Research On Online Reviews Sentiment Mining Based On Deep Learning And Sequential Three-way Decision

With the continuous advancement of Internet technology and e-commerce,consumers are becoming more and more enthusiastic about online shopping.Browsing online reviews of products has long been regarded as a vital procedure of consumer buying behavior.Online reviews are becoming more and more important to consumers' purchasing decisions,and merchants or companies are paying more and more attention to online review management.It is worth noting that consumers are more sensitive to negative reviews than positive reviews.Therefore,the establishment of a more accurate sentiment classification model for massive online reviews is a very realistic subject,and it is also a research hotspot in natural language processing.However,many sentiment classification researches previously only view it as a simple two-class problem,not considering that consumers' sensitivity to positive reviews and negative reviews are different.Which means that the misclassification cost of positive reviews is also different from the misclassification cost of negative reviews.Sequential three-way decision is an effective dynamic method to solve the problem of imbalance cost.Therefore,this study proposes a new text granulation method based on deep learning.In order to further improve the classification effect of the sentiment classification model,this paper introduce ensemble learning ideas and propose online reviews sentiment classification model based on deep learning and sequential three-way decision finally.The main work is divided into three parts:Firstly,we collects substantial hotel online reviews and computer online reviews from two perspectives of service-based and product-based.After text preprocessing,the review sentences are converted into low-dimensional and dense word vectors;Secondly,this study build a text granulation method based on Bi LSTM and CNN,which granulate the text into multi-granular text information as the input of the sequential three-branch sentiment classification model;Thirdly,introducing ensemble learning ideas,this paper constructs three methods based on relative voting,weighted voting and Stacking.This study can effectively solve the mi-classification cost of different sentiment polarity by introducing the sequential three-way decision theory,provide a new text granulation method,and extend the sequential three-way decision model.Through comparative experiments on the two online review datasets,the method built in this paper not only can enhance the accuracy of the sentiment classification model,but also can cut down the cost of mis-classification.It provides a reliable information for businesses to manage online reviews.