Toward Explainable Dialogue System Using Two-stage Response Generation

In recent years, neural networks have achieved impressive performance on dialogue response generation. However, most of these models still suffer from some shortcomings, such as yielding uninformative responses and lacking explainable ability. This article proposes a Two-stage Dialogue Response Generation model (TSRG), which specifies method to generate diverse informative based an interpretable procedure between stages. TSRG involves two-stage framework that generates candidate first then instantiates it the final response. The positional information resident token are injected into stabilize multi-stage framework, alleviating shortcomings in framework. Additionally, allows adjusting interpreting interaction pattern two generation stages, making somewhat controllable. We evaluate proposed three datasets contain millions single-turn message-response pairs web users. results show that, compared with previous models, can produce more maintain fluency relevance.