Transforming the Personality of a Dialogue System Using Variational Auto-Encoder and Attention Mechanism

Abstract

Dialogue systems such as chatbots and smart speakers are becoming ubiquitous in our daily lives. In order to achieve high user experience, it is important for dialogue agents to have attractive personality which enables friendly and human-like conversations. However, in most present cases, development of such dialogue systems involves hand-crafting of dialogue scenarios, and introducing the characteristics representative of certain personality to system utterances requires large time and human resources. This paper proposes a technique to automatically transform the characteristics of a sentence using variational auto-encoder and attention mechanism.