Research On Sentiment Analysis Of Online Reviews And Extraction Of Negative Representative Reviews

With the rapid development of the Internet,consumers are more and more interested in online shopping,and various e-commerce platforms have also opened evaluation functions which facilitate consumers to express their opinions on products and services.Therefore,a large number of online reviews containing users' sentimental tendencies and opinions on product evaluation have been generated.These reviews are very important for consumers to make purchase decisions and businesses to make marketing strategies and product optimization.Therefore,how to mine consumers' sentimental tendencies in massive online reviews and further dig out the key issues of commodities on this basis has become the current research hotspot.Aiming at the problem that the effect of traditional machine learning methods on sentiment analysis still has a large room for improvement and poor portability in the field,this dissertation designs an online review sentiment analysis model based on deep learning.At the same time,on the basis of sentiment analysis,this dissertation uses topic model and clustering algorithm to mine the key issues of negative reviews.The main research is as follows:First,for the problem that traditional machine learning models rely heavily on manual feature selection and parameter tuning,combining CNN and BiLSTM based on attention mechanism are proposed for sentiment analysis of online reviews.This method can use the CNN for local feature extraction,and can also use the BiLSTM to learn the sequence features of the text.At the same time,the attention mechanism also assigns greater weight to important words,thereby avoiding paying attention to the useless words for analyzing sentiment.The experimental results of the data sets in the five major fields show that the accuracy of the dissertation model for online reviews sentiment classification is higher than that of the traditional machine learning model.The F1 value can reach 95%,and it also proves that the model has good field portability.Second,on the basis of sentiment classification,for negative reviews containing rich hidden topics,the Gaussian LDA topic model is used for topic mining on negative reviews.Then this dissertation uses spectral clustering algorithm to cluster the mined topics,which clusters negative comments with similar topics into one category and takes the central comments of each cluster as representative comments of that category.Thereby it reduces the problem of information redundancy caused by massive negative reviews,so that consumers and businesses can quickly and comprehensively understand the key issues of commodities.The model proposed in this dissertation improves the accuracy of online reviews sentiment analysis,and mines the key issues contained in negative reviews based on sentiment analysis.The model reduces the information redundancy problem of negative reviews,which has certain practical guiding significance for consumer,e-commerce platforms and businesses.