Research on the optimization of tourist attraction route planning through multiple big data

Abstract

Tourism travel is a prevalent form of leisure and entertainment. This paper provides a brief overview of an attraction route planning algorithm based on multi-source data, which combines multi-source data to assess the effectiveness of the planned route and utilizes the particle swarm optimization (PSO) algorithm for path optimization. The genetic algorithm (GA) operations were incorporated with the PSO algorithm to enhance optimization performance. Subsequently, simulation experiments were conducted to compare the GA-PSO algorithm with using the PSO algorithm and GA. Moreover, a comparative analysis was performed on the performance of the path planning algorithm using single-source and multi-source data. The results demonstrated that the GA-PSO algorithm exhibited the fastest convergence in optimization search and achieved the best fitness value at stabilization. Among the three path schemes, the GA-PSO algorithm performed the best, followed by the GA, while the PSO algorithm was found to be the least optimal. Furthermore, path planning with multi-source data demonstrated better alignment with tourists’ landscape preferences, enabling personalized routes construction.

Keywords

Multi-source data tourism route planning particle swarm optimization

1. Introduction

With the advancement of the social economy and the progress of science and technology, tourism has become a pivotal means of leisure and entertainment for people [1]. When choosing attractions during travel, tourists no longer rely solely on their popularity or distance from each other [2]. Instead, there is a growing emphasis on enhancing both quality and depth when planning attraction visits and routes. The utilization of multi-source data to optimize attraction routes has emerged as a critical area for research, harnessing big data technology to drive rapid development within the tourism industry [3]. Wu et al. [4] put forward a tourism experience utility function and used it to establish an optimization model for tourism route planning. The experimental results revealed the substantial impact of preference for tourism attractions, attention to the timing of tourism, and cost on route planning. Tang et al. [5] introduced an enhanced hierarchical fuzzy reasoning approach based on the c-measure map matching algorithm and verified its effectiveness and superiority. Yu et al. [6] optimized the logistics distribution path algorithm based on actual demand. The resulting three-dimensional constraint model-optimized algorithm reduced distribution mileage by over 25% compared to the genetic algorithm (GA). This paper provides a brief overview of an attraction route planning algorithm based on multi-source data. The algorithm integrated multi-source data to assess the quality of the planned route. The particle swarm optimization (PSO) algorithm was used for path optimization, and GA operations were introduced to enhance optimization performance. Subsequently, simulation experiments compared the GA-PSO algorithm with the PSO algorithm and the GA. Furthermore, the performance of the path planning algorithm under single-source and multi-source data conditions was also compared. This article enhanced the PSO algorithm for tourist route planning by incorporating genetic operations from GA. The optimization process utilized multiple sources of data to evaluate the quality of the planning solutions, providing effective support for convenient travel for tourists. The novelty of this study lies in utilizing genetic operations from GA to address the issue of particles getting trapped in local optima in the traditional PSO algorithm, thereby enhancing its optimization capability. However, limitations include a lack of diversity in the types of data used to measure solution quality and a simplistic approach to constraining search ranges with simple conditions.

2. Algorithm for attraction route planning based on multi-source data

When planning routes for tourist attractions, if the number of attractions and the connections between them are relatively small, the traversal method can be employed to plan routes and select the optimal few [7]. However, when confronted with a large number of attractions and intricate routes between them, the computational load of the traversal method increases significantly. Furthermore, traditional route planning methods often evaluate the planning scheme based on a single criterion, such as shortest length or least time consumption, making it challenging to cater to individual tourist preferences. Therefore, this paper adopts a multi-objective optimization approach for attraction routes by leveraging multi-source data such as tourists’ visit history, road network structure, and attraction popularity [8].

For the problem of route planning for tourist attractions, it can be considered as a route optimization problem. The PSO algorithm can be used to solve it. The PSO algorithm is an optimization algorithm based on cooperation that simulates the foraging behavior of bird flocks. The basic principle of the PSO algorithm’s optimization process lies in the collaboration and information sharing among different individuals within the population. Each particle represents a potential solution, and its movement direction and speed are determined by its own experience and the collective experience of the group. The advantages of the PSO algorithm lie in its simplicity, ease of implementation, lack of many parameter adjustments, and fast convergence speed. Particles in the PSO algorithm possess two attributes: velocity and position. The stronger the fitness represented by a particle’s position, the greater its attraction to other particles. Velocity is determined by both individual experience and collective experience, which are derived from the quality of the particle’s current position. In other words, for the PSO algorithm, a particle’s position directly influences the movement trend of the particle swarm. Consequently, once a particle occupies a relatively optimal position, it will attract other particles and hinder their search for even better positions, potentially leading to being trapped in local optima.

GA is an optimization algorithm based on the principles of biological evolution that mathematically simulates natural selection and genetic mechanisms. In GA, operations such as replication, crossover, and mutation are performed on chromosomes. Replication preserves excellent solutions while crossover and mutation generate new solutions. These two operations, especially mutation, can be considered sudden changes for individual chromosomes. The dilemma of local optima faced by the PSO algorithm occurs because other particles are attracted to the local optimum particle, resulting in the overall velocity trend of the particle swarm pointing towards a local optimal solution. To solve this dilemma, it is necessary to allow other particles to break free from the influence of local optima after being attracted. Therefore, this paper introduces genetic operations from GA. After one iteration by the PSO algorithm, although a local optimal particle is not globally optimal, it is still a relatively excellent particle and should not be discarded directly. Hence, the replication operation in genetic operations can be used to preserve it. Other particles undergo position changes through crossover and mutation operations to escape its influence. During this process of change, on one hand, they can break free from the attraction of local optima particles, and on the other hand, these changed particles may end up in even better positions, thereby overcoming the dilemma of being trapped by local optima.

Figure 1.

Path planning flow of the GA-PSO algorithm based on multi-source data.

The planning flow of the GA-PSO algorithm for traveling paths combined with multi-source data is presented in Fig. 1.

➬

Multi-source data such as tourists’ visit history information, basic information for points of interest (POI), and road network information are collected.

➭

Based on the historical visit information of tourists to attractions, a user-attraction check-in matrix is constructed, as shown in Fig. 1 [9]. $U_{i}$ represents the $i$ -th visitor, $\textit{POI}_{j}$ represents the $j$ -th attraction, and $c_{\textit{ij}}$ represents the number of times that the $i$ -th tourist visits the $j$ -th attraction. Based on the basic information of POI, an attraction-landscape feature matrix is constructed. $T_{k}$ denotes the $k$ -th kind of landscape feature [10]. $\textit{It}_{\textit{jk}}$ denotes whether the $j$ -th attraction has the $k$ -th kind of landscape feature. If it has, it is denoted as 1; otherwise, it is denoted as 0. The user-landscape feature check-in matrix is obtained after two matrices are multiplied. $m_{\textit{ik}}$ denotes the number of times that the $i$ -th tourist visits the $k$ -th kind of landscape feature, which can reflect tourists’ preferences for landscape features.

➮

The relationship between attractions is calculated based on the three constructed matrices. The user-attraction check-in matrix reflects the proximity of attractions from the tourists’ perspective, while the attraction-landscape feature matrix represents the similarity between attractions in terms of landscape features. By combining these two matrices, the degree of relationship between any two attractions can be determined.

➯

Road network data and online evaluation data of attractions are input.

➰

Landscape values are assigned to the attractions in the road network as well as the edges (paths) between attractions. The landscape value of attractions can be divided into inherent landscape value and personalized landscape value. The inherent landscape value of an attraction is determined by online evaluations, including tourist ratings, attraction level, and number of uploaded photos. The higher these feature values are, the higher the inherent landscape value will be. Personalized landscape value is the landscape value viewed from the perspective of individual tourists, where the cosine similarity between the landscape feature vector of an atttraction and the tourist’s preferred landscape feature vector is used as a reward factor to enhance the inherent landscape value. The landscape value of an edge depends on the neighboring attractions. The greater the number of neighboring attractions, the higher the landscape value, the stronger their interrelationships, the higher the landscape value of the edge. The calculation using personalized landscape values can be conducted to obtain personalized landscape preferences.

➱

Based on user demands, the valid area is delineated within the road network to reduce the search scope for optimal pathfinding. In this context, user demands refer to reaching the destination within a specified time limit. A circle is drawn by taking the starting point and the endpoint given by the user as the center of the circle and a certain length as the radius during the division of valid areas. The intersection of the two circles is the search area. The length is determined by the speed at which tourists travel and the designated time limit.

➲

Population particles are randomly generated. The coordinates of each particle are a path planning scheme. The corresponding constraints also need to be adhered to when randomly generating the particle scheme:

$\displaystyle\left\{{\begin{array}[]{l}x_{i,j}\in\left\{{0,1}\right\}\\ \sum\nolimits_{j=1}^{k}{x_{0,j}}=\sum\nolimits_{i=0}^{k-1}{x_{i,k}}=1\\ \sum\nolimits_{i=1}^{k}{x_{i,0}}=\sum\nolimits_{j=0}^{k-1}{x_{k,j}}=0\\ \sum\nolimits_{i=1}^{k}{x_{i,j}}=\sum\nolimits_{j=0}^{k}{x_{i,j}}\leqslant 1\\ \end{array}}\right.,$ (1)

where $x_{i,j}$ is an indicator coefficient that takes the value of 1 when attraction $i$ moves towards attraction $j$ in the planning path and vice versa is 0. Equation (1) restricts that the planning path must start at the beginning and end at the ending point, be unidirectional, and pass through the attraction only once.

➳

The fitness value of the scheme represented by each particle is calculated:

$\displaystyle\left\{{\begin{array}[]{l}R_{T}=\sum\limits_{i=0}^{k}{\sum\limits% _{j=0}^{k}{\frac{x_{i,j}}{T(e_{i,j})}}}\\ R_{\textit{sc}}=\sum\limits_{i=0}^{k}{\sum\limits_{j=0}^{k}{x_{i,j}\cdot S_{% \textit{sc}}(e_{i,j})}}\\ \end{array}}\right.,$ (2)

where $R_{T}$ and $R_{\textit{sc}}$ denote the reciprocal of the time spent on path scheme $R$ and the path landscape value, respectively, and $T(e_{i,j})$ is the time spent on edge $e_{i,j}$ , depending on the length of the edge and the speed of movement on that edge.

➴

Whether the optimization search is terminated or not is determined. if it is terminated, the optimal path scheme is output; if not, proceed to the next step. The termination conditions include reaching the maximum number of optimization iterations and the convergence of population fitness values to stability.

➵

The population particles are iterated [14]:

$\displaystyle\left\{{\begin{array}[]{l}V_{i+1}=V_{i}+a_{1}\cdot c_{1}\cdot(% \textit{pbest}_{i}-P_{i})+a_{2}\cdot c_{2}\cdot(\textit{gbest}_{i}-P_{i})\\ P_{i+1}=P_{i}+V_{i+1}\\ \end{array}}\right.,$ (3)

where $P_{i}$ and $V_{i}$ are the position and velocity of particle $i$ , $\textit{pbest}_{i}$ and $\textit{gbest}_{i}$ are the individual historical and current global optimal positions, $a_{1}$ and $a_{2}$ is learning factors, $c_{1}$ and $c_{2}$ are random numbers, which are uniformly distributed between 0 and 1. Then, genetic operations are performed on the population particles, and the corresponding operations are as follows. The crossover operation is to randomly select two particles to exchange the values on the same axes, and the possibility of the axes’ values being exchanged depends on the crossover probability. The mutation operation is to randomly change the values on the axes of a particle under the constraints, and the possibility of the random change depends on the mutation probability [15]. After updating the population using genetic operations, return to step ➴.

3. Simulation experiments

3.1 Experimental data

Table 1
Statistics on the road network, attractions, and tourists

Indicator	Numerical value	Indicator	Numerical value
Number of attractions/n	4,213	Number of attraction feature types/n	10
Number of edges/n	32,569	Number of visitors/n	3,268

The road network data for Nanyang City, Henan Province, utilized in the simulation experiments were sourced from the OSM platform. Prior to formal usage, the data underwent a cleaning process. Attraction data were acquired through a crawler program from “Tongcheng Travel (www.ly.com)” and “Fliggy (www.fliggy.com)”. The statistical summary of the road network, attractions, tourists’ check-ins, and other relevant data is presented in Table 1. There were a total of 4,213 attractions connected by 32,569 path edges, ten different types of attraction features, and some attraction-related data for 3,268 tourists, including their evaluations and check-ins for each attraction.

3.2 Experimental setup

The relevant parameters of the GA-PSO algorithm for path planning were set as follows. The population size was 20, $c_{1}$ and $c_{2}$ were set to 1.5 and 1.0, respectively, the inertia weight was set to 0.6, and the maximum number of iterations was 100. The crossover probability was 0.9, and the mutation probability was 0.1. In addition, the relevant parameters of the PSO algorithm used for the comparison were the same as those of the PSO portion of the GA-PSO algorithm. The relevant parameters of the GA used for the comparison were set as follows. The population size was 20, the crossover probability was 0.9, the mutation probability was 0.2, and the maximum iteration count was 100.

In the process of the simulation experiment, taking into account the average speed of driving in daily life and the amount of time tourists are willing to spend on the road, the average speed of tourists’ driving was set to 25 km/h, and they were requested to reach the endpoint in 1.5 h. One starting point and one ending point were randomly selected from the attractions, and the above three path planning algorithms were used to obtain the planning scheme from the perspective of the same tourists, and the advantages and disadvantages of the path schemes under the three algorithms were compared.

Once again, one starting point and one ending point were randomly selected from the attractions, and the GA-PSO algorithm was used to plan the path scheme for one tourist. To verify the usefulness of multi-source data for personalized route planning, another path planning algorithm using only road network data was used to plan a path scheme for the same tourist. The difference between the path planning algorithm using single-source data and multi-source data was that only $R_{T}$ was used in constructing the fitness value formula. The advantages and disadvantages of the path schemes under the two algorithms were compared, and the cosine similarity was employed to measure the proximity of the planning scheme to the user’s preference.

3.3 Experimental results

The convergence curves of the three algorithms for the same tourist are depicted in Fig. 2. It can be observed that all three algorithms converged as the number of iterations increased during the optimization process. The landscape value of the planning schemes gradually rose and stabilized, while travel time gradually decreased and stabilized. Notably, among the three algorithms, the GA-PSO algorithm exhibited the swiftest convergence, achieving stability after approximately 25 iterations. Moreover, it attained both highest landscape value and shortest travel time once convergence was stabilized.

Table 2
Advantages and disadvantages of planning schemes under the three planning algorithms for the same visitor

Algorithm	Landscape value	Travel time/min	Average cosine similarity to users’ preferences	Computation time/s
PSO	489	31	0.88	5.2
GA	514	26	0.89	4.8
GA-PSO	632	21	0.90	3.6

Figure 2.

Convergence curves of three algorithms during optimization.

The comparative results of the algorithms for the same tourist route are summarized in Table 2. Analysis of Table 2 indicated that the GA-PSO algorithm performed the best in terms of landscape value, travel time, and computation time compared to the GA and PSO algorithms, which were ranked second and third respectively. The “average cosine similarity to users’ preferences” reflected the alignment between the landscape feature types of attractions in the planned path and users’ preferences. The data in Table 2 revealed that the cosine similarity of the planning scheme under the GA-PSO algorithm was slightly higher, but there was no significant difference compared to the other two planning schemes. This suggested that all three path planning schemes were well aligned with tourists’ preferences.

The comparison results of the GA-PSO algorithm, which uses only road network data as a single source for path planning, and the path solutions obtained from using multiple sources of data are shown in Table 3. Table 3 reveals that, concerning “travel time” and “computation time”, the scheme employing multi-source data held an advantage, but the advantage was not significant. In terms of “landscape value” and “average cosine similarity”, the path planning scheme incorporating multi-source data exhibited clear advantages. The scheme utilizing multi-source data not only boasted a higher landscape value but also aligned more closely with tourists’ preferences for landscape.

Table 3

Advantages and disadvantages of planning schemes under the GA-PSO algorithm using single-source data and multi-source data

Data type	Landscape value	Travel time/min	Average cosine similarity to users’ preferences	Computation time/s
Road network data	369	19	0.44	3.9
Multi-source data	632	21	0.90	3.6

4. Conclusion

This paper provides a brief introduction to an attraction route planning algorithm based on multi-source data. The algorithm combined multi-data sources to evaluate the effectiveness of the planned route. The PSO algorithm was used for path optimization, and GA operations were incorporated within the PSO algorithm to enhance optimization performance. Subsequently, simulation experiments compared the performance of the GA-PSO algorithm with both the PSO algorithm and the GA. The performance of the path planning algorithm using single-source and multi-source data was also compared. All three algorithms exhibited convergence with an increase in iterations. The landscape value of the planning scheme gradually increased and stabilized, while the travel time decreased and stabilized. Notably, the GA-PSO algorithm demonstrated the fastest convergence, achieving the highest landscape value and the lowest travel time after stabilization. In terms of landscape value, travel time, and algorithm computation time, the GA-PSO algorithm performed the best, followed by the GA and the PSO algorithm. However, their difference in “average cosine similarity to users’ preferences” was not significant, and all the schemes aligned well with tourists’ preferences. Regarding “travel time” and “computation time”, the scheme using multiple data sources for planning held a slight advantage. In terms of “landscape value” and “average cosine similarity”, the planning scheme that utilized multi-source data demonstrated superiority.

Footnotes

Funding

This work was supported by Key Specialized Research and Development Program in Henan province in 2022, Project name: Smart Tourism Technology Research and Development Based on Big Data Development and Machine Learning Algorithm (222102320115).

References

Zhang

. Research on experiential tourism route planning based on multi-source data algorithm in rich energy and cultural resources areas. Journal of Physics: Conference Series. 2020; 1648(2): 1-5.

Dayoub

Yang

Dayoub

Omran

. The role of cultural routes in sustainable tourism development: A case study of syria’s spiritual route. International Journal of Sustainable Development and Planning. 2020; 15(6): 865-874.

. Construction of tourism planning information system based on ant colony algorithm. Journal of Physics: Conference Series. 2020; 1533: 1-7.

Guan

Han

. A tour route planning model for tourism experience utility maximization. Advances in Mechanical Engineering. 2017; 9(10): 1-8.

Tang

Zhang

Zou

Liu

. An adaptive map-matching algorithm based on hierarchical fuzzy system from vehicular GPS data. PLoS ONE. 2017; 12(12): 1-11.

Zhou

. Optimization of e-commerce logistics distribution path algorithm under the background of big data. Journal of Physics: Conference Series. 2021; 1744(4): 1-7.

Xiang

. Cold chain logistics distribution routing optimization based on realistic delivery time and ant colony. IOP Conference Series: Materials Science and Engineering. 2020; 768: 1-5.

Wang

Lim

Tseng

Yang

. Promoting low carbon agenda in the urban logistics network distribution system. Journal of Cleaner Production. 2019; 211(FEB.20): 146-160.

Zhang

. Solving route optimisation problem in logistics distribution through an improved ant colony optimisation algorithm. International Journal of Services Operations and Informatics. 2017; 8(3): 218-230.

10.

Zhang

Cao

. Multi-objective clustering optimization algorithm for paper-making enterprises’ logistics distribution routes under e-commerce environment. Paper Asia. 2019; (4): 26-32.

11.

Feng

Timmermans

HJP

. Detecting spatial and temporal route information of GPS traces. Lecture Notes in Geoinformation & Cartography. 2015; 61-75.

12.

Wang

Liu

Tan

Zheng

. Multi-sensor data fusion identification for shearer cutting conditions based on parallel quasi-newton neural networks and the dempster-shafer theory. Sensors. 2015; 15(11): 28772-28795.

13.

Jagadeesh

Srikanthan

. Online map-matching of noisy and sparse location data with hidden markov and route choice models. IEEE Transactions on Intelligent Transportation Systems. 2017; 18(9): 2423-2434.

14.

Rajabi

Ataie-Ashtiani

. Efficient fuzzy Bayesian inference algorithms for incorporating expert knowledge in parameter estimation. Journal of Hydrology. 2016; 536: 255-272.

15.

Sobreira

Costa

Sousa

Rocha

Lima

Farias

PCMA

Costa

Moreira

. Map-matching algorithms for robot self-localization: a comparison between perfect match, iterative closest point and normal distributions transform. Journal of Intelligent & Robotic Systems. 2018 (2): 1-14.

Research on the optimization of tourist attraction route planning through multiple big data

Abstract

Keywords

1. Introduction

2. Algorithm for attraction route planning based on multi-source data

3.1 Experimental data

Table 1 Statistics on the road network, attractions, and tourists

3.3 Experimental results

Table 2 Advantages and disadvantages of planning schemes under the three planning algorithms for the same visitor

Footnotes

Funding

References

Table 1
Statistics on the road network, attractions, and tourists

Table 2
Advantages and disadvantages of planning schemes under the three planning algorithms for the same visitor