Sentiment Analysis For Product Reviews Based on Deep Learning

作     者：Shaozhang Xiao Hexiang Wang Zhi Ling Lanfang Wang Zhaoxia Tang 

作者机构：Faculty of Computer and Software Engineering Huaiyin Institute of Technology Huai'an Jiangsu 223003 China 

出 版 物：《Journal of Physics: Conference Series》 

年 卷 期：2020年第1651卷第1期

学科分类：07[理学] 0702[理学-物理学] 

摘      要：With the popularity of the Internet and e-commerce, the sentiment analysis of text can help users to quickly and accurately obtain effective information they are interested in from massive product reviews to purchase satisfactory products. In this paper, a sentiment analysis system for product reviews was designed based on deep learning, and the digital and electronic products on Jingdong Mall with at least 100,000 reviews were crawled as the training data set. After data pre-processing operations such as word segmentation and removal of stop words for product reviews to remove useless features, the feature vectors were constructed based on the bag of words model and word2vec method, and then three classification algorithms, namely LSTM, Naive Bayes and logistic regression were used to model reviews. The LSTM algorithm is significantly superior to Naive Bayes and logistic regression algorithm in the training stage, and provides a reliable reference for the analysis of product reviews.