Identification Emotion Cause in Conversation for Intelligent Customer Service

Most of previous research focus on emotion classification and extraction. However, there is a lack of research and discussions on the core aspect behind emotion, i.e., emotion causes. Therefore, this study introduces the task of emotion cause identification (ECI) to environment, targeting the conversation text between intelligent customer service(ICS) and user. This study constructs an accompanying dataset named REC-JD for recognizing emotion causes in conversation for ICS. Then we apply the multi-head attention mechanism and informer model to address two different sub-tasks of ECI: the causal span extraction (CSE) and causal emotion entailment (CEE). This research resolves the issue of insufficient user emotion comprehension caused by the lack of ECI technology in ICS systems. The experimental results show that the proposed model outperforms other baseline models.