Research On Text Sentiment Analysis Based On BERT-BIGRU Model

With the popularization and application of the mobile Internet,various online service software makes people’s lives more convenient.People can make plans and arrangements in advance when purchasing,eating,living,and transporting through various apps.Social commerce that assists the purchase and sale of goods through social interaction and user-generated content has also become an important business model.Some online service software leverages product reviews and customer interactions as an important way of guiding potential customers.Through sentiment analysis of customer reviews,it can not only help service software platforms to fully understand customer needs,but also help customers quickly find desirable services and products.Sentiment analysis is a hot topic in the field of natural language processing.The mainstream research methods mainly rely on the word vector model to transform the content of the reviews,and then using the neural network model of deep learning to classify the reviews.This article selects hotel customer reviews from Ctrip.com,and employs crawler technology to collect data from nearly 90,000 hotel customer reviews.After data cleaning,stop-words removal,text segmentation,and sentiment tagging,an experimental data set is obtained.This article first introduces the Two-way gated recurrent unit model(BiGRU model).The BiGRU model has strong learning and generalization capabilities.The classification results of this model makes full use of the contextual information of the input sequence,and is more suitable for text-types data.Secondly,the BERT model is introduced.As a pre-training language model,the BERT model only needs to fine-tune according to the downstream tasks in the natural language.The corresponding downstream task in the natural language in this article is sentiment analysis.The chinese＿roberta＿wwm＿large model suitable for Chinese in the BERT model is selected to transform the hotel customer reviews verbatim into a vector format that the computer can understand.Based on the BERT model and the BERT-BiGRU model,this paper conducted two sentiment analysis experiments.During the experiment,the parameter were adjusted many times so as to obtain a model with higher classification accuracy.In the two experiments,the BERT layer is a word vector,and the input text sequence is vectorized and transformed;the softmax activation function is added to the output layer of the experiment to classify the vector.Through many experiments,it is concluded that the accuracy rate of the BERT model can reach 91.36% after the optimal parameter selection,and the accuracy rate of the BERT-BiGRU can reach 91.41% after the optimal parameter selection.The classification effects of the two models are excellent,and the latter better.