Research On Sentiment Analysis Of Review Text Based On Deep Learnin

Along with the use of application software,various kinds of review data come into being.Some of these review data are consumers’ subjective evaluation of a product or service,and some are netizens’ opinions on a certain phenomenon or event.Reasonable processing of these review data is not only helpful to provide direction for businesses to improve the quality of their products and services,and help more consumers to provide valuable references when making product choices,but also to help the government when formulating or adjusting policies.This paper focuses on using deep learning methods to analyze the sentiment of consumers’ review texts after staying in hotels.The main research elements are as follows:(1)To address the problem that review texts are mostly unstructured data,this paper uses regular expressions to process the noise in the original hotel review text dataset,and then carries out Chinese word separation by jieba word separation tool,and finally uses integrated deactivation words for deactivation word processing,etc.The traditional word embedding method(Word embedding)in neural network has the problem of poor classification accuracy when dealing with domain-specific text data.In this paper,we use the Skip-Gram method in Word2 Vec model to do vectorized representation of hotel review text to generate a word vector representation exclusive to the domain of hotel review text,and through comparison experiments,we show that this word vector The comparison experiments show that this word vector representation is effective in improving the accuracy of hotel review text sentiment classification.The central themes and keywords of the dataset are better presented by the word cloud map.(2)To address the problem that convolutional neural network(CNN)cannot accurately extract historical and future information of hotel review text,this paper adds bi-directional gated recurrent unit(Bi GRU)to the CNN model and constructs a CNN-Bi GRU model,which first uses the convolutional layer to obtain useful information of hotel review text,finds the matrix with the most sentiment feature vector in the matrix by pooling layer,then adding Bi GRU layer for global information extraction,and finally outputting sentiment analysis prediction by output layer.Multiple sets of parameter comparison experiments are set,and some optimal parameters are selected to compare with the traditional deep learning model on the hotel review text dataset,and the results prove that the CNN-Bi GRU model has better classification effect.(3)Because some review texts are long and the distance between words in the sentences is long,the CNN-Bi GRU model can only simply capture the pre-and post-textual information of each word vector in the input sequence,and cannot adjust the important information in the text.Therefore,in this paper,based on the CNN-Bi GRU model,an improved attention mechanism is added to construct a CNN-Bi GRU-Self Attention model for sentiment analysis of hotel review texts,with Accuracy(Accuracy)and F1 value as evaluation indexes,and the experimental results show that the CNN-Bi GRU-Self Attention model can ignore the distance between words and capture the dependencies between long-distance words to further improve the performance of text sentiment analysis.