Research On E-commerce Product Review Sentiment Analysis Based On Machine Learning

The rapid development of e-commerce is profoundly changing and affecting the production and life style of the whole society.But the quality problems of e-commerce products are still prominent.With the expansion of the scale of online transactions,a large number of product reviews,as an important public opinion data,contain a lot of valuable information.First of all,consumers can understand the product quality reputation through product reviews to make a reasonable purchase decision;secondly,businesses can find the problems of products through reviews,and improve and perfect them in time.Finally,it can provide some reference information for quality supervision department to grasp and solve the quality problems of e-commerce products.Because the results of emotional analysis of e-commerce product reviews can provide valuable decision-making information to consumers and businesses,so how to better complete the emotional analysis of e-commerce product reviews has great research significance.Therefore,in view of the credibility of the results of garbage comment impact analysis and the higher value of emotional analysis based on product features,this paper takes e-commerce product reviews as the research object,and conducts research in three aspects: garbage comment recognition,product feature extraction and comment text emotional analysis.The main work is as follows:(1)Aiming at the problem of spam comments in e-commerce product review data,this paper proposes a fusion clustering algorithm based on DBSCAN and Mean Shift.The experimental results show that by combining,it can can effectively solve the problem that Mean Shift has a long running time and the accuracy is affected by the random selection of the initial centroid due to the number of iterations.(2)Aiming at the fact that LDA can not effectively extract product features from e-commerce product reviews,this paper proposes the AP-LDA method.Experiments show that the improved algorithm can not only recognize the dominant product features more accurately,but also extract the recessive product features effectively,which makes the product features more efficient and accurate.(3)In order to analysis the emotional tendency of e-commerce product reviews,this paper proposes a Bert-Bi GRU deep learning algorithm,which uses Bert to extract text features,and then uses the bidirectional GRU plus full connection layer,and add sparse attention mechanism to discriminate the emotional tendency of the text.Theexperimental results show that the Bert-Bi GRU deep learning algorithm proposed in this paper has better performance than the traditional cyclic neural network model in the emotional analysis task for e-commerce products.