Study On Aspect-level Sentiment Classification Algorithm Based On Dependency Parsing

Although machine learning technology reduces the dependence of aspect-level sentiment classification on sentiment dictionaries,its ability to extract deep contextual features is still insufficient.The rapid development of deep learning technology has solved this problem in recent years.It does not rely on manual feature extraction and excavates deeper semantics through endto-end training and learning.However,aspect-level sentiment classification has high requirements for semantic understanding.Mining the deep semantics of text is conducive to improving the accuracy of fine-grained sentiment classification,in which the dependent parsing structure information is essential.Therefore,this thesis' s problem is to mine deep semantic information through deep learning technology and the effective use of dependent parsing information.For the above findings,this thesis focuses on improving the classification ability of the aspect-level sentiment classification algorithm and has made progress in the following two aspects:1.Based on dependency graph convolution and transferring part of speech tagging for an aspect-level sentiment classification model is proposed,namely LCF-TDGCN.It is an aspect-level sentiment classification based on multi-head self-attention,local context focus and attention mechanism,transfer part of speech tagging assistance,and dependent graph convolution network.It can correctly capture parsing information and long-distance words and solve the problem of long-distance multi-word dependence.The model learns features from the sequence level global context and the local context words related to aspect words.The local context part calculates the local context features according to the semantic relative distance.In addition,the model takes the transfer part of speech tagging as an auxiliary task to enhance the feature quality and improve the accuracy.The experimental results on five benchmark data sets show that the LCF-TDGCN model achieves an excellent sentiment classification effect and effectively solves the problem of parsing constraints in the local context.2.An aspect-level sentiment classification model based on local context and improved dependency parsing information capture is proposed,namely LCF-KVMN.It is an aspect-level sentiment classification model based on multi-head self-attention,local context focus and attention mechanism,and key-value memory network.The model also focuses on solving the problem of parsing constraints in the local context.Furthermore,we focus on effectively extracting local context information and improving the method of capturing dependent parsing information.The model uses the key-value memory network to process the dependency tree,extract the parsing information of sentence structure,correctly capture the dependency of dependency parsing,the type of dependency parsing,and long-distance words,and solve the problem of long-distance dependency.In addition,this thesis integrates the pre-trained BERT model and uses the local context attention mechanism and self-attention mechanism to improve the performance of aspectlevel sentiment classification task significantly.The experimental results on the benchmark dataset show that LCF-KVMN effectively focuses on local context and enhances the capture of contextdependent parsing information.