Research On Sentiment Analysis Method Based On Heterogeneous Graph Convolutional Network

With the advent of the Internet era,many trendy network applications have also emerged,such as large online e-commerce platforms including Taobao and JD,or broadcast social network platforms including Xiaohongshu and Weibo.Under the influence of this trend of the times,more and more young people and even old people have turned into senior netizens.Compared with the past,netizens prefer to publish their opinions and opinions on the Internet,and these contents often contain great value.In view of this situation,methods such as user sentiment analysis and interest mining have gradually become current hot spots,so how to analyze and judge the user’s emotional polarity more accurately is an urgent issue.Aspect-level sentiment analysis aims to identify the sentiment polarity of a certain aspect in contextual sentences.Existing sentiment analysis methods simply combine the grammar dependency tree to construct graph convolution.Usually,a certain aspect of sentiment can sometimes be determined by a few words,and relying entirely on the grammar tree may distract the model’s attention.In addition,due to the limitation of the corpus,the model can only learn limited knowledge.In order to solve the above limitations,this paper proposes a heterogeneous graph convolutional network model based on aspect dependence for sentiment classification tasks.The model prunes the dependency tree and directly serves each sentence in the sentence through Multi-Head-Attention.The word selects the k words with the highest attention score,which can reduce the influence of other irrelevant information on the result.In addition,the model integrates multiple feature relationships between words by constructing heterogeneous graphs,and uses GCN to find meaningful representations for each node.It enables the model to further integrate a variety of information on the original basis,and no longer rely solely on a relationship.At the same time,this article will introduce the Sentic Net5 common sense knowledge base to participate in the construction of heterogeneous graphs,so that the model can learn knowledge outside the corpus,thereby improving the accuracy of sentiment classification.In this paper,a total of five data sets are selected,and experiments are performed on each data set to verify the classification effect of the model.In order to fully illustrate the effectiveness of the modules proposed in the thesis,this article also uses ablation experiments for in-depth demonstration.In addition,statistical analysis is performed on the choice of two parameters of the model.Finally,this article also visualized the experimental results.