The Method And Application Of Reliability Prediction Of Product Reviews In Social Media

With the rapid development of the Internet.Users can be active in the electronic community or shopping site publishing and provide a wealth of product information,while make comments for the property of a variety of products.Naturally,many consumers read the product review information to help decide whether to purchase the product.However,with the increase of online shopping crowd,the product online comment information is also more and more,for a large number of reviews of the same product,it is difficult to quickly obtain credible evaluation to really understand the true quality of the product.Likewise,product manufacturers need to obtain credible review information to help determine customer feedback on the product and improve the product.Therefore,the research on the credibility of product reviews is of great significance to the development of electronic commerce.This paper studies the predictive methods and applications of product reviews credibility in social media.The credibility of product reviews can be considered as a kind of short text classification problem.Compared with normal text classification problem,short feature space has higher dimensionality and eigenvector is more sparse.The classical text classification method in short text classification problem has a poor performance.At the same time,the factors that affect the credibility of product reviews can't be confined to a single comment text feature.Based on this,this paper puts forward the feature expansion and selection method for the credibility prediction of the commentary from the perspective of the text features,statistical features,semantic features and commentary meta information,and finally obtains the feature library for the credibility prediction of product reviews.Product review information in social media is often changing over time.Offline machine learning model is difficult to effectively adapt to the credibility of the information distribution changes,Therefore,based on the study of product reviews credibility prediction's feature expansion and selection.This paper propose the IOLR algorithm,it improve the traditional online learning algorithm from the aspects of enhancing error boundary learning,incremental iteration multiple times,introducing learning rate decay and regularization terms.Then,based on the text features of the product reviews and other aspects of the impact of the review's credibility build the classification model,and continue to optimize the model.Then compared with the existing online learning algorithm,The results show that IOLR is effective in improving modeling efficiency and prediction accuracy.Finally,a product credibility prediction application system is designed and implemented.The validity of the method is validated from the perspective of application.