Research On Aspect-Based Sentiment Analysis Model That Integrates Multiple Semantic Features

Aspect-based sentiment analysis is mainly applied to fine-grained sentiment classification of text data such as opinions and comments.Currently,deep learning methods are mainly used to analyze the sentiment attitudes towards multiple entities in text sentences.The results of aspect-based sentiment analysis can provide a basis for decision-making and better grasp of demand,and have high value.Currently,aspectbased sentiment analysis models that use graph neural networks combined with syntactic dependency trees have good performance,but there are still some issues with insufficient utilization of local contextual information.This paper addresses these shortcomings by adopting a feature interaction learning method to enable the model to more efficiently and accurately extract the sentiment features of each aspect word.As electric power customer service work order text data involves handling user sentiment issues,this paper applies the model to this data.Details are as follows:(1)An interactive relation graph attention network model is proposed to address the issues of existing research ignoring the type of syntactic dependency relationship information in text sentences,insufficiently exploring the potential semantic information contained in relationship types,and ignoring the relationship between dependency relationships and relationship types.The model extracts feature information of relationship types,allowing them to interactively learn with the context feature information extracted by the graph attention network,strengthening their respective feature representation abilities.Finally,the aspect attention mechanism is used to fuse features,and a classifier is used to capture sentiment classification results.This model effectively mines semantic features of relationship type information and improves the sentiment classification ability of the model.(2)A model for aspect-based sentiment analysis that combining multi-windows local information and graph attention networks is proposed to address the problem of models relying too much on sparse syntactic dependency tree learning for feature representation,leading to insufficient local information learning ability.First,the multiwindow local feature learning mechanism is used to learn local context features,mining potential local information contained in the text.Secondly,a graph attention network that can better understand the dependency tree is used to learn the syntactic structure information represented by the syntax tree,generating syntactic-aware context features.Finally,these two features representing different semantic information are fused to form a feature representation that includes both syntactic information from the dependency tree and supplemented local information.The model can efficiently discriminate the sentiment polarity of aspect words and more accurately predict the sentiment polarity of aspect words.(3)Based on the above research,a model for aspect-based sentiment analysis that fuses relationship type features with local information features is proposed.The model simultaneously mines semantic features of relationship types and local information,and effectively fuses these two semantic features with context features through an interactive attention mechanism,improving model performance and applying the model to electric power customer service work order text data.Finally,experiments verify the effectiveness of the model and achieve aspect-based sentiment analysis of text data in the electric power field.