An affective chatbot with controlled specific emotion expression

Endowing a chatbot with the capability of specific emotion expression will significantly improve both chatbot’s usability and users’ satisfaction. Recently, many studies on open-domain neural emotional, conversational models (chatbots) have been conducted. However, enabling a chatbot to control what kind of emotion to respond to in conversation explicitly is still under exploration. This paper proposes a novel affective chatbot based on the sequence-to-sequence framework, responding with appropriate emotion like a human. In particular, a new module called single emotion generator is designed in the new chatbot model to address the existing issue of controlling over reacting emotion. It enables the chatbot to select the appropriate emotion for a response when interacting with users. In the decoder, an affective lexicon-based method generates emotion-awareness responses based on the specific emotion controlled by the single emotion generator. The proposed chatbot outperforms mainstream baseline algorithms for both semantic fluency and emotion consistence metrics through experimental evaluation. The experimental results also demonstrate that the new chatbot obtains the ability to control the emotion for response explicitly and responds emotionally with the specific emotion.