Research On Sentiment Analysis Method For Case Microblog Comments

In recent years,the work of courts has attracted widespread attention from online media and public opinion,and the impact of online public opinion has brought tremendous pressure to the work of courts.Emotion analysis of case microblog comments helps courts and other related departments understand and grasp case microblog public opinion in a timely manner,and then monitor and guide the dissemination of public opinion.Emotion analysis of case microblog comments is a domain-specific task with research problems such as the absence of a public dataset and the domain nature of emotion expression,which is worth studying.This paper studies the emotion analysis method for case microblog comments in the following aspects:(1)Construction of case microblog comment emotion analysis dataset: The construction of a dataset with annotation is the basis of supervised learning.This paper proposes a dataset construction method for case microblog comment emotion analysis.Firstly,the case microblog comment data are collected from Sina Weibo platform.Secondly,two data annotation methods are proposed for the different demands of case microblog comment emotion classification task and emotion-cause pair extraction task.Finally,the case microblog comment emotion classification dataset and the case microblog comment emotion-cause pair extraction dataset are constructed.The work plays an important supporting role for the later research work.(2)Emotion classification method of case microblog comments integrating emotional knowledge: Traditional emotion classification methods are difficult to effectively use the emotion knowledge such as emoticons,domain emotion new words,negation rules and degree adverb rules commonly used in case microblog comments.Therefore,this paper proposes an emotion classification method for case microblog comments by fusing emotional knowledge.Firstly,a microblog emotional knowledge base is constructed by integrating the existing emotion computing resources.Secondly,considering the role of the emotional knowledge base and part of speech,15 kinds of emotional knowledge are defined to construct the emotion knowledge representation of comments.Then,using convolutional neural networks and attention networks to obtain deep semantic features and emotional knowledge features.Finally,the emotion classification is completed by fusing the above two features.The experimental results on the case microblog comment emotion classification dataset show that our method is more effective and advanced compared with the benchmark method.(3)Emotion-cause pair extraction method based on joint learning for case microblog comments: The benchmark method of emotion-cause pair extraction is a two-step method with error transfer problem.The researchers propose the end-to-end ECPE method,called ECPE-2D,which can overcome the error transfer problem of the two-step method.However,ECPE-2D retains a hierarchical coding structure based on Bi-LSTM+Attention and Bi-LSTM stacking,which has problems such as non-parallel processing and gradient disappearance.In order to better solve these problems,this paper proposes a joint learning-based emotion-cause pair extraction method for case microblog comments.Firstly,based on pre-training BERT and Transformer,joint learning of emotion extraction and cause extraction is performed,and all possible emotion-cause pairs are represented by a two-dimensional representation scheme.Secondly,a two-dimensional Transformer module is used to model the interaction between different emotion-cause pairs.Finally,a Softmax layer is used to predict the real emotion-cause pairs.This method integrates two-dimensional representation,interaction,and prediction into a joint learning framework.The experimental results on the case microblog comment emotion-cause pair extraction dataset show that our method is more effective and advanced compared with the existing method.(4)Prototype system for emotion analysis of case microblog comments: This paper design and implement a prototype system for emotion analysis of case microblog comments.The system has the functions of data collection,emotion classification and emotion-cause pair extraction for case microblog comments.To facilitate users to access the service more quickly,the system also provides a visualization interface.