Predictive Models Based On EMD And Web Search Data

In the face of the increasingly competitive domestic book market in China,in order to obtain the predicted trend of book sales and better to help publishers make decision on publishing topics,especially in certain subsectors.It is essential to have an insight into the dynamics changes of the book market.In order to get the prediction trend of some subdivided book markets,we need to make use of advanced model and technology to carry on data analysis,model selection and market prediction.Analysis the book market could help publishing enterprises to make better production decisions in certain sub-areas.More scientific decisions could help publishing enterprises optimize limited resources,advance the layout of the production of books resource planning,further deepen the development of enterprises and improve the core competitiveness of publishers.With the continuous changes of the Internet and information technology,the information level of the whole publishing industry has been gradually improved,making it possible to collect a lot of data.These large amounts of book sales data are collected through platforms has became a very valuable reference data for publishing enterprise.If we can analyze and predict these time series data,and apply the mature model algorithm to describe the model,it will have a strong reference significance in the decision-making process for publishing enterprises.The book publishing industry's insight into the market segmentation is more valuable than the overall trend of the industry as a whole.The so-called book segmentation market refers to the diversification of the book industry according to the market demand and the differences in readers' purchasing behavior The whole book market is divided into a number of sub-markets with some kind of similarities.Each market segment is made up of readers with similar demand tendencies.Therefore,the readers who belong to different market segments have different needs and desires for book products,and the readers who belong to the same market segment have very similar needs and characteristics,which is the potential target market of publishing enterprises.Therefore,the analysis and prediction for the market segmentation in the book retail industry will help publishers to discover the potential opportunities of the market and find out the future publishing trends;It is beneficial to improve the influence and economic benefit of the publishing enterprise in the subdivided market.In this paper,by analyzing the the Total Selling Price for Japanese Book Market,this paper summarize forecasting model according to the time series sales data.The historical data of Japanese book market is decomposed into four components of highfrequency,medium-frequency,low-frequency and residual by EMD.At the same time,the network search data is introduced,and through the collection of the network search data of 20 key words related to the Japanese market influence factors,the search factor data of the Japanese Book Market Sales Code are obtained through the principal component analysis.By adding the search factor data into the prediction model,four components of high-frequency,medium-frequency,low-frequency and residual are applied to the prediction of different prediction models,and as a whole to compare with the actual values.The comprehensive evaluation of the effect of the model prediction with MAPE index and victory rate shows that the middle frequency component after EMD decomposition is combined with Baidu search factor to predict the result of the model is better than that only using simple moving average and ARIMA model.This paper provides a method to build a prediction model of the Total Selling Price for Japanese Book Market,which provides some ideas and references for the establishment of the future market prediction model.