract

Dialogue system has made great progress recently, but it is still in the initial stage of passive reply. How to build a dialogue model with proactive reply ability is a great challenge. This paper proposes an End-to-End dialogue model based on Memory network and Graph Neural Network, which uses memory network to store conversation history and knowledge, and uses Graph Neural Network to encode background knowledge. We propose a soft weighting mechanism to integrate the dialogue goal information into the query pointer, so as to enhance the dynamic topic transfer ability during decoding. Experimental results indicate that our model outperforms various kinds of generation models under automatic evaluations and can accomplish the conversational target more actively