Abstract
The rise of the Internet has promoted the vigorous development of online shopping. Evaluating and predicting the reputation of products is of great significance for sellers to identify the reputation trend of products and make sales plans in time. In this paper, based on the customer reviews data for hair dryer provided by Sunshine Company for Amazon Markets from August 2005 to August 2015, we propose five indexes: number of reviews, high confidence review rate, weighted average rating star, weighted average review score, and review score skewness. Then we use SSA-LSTM (Singular Spectrum Analysis-Long Short-Term Memory networks) time series method to predict the trend of product reputation. We find that the time series data processed by SSA has higher prediction accuracy than the unprocessed data, which proves that SSA processing can improve the accuracy of product reputation prediction. Moreover, we propose a product reputation evaluation method combining star rating and review data and use improved SSA-LSTM time series prediction method to predict product reputation. Overall, our results indicate that the above methods have achieved high prediction accuracy and have certain reference value for sellers' sales strategy formulation.