Tourism Demand Forecasting With Search Engine Data And Deep Learning Framework

Tourism demand forecasting is an attractive topic in the field of tourism management.Its accuracy and stability have a profound impact on the decision-making of managers and the planning of industry participants.This study try to expand the current situation of prediction from three aspects.Firstly,this study introduces powerful exogenous variable – search engine data,which has been applied in many fields and is closely related to prediction variables;secondly,this study uses an effective and stable feature processing method – stack autoencoder(SAE),which is a dimension reduction method based on deep learning;thirdly,this study trains highly efficient and accurate model – bi-direcitional gated recurrent unit neural network(Bi-GRU),which is an excellent deep learning model for processing time series information.Finally,an ensemble prediction model(SAE-Bi-GRU)is constructed.This paper takes Hong Kong tourist arrivals as an example,and verifies that the deep learning framework can play a satisfactory role in extracting and processing data features and improving the prediction accuracy under the results of a variety of evaluation indicators and statistical test.At the same time,the Bi-GRU is superior to other benchmark models without considering the search engine data,and if the search engines are attached,the ensemble prediction model SAE-Bi-GRU proposed in this paper is obviously better than all the benchmark models.