Text Sentiment Analysis And Reason-finding Research

In the field of natural language processing,text emotion analysis is an important research direction.But this task only focuses on emotion class,which is a shallower emotion analysis.But sometimes,we are more concerned about the causes of these emotions,which leads to a deeper task: emotion cause extraction.However,there are some problems with this task: The first is that emotion expressions must be annoated before emotion cause extraction,which limits the applications of ECE in real-world scenarios.The second is that the way to first annotate the emotion and then extract the cause ignores the fact that emotions and causes are mutually indicative.To solve these problems,a new task was proposed in last year: emotion-cause pair extraction.Based on news texts,this paper carries out a series of explorations and researches of two tasks:1.The task of emotion cause extraction,which aims at discovering the potential causes behind a certain emotion expression in the text.Most of the previous work considered this task as a set of independent clause classification problems and ignored the relations between multiple clauses in a document.This paper proposes a hierarchical network based on RNN-Transformer,to encode and classify multiple clauses synchronously,it is the first time that Transformer has been used to solve this task,the experiment demonstrates excellent performance in learning the correlation between multiple clauses in ECE.This paper further encode the relative position and global predication information into the Transformer framework.It can capture the causality between clauses and achieve extra improvements.2.The task of emotion-cause pair extraction,which aims to extract the potential pairs of emotions and corresponding causes in text without providing emotion annotation.Previous work proposed a two-step framework to solve this problem,but this method is not an end-to-end model,so some information may be lost at each step.In this paper,this task is definded as a multi-label classification problem,and build a one-step model that directly extract emotion-cause pairs in an end-to-end fashion.In addition,this paper designed two auxiliary tasks to further improve the performance of the model.