Résumé

Recent years show an increasing popularity of chatbots, with latest efforts aiming to make them more empathic and humanlike, finding application for example in customer service or in treating mental illnesses. Thereby, emphatic chatbots can understand the user’s emotional state and respond to it on an appropriate emotional level. This survey provides an overview of existing approaches used for emotion detection and empathic response generation. These approaches raise at least one of the following profound challenges: the lack of quality training data, balancing emotion and content level information, considering the full end-to-end experience and modelling emotions throughout conversations. Furthermore, only few approaches actually cover response generation. We state that these approaches are not yet empathic in that they either mirror the user’s emotional state or leave it up to the user to decide the emotion category of the response. Empathic response generation should select appropriate emotional responses more dynamically and express them accordingly, for example using emojis.

Détails

Actions

PDF