Sentiment Analysis Of Product Reviews Based On Text Mining

The increasingly mature Internet technology has built an efficient and diversified platform for the spread of online shopping,and there is a large amount of product review information on various forums and online shopping platforms.For the sales of a product,word-of-mouth has a significant impact,especially for consumer products such as cell phones,which have a high demand frequency.and the good or bad evaluation of the public will directly affect the revenue of cell phone manufacturers and the reputation of online shopping platfrms.Therefore,with the help of text mining technology.the sentiment analysis of review data is of great significance to cell phone manufacturers and platforms.This paper collects three brands of cell phone review data from Jingdong Mall as a dataset,and the sentiment analysis of commodity reviews is mainly divided into two parts.First,the sentiment classification,according to the brand of cell phone review s are divided into two categories of Apple and Android datasets,to build a sentiment classifier.Compare the classification effects of different models based on three evaluation indexes:accuracy.F1 value and running time,and select the sentiment classifier with the best effect,so as to accurately recognize the sentiment tendency of the reviews.The second is the implicit Dirichlet distribution(LDA)topic model,which extracts the positive and negative comment topic words of the two types of data according to the results of sentiment classification,to understand what users are satisfied and dissatisfied with the cell phone products.so as to provide e-commerce platforms and cell phone manufacturers with suggestions for improvement.Sentiment classification part.Firstly.the text data is preprocessed,and the word vectors are constructed by two methods,TF-IDF and Word2vec.Applied to the traditional machine learning model and Bert model testing,the experimental results found that:for the Apple dataset,the SVM and Logistic models combined with TF-IDF perform close to each other,but the SVM model is time-consuming.For the Android dataset the combination of TF-IDF+SVM performs best,which are significantly better than the other two models.In terms of deep learning,although the Bert model outperforms the machine learning model in both types of datasets,it takes much longer than the machine learning model.Finally,the TF-IDF+Logistic model combination was determined to be the best sentiment classifier for the Apple dataset.For the Android dataset,the TF-IDF+SVM classification model combination is the best sentiment classifier.LDA Topic Modeling Section.This paper finds that among the feature words extracted from negative reviews,Apple users are dissatisfied with the signal and battery life of their phones,while Android users are more concerned about quality control,the Jingdong platform,and after-sale issues.Based on the results of this analysis,some reference opinions are put forward to the Jingdong platform as well as cell phone manufacturers,such as suggesting that Apple manufacturers optimize cell phone charging,and Android manufacturers improve cell phone acceptance standards and after-sales service.For the Jingdong platform,it is necessary to ensure the service quality of Jingdong customer service and improve the logistics efficiency of the platform.