Personalized tourist route planning method based on knowledge graph

Abstract

The invention belongs to the field of tourism management and artificial intelligence, and provides a personalized tourism route planning method based on a knowledge graph, which comprises the following steps: constructing a scenic spot knowledge graph; calculating the mastering degree of the user on the scenic spot; based on the mastery degree of the user on the scenic spot, endowing mastery degree attributes to nodes of the scenic spot knowledge graph; performing breadth-first search on the to-be-recommended nodes from each first node of which the mastering degree is greater than a first preset value to form a second scenic spot knowledge graph; determining a current position of a user on the second scenic spot knowledge graph, and performing breadth-first search on a to-be-recommended node based on the current position of the user to obtain a first to-be-recommended node; and searching the to-be-recommended nodes from the first to-be-recommended node, and determining a plurality of to-be-recommended nodes based on user plan time and cost constraints to form a recommended route.

Description

Technical Field

[0001] This invention belongs to the field of tourism management, and specifically relates to a personalized tourism route planning method based on knowledge graphs. Background Technology

[0002] In recent years, with the rapid development of the tourism industry and the widespread application of information technology, personalized tourism recommendation systems have become an important means to improve user experience. Existing tourism recommendation methods mainly rely on rule-based recommendation, collaborative filtering recommendation, and deep learning recommendation. These methods have improved the accuracy of recommendations to some extent, but still have the following shortcomings: Traditional recommendation methods mainly rely on historical visit records or popular attraction rankings, but these methods fail to accurately measure a user's level of understanding of a particular attraction, i.e., whether the user has only browsed related information or has already gained in-depth knowledge and no longer needs recommendations.

[0003] Existing deep learning recommendation methods (such as neural network recommendation models and reinforcement learning recommendation models) typically require a large amount of user behavior data for training, including users' browsing history, rating feedback, search behavior, social interactions, etc.

[0004] However, in the context of travel recommendations, user access data is relatively sparse, and each user's travel behavior is highly personalized, making it difficult for training data to cover all personalized needs, which in turn affects the generalization ability of the recommendation system.

[0005] On the other hand, deep learning recommendation models have high computational complexity, involving large-scale matrix calculations, vector embeddings, and multi-layer neural networks, which makes the system face computational resource bottlenecks when processing real-time personalized recommendations.

[0006] Since most existing recommendation algorithms are based on collaborative filtering or similar user behavior, they are prone to duplicate recommendations, meaning that the system recommends attractions that the user has already visited multiple times. Summary of the Invention

[0007] To address the problems in the prior art, this invention provides a personalized travel route planning method based on knowledge graphs, comprising the following steps: S1, Construct a scenic spot knowledge graph, which includes nodes and relation edges. The nodes are scenic spots and are used to represent various accessible tourist destinations. Each scenic spot node contains attribute information. The relation edges are used to represent different types of connection relationships between scenic spots. S2, calculate the user's level of understanding of the attractions, which is calculated based on the user's historical access records, search behavior, social interactions and rating feedback information; S3, based on the user's level of understanding of the attractions, assign a level of understanding attribute to the nodes of the attraction knowledge graph; S4. Starting from the first node whose mastery level is greater than the first preset value, perform a breadth-first search. Mark nodes whose mastery level is less than the second preset value, whose distance from the first node is less than the third preset value, and whose attraction type is the same as the first node as nodes to be recommended, thus forming a second attraction knowledge graph. S5, determine the user's current location on the second scenic spot knowledge graph, and perform a breadth-first search on the node to be recommended based on the user's current location to obtain the first node to be recommended; S6, starting from the first node to be recommended, search for nodes to be recommended, and determine multiple nodes to be recommended based on the user's planned time and cost constraints, forming a recommendation route.

[0008] Furthermore, the attribute information includes: attraction type, geographical location, average cost, visit duration, opening hours, and visitor flow.

[0009] Furthermore, the connectivity relationships include: transportation connectivity, geographical proximity, tourist behavior correlation, theme similarity, and recommended routes.

[0010] Furthermore, the method for calculating the degree of mastery is as follows: Score_mastery(U, D) = weight_3 * Score_visit + weight_4 * Score_search + weight_5 * Score_social + weight_6 * Score_rating in: Score_mastery(U, D) represents the level of mastery user U has over attraction D; Score_visit is a score for the user's historical visits; Score_search is a rating of the user's search behavior for this attraction; Score_social is a rating of a user's interactions on social media platforms; Score_rating is the user's rating feedback for the attraction; weight_3, weight_4, weight_5, and weight_6 represent the weights of each factor.

[0011] Furthermore, the mastery level attribute is represented as follows: Mastery_Level(U, D): Represents the level of mastery user U has over attraction D; Value range: [0, 1], where: 0 indicates that the user is completely unfamiliar with the attraction; 1 indicates that the user is completely familiar with the attraction; The higher the value, the deeper the user's understanding of the attraction.

[0012] Furthermore, the method for determining the user's current location includes any of the following: Based on real-time GPS positioning; Based on user-defined starting points; Based on the user's historical access records.

[0013] Furthermore, the step of performing a breadth-first search on the nodes to be recommended based on the user's current location to obtain the first node to be recommended includes: In breadth-first search, a comprehensive score Score_recommend is calculated for all candidate nodes D_recommend that meet the search criteria. Finally, the node with the highest score is selected as the first node to be recommended. The calculation formula is as follows: Score_recommend = alpha * Distance_factor + beta * Accessibility_score in: Distance_factor: Calculates the geographical distance between D_recommend and the user's current location; the closer the distance, the higher the score. Accessibility_score: Based on the knowledge graph, calculates whether D_recommend is accessible by an efficient mode of transportation, prioritizing those with fewer transfers; Alpha and beta are weighting coefficients; The candidate node with the highest score is selected as the first node to be recommended.

[0014] Furthermore, the user-planned time and cost constraints include: Time constraints are imposed, and the expected stay time for users at each attraction is calculated, taking into account the travel time of different modes of transportation, to ensure that the total time does not exceed the user's plan; Cost constraints: Calculate attraction and transportation costs to ensure total expenses do not exceed the user's budget.

[0015] Furthermore, the search for recommended nodes starting from the first node to be recommended uses a breadth-first search or A* search algorithm.

[0016] Furthermore, after the recommended nodes are selected, the optimal tour order is calculated based on the Traveling Salesman Problem optimization algorithm to ensure the shortest path.

[0017] This invention provides a personalized travel route planning method based on knowledge graphs. By comprehensively calculating user interests, attraction characteristics, and route optimization strategies, it achieves accurate and efficient personalized travel recommendations. Compared with existing technologies, this invention has the following advantages: By calculating the user's familiarity with attractions, attractions that the user is already fully familiar with are excluded, ensuring the freshness of the recommendation results and preventing the system from continuously recommending attractions that the user has already visited or is not interested in, thereby improving the effectiveness of the recommendations.

[0018] By combining changes in users' interests in different types of attractions, the recommendation categories are dynamically adjusted to ensure that the recommendations match the users' latest preferences.

[0019] This study employs knowledge graph modeling to study user interests and attraction characteristics, avoiding the high dependence of traditional deep learning recommendation systems on large-scale behavioral data and improving the system's adaptability in data-sparse scenarios. Lightweight graph search and optimization algorithms replace computationally expensive deep neural networks, reducing the computational complexity and improving real-time performance.

[0020] By combining factors such as geographical proximity, thematic similarity, tourist behavior patterns, and transportation accessibility, we ensure that recommended attractions not only match user interests but are also practically feasible, avoiding recommendations based solely on distance or popularity. We employ a multi-type attraction expansion strategy to guarantee the diversity of recommended content, preventing users' itineraries from becoming too monotonous and enhancing the richness of their travel experience.

[0021] The optimal tour sequence is calculated using a route optimization algorithm, ensuring route rationality while satisfying user interests and reducing unnecessary transportation costs and time waste. Recommended attractions are filtered based on the user's planned time and budget constraints to ensure the recommended route aligns with the user's travel itinerary.

[0022] By employing real-time location awareness and dynamic interest modeling, the system can dynamically adjust recommendation schemes based on factors such as the user's current geographical location, weather conditions, and peak tourist seasons, making recommendations more flexible. By combining historical user behavior data, it can predict the types of attractions the user might be interested in at different stages, improving the intelligence of personalized recommendations. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0025] The invention will now be described in preferred form with reference to the accompanying drawings and specific embodiments.

[0026] This embodiment solves the above problems through the following steps: In one embodiment, reference Figure 1 This invention provides a personalized travel route planning method based on knowledge graphs. This method comprehensively calculates and generates the optimal personalized travel route based on user interests, historical behavioral data, destination characteristics, and route optimization strategies. Specifically, the method includes the following steps: S1. Construct a scenic spot knowledge graph, which includes nodes and relation edges. The nodes are scenic spots and are used to represent various accessible tourist destinations. Each scenic spot node contains attribute information. The relation edges are used to represent different types of connection relationships between scenic spots.

[0027] The tourist attraction knowledge graph is used to represent tourist attractions and their interrelationships to support personalized travel route planning. The knowledge graph includes nodes and relational edges, wherein: The nodes include different tourist attractions, representing various accessible tourist destinations. Each attraction node contains the following attribute information: Attraction Type: Used to characterize the category of attractions, including but not limited to natural landscapes, cultural heritage sites, city sightseeing, theme parks, adventure tourism, historical sites, leisure resorts, museums, shopping centers, religious sites, etc. Geographic location: includes the latitude and longitude coordinates of the attraction, the city it is located in, and administrative division information, to support route calculation and map visualization; Average cost: This describes the average amount of money tourists spend when visiting the attraction, including ticket prices, average spending levels, etc. Visit time: refers to the recommended visit time for the attraction, expressed in hours or minutes, representing the average stay time of a typical visitor at the attraction; Opening hours: Used to indicate the opening hours of attractions, such as 9:00-18:00 daily, so that time constraints can be taken into account when planning routes; Visitor traffic: This indicates the number of visitors to the attraction at different times, including historical visitor data and real-time crowd density. Rating: This represents the average rating of the attraction on the travel platform, calculated based on visitor review data; The relation edges are used to represent different types of connection relationships between attractions, including but not limited to: Transportation connections: This indicates the mode of transportation, shortest travel time, and transfer options between two attractions. For example: (Beijing Forbidden City → Direct subway access → Temple of Heaven) (Shanghai Bund → High-speed rail access → Hangzhou West Lake) Geographic proximity: used to represent the spatial proximity of two attractions. If the geographical distance between two attractions is less than a certain threshold (such as 5km), a geographic proximity connection is established in the knowledge graph. Tourist behavior correlation: Based on the historical travel paths of tourists, connect frequently visited attractions, for example: (After visiting the Great Wall, tourists usually go to the Summer Palace.) Theme similarity: Used to represent the similarity between two attractions in terms of culture, history, entertainment, etc. If the two attractions have a high degree of theme similarity, a corresponding connection is established in the knowledge graph. For example: (Both the Forbidden City and the Ming Xiaoling Mausoleum in Nanjing are royal buildings from the Ming and Qing dynasties.) (Both Disneyland and Universal Studios are theme parks) Recommended route: A recommended tour sequence based on expert recommendations or travel guides, for example: (Forbidden City → Summer Palace → Temple of Heaven) By constructing the aforementioned scenic spot knowledge graph, multi-dimensional connections between scenic spots can be realized, enabling subsequent travel route planning to be optimized based on factors such as user interests, geographical location, and transportation convenience.

[0028] S2, calculate the user's level of understanding of the attractions, which is calculated based on the user's historical access records, search behavior, social interactions, and rating feedback information.

[0029] The historical access records are used to indicate whether a user has actually visited a specific attraction, as well as the frequency and recent time of the visit. The data can be obtained through the user's travel orders, location data, check-in records, etc.

[0030] The impact of the historical access records can be calculated using a weighted model, as follows: Score_visit = weight_1 * visit_count + weight_2 * exp(-lambda * time_difference) in: Score_visit represents the user's rating of the attraction; visit_count represents the number of times a user has visited this attraction; time_difference is the time interval between the current time and the time of the most recent access; lambda is a time decay factor that controls the impact of access records on the degree of mastery.

[0031] The search behavior is used to indicate a user's level of interest in a particular attraction. The search behavior includes, but is not limited to, users' queries for relevant information on travel platforms, search engines, social media, and other channels.

[0032] The social interaction data is used to indicate whether users interact with content related to specific attractions on platforms such as social media, travel communities, and online forums. These interactions include, but are not limited to, liking, commenting, sharing, and saving.

[0033] The rating feedback information is used to measure the user's level of approval for the attractions they have visited or are familiar with. The rating data can be obtained based on the user's ratings, reviews, and other content on the travel platform.

[0034] Based on the above behavioral data, a quantitative model of users' mastery of specific attractions is constructed. The model adopts a weighted scoring method, comprehensively considering users' access frequency, search depth, social interaction intensity, and rating feedback. The calculation formula is as follows: Score_mastery(U, D) = weight_3 * Score_visit + weight_4 * Score_search + weight_5 * Score_social + weight_6 * Score_rating in: Score_mastery(U, D) represents the level of mastery user U has over attraction D; Score_visit is a score for the user's historical visits; Score_search is a rating of the user's search behavior for this attraction; Score_social is a rating of a user's interactions on social media platforms; Score_rating is the user's rating feedback for the attraction; weight_3, weight_4, weight_5, and weight_6 represent the weights of each factor. The specific weight values ​​can be adjusted based on the actual data.

[0035] Based on the above calculation method, a user's level of familiarity with each attraction can be obtained, and this can be used to determine whether the user is already familiar with a particular attraction. This information can then be used to exclude attractions that the user is already familiar with and to prioritize recommending attractions that the user is less familiar with but may be interested in.

[0036] S3. Based on the user's level of understanding of the attractions, assign a level of understanding attribute to the nodes of the attraction knowledge graph.

[0037] The mastery level attribute is used to characterize the user's cognitive depth of different attractions, in order to support the optimization calculation of subsequent personalized travel route planning.

[0038] The level of familiarity attribute is one of the key attributes of scenic spot nodes, used to quantify the user's familiarity with each scenic spot. The level of familiarity attribute can be expressed as: Mastery_Level(U, D): Represents the level of mastery user U has over attraction D.

[0039] Value range: [0, 1], where: 0 indicates that the user is completely unfamiliar with the attraction; 1 indicates that the user is completely familiar with the attraction; The higher the value, the deeper the user's understanding of the attraction.

[0040] Each attraction node D in the attraction knowledge graph will be appended with Mastery_Level(U, D) as an attribute to ensure that subsequent recommendation calculations can dynamically adjust the recommendation strategy based on the user's cognitive state.

[0041] S4. Starting from the first node whose mastery level is greater than the first preset value, perform a breadth-first search. Mark nodes whose mastery level is less than the second preset value, whose distance from the first node is less than the third preset value, whose attraction type is different from the first node, but whose node type is the same as any non-first node whose mastery level is greater than the first preset value as nodes to be recommended, thus forming a second attraction knowledge graph.

[0042] This step is used to filter suitable attractions from the attraction knowledge graph and build a second attraction knowledge graph based on the user's level of knowledge. This knowledge graph only contains nodes that meet the filtering criteria and their relationships, which will be used for subsequent travel route planning.

[0043] A breadth-first search (BFS) is performed starting from the first node whose mastery level is greater than the first preset value (Threshold_1), where: The first node refers to the attractions that the user has fully grasped, i.e., Mastery_Level(U, D) > Threshold_1; This node serves as the starting point for the search and aims to ensure that the recommended content is relevant to the user's existing interests and is located in the same area, thus avoiding recommending completely unfamiliar attractions that could cause problems such as climate or food incompatibility.

[0044] When performing a breadth-first search (BFS) starting from the first node, the node to be recommended, D_recommend, needs to satisfy the following three independent conditions: (1) The level of mastery is less than the second preset value (Threshold_2) Only select attractions where the level of mastery is lower than Threshold_2, i.e., Mastery_Level(U, D_recommend) < Threshold_2; The purpose of this condition is to exclude attractions that users are already familiar with, ensuring the freshness of the recommendations and enhancing the exploration experience.

[0045] (2) The distance between the node to be recommended and the first node is less than the third preset value (Threshold_3). Select only attractions whose geographical distance from the first node, Dist(D_first, D_recommend), is less than Threshold_3; This condition ensures that recommended attractions are close to the user's current main areas of interest, avoiding problems such as climate or food incompatibility caused by recommending completely unfamiliar attractions.

[0046] (3) The type of attractions of the node to be recommended must be the same as the type of any non-first node whose mastery level is greater than the first preset value, but different from the type of the first node.

[0047] When searching from the first node, surrounding nodes of the same type as the first node, if the user has not visited them in previous trips (i.e., has low familiarity with them), indicate that the user may not be interested in this type of attraction. Therefore, although these attractions are geographically or thematically similar to the first node, they should be excluded because the user's historical behavior indicates that they have not actively chosen to visit them. This avoids recommending attractions that the user may not be interested in, thereby improving the personalization and accuracy of the recommendation results.

[0048] For example, the user is very familiar with the Forbidden City, but has a lower level of knowledge about the Temple of Heaven: The user did not choose to visit the Temple of Heaven during previous trips. Although both the Temple of Heaven and the Forbidden City are cultural heritage sites and are geographically close, the user's behavior pattern indicates that they have a low interest in such sites, so the Temple of Heaven is not recommended.

[0049] However, if a user already knows a significant number of park-type attractions, such as the Summer Palace and the Old Summer Palace, it can be inferred that the user has a high interest in park-type attractions. Although the Forbidden City and the Summer Palace belong to different categories (cultural heritage vs. park), because the user has a high level of knowledge about park-type attractions (the user is interested), the system can expand and recommend park-type attractions to match the user's interests.

[0050] In this step, redundancy in recommendations is reduced and personalization accuracy is improved by eliminating similar attractions that the user has not actively visited. For example, if a user is very familiar with the Great Wall but has not visited the Juyongguan section, it indicates that the user has limited interest in Great Wall-related attractions, and other related categories should be recommended. Avoid overly focusing on recommendations of the same type of attraction to enhance the richness of the travel experience and make the itinerary more diverse. When recommending itineraries in Beijing, if the user has already visited Tiananmen Square (a landmark), the recommendations can be appropriately expanded to include the Summer Palace (historical and cultural) or the Yonghe Temple (religious architecture), rather than only recommending other landmarks.

[0051] S5, determine the user's current location on the second scenic spot knowledge graph, and perform a breadth-first search on the node to be recommended based on the user's current location to obtain the first node to be recommended.

[0052] This step determines the user's current location on the second scenic spot knowledge graph (i.e., the subgraph filtered based on the user's level of knowledge), and uses this location as the starting point to perform a breadth-first search (BFS) on the nodes to be recommended, selecting the first node to be recommended to support personalized travel route planning. The specific method includes the following steps: In the second site knowledge graph, the methods for determining the user's current location P_current(U) include, but are not limited to: (1) Based on real-time GPS positioning The user's current geographical location is determined by the GPS coordinates of the mobile device, and the closest attraction D_nearest in the second attraction knowledge graph is matched as P_current(U); This method is suitable for dynamic recommendation scenarios, ensuring that the recommended route adapts to the user's real-time location changes.

[0053] (2) Based on user-defined starting point If the user has not enabled GPS location, they can manually set the starting point P_manual(U), for example, by selecting a specific scenic spot as the starting point of their trip. This approach is suitable for pre-trip planning scenarios, where users plan their travel routes in advance, rather than relying on recommendations based on real-time location.

[0054] (3) Based on the user's historical access records If GPS is unavailable and the user does not manually specify the starting point, the most recently visited attraction by the user can be used as P_current(U); This method is applicable to offline recommendations or prediction of user location.

[0055] Starting from the user's current location, perform a breadth-first search (BFS). After P_current(U) is determined, a breadth-first search (BFS) is performed on the second site knowledge graph, starting from P_current(U), ​​to select the first node to be recommended.

[0056] The search criteria are as follows: Distance constraint: Ensure that the target node D_recommend and P_current(U) are within an acceptable travel time range; Transportation accessibility: Prioritize attractions that can be directly reached from P_current(U) or have lower transfer costs.

[0057] In breadth-first search, a comprehensive score Score_recommend is calculated for all candidate nodes D_recommend that meet the search criteria. Finally, the node with the highest score is selected as the first node to be recommended. The calculation formula is as follows: Score_recommend = alpha * Distance_factor + beta * Accessibility_score in: Distance_factor: Calculates the geographical distance between D_recommend and P_current(U). The closer the distance, the higher the score. Accessibility_score: Based on the knowledge graph, calculates whether D_recommend is accessible by an efficient mode of transportation, prioritizing those with fewer transfers; Alpha and beta are weighting coefficients.

[0058] S6, starting from the first node to be recommended, search for nodes to be recommended, and determine multiple nodes to be recommended based on the user's planned time and cost constraints, forming a recommendation route.

[0059] This step, based on the initial list of recommended nodes, further expands the search scope, filters out nodes that meet the user's planned time and cost constraints, and generates personalized travel recommendations based on an optimization algorithm. The specific method includes the following steps: Starting with the first node to be recommended, expand the search to include other nodes to be recommended. Starting from the first node to be recommended, D_first_recommend, search the knowledge graph of the second attraction to obtain multiple nodes D_recommend that meet the criteria. The search method can be: Breadth-first search (BFS): Prioritizes exploring nearby attractions to control the search scope.

[0060] A* search algorithm: Based on heuristic path optimization, it improves search efficiency.

[0061] The node to be recommended, D_recommend, needs to further align with the user's travel time and budget constraints. The specific calculation method is as follows: (1) Time constraints Assuming the user's planned total travel time is T_total, the visit time T_visit(D) and travel time T_travel(D_current, D_next) for each attraction must satisfy: Sum(T_visit(D) + T_travel(D_current, D_next)) ≤ T_total Right now: Tour Duration: Calculates the user's estimated time spent at each attraction.

[0062] Travel time: Consider the travel time of different modes of transportation (walking, bus, subway, etc.).

[0063] Ensure the total time does not exceed the user's plan.

[0064] (2) Cost constraints Assuming a user's travel budget is Cost_total, the total cost of the recommended route, Cost_route, must satisfy the following: Sum(Cost_visit(D) + Cost_travel(D_current, D_next)) ≤ Cost_total Right now: Cost_visit(D): entrance fees, activities, etc.

[0065] Transportation costs Cost_travel(D_current, D_next): Taxi, bus, high-speed rail, etc.

[0066] After the recommended nodes are selected, the optimal tour order is calculated based on the Traveling Salesman Problem (TSP) optimization algorithm to ensure the shortest path and best experience. The calculation method is as follows: Route_cost = Sum(Travel_time + alpha * Cost_factor + beta * Interest_matching_score): Travel_time: The shortest travel time between two attractions.

[0067] Cost_factor: Tour cost.

[0068] Interest_matching_score: User preference matching degree.

[0069] alpha, beta: Weight parameters, which can be adjusted according to user needs.

[0070] Ultimately, this method ensures that the recommended route conforms to: User interests and preferences (avoiding attractions already known).

[0071] Trip duration is subject to time constraints (not exceeding the planned duration).

[0072] Cost control (do not exceed travel budget).

[0073] Optimal route planning (reducing unnecessary transportation costs).

[0074] This step ultimately outputs the optimal travel route that meets individual needs, improving the user experience.

[0075] For any module structures not specifically defined in this invention, the existing technical descriptions shall prevail. The prior art mentioned in the foregoing background and specific embodiments sections can be considered part of this invention and used to understand the meaning of certain technical features or parameters.

Claims

1. A personalized travel route planning method based on knowledge graphs, characterized in that, The method includes the following steps: S1, Construct a scenic spot knowledge graph, which includes nodes and relation edges. The nodes are scenic spots and are used to represent various accessible tourist destinations. Each scenic spot node contains attribute information. The relation edges are used to represent different types of connection relationships between scenic spots. S2, calculate the user's level of understanding of the attractions, which is calculated based on the user's historical access records, search behavior, social interactions and rating feedback information; S3, based on the user's level of understanding of the attractions, assign a level of understanding attribute to the nodes of the attraction knowledge graph; S4. Starting from the first node whose mastery level is greater than the first preset value, perform a breadth-first search. Mark nodes whose mastery level is less than the second preset value, whose distance from the first node is less than the third preset value, and whose attraction type is the same as the first node as nodes to be recommended, thus forming a second attraction knowledge graph. S5, determine the user's current location on the second scenic spot knowledge graph, and perform a breadth-first search on the node to be recommended based on the user's current location to obtain the first node to be recommended; S6, starting from the first node to be recommended, search for nodes to be recommended, and determine multiple nodes to be recommended based on the user's planned time and cost constraints, forming a recommendation route.

2. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The attribute information includes: attraction type, geographical location, average cost, visit duration, opening hours, and visitor flow.

3. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The connections include: transportation connections, geographical proximity, tourist behavior associations, thematic similarity, and recommended routes.

4. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The method for calculating the level of mastery is as follows: Score_mastery(U, D) = weight_3 * Score_visit + weight_4 * Score_search +weight_5 * Score_social + weight_6 * Score_rating in: Score_mastery(U, D) represents the level of mastery user U has over attraction D; Score_visit is a score for the user's historical visits; Score_search is a rating of the user's search behavior for this attraction; Score_social is a rating of a user's interactions on social media platforms; Score_rating is the user's rating feedback for the attraction; weight_3, weight_4, weight_5, and weight_6 represent the weights of each factor.

5. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The level of mastery attribute is represented as follows: Mastery_Level(U, D): Represents the level of mastery user U has over attraction D; Value range: [0, 1], where: 0 indicates that the user is completely unfamiliar with the attraction; 1 indicates that the user is completely familiar with the attraction; The higher the value, the deeper the user's understanding of the attraction.

6. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The method for determining the user's current location includes any of the following: Based on real-time GPS positioning; Based on user-defined starting points; Based on the user's historical access records.

7. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The step of performing a breadth-first search on the nodes to be recommended based on the user's current location to obtain the first node to be recommended includes: In breadth-first search, a comprehensive score Score_recommend is calculated for all candidate nodes D_recommend that meet the search criteria. Finally, the node with the highest score is selected as the first node to be recommended. The calculation formula is as follows: Score_recommend = alpha * Distance_factor + beta * Accessibility_score in: Distance_factor: Calculates the geographical distance between D_recommend and the user's current location; the closer the distance, the higher the score. Accessibility_score: Based on the knowledge graph, calculates whether D_recommend is accessible by an efficient mode of transportation, prioritizing those with fewer transfers; Alpha and beta are weighting coefficients; The candidate node with the highest score is selected as the first node to be recommended.

8. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The user-based plan time and cost constraints include: Time constraints are imposed, and the expected stay time for users at each attraction is calculated, taking into account the travel time of different modes of transportation, to ensure that the total time does not exceed the user's plan; Cost constraints: Calculate attraction and transportation costs to ensure total expenses do not exceed the user's budget.

9. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, The search for recommended nodes starting from the first node to be recommended uses either breadth-first search or A* search algorithms.

10. The personalized travel route planning method based on knowledge graphs according to claim 1, characterized in that, After the recommended nodes are selected, the optimal tour order is calculated based on the Traveling Salesman Problem optimization algorithm to ensure the shortest path.

Images