Research And Implementation Of Aspect-based Sentiment Analysis Model Based On Deep Learning

With the rapid development of the Internet,the global business competition environment is constantly changing.Using Internet technology to collect user evaluation has become a compulsory course for most of the enterprises.In addition,with the continuous innovation of deep learning,using neural network for automatic analysis on customers’ reviews has become an important way to improve the efficiency of enterprises.But the coarse-grained sentiment classification of the whole review text is far from enough.The fine-grained sentiment analysis on the specific target of the review text can really help the relevant personnel to greatly improve efficiency.This requires the use of a more targeted subtask in text sentiment analysis,namely aspect-based sentiment analysis.Based on such background,this paper mainly completes the following work.Firstly,a new object level sentiment analysis model named SGCNN is designed.Object level sentiment analysis is also called aspect-based sentiment analysis.Its main goal is to identify the sentiment polarity of the input text for a specific sentiment target.It includes two subtasks:aspect-category sentiment analysis and aspect-term sentiment analysis.For this task,according to the design idea of parallel computing,this paper uses multiple convolutional neural networks combined with self-attention mechanism and gating mechanism to enhance the learning of internal features of the text.Experiments on multiple data sets verify the reliability of the model in accuracy and the rapidity of parallel design.Secondly,a sentiment aspect extraction model based on multiple CNN and attention mechanism is designed.Because the aspect-based sentiment analysis model needs labeled sentiment aspects,only the aspect-based sentiment analysis model is not enough to form an applicable end-to-end system.In this paper,we use the design of parallel computing to construct an affective aspect extraction model named Attention-MCNN based on multi-CNN and scaled dot product attention mechanism.The model only needs general word embedding and labeled corpus as input.By testing its effect on multiple data sets,it is proved that compared with previous research results,it can ensure the accuracy and avoid additional data collecting work.Finally,with the help of multiple development tools,an online sentiment analysis prototype system for customer reviews is designed and implemented,which combines with aspect-base sentiment analysis model and sentiment aspect extraction model.The system can provide fine-grained customer review analysis results,help users quickly understand the customer review status,quickly locate product problems and solve them.As an end-to-end system,the algorithm analysis process does not need other manual intervention,it realizes a one-stop service,and breaks through the application barriers of aspect-based sentiment analysis algorithms.