Study On Emotion Cause Pair Extraction Based On Fusion Word Vectors

In recent years,with the rapid development of Twitter,MSN,Weibo,BBS,We Chat and other social media,people have become more and more dependent on social networks and are used to expressing their views and emotions on social networks.Text sentiment analysis can effectively help government decision-making departments,research institutions and information consulting organizations grasp the trend of social sentiment changes.Text sentiment analysis refers to the process of analyzing,processing and extracting information on sentimental and subjective texts using natural language processing technology.Emotional reason extraction is an important task in text sentiment analysis.The purpose is to extract the potential causes of emotion through the given emotional information in the text.When treating emotion as an event,causality becomes the key relationship to be extracted.This paper conducts research on the extraction of emotional reasons,which can lay an important foundation for downstream research tasks such as cause-based reasoning,public opinion prediction,and cause detection in text sentiment analysis.However,the existing research on the extraction of sentimental reasons still suffers from the lack of a large-scale sentimental labeling corpus,insufficient mining of word sequence information,and insufficient use of location information in prediction.This article conducts in-depth research on these issues,and the main innovations are as follows:1?A corpus enhancement algorithm EDA-ECPE for emotional reasons is proposed.The algorithm is based on EDA operations to enhance the emotional reasons to extract annotated corpus,and solves the problem of lack of large-scale data sets in the existing research.Through the four operations of random insertion,random exchange,synonym replacement and random deletion for each document in the original corpus,the emotional reason corpus is enhanced.The new corpus can effectively alleviate the risk of overfitting caused by the small amount of data in the existing corpus.The proposed algorithm can effectively enlarge the amount of data and keeps the original label.2?Propose a short document emotional reason pairs extraction model SDF-ECPE based on fusion word vectors.This model utilizes BiLSTM and attention mechanism to extract the overall semantics of context and the context vectors of different text contents in terms of cause or emotion.For reason extraction,the accuracy of reason extraction is improved by fusing emotion context vectors and emotion prediction results.For the extraction of emotional reason pairs,the accuracy of extracting emotional reason pairs is improved by fusing the word vector of the position information.Through experiments,this paper verifies the effectiveness of the proposed EDAECPE algorithm and SDF-ECPE model on the extraction of annotated corpus based on the sentimental reasons of Weibo news.