Design And Implementation Of Product Reviews Sentiment Analysis System

In recent years,with the mature development of e-commerce,online shopping has been more popular.While browsing websites,consumers can post product reviews on the websites.How to fully mine these information is particularly important.Sentiment analysis technology is generated from it.The common sentiment analysis method,one is based on sentiment knowledge and the other is based on machine learning.Sentiment knowledge can study the sentiment tendency of words,which has limitations.Sentiment analysis methods based on machine learning mainly use traditional machine learning methods at present.With the rise of deep learning technology,more scholars have applied them to the field of sentiment analysis.Therefore,this paper explores using deep learning methods for sentiment analysis,to improve the accuracy of sentiment analysis.In this pape,the goal is to build an efficient sentiment analysis system of product reviews.The data set is partly crawled from JD.COM by Scrapy framework,partly from open data set on the Internet.The Chinese data operate segmentation,filtering of part of speech and removal of stop words.By improving Word2Vec model,the word vector matrixs input into the classifier.This paper focuses on the design of sentiment analysis classifier.The GRU model has a simple structure and a short training time.It can effectively solve the problem of long and short time series changes and obtain global text features.TextCNN can reduce the size and complexity of data,capture the n-gram features of text data.This paper tries to use the hybrid model TextCNN-GRU.The model first enters the embedding layer to complete the word vectorization training,then extracts the local optimal features of the text in CNN part,gets the global sentence expression in GRU part,and finally outputs the sentiment classification.In order to improve the effect of the model,this paper sets up many contrast experiments on parameter selection,such as word vector dimension,GRU layer number,sliding window size and so on.By comparing TextCNN-GRU with TextCNN,GRU and SVM,Naive Bayes,KNN,it shows that TextCNN-GRU has higher accuracy in sentiment analysis than the other five models.It shows that TextCNN-GRU can synthesize the respective advantages of TextCNN and GRU,achieve better experimental results.