Generating Consistent Multimodal Dialogue Responses with Emoji Context Model

Emoji plays a critical role in human-to-human conversations. It is straightforward to understand emotional feelings of others via the emojis in a dialogue response. However, incorporating emoji into a chatbot is a challenging task given the heterogeneity of the data sources. Motivated by previous work using emojis for sentiment analysis, we pave a novel way of creating a chatbot that can generate consistent multimodal responses. In this paper, we propose to develop an Emoji Context Model to generate emoji in dialogues. Also, considering dialogue historical information, responses generated by our proposed model achieve better performance in terms of coherency comparing with the state-of-the-art. The experimental results demonstrate our proposed model improves conversation consistency and provides effective multimodal responses with emojis.