The Online Auction Price Forecasting Based On LSTM

The rapid development of the Internet has brought new development to the traditional auction industry.So much so that there was an online auction.The online auction transaction price and the forecast of the auction price trend have the bigger influence to the online auction.Through prediction,both buyers and sellers and the auction platform can get the required information,which is helpful to the three parties.The study of online auction data has a long history.Domestic and foreign scholars have made research and contributions in this field.Early research on auctions usually used multivariate statistical regression.However,the bidding data is non-equidistant time series data.Linear models do not fit well.And most of the relevant research regards the auction process as cross-sectional data.As a result,the dynamic information in the auction process is ignored.Now,there are many popular research methods,such as Stochastic differential equation method using static data,FDA(Functional data analysis)model analysis method.In addition,point process approach is another approach to study online auction behavior.In this paper,we predicted and simulated the final transaction price and bidding trend using the relevant data of Microsoft-xBox on eBay.Using the dynamic information in the auction process and the research idea of point process,the auction bidding process is divided into the horizontal time interval and the study of price increment.Based on the basic panel data,the dynamic characteristics are constructed,which enables the model to make better use of the dynamic information in the bidding process.A compound prediction mechanism is constructed,in which the time interval between the bidding points is regarded as the life span of the bidding points,and the COX proportional risk model in survival analysis is used to judge whether the bidding points will live or not.LSTM composite model is used to predict the bidding time interval and price increment,and dynamic transaction price prediction and bidding curve simulation,and compared with the traditional weibull process and BP network model.