Research On Chinese Implicit Sentiment Analysis Method Based On Feature Fusion

With the rapid development of the mobile Internet,a large number of discussions will appear on social platforms in view of social facts.These discourses are characterized by large numbers,many topics,colloquial structure and insufficient semantic information,especially some implicit emotional sentences.There are linguistic methods such as irony and irony,and their emotional characteristics are not obvious.These characteristics pose a greater challenge to traditional sentiment analysis methods.Experiments show that simple text sequence information cannot meet the needs of implicit sentiment analysis.This article has focused on the task of identifying the polarity of implicit emotions.Based on the traditional classification based on sequence features,the two-dimensional text sequence has been transformed into a high-dimensional graph structure by introducing grammatical structure and context information.It has mapped text to a text graph,and the original text information has been classified by graph classification through the graph neural network.The main research contents are as follows:(1)Research on feature fusion network that introduces sequence information and dependent syntactic information: This article has proposed a network(SDNN)that integrates text sequence features and grammatical structure features,extracted text sequence information and syntactic information,and constructed it.It has been reflected in the text graph representation process,and finally the classification of sentiment polarity has been realized by classifying the text graph.(2)Research on the feature fusion network that introduces context information:This article has further integrated the features of contextual text on the basis of SDNN,and proposed a network that combines three features(CDHNN).The context information has been introduced into the graph representation through the model,and finally the classification of sentiment polarity has been realized.(3)Design and verification based on the Weibo public opinion analysis method:In order to further verify the effectiveness of the proposed model in practical applications,this article has proposed a method for judging Weibo public opinion sentiment,and brought the model proposed in this article into this method to verify the actual usability of this model.This article has conducted experiments on the implicit sentiment analysis data set provided by the SMP-ECISA2019 competition to verify the effectiveness of the model proposed in this article.It has used BERT+LSTM as the baseline model,the accuracy of the SDNN model has increased by 1.6%,and the accuracy rate of the CDHNN has increased by 3.3%.It can be seen that the model based on feature fusion proposed in this article can effectively improve the ability to discover hidden emotions.