| Tourism demand forecasting is a topic of great interest to both academia and industry,and accurate tourism forecasting directly affects the business operations and management of the destination tourism industry.In order to achieve a more accurate tourism demand forecast,this thesis extends the existing research in the following aspects: first,we propose a Holiformer deep learning model that introduces holiday schedules in an embedded manner and is able to forecast tourism demand in a future period using historical tourism data,historical and future timestamp information,and meteorological information;second,further In order to achieve more accurate prediction,we build a multimodal fusion framework based on the proposed model,which fuses data such as web comment text sentiment with timestamp information and meteorological information to achieve multimodal fusion of structured and unstructured data in tourism headcount prediction.In this study,the model was applied to the historical tourist number dataset of Jiuzhaigou,and the performance of the model was evaluated in the test set.From the empirical results,firstly,in the deep learning model,the model that considered holiday arrangements in an embedded manner outperformed the model that considered only historical data,which indicates that the introduction of holiday arrangement timestamps in an embedded manner is highly meaningful for tourism number prediction;Secondly,the forecasting performance of the model considering historical data,meteorological data and holiday schedule is significantly superior to that of the model without considering holiday schedule data or without holiday data and meteorological data,which indicates the vital role of meteorological data for tourism demand prediction;finally,on this basis,further consideration of online review texts can make the model more effective for tourism demand forecasting,which also indicates the importance of online text sentiment for tourism demand prediction. |
PDF Full Text Request