Research On Aspect Level Sentiment Analysis Method Of User Reviews Based On Deep Learning

Aspect level sentiment analysis is more fine-grained and can classify the sentiment polarity of different aspects in a text.Its key point is how to express the implicit relationship with the context according to the given aspect in the text.The method based on the attention mechanism can pay attention to the importance of aspects in the text,and the deep learning method can learn to extract features independently,and perform better in feature expression.In order to more effectively extract the emotional feature information of the text to enhance the classification effect,this paper conducts an aspect-level sentiment analysis research based on deep learning for the user comment text.The main work is as follows:(1)A sentiment analysis model of hybrid neural network based on multi-attention mechanism is proposed.This model introduces a multi-attention mechanism of aspect,part of speech,and location,and pays attention to the connection between various aspects and context in the text from multiple angles,and makes up for the shortcomings of relying only on the content-level attention mechanism.The improved network Bi GRU is used to make up for the large amount of LSTM parameters and slow model convergence.At the same time,it is fused with CNN to obtain an emotional semantic vector that has both CNN local features and Bi GRU global features.Experiments were performed on the laptop user reviews and restaurant user reviews data set in Sem Eval2014 Task4.The results show that the performance of this model is better than other benchmark models.(2)A sentiment analysis model based on BERT and dual attention mechanism is proposed.The model is based on the characteristics of BERT that is good at processing sentence pairs.It uses sentence pairs consisting of original sentences and auxiliary sentences constructed by them as network input.The BERT pre-training model is used to solve the existence of hybrid neural network models based on multi-attention mechanisms.problem.At the same time,in order to more fully process the textual information,aspect word-level attention and aspect sentence-level attention are added to the fine-tuning structure of BERT.Aspect word-level attention is used to solve the problem of different words and contexts in the aspect,and aspect sentence-level attention is used to solve the problem of the model's attention to the aspect in the sentence.Comparing experiments with other models on three different data sets verifies the effectiveness of the network based on BERT and dual attention mechanism.