Rating and reviews provided by customers can send many signals to the market, especially in online marketplace. How to formulate an outstanding sales strategy and show users the most useful information of the product is particularly important. In this paper, we study the review information cascade and build a MLP model to distinguish the spam and informative reviews from the noise reviews. The information cascade phenomenon in E-commerce reviews is identified by Multinomial Logistic Regression. After seeing a series of one/five star ratings, consumers are more likely to write low/high score reviews and the effect caused by low star rating is more significant than that of high star rating. Therefore, we use the helprate of the review as the label to distinguish the informative reviews and spam reviews. The one-hot review vectors is reduce to 136 principal components components as the input variables to train the MLP Model. The accuracy of the model on the test set is 78.5%, and AUC is 0.713. The E-commerce companies can evaluate the informative review, improve online sales strategies, enhance the product desirability and identify customers’ preferences through the model proposed in this paper.
Zhang, Haoran and Chen, Yin, "Information Cascade and Spam E-commerce reviews filtering" (2022). WHICEB 2022 Proceedings. 5.