Research And Design Of Character Stance Portrait Based On Graph Neural Network

In the era of big data,the content of social media platforms is also becoming richer and more diverse.The importance of user-created content in the data is becoming more and more prominent.The purpose of this thesis is to study the user’s persona stance by processing the data and content created by the user and mining the user’s persona stance.The traditional persona stance analysis relies on sentiment analysis techniques to obtain the sentiment polarity of specific aspect words in the review text through fine-grained sentiment analysis tasks,i.e.Aspect-based Sentiment Classification(ASC).The main problem of this task is how to classify sentiment by using the connection between aspect words and sentence context.Traditional aspect-level sentiment analysis models predictions by studying the syntactic structure of sentences,since the structural relationship between aspect words and corresponding viewpoint words often indicates sentiment orientation.Thanks to recent improvements in neural network-based syntactic dependency parsing,ASC models based on more accurate syntactic trees have improved significantly.Recently,graph convolutional network approaches based on dependency trees have received increasing attention,so this paper explores the use of graph neural networks and syntactic dependency trees for fine-grained aspectual word sentiment analysis,and investigates methods for portraying character stances by analyzing users’ social graph networks.First,this paper proposes an aspect-level sentiment classification algorithm based on graph attention networks,which enriches the feature representation of sentences by encoding and extracting syntactic features and syntactic meaning features embedded in the sentences.For these two features,this paper also proposes a more efficient way to fuse and extract them,solving the problem of information loss when features are fused.Through experiments,the necessity of introducing these two features and the effectiveness of the feature fusion approach are verified.Secondly,this paper proposes a character stance algorithm using social graphs for inference on aspect-level sentiment classification,by extracting and modeling the feature information of users and inter-users,and introducing social influence indices to construct a social graph of users.By combining the feature information in the graph,the problem of inaccurate judgment of stance caused by missing features in the construction of persona stance is solved.Through experiments,the effectiveness of the model is verified.Finally,this paper follows the design ideas and implementation process of software engineering,and designs and develops a character stance portrait analysis system,which implements functional modules such as character stance analysis,character social graph,and speech prediction analysis,thus landing the research into practical applications.