We know an emotion there, but what type is it and what triggers it? Towards emotion-cause triplet extraction

A recently emerging emotion analysis task is emotion-cause pair extraction (ECPE), which aims to simultaneously obtain emotions and their corresponding causes expressed in documents. While ECPE is appealing, it does not take into account the types of emotions when connecting them with causes. The emotion type can reflect affective states and subjective information being recognized. Without it, an emotion-cause pair is of limited usefulness in practical applications such as strategy formulation, decision making, etc. To address this issue, we propose a new task, named emotion-cause triplet extraction (ECTE). It not only extracts emotion-cause pairs but at the same time also distinguishes emotion types. Further, we establish a biaffine attention-based multi-task learning approach to the ECTE task. Experiments are carried out on a benchmarking emotion-cause corpus that is slightly modified to suit the ECTE task. The results demonstrate the feasibility of the new task and the effectiveness of our proposed approach.