Commodity Review Analysis Based On LDA And Transformer Model

The popularity of the Internet provides a platform for the development of e-commerce.With the help of these favorable factors,a large number of online orders will be generated every day,which has greatly promoted the prosperity of China’s economy.Meanwhile,the shopping platforms have gathered huge amounts of review,which are like a huge treasure house,which make complaints about customers’ products and customers’ expectations for future products.How to dig out the value behind the review and turn it into commercial wealth is a topic worth pondering.The analysis of commodity review in this paper is mainly divided into two parts.Firstly,the LDA theme model is used to extract the keywords from the established commodity review.By extracting the favorable topic words,we can make complaints about the customers’ satisfaction with the products,extract the difference theme words,and understand the customers’ place for the products to tucks.Secondly,the emotion dictionary and Transformer model are used to classify the review,select the appropriate strategy during the model training,and adjust the value of some parameters according to the specific training task,which not only solves the problem that the traditional cyclic neural network can not carry out parallel operation,but also greatly improves the accuracy and shows good performance in the test set.When using Transformer’s variant model——BERT model for classification,two schemes are adopted: treat the problem as a three classification problem and directly predict the positive,medium and negative comments;Adopt the "two-step" strategy,first predict the star rating(1-5stars)evaluated by customers,which is a five classification problem,and then map 1 star to poor rating,2-3 stars to medium rating and 4-5 stars to high praise through a nonlinear function.The experimental results show that under the same number of samples,the time cost of the above two schemes is almost the same,but the accuracy of the second scheme is higher.It is of great practical significance for buyers to make shopping decisions,merchants to improve product performance and service quality,and the market to strengthen supervision.