Study On The Method Of Ironic Identification Based On Network Text

With the rise of network culture such as online forums and paste bars,users are using irony more and more frequently as an expression of strong emotion.Irony,also known as an antithesis,is an ironic statement when speaking or writing,which is not only literally unable to understand what it really means,and it's true meaning is the opposite of the literal meaning,most of the time from the context and context,or even current affairs to understand.This makes it difficult for webmasters to quickly and accurately obtain the true intentions of a large number of users.To network regulation and public opinion collection has brought no small obstacles.In order to solve this problem,this paper uses deep learning to process and analyze the satirical text in the network,and then obtain the user's true intentions and views.The study of irony identification is mainly faced with the problem of(1)not mature,complete and accurate Chinese the irony prediction library for our study and research;(2)Most of the network text is short,refined language-based,it is difficult to obtain context and context ug time information,but the recognition of irony precisely needs to analyze its intrinsic meaning through context and context;(3)Due to cultural differences and language habits,the current relatively mature Method of English ironic identification cannot be directly used.The purpose of this thesis is to use computer technology to understand the true meaning behind the user's satirical sentence.Previous researchers have experimented with the complex features of many people's work-labeling,and a variety of classic machine learning methods.This thesis explores how to apply the deep learning model of word embedding and neural network to the recognition of irony.In this paper,four different models are used,namely convolutional neural network,convolutional neural network based on network topics and feature words,circular neural network with attention mechanism based on network topic situ and feature words,and a circular neural network with attention mechanism combined with the subject pattern processing context.Breaks through the limitations of previously identified irony ignoring the context.Among them,convolutional neural network,convolutional neural network based on network topics and feature words,circular neural network with attention mechanism based on network topics and feature words,these three models are for the contextless network short text of the ironic recognition,such as Twitter and Weibo.A circular neural network with attention mechanism that combines the subject pattern processing context for ironic recognition of comments on context-based web articles,such as comments posted below.This thesis artificially marked two Chinese satirical materials,one is a contextless network short text of a total of 20,000 words,of which 1933 is a sentence containing irony,the other is a context-based network article comments a total of 2000 articles and17067 related comments.Results In contextless languages,the long-and long-term memory network based on attention mechanism performed at the best F1 at 85.9%,better than the benchmark value.In addition,in the context of the material,a circular neural network combining the attention mechanism of the subject pattern processing context is proposed,which breaks through the short board of analyzing only single sentences and ignoring the context,and the F1 value of the ironic recognition is 87.8%.