Research On Risk Public Opinion Monitor Technology Of E-Commerce Product Quality Based On Data Mining

In recent years,the rapid development of E-commerce industry,the resulting problem has become increasingly prominent.One important issue is the quality of E-commerce products.In order to pursue short-term economic interests,some enterprises cut corners,shoddy,product fake and shoddy products in the production.The data of risk public opinion can provide the benchmark to help quality supervision solve E-commerce products quality problem,so that E-commerce product quality problems can become "Accurate monitoring,intelligent early warning".The information and comments of the electronic commerce products are used as a kind of risk public opinion data,which can reflect the trend of the electronic commerce product quality problem so that it has great reference value.Starting from the perspective of text classification and the field of data mining,and combining with the text representation model,feature dimension reduction,classification algorithm design and selection,this paper has completed the following work and research results:(1)Constructing the text model of E-commerce product information and comment.Product information and comments as a short text whose structure is complex.The computer cannot understand their meaning,so they need transformed them into the structured model through the segmentation,deleting stop words.In order to represent product information or comments expediently,VSM space vector model is introduced.In this model,each column represents a sample and each row represents a feature value with 0 or 1 representing the presence or not,then all non 0 eigenvalues are composed of a complete text message,so that the computer can identify the storage.(2)Dimension reduction to improve classification efficiency of E-commerce product information and comment feature set.Even if punctuation and stop words have been removed,the number of feature value in a text is still much.The entire text content feature set scale will greatly affect the implementation of classification algorithm.IG information gain algorithm delete the feature which have no contribution by sorting out the feature value information entropy and it have a threshold value for feature set.Compared with feature selection algorithm,feature extraction algorithm can preserve the contents of the original feature set better.(3)By constructing a fast auto-encoder neural network,RELM can quickly calculate the node weights of the hidden layer,so as to realize the feature dimension reduction.The original RELM regularization extreme learning machine is commonly used in data classification,and it achieves the feature extraction of the improved RELM with the principle of auto-encoder neural network applied to RELM.Auto-encoder neural network inherits the advantages of RELM,and effectively improves the speed of feature extraction.(4)Selecting classification algorithm to classify the E-commerce product information and comment after dimension reduction.Support vector machine makes the experiment achieved good results in processing nonlinear data,but compared with RELM,its classification speed is relatively poor.The results shows that the classification efficiency of RELM is better than SVM,and the classification accuracy is equivalent to SVM.