Research On Sentiment Classification Of Weibo Comments Based On LSTM Dual-channel Word Feature Fusion

With the rise of the Internet era,people are more inclined to express personal opinions on the Internet,and at the same time,negative public opinion is spreading rapidly on the Internet like a virus.As a widely used information acquisition and communication tool,Weibo is an important medium that reflects online public opinion.The comments below its hot topics reflect everyone's emotional tendencies on the topic.The sentiment classification of Weibo comments is conducive to public opinion supervision,maintaining the security of the network environment,and at the same time providing a reference for the government to formulate relevant policies.This paper studies the sentiment classification of Weibo comments based on deep learning.The main work is as follows:First,in view of the problem of using pictures to comment on Weibo to avoid system detection,the tesseract-based optical character recognition technology is used to convert the text in the picture into text information,which solves the problem that microblog picture comments cannot be directly detected by the text recognition system.At the same time,the basic principles of traditional text representation methods,word2 vec and term frequency–inverse document frequency are introduced.Aiming at the high feature dimension in traditional text representation methods and the problem of not considering contextual semantic relations,word2 vec combined with term frequency–inverse document frequency is proposed.The text representation method has stronger feature extraction and representation capabilities compared with a single word2 vec or word frequency inverse text frequency,and achieves feature dimensionality reduction.Then,in order to solve the problem of low classification accuracy in the traditional sentiment classification model,a two-channel word feature fusion based on long-and short-term memory network is proposed.The model uses word frequency and inverse text frequency weighted word vectors and word vectors as dual-channel features to be input to the bidirectional long-term short-term memory network for learning,and the feature fusion results are input into the long-term short-term memory network for further learning,and finally the sentiment classification results are output through the fully connected layer.Experimental verification shows that this model has a higher classification accuracy than other models in the classification of Weibo comments.Finally,based on the classification model,the sentiment classification software of Weibo comments is designed.The software obtains the text and picture comments of Weibo through web crawlers,and enters the proposed two-channel word fusion Weibo comment sentiment classification model based on the long and short-term memory network to classify,and determine whether the comments are positive or negative comments one by one,and make statistics obtain the proportion of sentiment classification,and at the same time count the word frequency according to the word segmentation result,and draw the word cloud.