Sentemoji: A dataset to generate empathising conversations

Emojis are gaining popularity in day-to-day computer-mediated conversations, resulting in more interactive conversations. On the other hand, traditional chatbots lack the ability to use emojis effectively for creating an engaging and empathising conversation even after recognising feelings of the conversation partner, an essential communicative skill. This inability is majorly due to the paucity of any such suitable publicly available datasets and framework for training and evaluation of chatbot. Prior work has either classified the emojis or generated empathy dialogue without the use of emojis. Through this work, we propose a new dataset SentEmoji, generated using public dataset EmpathyDialogues, and its mapping to relevant emojis using EmojiNet dataset. We present a novel approach to generate dialogue with emojis to express empathy. A study will be conducted to get user rating on three aspects - empathy/sympathy, relevance and fluency. The comparison of this user-study with prior studies will reflect the effectiveness of this approach.