Research On Tourism Demand Forecast Based On Consumer Search Within Internet Environment

The scientific prediction of tourism demand has been one of the hot issues in the field of tourism research,and also been the critical link in the safety emergency management of tourist scenic spots,and also it can provide some essential references for decision-making of tourism-related industries.However,the tourism demand curve presents complex nonlinear characteristics,because of being affected by seasonal and other external factors,such as emergencies;while,on the other hand,the development of tourism industry in China starts late with a short history,which leads to the relatively small amount of data available.Therefore,traditional nonlinear forecast techniques can not fully capture the dynamic characteristics of tourism demand.Along with the popularization of the Internet,a large number of search records generated from consumers information search provide a new train of thought to solve this problem.This study conducts tourism demand forecast based on web search queries,and the main contents are presented as follows:(1)Through the reviewing and summarizing on the evolution of tourism demand forecasting methods and tourism demand forecasting based on web search data,this study points out the rationality of existing tourism demand forecasting methods in both domestic and aboard and the problem still need to be improved,and based on this,this paper puts forward some problems which are needed to be solved.(2)By summarizing and analyzing the motivation theory and information search behavior theory for consumer,this paper puts forward the dynamic relationship between network information search,tourist decision making and fulfillment of tourism demand,and applies statistical analysis methods for the systematic establishment of web data acquisition and construction of experimental data set,an empirical research framework is also constructed.(3)For purpose of exploring the potential relationship between consumer information search and tourism demand,this study regards 30 excellent tourism cities as research object to implement panel data analysis.According to the classification of basic element of tourism,the search data obtained are constructed into tourism factor search indexes,and related control variables are added.Results of empirical analysis based on the whole sample and the sub-sample extracted according to the legal holidays indicate that related information search of “restaurant”,“traveling” and “entertainment” are more robust,all have a significant positive impact on tourist flow;however,the information search related to “accommodation”,“transport” and “shopping” have no significant effect on the explained variables;The influence of the entrance ticket price index on the tourists' decision-making presents enormous difference between the two samples,in the whole sample,the relationship between the tourist flow and the price index are inverted “U” shaped,while sample restricted to the statutory holidays,the price index has no significant effect on the tourist flow.Finally,the predictive power of the web search data on the tourist flow is preliminarily verified,the forecast results based on the training set and the testing sample show that the introduction of web search data has the potential to improve the prediction accuracy of the constructed model.(4)In view of the nonlinearity in the tourist flow and in the case of small sample size of experimental dataset available,this dissertation establishes the BA-SVR&CS hybrid forecast model,and tries to apply consumer search data(CS)to construct the input set of the model.In this model,Bat algorithm(BA)is employed for calibrating the SVR's free parameters,the mark “&” aims at emphasizing the combination of CS and SVR forecast model.Results of empirical analysis based on tourist flow data from Jan,2009 to Oct,2010 in Hainan scenic show that the forecast performance of the proposed model is better than that of BA-SVR,PSO-SVR&CS and GA-SVR&CS benchmark models,which verifies that the BA has good capability of parameter optimization when compared with Particle Swarm Optimization(PSO)and Genetic Algorithm(GA);meanwhile,it also confirms that the introduction of web search data can effectively improve the prediction accuracy of the proposed hybrid model.(5)Seasonal fluctuation of tourism demand can to some extent affect the forecast precision,whereas a single forecasting method can not make accurate prediction for this seasonal characteristics of tourism demand.On the other hand,similar to SVR,relevance vector machine(RVM)has good nonlinear prediction ability for small samples,however,it has its unique advantages in forecast complexity as well as predicted output.To these mentioned above,SI-BA-RVM&CS hybrid model is developed to forecast occupancy for star-rated tourist hotels in Beijing.In this model,BA is applied to optimize the parameters of RVM,and seasonal index(SI)aims at the post correction of forecast bias caused by seasonal fluctuations,and CS is used to construct the inputs of the developed model.With the prediction results indicating that the introduction of SI can effectively correct the seasonal prediction bias,and incorporating the search query data into forecast model can significantly improve the predictive ability of the established model;additionally,there exists no significant difference between the developed hybrid model and the SI-BA-SVR&CS model in terms of significance test results of prediction precision based on the forecasts.(6)Short-term tourism demand forecasting can make up for some deficiencies of medium-and long-term forecast and provide some more comprehensive and real-time information for decision-making.However,the short-term tourism demand is more sensitive to some external events,so a single nonlinear model can not fully approximate its dynamic characteristics.In addition,click-through data of official website of tourist destination to a certain extent reflects the potential tourism demand of tourist.In view of this situation,this paper introduces multi-modal web data(MD)and establishes BA-RVM-ARIMA&MD hybrid forecast model.In this model,the MD is used to construct the inputs of RVM;the ARIMA model is used to further fit the residual sequence of BA-RVM&MD prediction,then,the fitted value and the predictive value of BA-RVM&MD are summed up to get the final prediction value.Taking Jiuzhai Valley in Sichuan province as an numerical example,this paper applies the proposed approach to forecast its daily tourist flow,with simulation results indicating that the hybrid model is significantly higher in forecast precision than the benchmark forecast models including BA-RVM-ARIMA and BA-RVM&MD.