Identification And Sentiment Discrimination Of Rhetorical Questions Based On Deep Learning

With the rapid development of online social media and e-commerce,people are not only limited to simply browsing the web,more and more people are beginning to participate in the construction of the network,such as comments on the purchase of goods,express one's own opinions and comments on the person or thing he sees,so on.This kind of behavior of users will produce a lot of text information,these information often contains a lot of valuable information about the user's experience,attitude,and opinion on the product,person or service.The sentiment analysis of the text is to find the valuable texts with subjective from these massive information.When users post information on the Internet,they do not always express their opinions simply and directly.They often use a large number of rhetorical devices such as parallelism,metaphor,personification,irony and rhetorical question,etc.The application of these rhetorical device undoubtedly produces a great challenge to the sentiment analysis of the text.This paper studies the rhetorical question,and taking the real Chinese Weibo as the experimental corpus to carry out the research on the identification and sentiment discrimination of the rhetorical questions.The main research work of this paper is as follows:(1)Analysis of the problem of the rhetorical question.When users express opinion in a rhetorical approach,the sentence may not contain obvious emotional vocabulary,we refer to sentences that do not contain explicit words but can express subjective information as implicit sentiment sentences.According to our statistics,rhetorical questions account for about12.31% of implicit emotional sentences.If the sentiment of the rhetorical question can be effectively recognized,the overall efficiency of the implicit sentiment analysis will be improved.Therefore,this paper takes the rhetorical questions as the research object,and analyzes the sentencestructure and semantic features of the rhetorical questions.In addition,2,580 the rhetorical questions were collected and compiled from the annotations of Chinese Micro-Blogs.Combined with the results of the annotations and the linguists' research results on the rhetorical questions,28 kinds of specific sentence structure were collected.(2)Rhetorical questions recognition.According to the structure of the sentence structure of the Rhetorical Question,this paper integrates the sentence structure characteristics of the Rhetorical Question into the construction of the convolutional neural network,and proposes a method of combining the sentence structure of Rhetorical Questions with the construction of convolutional neural network.In order to verify the effectiveness of the method,the comparison experiments with SVM,CNN and LSTM were carried out on the labeled Micro-Blog corpus.The results show that my method works best.Using this method,the recognition Precision,Recall and F1 value of the Rhetorical Questions in Weibo reached89.5%,84.2% and 86.7%,respectively.(3)Rhetorical Question sentiment discrimination.Rhetorical Questions are different from other rhetorical devices,which often emphasize certain aspects in semantics and also express subjective views on the emphasized information.Based on the characteristics of the Rhetorical Questions and the sentiment information contained in the sentence,this paper designs a attention loop neural network model based on multi-feature fusion for the sentiment discrimination of Rhetorical Questions.In this paper,the sentiment information of the Rhetorical Question is marked as positive,negative and neural,respectively.The comparison experiment with other methods is carried out on the Micro-Blog Rhetorical Questions data set.The experimental results show that the method is used to analyze the sentiment of Weibo Rhetorical Questions.Macro-Precision,Macro-Recall and Macro-F1 values can be achieved 90.5%,84.7% and 87.5%,respectively.