This paper proposes a new hybrid deep learning framework that combines search query data, autoencoders (AE) and stacked long-short term memory (staked LSTM) to enhance the accuracy of tourism demand prediction. We use data from Google Trends as an additional variable with the monthly tourist arrivals to Marrakech, Morocco. The AE is applied as a feature extraction procedure to dimension reduction, to extract valuable information and to mine the nonlinear information incorporated in data. The extracted features are fed into stacked LSTM to predict tourist arrivals. Experiments carried out to analyze performance in forecast results of proposed method compared to individual models, and different principal component analysis (PCA) based and AE based hybrid models. The experimental results show that the proposed framework outperforms other models.

No abstract

No abstract

Online review data attracts the attention of researchers and practitioners in various fields, but its application in tourism is still limited. The social media data can finely reflect tourist arrivals forecasting. Accurate prediction of tourist arrivals is essential for tourism decision-makers. Although current studies have exploited deep learning and internet data (especially search engine data) to anticipate tourism demand more precisely, few have examined the viability of using social media data and deep learning algorithms to predict tourism demand. This study aims to find the key topics extracted from online reviews and integrate them into the deep learning model to forecast tourism demand. We present a novel forecasting model based on TripAdvisor reviews. Latent topics and their associated keywords are captured from reviews through Latent Dirichlet Allocation (LDA), These generated features are then employed as an additional feature into the deep learning (DL) algorithm to forecast the monthly tourist arrivals to Hong Kong from USA. We used machine learning models, artificial neural networks (ANNs), support vector regression (SVR), and random forest (RF) as benchmark models. The empirical results show that the proposed forecasting model is more accurate than other models, which rely only on historical data. Furthermore, our findings indicate that integration of the topics extracted from social media reviews can enhance the prediction.