Research On Fine-Grained Text Sentiment Analysis Technology Based On Syntactic Trees

With the rapid development of computers,digital data has shown an exponential accumulation trend in recent years.The massive information in it has made data mining technology emerge as the times require.As a specific application of data mining,natural language processing has received extensive attention from the industry and academia.Text sentiment analysis refers to using natural language processing and text mining technology to identify and extract subjective information in original natural language texts.From the perspective of the granularity of the analysis object,text sentiment analysis can be divided into three levels: document level,sentence level,and aspect level.Aspect-level finegrained text sentiment analysis is more suitable for feedback,reference,and vertical analysis since people usually need to be clear about the specific drivers of sentiment.Existing studies have shown that mining and utilizing syntactic information in text materials can help improve the performance ceiling of fine-grained text sentiment analysis.This paper focuses on fine-grained text sentiment analysis based on syntactic trees and studies the two main problems in the existing work.On the one hand,the syntactic tree is a tree diagram obtained through syntactic analysis.The structural,hierarchical,and functional relationships of the components(aspects and their contexts)are encoded.How to effectively mine the syntactic information contained in the syntactic tree and introduce it into fine-grained text sentiment analysis is an urgent problem to be solved.On the other hand,due to the differences in the construction of the syntactic tree and the angle of syntax capture,the same sentence often has syntactic information from different views.Considering only syntactic information under a single view is difficult to achieve optimal aspect prediction performance,so how to effectively introduce syntactic information from multiple views into fine-grained text sentiment analysis is a problem that needs to be solved.In response to the above problems,this paper has carried out the following two aspects of work:(1)Aiming at how to effectively mine and utilize the syntactic information contained in syntactic trees,this paper proposes a new fine-grained text sentiment analysis model,namely the syntactic tree-driven graph neural network(STD-GNN)model.Specifically,STD-GNN contains a new induced syntactic graph construction strategy,which reconstructs the original dependency syntactic tree into a star-like topology centered on the aspect set based on the minimum tree distance.This strategy alleviates the lack of syntactic information and inefficiency that may exist in existing methods.Then,STD-GNN obtains the initial features of the sentence through BERT and applies the graph neural network to learn the star-like topology graph,thereby optimizing the feature representation of the aspect words.Finally,the learned aspect representations are fed into a classifier to achieve sentiment polarity prediction.(2)Aiming at how to effectively introduce syntactic information under multiple views in fine-grained text sentiment analysis,this paper proposes a new text sentiment analysis model: syntactic tree-driven multi-view graph neural network(STD-MVGNN).In addition to the above-mentioned induced syntactic graph belonging to the minimum distance view,STD-MVGNN also introduces the topology structure graph of the original dependency syntactic tree and the PLM induced syntactic graph constructed based on BERT,and uses a multi-view graph neural network for synchronous learning.Then,the features of different views are weighted and fused using the attention mechanism to obtain high-quality aspect feature representations that contain three kinds of syntactic information.This paper conducts sufficient experiments and analyses on STD-GNN and STDMVGNN on three public datasets to verify the effectiveness of the proposed new syntactic tree-graph transformation strategy and multi-view syntactic fusion framework.