Research On Fine-Grained Sentiment Analysis Based On Catering Industry Evaluation

Sentiment analysis(SA)is an important part in the field of natural language processing.Its purpose is to analyze and excavate the emotional tendency contained in subjective texts.Sentiment analysis tasks can be divided into coarse-grained sentiment analysis and fine-grained sentiment analysis.Coarse-grained sentiment analysis can judges the emotional tendency of entity level,and fine-grained sentiment analysis can judge the emotional tendency of entity attribute level.With the improvement of people’s living standards,people begin to advocate individualization.The user evaluation on the network also tends to be multidimensional,which makes the evaluation text more complex.Only the evaluation at the entity level can not meet the actual needs of the public.Therefore,refining the granularity of sentiment analysis has become an inevitable trend.However,the refinement of the granularity of sentiment analysis leads to a sharp increase in the difficulty of capturing key information,and makes sentiment analysis from a single task problem to a multi task problem.Many simple sentiment analysis models are difficult to deal with finegrained sentiment analysis tasks,which greatly reduces the efficiency of sentiment analysis.In view of this situation,this paper designs a fine-grained sentiment analysis model which can not only improve the classification accuracy but also save the calculation time.The main research contents are as follows:Firstly,select the evaluation data set of catering industry,and preprocess the data to remove stop words,invalid characters and so on.Secondly,select part of the data,select the original deep learning model,and use the methods of Single-task learning and Multi-task learning to conduct sentiment analysis and comparison experiments on the selected data.Through the analysis of the experimental results,it is concluded that Multi-task learning is more suitable for fine-grained sentiment analysis.Thirdly,a method of sentiment analysis using Pre-training model alone is proposed.The BERT model and BERT + ERNIE model are used to conduct fine-grained sentiment analysis on some data on the basis of Single-task learning and Multi-task learning respectively.The experimental results verify that the Pre-training model has sufficient ability to extract features and capture context information,and can be competent for natural language processing tasks without adding downstream models,It is verified that ERNIE model can make up for the deficiency of BERT model in solving Chinese sentiment analysis problems,and verifies the ability of ERNIE model.At the same time,comparing the experimental results of the Pre-training model when using Single-task learning or Multitask learning method,it is verified that the Multi-task learning method is also suitable for the Pre-training model.Finally,the BERT-ERNIE-LSTM-Attention parallel computing model is proposed.The Long Short Term Memory networks and Attention mechanism are added downstream of the BERT + ERNIE model in order to improve the accuracy of the model,and the parallel computing structure is introduced to reduce the time spent in the training of the model.Using all the data of the data set,the model proposed in this paper is compared with several deep learning models for sentiment analysis.The experimental results show that the model proposed in this paper can improve the accuracy of sentiment analysis and reduce the training time.