A survey on emotion–cause extraction in psychological text using deep learning methods

The study that has been done in this survey paper mainly focuses on the recent work in detecting emotions based on causes. The emotion–cause extraction approach is discussed in depth, and many variants are compared to achieve a contrasting result. The use of causes when extracting emotions in a sentence requires following the casual relationship at each process stage. The O(n2) drawback of the emotion–cause pair extraction approach is contrasted with the transition approach of following the casual relationship of emotion and causes. A new approach that works on the emotion–cause pair extraction technique uses the attention, and transition methods are discussed in great depth by introducing all the related technologies. Enhanced deep learning methods that can easily replace the current state-of-the-art models by fine-tuning pretrained models like bidirectional encoder representations from transformers are studied, and results are compared on various use cases. These models are helping in the emergence of hierarchical networks for the optimization of preexisting methods. The psychological dataset is used to compare various techniques. The dataset requires annotation with the emotion clauses and cause clauses to train the models. This survey has reviewed many papers, and the advantages and disadvantages of the approaches introduced in these papers have been discussed.