Fine-grained Sentiment Analysis Based On Graph Neural Network

The development of network technology has made users more frequent online communication.Users often express their feelings about certain events or things by posting information on blogs,forums,and e-commerce websites.The analysis of these texts has very important research significance: on the one hand,it can help the government and other regulatory agencies to understand the social mood fluctuations and analyze the situation,further judge the development of the situation and make reasonable guidance for decision-making to maintain social stability;on the other hand,with With the rise and popularization of e-commerce platforms,user reviews promote more transparent product purchase information,and product reviews will greatly affect users' desire to buy.Therefore,merchants can conduct sentiment analysis on user reviews and make decisions to maximize commercial profits.Traditional sentiment analysis tasks believe that a document or sentence contains only one sentiment,and the granularity is relatively coarse.With the development of technology and the social demand for sentiment analysis,more fine-grained sentiment analysis for attribute words has gradually become a research hotspot in the field.However,the existing fine-grained sentiment analysis methods mostly treat text as a sequence of words for processing,and often ignore the syntactic structure of sentences.In order to solve the above problems,this topic proposes a graph neural network model for fine-grained sentiment analysis.It has a strong ability to fuse graph structures.Experimental results show that the integration of syntactic information through graph neural networks can improve fine-grained sentiment analysis.In order to better characterize and integrate syntactic information and further optimize for fine-grained sentiment analysis tasks,this paper also proposes a finegrained sentiment analysis method enhanced by graph information,which is improved from the two perspectives of graph neural network model and graph structure.Experimental results show that our method can not only improve the performance of fine-grained sentiment analysis,but also improve its robustness.Furthermore,this paper also proposes a cross-domain fine-grained sentiment analysis framework based on perturbation masks.We use this framework based on graph neural network to fuse the graph structure containing domain information and the graph structure containing syntactic information.The experiment proves this The framework can improve the performance of cross-domain fine-grained sentiment analysis.