Sentiment Analysis Of E-commerce Users’ Comments Based On Text Mining

Due to the vigorous development of modern information technology,a very potential e-commerce industry has emerged.Due to its low cost and quick profit,more and more people are engaged in it,and the competition pressure is also increasing.If merchants want to stand out from it,Only by accurately grasping the needs of consumers is the way out.Nowadays,more and more consumers are shopping online on e-commerce platforms,and they have left a lot of user comment information with rich practical value on the platform.If they can be accurately mined,businesses can quickly understand users.Voice,grasp the user’s concerns and needs of this product,so as to enhance their competitiveness.Therefore,this thesis takes the electric water heater industry of Jingdong Mall as an example,and studies and analyzes the application value of user comments to brand merchants through data crawling,data preprocessing,user comment sentiment analysis and theme analysis.This thesis firstly uses python web crawler technology to crawl user review data of many home appliance water heater merchants on Jingdong Mall.Second,this thesis constructs multiple sentiment classification models.For the crawled user comment data,multiple processes such as data deduplication and denoising are carried out.At the same time,the preprocessed comment text is segmented into Chinese with the help of Jieba Library,and the Word2 Vec model in the gensim library is used to vectorize Chinese words as model input;then,construct multiple sentiment classification models from traditional machine learning and deep learning.The former includes naive Bayes,random forest,and support vector machine models,and the latter includes long-term memory neural network(LSTM),threshold recurrent unit network(GRU),and variant models Bi-LSTM and Bi-GRU,according to the fitting on the same dataset,the models with the optimal parameters are trained,and are carried out with the help of multiple classification evaluation indicators.The classification effects of the models were compared,and the best LSTM model was selected with the highest accuracy rate,and the best model was used to extrapolate the new review data.Then,LDA theme analysis was carried out,and high-frequency keyword word frequency statistics were carried out on the user praise and negative review data of Midea and Haier respectively,and the key keywords were displayed intuitively with the help of word cloud map,and then an LDA model was constructed to mine the data.Find out the potential themes in the good and bad reviews,summarize the potential themes,analyze the marketing focus of each brand,and give some feasible suggestions on this basis.Finally,by summarizing the results of the full text,the following research conclusions are drawn:(1)Compared with the traditional machine learning algorithm model,the modern deep learning model is better and more accurate in the emotion classification of electric water heater user comments.The highest rate reaches 0.95;(2)The sentiment classification model constructed in this thesis also has a good effect on extrapolation prediction,which can help merchants to quickly grasp the user’s emotional tendencies towards products;(3)The subject analysis of electric water heater user reviews found that two themes that appear in the comments of big brands are reflected in the quality,packaging,price,after-sales and logistics of the products,occupying a high weight,reflecting the user’s focus and demand for electric water heater products,and businesses need to focus on improving product quality,reasonable pricing,and strict professional level assessment of installation masters and customer service.