CoEmoCause: A Chinese Fine-Grained Emotional Cause Extraction Dataset

Emotion Cause Extraction (ECE) is an emerging hot topic in the field of sentiment analysis. The purpose of the ECE task is to extract the causes of emotions in the text according to the text and given emotions. Emotional Cause Pair Extraction (ECPE) is a brand-new research problem on ECE, whose main purpose is to extract the emotion clauses and emotion cause clauses in texts at the same time. At present, the ECPE task has received extensive attentions from both academia and industry communities. The existing ECPE researches are mainly carried out on online news report corpus. However, this kind of corpus has several limitations, such as the small scale and rough tagging granularity. In this paper, we construct an emotion cause extraction dataset containing 5,195 COVID-19 pandemic-related discussion posts collected from Sina Weibo, and leverage 10 state-of-the-art models to perform ECE and ECPE tasks on this new dataset. We find that the performances of most existing models on our newly constructed dataset decrease dramatically compared with the reported results in the online news dataset. We further analyze the causes of this phenomenon.