Research On Aspect-Level Sentiment Classification Method Based On Syntax Enhancement

With the rapid development of internet technology and the growing popularity of artificial intelligence,China’s AI and economy and society are fully integrating,with the internet becoming the lifestyle of choice for people in the new era.The emotional tendencies contained in data such as microblogging one’s life,posting comments on products on Taobao,and leaving records of medical visits in internet hospitals can be used to help grasp social opinion,understand user tendencies,and summarise patient information,and have huge development prospects and commercial value.As a result,sentiment analysis techniques that can tap into the sentiment of text are gaining the attention of the entire community,and presenting opportunities and challenges for the field of aspect-level sentiment analysis.Aspect-level sentiment analysis aims to identify the sentiment polarity of an aspect in a contextual sentence.Existing sentiment analysis methods introduce syntactic dependencies by combining syntactic dependency trees and graph neural networks,but are often susceptible to syntactic parsing errors.A syntactic parser cannot guarantee correct parsing,which is the basis for the construction of the model graph,and if an error occurs,the impact on the subsequent classification results will be enormous.In order to solve the above problems and enhance the parsing results of the syntactic dependency trees generated by the parser,a syntactically enhanced graph attention network model is proposed in this paper for the aspect-level sentiment analysis task.The model uses multiple external parsers to syntactically parse the text to obtain multiple syntactic dependency trees,and constructs a corresponding graph structure for each syntactic dependency tree,with each edge labelled by the type of corresponding dependency relationship between words.The multiple parsing trees are fused and filtered prior to input to the model,resulting in a syntactically enhanced graph with higher syntactic parsing accuracy.Furthermore,in order to encode the augmented graph,the sentences and the augmented graph are fed into a densely connected graph attention network,which allows for the capture of richer hierarchical features.The aspectual attention mechanism is also introduced to capture the semantic features of aspectual words to enhance the model’s focus on aspectual expressions,thus improving the classification effectiveness of the model.In this paper,Four datasets were selected to evaluate the model,experiments were conducted on each dataset to validate the classification effectiveness of the proposed model,and statistical analysis of the model’s parameter selection was carried out.In order to fully illustrate the contribution of the module proposed in the paper,an ablation experiment is used for an in-depth demonstration.The experimental results validate the effectiveness of the syntactic enhancement approach,with improved classification accuracy on all three benchmark datasets and better performance in the area of aspect-level sentiment analysis.