Festival daily passenger flow analyzing and forecasting of Sanxingdui based on the Baidu Index

Abstract

As the main representative of ancient Shu culture and one of the precious Chinese cultural heritages, Sanxingdui Ruins has gradually been favored by Chinese and foreign tourists in recent years. Due to the influence of time factors such as festival and weekends, the daily passenger flow of Sanxingdui shows a large fluctuation characteristic. The sudden increase in the number of tourists on festival compared with ordinary days has brought challenges to the daily management of the scenic spot. At the same time, the development of new generation information technology makes network search become one of the main behaviors of tourists before travel. Therefore, this paper analyzes the correlation between Baidu Index and Sanxingdui festival daily passenger flow, and establish prediction model by using SVR prediction method combined with Baidu Index and Sanxingdui festival daily passenger flow, the model has an accurate prediction of Sanxingdui holiday passenger flow. The prediction can provide a scientific reference for the management decision-making of scenic spots on festival.

Keywords

daily passenger flow forecasting Baidu Index SVR Sanxingdui

Introduction

With the rapid development of the social economy and the increasing demand for spiritual enjoyment, tourism has become an indispensable component of people’s daily life. At the same time, the role and contribution of tourism in the tertiary industry and the national economy are also increasing. At the beginning of 2019, the Ministry of Culture and Tourism announced the basic situation of the tourism industry in 2018 (Ministry of Culture and Tourism of the People’s Republic of China. https://www.mct.gov.cn/), showing that the number of domestic tourists in China reached 5.539 billion in 2018, an increase of 10.8% compared to the same period in 2017. The total number of inbound and outbound tourists reached 291 million, a year-on-year increase of 7.8%. The total tourism revenue for the year was 5.97 trillion yuan, a year-on-year increase of 10.5%. After preliminary calculations, the comprehensive contribution of the tourism industry to GDP in 2018 was 9.94 trillion yuan, with a contribution rate of 11.04%. Affected by the COVID-19 in 2020–2022, China’s tourism industry suffered a heavy blow. However, with the lifting of epidemic prevention and control, the tourism industry will quickly recover to the state before the epidemic. Sichuan is rich in tourism resources and has developed a large number of famous representative tourism resources such as Jiuzhaigou Valley Scenic and Historic Interest Area, Mount Emei, and giant pandas. The Sanxingdui Ruins is known as one of the greatest archaeological discoveries of the last century and is a representative of the brilliant ancient Shu civilization. Like the Yangtze River and Yellow River basins, it belongs to the mother body of Chinese civilization and has historical, scientific, cultural, and artistic value, as well as the most ornamental cultural relic group. It is known as the “source of the Yangtze River civilization.” In recent years, with the further development and improvement of Sanxingdui and people’s increasing love for cultural tourism, the number of tourists in Sanxingdui has also shown a trend of increasing year by year.

On the one hand, the increasing flow of tourists to Sanxingdui indicates that Sanxingdui culture is increasingly popular among tourists, playing a positive role in promoting the Sanxingdui Ruins and ancient Shu civilization, thereby promoting the further development and protection of the Sanxingdui Ruins. On the other hand, the rapid growth of tourist flow in Sanxingdui and the fluctuating daily passenger flow, especially during holidays, not only pose obstacles to the daily management of the scenic area, but more importantly, the resulting inadequate management can easily cause destructive damage to the “fragile” Ruins tourism resources. Meanwhile, through relevant literature search and sorting, it has been found that there are currently few scholars analyzing and predicting the daily passenger flow of Sanxingdui. Therefore, it is necessary to use relevant data combined with prediction methods to analyze and predict the passenger flow of the Sanxingdui Festival, providing scientific reference for scenic area management.

Literature review

At present, there is not only limited research on the analysis and prediction of passenger flow in Sanxingdui, but most of it reflects the development, promotion, and protection of the Sanxingdui Ruins. At the same time, there is also little research on passenger flow prediction related to historical and cultural tourism, and a large amount of research is focused on predicting the demand for other types of tourism. Wu et al. proposed a new hybrid prediction method, SARIMA + LSTM, which combines Seasonal Autoregressive Integrated Moving Average (SARIMA) and Long Short-Term Memory (LSTM), to predict the daily tourist volume of the Macau Special Administrative Region in China, and the results show that the prediction performance of this mixed method is superior to other methods.¹ Eugenio-Martin and Cazorla-Artiles used the random parameter logit model to predict the potential tourism demand of 28 European Union countries and proposed the design strategy of targeted marketing activities.² An adaptive multi-scale ensemble (AME) learning method is proposed to predict the season and trend of the number of tourists. ARIMA used to predict the trend component, and SARIMA is used to predict the seasonal component. The empirical results show that AME method can achieve higher prediction accuracy.³ Vergori et al. used the Gini coefficient as a seasonal indicator and combined it with a beta regression model to confirm that the increase in the proportion of tourists arriving by plane determines the overall seasonal decrease in international tourism demand in Italy.⁴ At present, in addition to econometric models, there are also time series methods,⁵ neural networks,⁶ and support vector machine regression (SVR)⁷ for predicting tourism demand. The SVR prediction method is widely used in tourism demand prediction research due to its advantages in solving small samples and strong nonlinear feature data, and it is not easy to fall into local optima during the solving process. Hong et al. proposed combining chaotic genetic algorithm (CGA) with SVR. Through empirical results, it was found that the SVRCGA model has better prediction accuracy in tourism demand prediction compared to traditional genetic algorithm optimized SVR models.⁸ On the other hand, search engine data are of considerable interest to researchers for their utility in predicting human behavior, and research has shown that the rapid development of information technology and the Internet has led to people becoming increasingly loyal to conducting online searches, which in turn can affect people’s choices and specific behaviors towards tourism destinations.⁹ Dergiades et al. think that search engines are the main tool for planning vacations, and they correct the overall search engine index by defining language bias and platform bias, which helps predict international tourist numbers.¹⁰ Chen et al.¹¹ also think that using search engine data (SED) to predict tourist traffic is crucial for the management and safety warning of tourist attractions, and the CNN BiLSTM model is used to denoise SED, resulting in a significant reduction in RMSE and MAPE. Sun found through research that the KELM model, which combines the Baidu index and Google index, can significantly improve the predictive performance of the tourism demand model, outperforming other benchmark models in terms of prediction accuracy.¹²

It can be seen that there is currently limited research on the analysis and prediction of Samsung’s passenger flow both domestically and internationally, and the network search index is beneficial for improving the prediction accuracy of the prediction model. Based on this, this paper intends to analyze the correlation between the Baidu search index with the word “Sanxingdui” as the root and the passenger flow data of Samsung holiday days. Based on this, combined with two types of data and other related data, support vector machine regression is used to establish a holiday day passenger flow prediction model and ultimately achieve accurate prediction of holiday day passenger flow.

The research in this paper has the following two research significance: (1) In practical application, it can provide a reference basis for the holiday management of Sanxingdui Scenic Area. By predicting the passenger flow in advance, the management department can make plans for the opening hours and travel routes of the tourist area in advance, and can scientifically dispatch and coordinate the transportation of materials and personnel in the tourist area in advance. (2) In terms of theoretical innovation, there is currently almost no research on predicting the passenger flow of Sanxingdui, and this study can slightly fill the gap in this regard.

Prediction method SVR

Support Vector Machine (SVM) was proposed by Vapnik and his collaborators and gained comprehensive development in the mid-1990s. As a regression application of SVM, SVR is widely used in system modeling and time series prediction due to its strong nonlinear data processing ability. The principle is, for a given set of training sample data, $X = {(x_{i}, y_{i}) | x_{i} \in R^{n}, y_{i} \in R, i = 1, 2, \dots, l}$ . The regression problem can be reduced to finding a function $f \in F = {f | f : R^{n} \to R}$ (F is the set of assumed functions), ensure that the error between the function value $f (x)$ , and the expected value y on the training sample is not greater than the set value $ε$ . Now assuming $f (x) = w * φ (x) + b$ , where w is the weight vector and b is the threshold. Unsensitive loss function $ε$ is utilized, and SVR can be represented as the following planning problem.

Min \frac{1}{2} {‖ w ‖}^{2} + C \sum_{i = 1}^{l} (ξ_{i} + ξ_{i}^{*})

(1)

s . t {\begin{cases} y_{i} - w \cdot φ (x) - b \leq ε + ξ_{i}, ξ_{i} \geq 0 \\ w \cdot φ (x) + b - y_{i} \leq ε + {ξ_{i}}^{*}, {ξ_{i}}^{*} \geq 0 \end{cases}

C > 0

is the penalty parameter representing the compromise between the complexity of the regression model and the fitting accuracy of the sample, the larger the value, the better the fitting effect, and

ξ_{i}

and

{ξ_{i}}^{*}

are slack variables. Using the principle of duality, simultaneously introducing Lagrange multipliers

α_{i}

and

α_{i}^{*}

and kernel functions, equation (1) is transformed into its dual problem.

Max - \frac{1}{2} \sum_{i = 1}^{l} \sum_{j = 1}^{l} (α_{i} - {α_{i}}^{*}) (α_{j} - {α_{j}}^{*}) K (x_{i}, x_{j}) - ε \sum_{i = 1}^{l} (α_{i} + {α_{i}}^{*}) + \sum_{i = 1}^{l} y_{i} (α_{i} - {α_{i}}^{*})

(2)

s . t {\begin{cases} \sum_{j = 1}^{l} (α_{i} - {α_{i}}^{*}) = 0 \\ C \geq α_{i}, {α_{i}}^{*} \geq 0 \end{cases}

The characteristic of $K (x_{i}, x_{j}) = φ (x_{i}) * φ (x_{j})$ is to convert the inner product operation of high-dimensional space into the kernel function calculation of low-dimensional space, and it avoids the “curse of dimensionality” problem that occurs during calculations in high-dimensional spaces. The Gaussian kernel function is the most commonly used and its expression is as follows:

K (x_{i}, x_{j}) = \exp (- {‖ x_{i} - x_{j} ‖}^{2} / σ^{2})

(3)

It enhances the nonlinear prediction ability of SVR. This paper will use it to construct SVR, and the final support vector machine regression function is

f (x) = \sum_{i = 1}^{l} (α_{i} - {α_{i}}^{*}) K (x_{i}, x_{j}) + b

(4)

Empirical analysis

Data analysis

Passenger flow

The data source of this paper mainly consists of two parts. One part is the actual number of tourists per day in Sanxingdui Scenic Area from 2016 to 2018, which is calculated by the scenic area’s ticketing system. Therefore, it is very objective and accurate, as shown in Figure 1. It should be noted that due to the data security requirements of Sanxingdui Scenic Area, all data has been normalized to [0.15, 0.85], and this processing method will not change the fluctuation characteristics of the data. From the graph, it can be seen that the daily passenger flow fluctuations and jumps of Sanxingdui are particularly prominent, indicating that the data has strong nonlinear characteristics. The major peaks are all holiday passenger flow, indicating that holidays have a higher contribution rate to passenger flow. The circular marked area belongs to the peak of all passenger flow during the Spring Festival and is the peak of all passenger flow. Through research, it is found that the main passenger flow comes from Guanghan and its surrounding areas, with relatively few from other places. At the same time, it is found that the passenger flow on New Year’s Eve is very small, almost belonging to the day with the lowest annual passenger flow. This is influenced by the traditional Chinese New Year tradition of enjoying family reunions at home. The elliptical area represents the passenger flow during the National Day holiday, which is at the second peak of the current year, but it is also significantly higher than other times except for the Spring Festival. In the rectangular area, there are four festivals: New Year’s Day, Tomb sweeping Day, May Day, and Dragon Boat Festival. In addition, the peak of Mid-Autumn Festival before the National Day is the third level of peak passenger flow. In addition, most of the small peaks in the figure are on weekends, and although the passenger flow is significantly higher than on weekdays, it is much lower than on holidays. Through statistics, it has been found that the contribution rate of holiday passenger flow in these 3 years has reached 25%, while the proportion of time is only 8%. Therefore, predicting holiday passenger flow has important practical significance.

Figure 1.

Daily passenger flow of Sanxingdui from 2016 to 2018.

Baidu Index

Another source of data for the article is the Baidu index related to the keyword “Sanxingdui.” As the world’s largest Chinese search engine, Baidu Search has an absolute advantage in the entire Chinese search engine market, and it is the best-loved of public. The Baidu Index is open to the public in China, so many scholars use it for predictive research.¹³ Therefore, using the Baidu search index instead of the online search index has strong representativeness, and keywords are the main search method of the Baidu index. This article mainly selects three types of keywords: “Sanxingdui.” “Sanxingdui Museum,”, and “Sanxingdui Ruins” to calculate the Baidu index related to Sanxingdui, which largely reflects the overall trend of online search. Due to the fact that online search behavior is more time-consuming than tourism behavior, and considering the relatively short timeliness of online search, we mainly consider the Baidu index for the first 3 days, 2 days, and 1 day before the holiday. Figure 2 shows the Baidu index for the three types of keywords on the day before the holiday. From Figure 2, it can be observed that the annual trend of Baidu Index is similar, and the changing trends of the three types of Baidu Index are also quite similar. On the other hand, more importantly, through careful comparison, it can also be found that the trend of changes in the three types of Baidu indices is roughly the same as that of actual passenger flow. Next, we will conduct a correlation analysis between the Baidu index and passenger flow to ensure that we can use the Baidu index to assist in predicting holiday passenger flow.

Figure 2.

Baidu Index of three kinds of keywords on the day before holidays.

Correlation analysis

Table 1 shows the correlation between the Baidu index and holiday passenger flow

Y_{t}

in the Sanxingdui Scenic Area, which is searched for keywords such as “Sanxingdui,” “Sanxingdui Museum,, and “Sanxingdui Ruins,” From the table, it can be seen that the Baidu index and passenger flow have a certain degree of correlation, with bold words indicating the best correlation and reaching a moderate correlation. It can be used as one of the variable factors in the passenger flow prediction model. Among the three types of keywords, The Baidu index of “Sanxingdui”

S_{t}

and “Sanxingdui Ruins”

{S R}_{t}

on that day had the highest correlation with the passenger flow of the day after tomorrow

Y_{t + 2}

, while the Baidu index of the keyword “Sanxingdui Museum”

{S M}_{t}

on that day had the best correlation with the passenger flow of tomorrow

Y_{t + 1}

. This indicates that tourists will search for information related to the scenic area before departure. Table 2 shows the correlation between holiday passenger flow and historical passenger flow. It can also be seen that historical passenger flow is highly correlated with the passenger flow of the scenic area on that day, making it a very important predictive variable.

Table 1.

Correlation between Baidu Index and daily passenger flow of scenic spots.

Degree of correlation	2016			2017			2018
Degree of correlation	$S_{t}$	${S M}_{t}$	${S R}_{t}$	$S_{t}$	${S M}_{t}$	${S R}_{t}$	$S_{t}$	${S M}_{t}$	${S R}_{t}$
$Y_{t}$	0.3003	0.4015	0.3237	0.4038	0.4211	0.3134	0.4063	0.4679	0.4033
$Y_{t + 1}$	0.4221	0.5127	0.4675	0.4917	0.5274	0.4714	0.4863	0.5303	0.4561
$Y_{t + 2}$	0.5369	0.4368	0.5437	0.5569	0.4383	0.5337	0.5274	0.4517	0.5414
$Y_{t + 3}$	0.4576	0.3633	0.4676	0.4308	0.3607	0.4479	0.4319	0.3631	0.4729

Bold indicates the maximum degree of correlation.

Table 2.

Correlation between scenic spots and historical passenger flow from 2016 to 2018.

Degree of correlation	$Y_{t - 1}$	$Y_{t - 2}$
$Y_{t}$	0.6132	0.5271

SVR model training prediction

Data input variable determination

Through the previous correlation analysis, the final input variables are determined to include five input variable, Baidu Index $S_{t - 2}$ , Baidu Index ${S M}_{t - 1}$ , Baidu Index ${S R}_{t - 2}$ , history passenger flow $Y_{t - 1}$ , and $Y_{t - 1}$ . The number of samples in the model is 85, so the entire input is an 85 × 5 matrix. If the predicted passenger flow is denoted as $Y_{t}$ , the function equation of model A can be expressed as,

Y_{t} = F (Y_{t - 1}, Y_{t - 2}, S_{t - 2}, {S M}_{t - 1}, {S R}_{t - 2})

Model prediction

The training and learning process of the SVR model was completed on the MATLAB experimental platform. Among the 85 data samples, the first 56 were used as the training set, and the last 29 were used as the testing set. The training set mainly included passenger flow from 2016 to 2017, while the testing set mainly consisted of passenger flow from 2018.

During the model training process, genetic algorithm was used to select the optimal loss function, penalty coefficient, width coefficient, threshold, and other parameters for the SVR model, in order to obtain the optimal prediction model for SVR. According to the GA process steps, before inputting data, normalize all types of data in the sample to the range of [0.15, 0.85]. Set the relevant parameters of the initial population: population size is 100, maximum evolution generation is 300, crossover probability, and mutation probability are 0.5 and 0.005, respectively, and cross validation fold is set to 10. Cross validation method divides the sample data into K subsets, where K-1 subsets are trained to obtain decision criteria, and the obtained decision criteria are used to validate the remaining subsets. Repeat the process until all subsets are used as validation sets, and the convergence criterion is Root Mean Square Error (RMSE), which is the fitness function. The fitness function is better when the RMSE value in the training set is smaller, and it is more likely to be retained in the next iteration. RMSE is defined in equation (5), $Y_{t}$ and $Y_{t}^{\sim}$ are the actual and predicted values of passenger flow, respectively, and n represents the number of samples.

RMSE = \sqrt{\frac{1}{n} \sum_{t = 1}^{n} {(Y_{t} - Y_{t}^{\sim})}^{2}}

(5)

After training, the optimal parameters for the SVR model are loss function parameter $ε = 0.0208$ , penalty coefficient $C = 72.8635$ , width coefficient $σ = 3.8817, b = - 0.5169$ , and the number of support vector machines is 13. The optimal prediction model for SVR was obtained through the above process, and then the prediction model was used to predict the 29 sample data in the test set, obtaining the final prediction value.

Comparative analysis of prediction results

In order to verify the positive effect of Baidu Index on passenger flow prediction, we established a comparative prediction model that uses only one or two of the three types of Baidu Index at the same time. Therefore, the final comparative models have 7 types, namely, B (excluding Baidu Index), C (excluding

S_{t}

), D (excluding

{S M}_{t}

), E (excluding

{S R}_{t}

), F (excluding

S_{t}

and

{S M}_{t}

), G (excluding

S_{t}

and

{S R}_{t}

), and H (excluding

{S M}_{t}

and

{S R}_{t}

). The commonly used methods for evaluating the accuracy of model predictions include mean error (ME), mean percentage error (MPE), mean absolute percentage error (MAPE), and RMSE. However, ME and MPE methods are not very effective because prediction errors usually have positive and negative values, which can lead to the mean value approaching 0. Here, MAPE is mainly used, which reflects the percentage deviation between predicted and actual values. The smaller the value, the smaller the deviation, indicating better prediction performance. MAPE is defined as equation (6), where is the actual value, is the predicted value, and is the number of evaluated samples. The comparison results are shown in Table 3. It can be seen that the prediction effect of adding three types of Baidu index is the best, both in the training set and the testing machine. Model B without Baidu index has the highest MAPE, and the MAPE values of both the training and testing sets exceed 30%. Among the prediction models with Baidu index added, only the MAPE of the testing set of Model H is slightly greater than 30%. It can also be found that the more variables predicted by Baidu index added, the better the prediction effect. In summary, empirical comparison shows that the Baidu Index related to Sanxingdui is beneficial for predicting customer traffic.

MAPE = \frac{1}{n} \sum_{t = 1}^{n} | \frac{Y_{t} - Y_{t}^{\sim}}{Y_{t}} | \times 100 %

(6)

Table 3.

Comparison of passenger flow prediction results of different prediction models (%).

MAPE	A	B	C	D	E	F	G	H
Training set	20.36	32.18	21.98	22.72	22.37	27.81	27.05	28.42
Test set	22.48	35.64	24.31	24.15	24.79	29.90	29.32	30.56

Conclusion

The tourism industry has become one of the pillar industries of the national economy and has important economic significance for regional development. Good tourism demand prediction models play an increasingly important role in the development of the tourism industry. Accurate prediction of short-term passenger flow, especially holiday passenger flow, is of great significance to the management decision-making of scenic spot tourism management departments, prevent the occurrence of unsafe events in the scenic area, and promote the safe construction of the scenic area. This article takes the representative of the ancient Shu civilization, Sanxingdui, as the research background, explores the current trend of tourism development, sorts out the research status at home and abroad, and based on this, uses the holiday passenger flow data of Sanxingdui Festival from 2016 to 2018 and Baidu search index as data samples, and uses SVR prediction method to establish a holiday day passenger flow prediction model. The empirical results indicate that the Baidu index, as a predictive variable, can effectively improve the predictive performance of the model. At the same time, this study can provide scientific suggestions for management decision-making and operation promotion for scenic spot tourism management departments and tourism-related enterprises, which is of great significance for improving the quality of scenic spot services, rational allocation of tourism resources, and increasing tourism revenue.

Of course, due to the limitation of limited data samples, the prediction error is still relatively large. Future research can further enrich the prediction variables of the prediction model by collecting relevant data on transportation, hotel accommodation, weather, and mobile communication in scenic areas. At the same time, the prediction method can be improved and the data processing methods optimized, which can greatly improve the prediction effect and provide more scientific reference basis for scenic areas and relevant departments.

Development strategy suggestions

On July 26, 2023, President Xi Jinping visited the Sanxingdui Museum to learn about the excavation process of the Sanxingdui site and the achievements of ancient Shu civilization, and provided important instructions for Chinese archaeological work. In November 2018, Peng Qinghua, Secretary of the Sichuan Provincial Party Committee, attended the “Sichuan Hong Kong Macao Cooperation Week - Entering Hong Kong” Economic and Trade Cooperation Forum held in Hong Kong. He first proposed the concept of the “San Jiu Da (Chinese Pinyin)” and promoted Sichuan’s tourism business card to the world. “San” refers to the representative of ancient Shu civilization—Sanxingdui, and “Three” also expresses the historical civilization of Tianfu. However, the current tourism development in Sanxingdui needs to be vigorously promoted, with huge potential for development. Attracting and increasing passenger flow is one of the most crucial steps. Therefore, suggestions for promoting Sanxingdui culture and attracting customer traffic are proposed in two aspects.

Deep integration of information technology and scenic area development

On the one hand, the application of information technology can enhance the tourism experience of tourists. For example, virtual cultural relics developed using information technology such as virtual reality and augmented reality can present the collection of cultural relics in a close, real, and all-round manner in front of tourists, providing an interactive visiting mode that real cultural relics cannot provide. On the other hand, the new generation of information technology represented by big data can fully explore and analyze tourist information, explore their consumption and tourism behaviors, attract more potential tourists, and develop personalized tourism products. This not only effectively improves tourist satisfaction but is also very conducive to the management and operation of scenic spots. More importantly, the application of information technology, especially the new generation of information technology, is the main content and key technology in the construction of smart scenic spots.

Organic integration of rural tourism and cultural tourism

Due to the geographical limitations of Sanxingdui, the development of Sanxingdui scenic area should be organically integrated with the development of surrounding rural tourism, forming a coordinated regional tourism. By establishing theme cultural parks with Sanxingdui characteristics in relevant rural areas, developing a number of tourism performance programs on ancient Shu culture, and carrying out rich and colorful ancient Shu culture tourism festivals on this basis, we will change the traditional single tour explanation style tourism method. Through rural tourism as a medium, we will vigorously promote and promote ancient Shu culture, thereby attracting more tourists to join the regional tourism model of “rural tourism + ancient Shu culture tourism” or “ancient Shu culture tourism + rural tourism,” and improving the regional tourism economy.

Footnotes

Acknowledgments

Author would like to express the sincere appreciation to the editor and the three independent referees in making valuable comments and suggestions to this paper. Their comments and suggestions have improved the quality of the paper immensely.

Funding

The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the project of The Research Center of Sichuan Province on Preventing and Controling Crime, China (Project No. FZFK24-10).

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

DCW

, et al. Forecasting tourist daily arrivals with a hybrid sarima-lstm approach. J Hosp Tour Res 2021; 45(1): 52–67.

Eugenio-Martin

Cazorla-Artiles

. The shares method for revealing latent tourism demand. Ann Tourism Res 2020; 84: 102969.

Xing

Sun

, et al. Seasonal and trend forecasting of tourist arrivals: an adaptive multiscale ensemble learning approach. Int J Tourism Res 2022; 24(3): 425–442.

Vergori

Arima

. Transport modes and tourism seasonality in Italy: by air or by road? Tour Econ 2022; 28(3): 583–598.

Chen

, et al. Forecasting seasonal tourism demand using a multi-series structural time series method. J Trav Res 2019; 58(1): 92–103.

Silva

Hassani

Heravi

, et al. Forecasting tourism demand with denoised neural networks. Ann Tourism Res 2019; 74: 134–154.

Xie

Qian

Wang

. Forecasting Chinese cruise tourism demand with big data: an optimized machine learning approach. Tour Manag 2021; 82: 104208.

Hong

Dong

Chen

, et al. SVR with hybrid chaotic genetic algorithms for tourism demand forecasting. Appl Soft Comput 2021; 11(2): 1881–1890.

Song

. Data source combination for tourism demand forecasting. Tour Econ 2020; 26(7): 1248–1265.

10.

Dergiades

Mavragani

Pan

. Google trends and tourists' arrivals: emerging biases and proposed corrections. Tour Manag 2018; 66: 108–120.

11.

Chen

Ying

Zhang

, et al. Forecasting tourism demand withsearch engine data: a hybrid CNN-BiLSTM model based on Boruta feature selection. Inf Process Manag 2024; 61(3): 103699.

12.

Sun

Wei

Tsui

, et al. Forecasting tourist arrivals with machine learning and internet search index. Tour Manag 2019; 70: 1–10.

13.

Chen

Wang

, et al. Effective tourist volume forecasting supported by PCA and improved BPNN using Baidu Index. Tour Manag 2018; 68: 116–126.