Transition-based Directed Graph Construction for Emotion-Cause Pair Extraction

Abstract

Emotion-cause pair extraction aims to extract all potential pairs of emotions and corresponding causes from unannotated emotion text. Most existing methods are pipelined framework, which identifies emotions and extracts causes separately, leading to a drawback of error propagation. Towards this issue, we propose a transition-based model to transform the task into a procedure of parsing-like directed graph construction. The proposed model incrementally generates the directed graph with labeled edges based on a sequence of actions, from which we can recognize emotions with the corresponding causes simultaneously, thereby optimizing separate subtasks jointly and maximizing mutual benefits of tasks interdependently. Experimental results show that our approach achieves the best performance, outperforming the state-of-the-art methods by 6.71% (p < 0.01) in F 1 measure.

A Mutually Auxiliary Multitask Model With Self-Distillation for Emotion-Cause Pair Extraction

“…Ding et al [43] designed a 2D Transformer and its two variants to model the interaction between emotion-cause pairs. Fan et al [44] proposed a novel method, transforming the relationship classification into the process of constructing directed graphs. These methods all solved multiple tasks in a framework, which achieved good performance.…”

A Unified Sequence Labeling Model for Emotion Cause Pair Extraction

“…• TDGC-LSTM: proposed by Fan et al (2020) recently, they formulated the recognition of the ECPs as a set of actions and transitions and built a directed graph to model the transition process. Compared with the below baseline, this one uses LSTM to encode words' embedding vectors into clause embedding vectors.…”

“…All of the above works are constructed based on calculating the attention value between any two clauses of the given document. Fan et al (2020) proposed to address the ECPE task from a different angle by modeling the extraction of the ECP as performing a sequence of transitions and actions. Specifically, they constructed a Transitional Directed Graph for each given document and transformed the original dataset into sequences of transitions and actions, based on which they trained a model to predict the next state of the sequence given the current state and the predicted transition.…”

Multimodal Emotion-Cause Pair Extraction in Conversations

“…In the field of textual emotion analysis, previous research mostly focused on emotion recognition. In recent years, emotion cause analysis, a new task which aimed at extracting potential causes given the emotions Gui et al, 2016b) or jointly extracting emotions and the corresponding causes in pairs Ding et al, 2020a;Wei et al, 2020;Fan et al, 2020), has received much attention. These studies were normally carried out based on news articles or microblogs.…”