Research On Sentiment Trend Prediction Method For Group User

Online shopping has become one of the most mainstream consumption methods,and users' comments the e-commerce platform tend to have a commodity purchase experience and sentiment opinion on the product.However,the dynamic nature of group users' comments on commodities leads to the change of sentiment trend over time.How to efficiently mine the potential sentiment views of comments,analyze and predict the real sentiment tendency of group user to the commodities,thus assisting merchants to accurately locate the user requirements,which is the key issue to promote the development of e-commerce.The existing sentiment trend prediction method mainly acquires sentiment trend from the perspective of a single user,which makes it difficult to deeply explore the sentiment trend changes of group user comments on commodities,resulting in low prediction accuracy.Starting with three aspects of feature extraction,sentiment analysis and time series sentiment trend prediction,this paper proposes a method of sentiment trend prediction for group user.Its main research work includes the following aspects:(1)In order to improve the accuracy of feature word extraction,a text feature extraction method based on multi-feature factor fusion is proposed to solve the problem of over-reliance on word frequency to calculate the weight of feature words in the traditional TF-IDF algorithm.Firstly,the traditional algorithm is used to calculate the weight of feature words.Secondly,the feature words position and part of speech factor are introduced to redistribute and sort the weight of TF-IDF algorithm.Finally,the weight of the updated feature words are calculated by the fusion of the three results.Experiments show that the accuracy of the feature words extraction of the optimized TF-IDF algorithm is relatively improved by 1.6%.(2)In order to improve the accuracy of sentiment classification,a multi-layer perception(MLP)network model based on multi-dimensional sentiment feature vector is designed.Firstly,the optimized TF-IDF algorithm is used to mine group user comments and obtain multi-dimensional sentiment feature.Secondly,the multi-layer perception(MLP)is used for sentiment analysis,so as to obtain the sentiment tendency value of group user.Finally,Support Vector Machine(SVM),Decision Tree(DT)and Naive Bayesian(NB)model are introduced for comparison.The experiments show that the accuracy of the proposed sentiment classification model is increased by 4.9% on average,and the F value increases by 1.7%,which can provide a good basis for the sentiment trend prediction stage.(3)In the process of time series sentiment trend prediction,the group user sentiment tendency value obtained in the previous stage is integrated with corresponding comment time,to construct time series sentiment tendency sequence of group user,and an sentiment trend prediction method based on multi-layer long short-term memory network(ML-LSTM)model is proposed.The experimental results on large-scale real data sets show that compared with the existing autoregressive(AR)model,the average MSE value of the long short-term memory network(LSTM)model is reduced by 0.5% and 0.02% respectively.In addition,the percentage of ML-LSTM over AR model and LSTM model reached 82.9% and 13.9% respectively,which can achieve more accurate prediction results.