Personalized tourist route planning method and system based on user portraits

Abstract

The invention discloses a personalized travel route planning method and system based on a user portrait, and aims to solve the problem of recommendation homogenization caused by single user portrait dimension and lack of dynamic adaptability of a route in the prior art. The method comprises the following steps: fusing dominant preferences, recessive behaviors and social attribute data, and constructing a multi-dimensional dynamic user intention state vector; constructing a tourism interest point knowledge graph fusing static information and dynamic attributes such as real-time people flow and weather; in a unified vector space, a personalized route is generated through a multi-constraint optimization model and an improved genetic algorithm. The system comprises a data acquisition module, an intention vector construction module, a knowledge graph construction module and a route optimization module, and supports dynamic re-planning based on position and environment change. According to the method, deep personalized, context awareness and real-time self-adaptive tourist route recommendation can be realized, and the user experience is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent tourism route planning technology, specifically to a personalized tourism route planning method and system based on user profiles. Background Technology

[0002] With the rapid development of online travel services and intelligent recommendation systems, the demand for personalized travel is growing. Modern travel planning has gradually shifted from traditional standardized itinerary recommendations to customized services based on user preferences. As a core component of customized travel, travel route planning aims to generate efficient, comfortable itinerary plans that match user interests by comprehensively considering user preferences, behavioral habits, time and space constraints, and destination resources. However, existing travel recommendation systems largely rely on static tags (such as age, gender, and historical orders) or simple collaborative filtering algorithms, lacking in-depth modeling of users' multi-dimensional dynamic characteristics, making it difficult to accurately depict users' true intentions and potential needs in different contexts.

[0003] Personalized travel route planning technology based on user profiles aims to build fine-grained, updatable user profiles by integrating users' historical behavioral data, social attributes, interests, spending power, and real-time contextual information, and then use these profiles to drive route generation logic. The core of this technology lies in achieving precise matching between user characteristics and tourism elements (such as attraction type, mode of transportation, length of stay, and dining preferences), thereby enhancing the personalization of the itinerary and improving user experience satisfaction.

[0004] In existing technologies, most systems still rely on rule-driven or shallow machine learning models for route recommendation, resulting in problems such as limited user profile dimensions, delayed updates, and poor scenario adaptability. Specifically, user profiles are often built solely based on explicit feedback, ignoring the deep preferences implied by implicit behaviors (such as page dwell time, search keywords, and cancellation actions). The route planning process typically involves simply ranking attractions or using fixed templates, without considering dynamic factors such as the spatial and temporal accessibility between attractions, user physical exertion, and interest decay. Furthermore, the system lacks the collaborative processing capability for multi-objective optimization (such as shortest time, lowest cost, and optimal experience), making it difficult to generate high-quality routes that balance feasibility and personalization under complex constraints. These shortcomings lead to highly homogenized recommendation results and insufficient practicality. Especially when facing independent travelers with high freedom and changing needs, existing solutions struggle to meet their dual requirements for itinerary flexibility and experience depth. Therefore, there is a need for a personalized travel planning method that can deeply integrate dynamic user profiles with intelligent route optimization mechanisms. Summary of the Invention

[0005] This invention provides a personalized travel route planning method and system based on user profiles. Its purpose is to solve the technical problems of existing travel route planning technologies, such as the single and static dimensions of user profiles, which cannot deeply understand the user's real and potential complex travel intentions, and the lack of adaptive integration of real-time dynamic environmental factors and the user's individual travel rhythm in the route generation process, resulting in homogenized recommended routes and poor user experience.

[0006] To address the aforementioned technical problems, this invention proposes a personalized travel route planning method and system based on user profiles. The core technical solution of this invention lies in: By using multimodal data fusion technology, a multidimensional dynamic user intent state vector containing explicit interests, implicit associations, and deep psychological traits is constructed from explicit preferences actively provided by users, implicit behavioral trajectories left on various online platforms, and basic sociodemographic information. In parallel, a dynamic multidimensional attribute knowledge graph of tourist interest points is constructed. This knowledge graph not only contains static descriptions of interest points, but also integrates dynamic environmental attributes such as population density, weather conditions, and online public opinion in real time. By placing the user intent state vector and interest point attribute vector in a unified vector space, a complex optimization problem is constructed with the goal of maximizing the overall route matching degree while satisfying multiple constraints such as spatiotemporal accessibility, budget, and individual pace preferences. A heuristic intelligent optimization algorithm is then used to solve the problem, generating a deeply personalized, context-aware, and dynamically adaptive travel route.

[0007] According to one aspect of the present invention, a personalized travel route planning method based on user profiles is provided, comprising the following steps: Acquire multi-source heterogeneous user data, which includes explicit preference data actively input by users through interactive interfaces, implicit behavioral data generated by users on social media, content sharing platforms, and online travel service platforms, as well as users' sociodemographic attribute data; the explicit preference data includes the user's set destination, travel theme tags, points of interest types, and budget range; the implicit behavioral data includes the user's published text and image content, likes and comments records, historical orders, and route browsing history; the sociodemographic attribute data includes the user's age, gender, place of residence, and family composition; Based on the aforementioned multi-source heterogeneous user data, a multi-dimensional dynamic user intent state vector is constructed. Construct a multi-dimensional attribute knowledge graph of tourist interest points, wherein the knowledge graph generates a corresponding multi-dimensional attribute vector for each tourist interest point; Based on the multidimensional dynamic user intent state vector and the multidimensional attribute knowledge graph of the tourist interest points, a personalized tourist route is generated through a constrained optimization model.

[0008] As one embodiment of the present invention, the specific steps for constructing a multidimensional dynamic user intent state vector include: The explicit preference data is structured to extract keywords and values, forming an initial explicit interest feature set. Using a pre-defined text sentiment analysis model and image content recognition model, deep semantic analysis and visual theme feature extraction are performed on the text and image content in the implicit behavioral data to generate an implicit association interest feature set. The text sentiment analysis model is a model based on a bidirectional long short-term memory network, used to extract the emotional polarity and fine-grained emotional objects contained in the text. The image content recognition model is a scene classification and object detection model based on a convolutional neural network, used to identify landmarks, natural scenery, activity scenes, and objects in the image. Based on the explicit interest feature set, the implicit related interest feature set, and the sociodemographic attribute data, a user travel psychological trait vector is generated through a pre-trained travel psychological trait mapping classifier. The travel psychological trait mapping classifier is a gradient boosting decision tree model, whose input is the aforementioned fused features, and whose output is a standardized multi-dimensional vector. The user travel psychological trait vector contains four preset orthogonal dimensions, namely exploration degree, rhythm preference degree, social tendency degree, and consumption sensitivity degree. Exploration degree represents the user's preference for novel and niche attractions, rhythm preference degree represents the number of interest points that the user expects to visit per unit of time, social tendency degree represents the user's preference for lively or quiet environments, and consumption sensitivity degree represents the user's attention to price factors. The explicit interest feature set, the implicit related interest feature set, and the user travel psychological trait vector are hierarchically concatenated and then dimensionality reduction and feature reconstruction are performed through a variational autoencoder to finally generate the multidimensional dynamic user intent state vector. The user intent state vector is a fixed-dimensional floating-point vector, and each dimension corresponds to a specific travel intent sub-component.

[0009] As one embodiment of the present invention, the specific steps for constructing a multi-dimensional attribute knowledge graph of tourist interest points include: Collect static basic information about each tourist attraction, including its geographical coordinates, category, historical background, official description text, and opening schedule; By accessing third-party data application interfaces, dynamic environmental attribute data of various tourist interest points can be obtained in real time. The dynamic environmental attribute data includes real-time crowd density index updated every minute, weather forecast data for the next few hours, traffic congestion index for a specific time period, and real-time public opinion heat value on social media platforms. Natural language processing is performed on the text data in the static basic information to extract topic keywords and entities, and these are mapped to the same four orthogonal dimensions as the user's travel psychological trait vector to generate a static trait score for each point of interest. The static basic information, the dynamic environmental attribute data, and the static trait score are integrated to construct a structured multidimensional attribute vector for each tourist interest point. The dimension of this vector is consistent with the dimension of the user intent state vector to ensure the comparability of the two within the same vector space.

[0010] As one embodiment of the present invention, the specific steps of generating a personalized travel route based on the multidimensional dynamic user intent state vector and the multidimensional attribute knowledge graph of tourist interest points through a constrained optimization model include: Define a solution space for route planning, where each solution in the solution space is an ordered sequence of multiple tourist points of interest; A constrained optimization model is constructed with the goal of maximizing the total route matching degree. The total route matching degree is determined by the sum of the weighted cosine similarity between the multidimensional dynamic user intent state vector and the multidimensional attribute vectors of each point of interest in the route sequence. The weights are proportional to the user's estimated dwell time at the point of interest. The constrained optimization model sets multiple constraints, including: spatiotemporal accessibility constraints, ensuring that the travel time between any two consecutive points of interest plus the dwell time at the point of interest is less than the available time; total budget constraints, ensuring that the sum of the ticket costs and estimated transportation costs for all points of interest does not exceed the user's set budget; individual rhythm preference constraints, ensuring that the total number of planned points of interest per day matches the rhythm preference value in the user's travel psychological trait vector; and point of interest opening time constraints, ensuring that the time to reach any point of interest is within its opening hours. An improved genetic algorithm is used to solve the constrained optimization model. The improved genetic algorithm defines each legal interest point sequence as a chromosome, and the fitness function is the total matching degree of the route. The algorithm performs multiple generations of iterative evolution through specially designed heuristic rule-based selection operators, partial matching crossover operators that maintain the legality of sequences, and neighborhood exchange mutation operators designed to enhance local search capabilities, until the algorithm converges or reaches the preset maximum number of iterations. The chromosome with the highest fitness generated in the final iteration is decoded, and the ordered sequence of interest points represented by the chromosome is the generated personalized travel route.

[0011] Furthermore, the method also includes a dynamic route replanning step, specifically including: During the process of a user executing a pre-generated personalized travel route, the location module of the mobile terminal monitors the user's geographical coordinates in real time at a preset time frequency, and records the actual dwell time of the user at each visited point of interest. At the same time, we continuously acquire real-time dynamic environmental attribute data for all unvisited points of interest along the route; When the deviation between the user's current location and the planned route exceeds the preset geofence threshold, or the difference between the user's actual stay time at a point of interest and the estimated stay time exceeds the preset time deviation threshold, or any unvisited point of interest's dynamic environmental attributes, such as crowd density or weather conditions, undergo a sudden change that causes a significant change in its multidimensional attribute vector, the dynamic replanning mechanism is triggered. The dynamic replanning mechanism takes the user's current location as the new planning starting point, the remaining travel time and budget as the new constraint boundaries, and updates the rhythm preference component in the multidimensional dynamic user intent state vector based on the user's behavioral deviations. It then re-executes the steps of constructing the constraint optimization model and solving it using an improved genetic algorithm, performs real-time optimization and adjustment on the subsequent unexecuted route parts, and pushes the updated route to the user's mobile terminal.

[0012] According to another aspect of the present invention, a personalized travel route planning system based on user profiles is provided, comprising: The multi-source heterogeneous user data acquisition module is used to acquire explicit preference data actively input by users through the interactive interface, implicit behavioral data generated by users on social media, content sharing platforms and online travel service platforms, as well as users' sociodemographic attribute data. The user intent vector construction module is used to construct a multi-dimensional dynamic user intent state vector based on the multi-source heterogeneous user data. The Interest Point Knowledge Graph Construction Module is used to construct a multi-dimensional attribute knowledge graph of tourism interest points, generating a corresponding multi-dimensional attribute vector for each tourism interest point. The route generation and optimization module is used to generate a personalized travel route based on the multidimensional dynamic user intent state vector and the multidimensional attribute knowledge graph of the tourist interest points through a constrained optimization model.

[0013] As one embodiment of the present invention, the user intent vector construction module includes: The explicit feature extraction unit is used to perform structured processing on the explicit preference data to form an initial explicit interest feature set; The latent feature parsing unit is equipped with a preset text sentiment analysis model and image content recognition model, which are used to perform deep semantic parsing and visual topic feature extraction on the latent behavior data to generate a latent association interest feature set. The psychological trait inference unit is equipped with a pre-trained travel psychological trait mapping classifier, which is used to infer and generate a user travel psychological trait vector containing dimensions of exploration degree, rhythm preference degree, social tendency and consumption sensitivity based on each feature set. The feature fusion and dimensionality reduction unit is equipped with a variational autoencoder, which is used to hierarchically concatenate various features, perform dimensionality reduction and feature reconstruction, and finally generate the multidimensional dynamic user intent state vector.

[0014] As one embodiment of the present invention, the interest point knowledge graph construction module includes: The static information collection unit is used to collect basic static information such as the geographical coordinates, category, historical background, introductory text, and opening schedule of each tourist interest point; The dynamic information access unit is used to obtain dynamic environmental attribute data such as real-time crowd density index, weather forecast data, traffic congestion index, and social media sentiment heat value of various tourist interest points through third-party data application interfaces. The attribute vector generation unit is used to process text data, extract themes and map them to psychological trait dimensions, and then integrate all static and dynamic information to construct a structured multidimensional attribute vector for each tourist interest point.

[0015] As one embodiment of the present invention, the route generation and optimization module includes: The constrained optimization model definition unit is used to define a mathematical model with the goal of maximizing the total route matching degree, and integrates multiple constraints such as spatiotemporal accessibility, total budget, individual rhythm preferences, and point of interest opening time; The intelligent algorithm solving unit contains an improved genetic algorithm program, which includes a specific chromosome encoding scheme, fitness function, and selection, crossover, and mutation operators, used to iteratively solve the constrained optimization model and search for the optimal sequence of interest points. The route decoding output unit is used to decode the optimal solution obtained by the algorithm, i.e. the chromosome with the highest fitness, into a specific travel route that includes the name of the point of interest, the order of visits, and the estimated arrival and stay time.

[0016] Furthermore, the system also includes a route dynamic replanning module, which is connected to the route generation and optimization module, and is used to: The real-time status monitoring unit is used to monitor the user's location and dwell time through the positioning module of the mobile terminal, and continuously acquire the dynamic attributes of unvisited points of interest; The replanning trigger is used to activate the replanning process when user behavior deviations or environmental changes exceed a preset threshold. The dynamic adjustment execution unit is used to call the route generation and optimization module, take the current state as a new starting point, recalculate and optimize the subsequent routes, and generate an updated route plan.

[0017] Compared with the prior art, the advantages and positive effects of the present invention are as follows: First, by constructing a multidimensional dynamic user intent state vector that integrates explicit preferences, implicit behaviors, and deep psychological traits, this invention achieves a deep, comprehensive, and dynamic understanding of users' travel needs, surpassing the shallow profiles based on static and sparse labels in existing technologies, thereby significantly improving the accuracy of personalized matching.

[0018] Secondly, the multi-dimensional attribute knowledge graph of tourist interest points created by this invention combines the static information of interest points with the dynamic data of the real-time changing environment, enabling route planning to have context awareness capabilities and dynamically adjust according to real-time pedestrian flow, weather and other conditions, thus avoiding the inapplicability of static routes generated by existing technologies in actual execution.

[0019] Third, this invention models the route generation process as a complex optimization problem with multiple constraints and uses an improved genetic algorithm for global optimization, which can generate a globally optimal route that achieves a comprehensive balance in terms of matching degree, feasibility, rhythm and cost-effectiveness, rather than the simple accumulation of interest points or local optimal planning in the prior art.

[0020] Fourth, this invention introduces a closed-loop dynamic replanning mechanism, which can adjust and optimize unfinished itineraries in real time and automatically based on the user's actual behavioral deviations and sudden changes in the environment during the journey, providing users with a continuously adaptive intelligent guide service, which greatly improves the smoothness and satisfaction of the travel experience. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a personalized travel route planning method based on user profiles proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework for constructing multidimensional dynamic user intent state vectors in this invention; Figure 3 This is a logical flowchart of the construction of a multi-dimensional attribute knowledge graph of tourist interest points in this invention; Figure 4 This is a flowchart illustrating the logical process of generating personalized travel routes based on a constrained optimization model in this invention. Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the terminal and the cloud in this invention. Detailed Implementation

[0022] This invention provides a method and system for personalized travel route planning based on user profiles. The method constructs a multi-dimensional dynamic user intent state vector by fusing multi-source heterogeneous user data, combines it with a real-time updated multi-dimensional attribute knowledge graph of travel interest points, establishes a constrained optimization model within a unified vector space with the objective of maximizing the overall route matching degree, and uses an improved genetic algorithm to solve the model, ultimately generating a deeply personalized, context-aware, and dynamically adaptive travel route. The specific implementation steps of this method will be described in detail below.

[0023] The method first executes step S1: acquiring multi-source heterogeneous user data. This step simultaneously collects three types of data through the user's terminal device's interactive interface, third-party platform authorized interfaces, and user account information database. The first type is explicit preference data actively input by the user through the interactive interface, including the user's explicitly specified destination city or region, travel theme tags such as history and culture, natural scenery, family entertainment, food exploration, etc., interest types such as museums, parks, restaurants, shopping malls, and the user's set total budget range. The second type is implicit behavioral data generated by the user on social media platforms, content sharing platforms, and online travel service platforms, specifically including image and text content with geolocation tags posted by the user on social media platforms, likes and comments on others' posts, historical order information on online travel platforms such as hotel bookings, ticket purchases, transportation tickets, and the user's browsing history and duration of various travel route detail pages. The third type is the user's sociodemographic attribute data, including the user's age, gender, administrative division code of permanent residence, and family composition such as whether they are traveling with children or elderly people. All data is encrypted using a secure data transmission protocol before being uploaded to the cloud processing server. The server performs timestamp alignment and data format standardization to ensure data consistency and timeliness in subsequent processing.

[0024] After completing step S1, step S2 is executed: based on the multi-source heterogeneous user data, a multi-dimensional dynamic user intent state vector is constructed. This step is further divided into four sub-steps.

[0025] First, the explicit preference data is structured by extracting keywords and numerical values ​​to form an initial set of explicit interest features. For example, destination cities are mapped to geocoding, travel topic tags are converted into index values ​​in a predefined topic dictionary, point-of-interest types are categorized into class codes in a standard classification system, and budget ranges are converted into numerical intervals.

[0026] Secondly, a pre-defined text sentiment analysis model and image content recognition model are used to perform deep semantic analysis and visual theme feature extraction on the text and image content in the implicit behavioral data, generating an implicit association interest feature set. The text sentiment analysis model is a deep learning model based on a bidirectional long short-term memory network. Its input is the text content posted by the user, and its output is the sentiment polarity score contained in the text, as well as fine-grained sentiment objects, such as the sentiments of "exhibits" and "shock" in "felt awe at the exhibits in a certain museum". The image content recognition model is a scene classification and object detection model based on a convolutional neural network. Its input is the image uploaded by the user, and its output is the visual theme tags and their confidence scores for landmarks, natural scenery types, activity scenes such as hiking and cycling, and items such as local delicacies and souvenirs identified in the image.

[0027] Furthermore, based on explicit interest feature sets, implicit association interest feature sets, and sociodemographic attribute data, a user travel psychological trait vector is generated through a pre-trained travel psychological trait mapping classifier. This travel psychological trait mapping classifier is a gradient boosting decision tree model, and its training data comes from large-scale user behavior logs and expert-annotated psychological trait labels. The input of this model is the aforementioned fused feature vector, and the output is a standardized four-dimensional floating-point vector, corresponding to four orthogonal dimensions: exploration degree, rhythm preference, social inclination, and consumption sensitivity. The exploration degree ranges from zero to one, with higher values ​​indicating that users are more inclined to visit novel, niche, and unpopular interests; the rhythm preference ranges from 0.5 to 3.0, representing the number of interests that users expect to visit per unit of time; the social inclination ranges from -1 to 1, with positive values ​​indicating a preference for lively, crowded environments and negative values ​​indicating a preference for quiet, less crowded environments; and the consumption sensitivity ranges from 0 to 1, with higher values ​​indicating that users are more sensitive to price factors.

[0028] Finally, the explicit interest feature set, implicit related interest feature set, and user travel psychological trait vector are hierarchically concatenated and then subjected to dimensionality reduction and feature reconstruction through a variational autoencoder to generate the multidimensional dynamic user intent state vector. The encoder part of this variational autoencoder maps the high-dimensional sparse original features to a low-dimensional continuous latent space, while the decoder part ensures that the reconstructed features retain the original semantic information. The final output user intent state vector is a fixed-dimensional floating-point vector, with each dimension corresponding to a specific travel intent sub-component, and the number of dimensions is 128. Each dimension of the 128-dimensional intent state vector is trained and learned, representing specific travel intent sub-components such as preference intensity for historical sites, preference intensity for food-related points of interest, preference for quiet environments, and acceptance of high-budget activities. Their specific semantic labels can be inferred and recorded through model interpretability analysis tools.

[0029] After completing step S2, proceed to step S3: construct a multi-dimensional attribute knowledge graph of tourist interest points. This step is also divided into three sub-steps.

[0030] First, we collected static basic information about each tourist attraction. This information came from official tourism databases, open data platforms, and web crawlers, including the precise geographical coordinates of the attraction, its standard category, a brief historical background, official introductory text, and detailed opening schedules such as daily opening hours and special arrangements for holidays.

[0031] Secondly, by accessing third-party data application interfaces, dynamic environmental attribute data for various tourist attractions is obtained in real time. These interfaces include real-time pedestrian density indices updated minute-by-minute from urban pedestrian monitoring platforms, hourly weather forecasts for the next 24 hours from meteorological service providers, traffic congestion indices for specific time periods released by traffic management departments, and real-time public opinion heat values ​​calculated by social media sentiment analysis platforms. All dynamic data is accurately timestamped and refreshed every five minutes.

[0032] Finally, natural language processing is performed on the text data in the static basic information to extract thematic keywords and entities, and these are mapped to the same four orthogonal dimensions as the user's travel psychological trait vector, generating a static trait score for each point of interest. For example, keywords such as "rarely visited" and "pristine" in the text description of an ancient village in a remote mountainous area are mapped to a high exploration score; "abundant attractions" and "long queues" in the description of a large theme park are mapped to a high pace preference score; "bustling crowds" and "lively atmosphere" in the description of a popular night market are mapped to a high social inclination score; and a free park is assigned a low consumption sensitivity score. Subsequently, the static basic information, dynamic environmental attribute data, and static trait scores are integrated to construct a structured multidimensional attribute vector for each tourist point of interest. The dimension of this vector is consistent with the dimension of the user's intention state vector, both being 128 dimensions, to ensure comparability within the same vector space. Dynamic environmental attribute data is mapped to the 0-1 interval through a preset normalization function and serves as a weighting factor for the corresponding dimension, adjusting the final attribute vector of the point of interest in real time.

[0033] After completing step S3, step S4 is executed: based on the multidimensional dynamic user intent state vector and the multidimensional attribute knowledge graph of tourist interest points, a personalized travel route is generated through a constrained optimization model. This step is the core of the invention, and its specific implementation process is as follows.

[0034] S4-1 defines a solution space for route planning. Each solution in this solution space is an ordered sequence of multiple tourist points of interest, the length of which is determined by the user's total number of travel days and pace preference.

[0035] S4-2, construct a constrained optimization model with the objective of maximizing the total route matching degree. The formula for calculating the total route matching degree is: , Where M is the total route matching degree, n is the total number of points of interest in the route, u is the multidimensional dynamic user intent state vector, and pi is the multidimensional attribute vector of the i-th point of interest. Let be the cosine similarity function, and wi be the weight of the i-th point of interest. This weight is proportional to the user's estimated dwell time at that point of interest. The estimated dwell time is calculated jointly by the category of the point of interest, the user's rhythm preference, and the average dwell time of similar users in the past.

[0036] S4-3 sets multiple constraints in the constrained optimization model. The first constraint is spatiotemporal accessibility: for any two consecutive points of interest (POIs) A and B on the route, the departure time from A plus the travel time from A to B must be less than or equal to the arrival time at B, and the arrival time at B plus the estimated stay at B must be less than the user's total available time. Travel time is calculated in real-time through the route planning interface of the high-precision map service, taking into account real-time traffic conditions. The second constraint is total budget: the sum of the entrance fees for all POIs and the estimated transportation costs must not exceed the user's set budget. Entrance fees are derived from static information in the POI knowledge graph, and transportation costs are estimated based on route distance and the user's chosen mode of transportation, such as walking, public transport, or taxi. The third constraint is individual rhythm preference: the total number of POIs planned each day must fall within a reasonable range determined by the user's rhythm preference value. For example, if the rhythm preference value is 1.5, the number of POIs per day should be controlled between 1.2 and 1.8. The fourth constraint is POI opening time: the planned arrival time at any POI must strictly fall within its operating hours.

[0037] S4-4 employs an improved genetic algorithm to solve the constrained optimization model. This algorithm defines each valid sequence of interest points satisfying all constraints as a chromosome. The fitness function is the total route matching degree M defined above. The algorithm uses a specially designed heuristic-based selection operator, prioritizing individuals with high matching degrees and low constraint violation rates; it employs a partial matching crossover operator to maintain sequence validity, ensuring that offspring chromosomes do not generate duplicate interest points and satisfy basic spatiotemporal order during crossover; and it uses a neighborhood exchange mutation operator designed to enhance local search capabilities, randomly selecting two adjacent genes in the chromosome for exchange, or inserting a high-potential unvisited interest point locally. The algorithm undergoes multiple generations of iterative evolution, with a population size of 200 per generation and a maximum of 500 iterations, or terminates early when the fitness value does not significantly improve for 50 consecutive generations. Finally, the chromosome with the highest fitness generated by the iteration is decoded. The ordered sequence of interest points represented by this chromosome constitutes the generated personalized travel route. Route information includes the name of each interest point, the visiting order, the estimated arrival time, the estimated stay duration, and the recommended mode of transportation.

[0038] After generating the initial route, the method further includes step S5: dynamic route replanning. This step continues to run as the user executes the generated personalized travel route. First, the user's geographic location coordinates are monitored in real time at five-minute intervals using the GPS and network positioning modules built into the user's mobile terminal, and the actual dwell time at each visited point of interest is recorded. Simultaneously, the background service continuously acquires real-time dynamic environmental attribute data for all unvisited points of interest along the route, including crowd density and weather conditions. The dynamic replanning mechanism is triggered when any of the following conditions are met: First, the straight-line distance deviation between the user's current location and the next point of interest to be visited in the planned route exceeds the preset geofence threshold. This threshold is dynamically set according to the density of urban areas, with 500 meters in urban areas and 1,000 meters in suburban areas. Second, the absolute difference between the actual duration of a user’s stay at a point of interest and the estimated duration exceeds a preset time deviation threshold, which is 30% of the estimated duration. Third, if the dynamic environmental attributes of any unvisited point of interest undergo a sudden change, such as the population density index increasing by more than 50% within ten minutes, or the weather forecast data being updated to a severe weather warning, the cosine similarity of its multidimensional attribute vector will decrease by more than the preset threshold of 0.2 compared to the initial planning.

[0039] Once the replanning mechanism is triggered, the system will use the user's current location as the new planning starting point, the remaining travel time and budget as the new constraint boundaries, and update the rhythm preference component in the user's multi-dimensional dynamic user intent state vector based on the user's exhibited behavioral deviations. For example, if the user's actual stay is generally longer than estimated, the rhythm preference value will be adjusted downwards. Subsequently, the system will re-execute the constraint optimization model construction and improved genetic algorithm solution process in step S4, but only optimize and adjust the subsequent unexecuted route portion. Finally, the updated route plan will be sent to the user's mobile terminal via push notification, and the changed parts will be highlighted in the map application.

[0040] At the system level, the personalized travel route planning system of this invention includes four core functional modules. The multi-source heterogeneous user data acquisition module is responsible for executing step S1. It interfaces with various data sources through a standardized data access layer to ensure the integrity and security of data collection. The user intent vector construction module is responsible for executing step S2. It integrates an explicit feature extraction unit, a latent feature parsing unit, a psychological trait inference unit, and a feature fusion and dimensionality reduction unit. These units communicate efficiently through an internal data bus. The interest point knowledge graph construction module is responsible for executing step S3. It includes a static information acquisition unit, a dynamic information access unit, and an attribute vector generation unit. This module maintains a distributed graph database for storing and querying multi-dimensional attribute vectors of interest points. The route generation and optimization module is responsible for executing step S4. Its internal constraint optimization model definition unit and intelligent algorithm solution unit are tightly coupled, enabling efficient handling of combinatorial optimization problems involving large-scale interest point sets. Furthermore, the system includes an independent route dynamic replanning module. This module includes a real-time status monitoring unit, a replanning trigger, and a dynamic adjustment execution unit. It interacts with the aforementioned modules through an event-driven mechanism to ensure the timeliness and accuracy of the replanning process. All modules are deployed on a cloud server cluster, achieving high availability and scalability through a microservice architecture. User terminals interact with cloud services through lightweight applications, which are only responsible for data display and user input, while the core computing logic is completed on the server side.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A personalized travel route planning method based on user profiles, characterized in that, include: Acquire multi-source heterogeneous user data, which includes explicit preference data, implicit behavioral data, and sociodemographic attribute data; Based on the aforementioned multi-source heterogeneous user data, a multi-dimensional dynamic user intent state vector is constructed, which includes explicit interests, implicit associations, and deep psychological traits. The deep psychological trait vector includes four dimensions: exploration degree, rhythm preference degree, social tendency, and consumption sensitivity. Construct a multi-dimensional attribute knowledge graph of tourist interest points. The graph integrates the static basic information of the tourist interest points and the real-time dynamic environmental attribute data to generate a corresponding multi-dimensional attribute vector for each tourist interest point. A personalized travel route is generated by a constrained optimization model. The constrained optimization model aims to maximize the sum of the weighted cosine similarity between the multidimensional dynamic user intent state vector and the multidimensional attribute vectors of each point of interest in the route sequence, and sets multiple constraints such as spatiotemporal accessibility, total budget, individual rhythm preference, and opening time of points of interest. as well as During the execution of the personalized travel route, user behavior and environmental attributes are monitored. When a deviation in user behavior or a sudden change in environmental attributes is detected, dynamic replanning is triggered to optimize and adjust the subsequent unexecuted parts of the route.

2. The method according to claim 1, characterized in that, The construction of the multidimensional dynamic user intent state vector includes: The explicit preference data is structured to extract keywords and numerical values, forming an explicit interest feature set; A bidirectional long short-term memory network text sentiment analysis model and a convolutional neural network image content recognition model are used to perform semantic parsing and visual theme feature extraction on the text and image content of implicit behavioral data, generating an implicit association interest feature set. By using a gradient-boosted decision tree psychological trait mapping classifier, and based on the explicit interest feature set, the implicit association interest feature set, and sociodemographic attribute data, a user travel psychological trait vector is generated; and The explicit interest feature set, the implicit related interest feature set, and the user travel psychological trait vector are hierarchically concatenated, and then dimensionality reduction and feature reconstruction are performed through a variational autoencoder to finally generate the multidimensional dynamic user intent state vector, where the vector is a fixed-dimensional floating-point vector.

3. The method according to claim 1, characterized in that, The construction of the multi-dimensional attribute knowledge graph of tourism interest points includes: Collect static basic information about tourist attractions of interest, including geographical coordinates, category, historical background, official description text, and opening schedule; Through third-party data application interfaces, dynamic environmental attribute data of tourist interest points can be obtained in real time, including real-time crowd density, weather forecast, traffic congestion index, and social media sentiment. Natural language processing is performed on the static basic information text data to extract topic keywords and entities, and these are mapped to the same four dimensions as the user's travel psychological trait vector to generate a static trait score for each point of interest; and By integrating the static basic information, the dynamic environmental attribute data, and the static trait score, a multidimensional attribute vector of the tourist interest point is constructed, wherein the dimension of the multidimensional attribute vector is consistent with the dimension of the multidimensional dynamic user intent state vector.

4. The method according to claim 1, characterized in that, The process of generating personalized travel routes through a constrained optimization model includes: Define a route planning solution space, where each solution is an ordered sequence of multiple tourist points of interest; The constrained optimization model is constructed, and the total route matching degree is determined by the sum of the weighted cosine similarity between the multidimensional dynamic user intent state vector and the multidimensional attribute vector of each interest point in the route sequence. The weight is proportional to the user's estimated dwell time at the interest point. The constrained optimization model sets multiple constraints, including spatiotemporal accessibility, total budget, individual rhythm preferences, and point-of-interest opening times; and An improved genetic algorithm is used to solve the constrained optimization model. The algorithm defines each legal interest point sequence as a chromosome, and the fitness function is the total matching degree of the route. The algorithm performs multi-generation iterative evolution through heuristic selection operators, partial matching crossover operators, and neighborhood exchange mutation operators to obtain the chromosome with the highest fitness as the personalized tourism route.

5. The method according to claim 1, characterized in that, The triggering of dynamic replanning and the optimization and adjustment of the subsequent unexecuted parts of the route include: Monitor users' geographical location and actual dwell time, and continuously acquire real-time dynamic environmental attribute data of unvisited points of interest; Dynamic replanning is triggered when the deviation between the user's current location and the planned route exceeds a preset geofence threshold, or the difference between the user's actual stay time at a point of interest and the estimated stay time exceeds a preset time deviation threshold, or a sudden change in the dynamic environmental attributes of any unvisited point of interest causes a decrease in the cosine similarity of its multidimensional attribute vector compared to the initial planning time exceeding a preset threshold; and Using the user's current location as the new planning starting point, the remaining travel time and budget as the new constraint boundaries, and combining user behavior deviations to update the rhythm preference component of the multidimensional dynamic user intent state vector, the personalized travel route generation step described in claim 1 is re-executed, the subsequent unexecuted parts are optimized and adjusted, and the updated route plan is pushed to the user's mobile terminal.

6. A personalized travel route planning system based on user profiles, characterized in that, include: A multi-source heterogeneous user data acquisition module is used to acquire multi-source heterogeneous user data, which includes explicit preference data, implicit behavioral data, and sociodemographic attribute data. The user intent vector construction module is used to construct a multi-dimensional dynamic user intent state vector that includes explicit interests, implicit associations, and deep psychological traits based on the multi-source heterogeneous user data. The knowledge graph construction module for interest points is used to construct a multi-dimensional attribute knowledge graph of tourism interest points. The graph integrates the static basic information of the tourism interest points and real-time dynamic environmental attribute data to generate a corresponding multi-dimensional attribute vector for each tourism interest point. The route generation and optimization module is used to generate a personalized travel route based on the multidimensional dynamic user intent state vector and the multidimensional attribute knowledge graph of the tourist interest points through a constraint optimization model. The constraint optimization model aims to maximize the total matching degree of the route and sets multiple constraints such as spatiotemporal accessibility, total budget, individual pace preference and opening time of interest points. as well as The route dynamic replanning module is used to monitor user behavior and environmental attributes during the execution of the personalized travel route. When a deviation in user behavior or a sudden change in environmental attributes is detected, dynamic replanning is triggered to optimize and adjust the subsequent unexecuted part of the route.

7. The system according to claim 6, characterized in that, The user intent vector construction module includes: The explicit feature extraction unit is used to perform structured processing on explicit preference data, extract keywords and values, and form an explicit interest feature set. The latent feature parsing unit is equipped with a bidirectional long short-term memory network text sentiment analysis model and a convolutional neural network image content recognition model. It is used to perform semantic parsing and visual theme feature extraction on the text and image content of latent behavioral data to generate a latent association interest feature set. The psychological trait inference unit is equipped with a gradient-boosting decision tree psychological trait mapping classifier, used to infer and generate user travel psychological trait vectors based on the explicit interest feature set, the implicit association interest feature set, and sociodemographic attribute data. These vectors include four dimensions: exploration degree, pace preference, social inclination, and consumption sensitivity. The feature fusion and dimensionality reduction unit is equipped with a variational autoencoder, which is used to hierarchically concatenate the explicit interest feature set, the implicit association interest feature set, and the user travel psychological trait vector, and then perform dimensionality reduction and feature reconstruction through the variational autoencoder to finally generate the multidimensional dynamic user intent state vector, wherein the vector is a fixed-dimensional floating-point vector.

8. The system according to claim 6, characterized in that, The interest-point knowledge graph construction module includes: The static information collection unit is used to collect static basic information about tourist attractions of interest, including geographical coordinates, category, historical background, official introductory text, and opening schedule. The dynamic information access unit is used to obtain dynamic environmental attribute data of tourist interest points in real time through a third-party data application interface. The data includes real-time crowd density, weather forecast, traffic congestion index and social media sentiment. The attribute vector generation unit is used to perform natural language processing on the text data of static basic information, extract topic keywords and entities, and map them to the same four dimensions as the user's travel psychological trait vector. It generates a static trait score for each point of interest, and then integrates the static basic information, the dynamic environmental attribute data and the static trait score to construct a multi-dimensional attribute vector of the travel interest point. The dimensions of the multi-dimensional attribute vector are consistent with the dimensions of the multi-dimensional dynamic user intent state vector.

9. The system according to claim 6, characterized in that, The route generation and optimization module includes: The constrained optimization model definition unit is used to define a mathematical model that maximizes the total route matching degree and integrates multiple constraints such as spatiotemporal accessibility, total budget, individual rhythm preferences, and point of interest opening time. The intelligent algorithm solution unit internally contains an improved genetic algorithm program. This program includes heuristic selection operators, partial matching crossover operators, and neighborhood exchange mutation operators, used to iteratively solve the constrained optimization model and search for the optimal sequence of interest points; and The route decoding output unit is used to decode the optimal solution obtained by the algorithm into a travel route, which includes the name of the point of interest, the order of visits, the estimated arrival time, the estimated length of stay, and the recommended mode of transportation.

10. The system according to claim 6, characterized in that, The route dynamic replanning module includes: The real-time status monitoring unit is used to monitor the user's location and actual dwell time through the positioning module of the mobile terminal, and continuously acquire real-time dynamic environmental attribute data of unvisited points of interest; A replanning trigger is used to activate the replanning process when the deviation between the user's current location and the planned route exceeds a preset geofence threshold, or the difference between the user's actual stay time at a point of interest and the estimated stay time exceeds a preset time deviation threshold, or a sudden change in the dynamic environmental attributes of any unvisited point of interest causes a decrease in the cosine similarity of its multidimensional attribute vector compared to the initial planning time exceeding a preset threshold; and The dynamic adjustment execution unit is used to call the route generation and optimization module. It takes the user's current location as the new planning starting point, the remaining travel time and budget as the new constraint boundaries, updates the rhythm preference component of the multi-dimensional dynamic user intent state vector, recalculates and optimizes the subsequent route, and generates an updated route plan.

Images