Emotional Analysis Of Product Reviews Of Online Shopping Platform Based On Machine Learning

With the development of the Internet,e-commerce has sprung up,and many consumers choose online shopping and express their own ideas and suggestions for the quality of the products purchased and the shopping experience.The product reviews of these online shopping platforms often contain a lot of useful information.It can provide reference for shoppers,which indirectly affects consumers' purchasing decisions;these product reviews also help merchants to further improve product quality and customer service,etc.At the same time,online shopping platforms can further manage businesses based on this information.Therefore,emotional analysis of product reviews on online shopping platforms has certain practical significance.In recent years,text sentiment analysis has received more and more attention and research.Some research results have been obtained in the existing research.Based on the previous studies,this paper puts forward some new research ideas.The specific research work is as follows:First,in the Chinese word segmentation problem,this paper based on Hidden Markov Model for Chinese word segmentation,and uses Word2 vec to extract text feature generation word vector.Finally,the machine learning algorithms commonly used in text categorization,namely,support vector machine(SVM),neares t neighbor classification(KNN),naive Bayes(NB),and long-term and short-term memory network(LSTM)in deep learning are used for classification.The ROC curve and AUC are used to evaluate the performance of the model.According to the results of Python experiments,the classification performance of LSTM mo del is obviously better than that of other three machine learning classification algorithms,but the training time is also the most;the classification effect of SVM and KNN is second,and the classification effect of NB is compared.Worst.In general,the classification effect of the four classification algorithms can be expressed as follows: LSTM > SVM > KNN > NB.Second,on the basis of the above,this paper introduces there methods of integrated learning.This dissertation introduces commonly used integrated learning methods including Boosting,Bagging and Stacking,and proposes a new model integration idea: firstly,holding vector machine(SVM),nearest neighbor classification(KNN),and naive Bayes(NB)as the first layer of the basic learner of stacking.Then this dissertation uses its predicted value as the input feature of the second layer.The second layer uses Boosting,Bagging,and LSTM as the meta-classifier.In the third layer,the prediction result of the second layer is voted as the final output of the total model.The experimental results show that the idea of this model integration can effectively improve the accuracy of the classifier and verify the effectiveness of this method.