Trout map searching system and method based on deep learning and Beidou database

Abstract

The invention discloses a travel map searching system and method based on deep learning and a Beidou database, and the method comprises the following steps: obtaining the position and track data of a user, loading a geographic object, and constructing a heterogeneous map; receiving a natural language request, extracting a semantic vector and writing the semantic vector into the graph; constructing a graph comparison learning network, generating a view and executing embedded learning; optimizing structural parameters by using a pigeon inspired algorithm, and outputting an optimal model; constructing a current position node, extracting a local sub-graph, and generating a sorting list; executing path reasoning, generating a recommendation result, and feeding back and updating. According to the method, deep semantic understanding and geographical trajectory modeling are fused, so that high-precision, interpretable and self-learning travel map intelligent search recommendation under the driving of the current position of the user is realized.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a cultural and tourism map search system and method based on deep learning and the Beidou database. Background Art

[0002] With the development of geographic information systems, big data analytics, and intelligent recommendation technologies, user location-based map search and recommendation services have gained widespread application in transportation, cultural tourism, and lifestyle navigation. In the context of cultural and tourism integration, tourists are demanding a more personalized and intelligent destination search experience. They not only expect the system to understand their complex queries expressed in natural language, but also expect search results to be location-relevant, interest-matched, and transparent, thereby improving overall travel efficiency and user satisfaction.

[0003] Current mainstream map search recommendation methods rely primarily on keyword matching, rule-based retrieval mechanisms, or center point recommendations based on points of interest. While these methods offer relatively fast response times for simple path navigation, they exhibit significant limitations in processing query requests containing complex semantic instructions, considering users' historical behavior paths, or integrating real-time location scenario modeling. On the one hand, existing systems are mostly based on vector space models and rely solely on keywords or tags for text query matching. They lack the ability to deeply understand the semantics of natural language search requests and are unable to identify implicit location preferences, scenario preferences, or behavioral needs in user intent. On the other hand, many recommendation algorithms lack effective modeling of users' historical behavior, geographic movement trajectories, and visit timing, resulting in a lack of pertinence and contextual adaptability in recommendation results, and are prone to semantic matching errors, redundant location recommendations, and unreasonable path planning.

[0004] Furthermore, existing map recommendation systems generally employ a flattened graph modeling approach, which is unable to express the complex relationships between the multi-type nodes and multi-dimensional edges found in cultural tourism data. In the cultural tourism sector, user behavior, geographic objects, and path connections often exhibit heterogeneity and semantic intersection. Without the ability to construct a graph structure that can express the fusion of semantic, spatial, and behavioral information, accurate and efficient multi-dimensional intelligent recommendations will be difficult to achieve. Furthermore, while graph neural networks have been gradually introduced into recommendation systems, most employ fixed structures and single parameter configurations, lacking structural tuning mechanisms tailored to specific task objectives. This makes them prone to overfitting, embedding failure, or poor model robustness. Summary of the Invention

[0005] One purpose of the present invention is to propose a cultural and tourism map search system and method based on deep learning and Beidou database. The present invention integrates heterogeneous graph modeling, graph comparative learning and pigeon flock optimization algorithm, and captures user historical trajectories and geographical relationships by constructing a heterogeneous graph structure containing users, geographic objects and path nodes; uses the optimized graph comparative learning network to perform semantic embedding calculation and relevance sorting, realizes map intelligent search recommendation driven by the user's current location, and realizes adaptive update of graph structure and model through user feedback behavior data. It has the advantages of deep semantic understanding, strong location association, high recommendation accuracy and strong system interpretability.

[0006] According to an embodiment of the present invention, a cultural and tourism map search method based on deep learning and Beidou database includes the following steps: Obtain the user's current location data and historical movement trajectory data from the Beidou database, load a static information set containing multiple cultural and tourism geographic objects, and construct a heterogeneous graph structure; Receive a natural language search request input by the user, perform context semantic vector extraction on the natural language search request, obtain a semantic representation result, and write it into the heterogeneous graph structure as the initial feature of the user node; A graph contrast learning network is constructed based on heterogeneous graph structures. Graph structure perturbations are performed to generate two sets of views. These two sets of views are input into the graph neural network for node embedding learning. The structural parameters of the graph contrastive learning network are optimized through the pigeon flock optimization algorithm, and the optimal structure and parameter configuration of the graph contrastive learning network are finally output to form the optimized graph contrastive learning network; When a user initiates a map search request, the user's current location coordinates are extracted from the BeiDou database. A graph node is constructed based on the user's current location coordinates and added to the heterogeneous graph structure. The local subgraph is extracted from the heterogeneous graph structure as the central node. The optimized graph contrastive learning network is called to perform node embedding representation calculation on the local subgraph to generate a semantic relevance ranked list. Based on the connection relationship between the semantic relevance sorted list and the path nodes in the heterogeneous graph structure, multi-hop path reasoning is performed to generate a recommendation result containing the optimal path sequence from the user's current location to the target geographic object and recommended explanation information.

[0007] Optionally, the Beidou database is a navigation data set that records the user's real-time location and movement path, including latitude and longitude coordinates, timestamps, speed and direction information, and is used to provide the user's spatial position and time-series trajectory information on the map, supporting graph node construction, subgraph extraction and path recommendation calculation; The cultural and tourism geographic objects include cultural and tourism related entities with geographic locations, label classifications, and rating records, which are used to construct geographic entity nodes in the graph, supporting user interest matching, semantic association calculation, and path recommendation generation; The heterogeneous graph structure includes user nodes, geographic object nodes and path nodes. Access edges, spatial adjacency edges and trajectory sequence edges are established between the nodes to represent user behavior, geographic location relationships and travel trajectories, and support graph comparative learning and personalized recommendation calculations.

[0008] Optionally, the initial features of the user node are written into the heterogeneous graph structure, including: receiving a natural language search request input by a user through a terminal, wherein the natural language search request includes geographic preference, behavioral intention, travel purpose, and topic keywords; Perform text preprocessing on natural language search requests, including word segmentation, stop word removal, word form unification, and invalid symbol removal, to generate standardized sentence input; Perform context semantic vector extraction processing on the preprocessed natural language input text input to generate a context-dependent semantic vector representation; The extracted semantic vector representation is dimensionally normalized and feature aligned to meet the dimensional consistency and structural constraint requirements of node feature representation in heterogeneous graph structures; Bind the semantic vector representation with the user identity to generate a complete user node feature representation data structure; The generated user node features are written into the corresponding user node positions in the heterogeneous graph structure, completing the semantic embedding mapping process from natural language search requests to the graph structure.

[0009] Optionally, constructing a graph contrastive learning network based on a heterogeneous graph structure includes: Constructing the original graph based on heterogeneous graph structure ,in, Represents a node collection, including user nodes, geographic object nodes, and path nodes. Represents the edge set, including visited edges, spatial adjacency edges, and trajectory order edges; Introducing geographic location similarity weighting coefficient , constructing a location-aware similarity function , used to fuse the spatial relationships between nodes in the contrastive loss; Introducing sub-graph-aware scale control factor , used to determine the adjacency depth of the local subgraph extracted from the central node, in the first view and the second view In the construction of The neighborhood of is used as the subgraph extraction range, and node embedding extraction is performed at the subgraph level; Introducing node shading ratio adjustment factor , used to control the masking ratio of node features during the view perturbation process, when constructing the first view and the second view When, according to the ratio Mask the node feature dimensions in the original feature matrix; The first view and the second view Input the graph neural network encoder to obtain the embedded representation of each node in the two sets of views 、 ; Constructing a location-aware contrastive loss function , used to optimize the consistency of embedding representation of the same node in two views; The graph neural network is trained based on the location-aware contrastive loss function, and finally a graph contrastive learning network representation structure is obtained that integrates geographic awareness, adjustable granularity and dynamic perturbation strategy.

[0010] Optionally, forming an optimized graph contrast learning network includes: Set the initial population size of the pigeon swarm optimization algorithm to , initialize the pigeon flock individual set , each individual Represents the structural parameter combination of the graph contrast learning network, where , represents the weighted coefficient of geographical location similarity corresponding to the i-th individual, is the subgraph perception scale control factor corresponding to the i-th individual, is the node shielding ratio adjustment factor corresponding to the i-th individual; Randomly initialize the velocity vector for each individual , and set the individual memory vector , used to record individual historical optimal parameter configuration; For each individual parameter combination Based on the constructed graph contrastive learning network structure, node embedding calculation and contrastive learning training process are performed to obtain the corresponding contrast loss function value. Node sorting recommendation accuracy , weighted fitness function ; All individuals are sorted from high to low according to their fitness values, and divided into sub-populations, and the individual with the highest fitness value in each sub-population is selected as the local pilot , the remaining individuals perform local guidance updates around the local pilot body; Calculate the current iteration round Dynamic inertia factor ; According to the dynamic inertia factor Calculate the candidate values ​​of individual geomagnetic navigation positions respectively Candidate values ​​for landmark-guided locations : Geomagnetic navigation update: ; Landmark guidance update: ; in, is the position vector of the i-th individual at the t-th iteration, is the velocity vector of individual i in round t, is the bootstrap weight coefficient, is the individual memory weight coefficient, is the position of the individual with the best fitness in the current population, is the historical optimal memory position of the i-th individual; Introducing stage control factors ,dynamically balance the two candidate positions through the fusion mechanism, perform position update, and update the individual speed of the next round according to the position update results of this round; Calculate the new round of individual positions Corresponding contrastive loss of graph contrast learning network and recommendation accuracy , update the fitness function ; If the current individual Fitness , then update the memory vector to the current solution ; Repeat until the maximum number of iterations is met Or fitness convergence threshold condition, and finally output the individual with the best fitness , as the optimal structural parameter combination of graph contrastive learning network.

[0011] Optionally, generating a semantic relevance ranked list includes: Receive map search requests initiated by users on the terminal and trigger the initialization of the location-aware recommendation process; Extract the latitude and longitude coordinates of the user's current location from the Beidou database, and use the latitude and longitude coordinates of the user's current location as the location node to construct a graph node entity; The location node is used as the user's current location node and added to the currently maintained heterogeneous graph structure to expand the spatial representation capability of the graph structure; Taking the user's current location node as the central node, the associated local subgraph structure is extracted from the heterogeneous graph structure based on spatial adjacency and access relations; The optimized graph contrast learning network is called to perform embedding calculations on the extracted local subgraphs to generate semantic representations for each candidate node. Based on the semantic relevance between the user's current location node and the embedded representation of the candidate node, a ranked result list is generated and output as a candidate result set for search recommendation.

[0012] Optionally, the user feedback behavior data specifically includes click records, browsing dwell time, historical search paths and actual navigation behaviors, which are used to optimize the heterogeneous graph structure and the retraining process of the graph comparison learning network.

[0013] The cultural and tourism map search system based on deep learning and Beidou database includes the following modules: A request processing module, configured to receive a user map search request and extract a semantic representation as a user node feature; The mapping module is used to extract user location and trajectory data from the Beidou database and build a heterogeneous graph structure based on geographic object information; The graph learning optimization module is used to generate perturbation views based on heterogeneous graphs, train graph contrastive learning networks, and adjust structural parameters using the pigeon flock optimization algorithm; Subgraph calculation module, used to extract local subgraphs and generate node embedding representations; The recommendation module is used to generate ranking results based on the semantic relevance of the embedded vectors and output recommended paths and explanation information; The feedback update module is used to collect user feedback behavior data and use it for graph structure update and model retraining.

[0014] The beneficial effects of the present invention are: By integrating the Beidou database, graph comparative learning network and pigeon flock optimization algorithm, this invention addresses the problems of existing map search and recommendation technologies, such as weak semantic understanding ability, insufficient graph structure expression, low recommendation accuracy and lack of adaptive model updating. It proposes an intelligent search method with an overall closed-loop structure and multi-source information fusion capability, which significantly improves the system's response quality and recommendation accuracy in complex user query contexts.

[0015] This paper constructs a heterogeneous graph structure consisting of user nodes, geographic object nodes, and path nodes, integrating real-time user location and historical trajectory data provided by the Beidou database. This effectively preserves spatial adjacency, access timing, and path evolution relationships, enhancing the system's ability to model geographic information and behavioral data. By introducing a graph contrast learning mechanism, constructing perturbation views based on the heterogeneous graph structure and performing node embedding alignment, the robustness of semantic representation and the ability to express node similarity are effectively enhanced, overcoming the shortcomings of traditional graph learning models, which tend to lose structural semantics when fusing multi-type data.

[0016] To further enhance the adaptability and optimization of graph neural network architectures, this paper introduces a pigeon flock optimization algorithm. By setting geographic perception weights, subgraph scale control factors, and feature perturbation ratios as optimization variables, the algorithm constructs individual parameter vectors and uses a swarm search approach to optimize the structure of the graph contrastive learning network. The algorithm utilizes a dynamic inertia factor, an individual memory mechanism, and a staged fusion guidance strategy to achieve a smooth transition between global exploration and local convergence, significantly improving the adaptability of structural parameter configuration and the discriminative power of graph embedding.

[0017] The present invention also constructs a graph node by extracting the user's current location in real time. Using this node as the central node, the system extracts its spatially adjacent subgraph and performs embedding calculations. It then combines semantic relevance and path accessibility to perform recommendation sorting, outputting search results that are geographically meaningful and semantically consistent. After generating the recommendation results, the system collects the user's click history, dwell behavior, and actual navigation path as feedback data. This data is then used to dynamically update the heterogeneous graph structure and retrain the graph comparison learning model, thereby constructing a recommendation system with continuous learning and dynamic adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flowchart of the cultural and tourism map search method based on deep learning and Beidou database proposed by the present invention; Figure 2 This is a structural diagram of the cultural and tourism map search system based on deep learning and Beidou database proposed in the present invention. DETAILED DESCRIPTION

[0019] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0020] refer to Figure 1-Figure 2 , a cultural and tourism map search method based on deep learning and Beidou database includes the following steps: Obtain the user's current location data and historical movement trajectory data from the Beidou database, load a static information set containing multiple cultural and tourism geographic objects, and construct a heterogeneous graph structure; Receive a natural language search request input by the user, perform context semantic vector extraction on the natural language search request, obtain a semantic representation result, and write it into the heterogeneous graph structure as the initial feature of the user node; A graph contrast learning network is constructed based on heterogeneous graph structures. Graph structure perturbations are performed to generate two sets of views. These two sets of views are input into the graph neural network for node embedding learning. The structural parameters of the graph contrastive learning network are optimized through the pigeon flock optimization algorithm, and the optimal structure and parameter configuration of the graph contrastive learning network are finally output to form the optimized graph contrastive learning network; When a user initiates a map search request, the user's current location coordinates are extracted from the BeiDou database. A graph node is constructed based on the user's current location coordinates and added to the heterogeneous graph structure. The local subgraph is extracted from the heterogeneous graph structure as the central node. The optimized graph contrastive learning network is called to perform node embedding representation calculation on the local subgraph to generate a semantic relevance ranked list. Based on the connection relationship between the semantic relevance sorted list and the path nodes in the heterogeneous graph structure, multi-hop path reasoning is performed to generate recommendation results containing the optimal path sequence and recommended explanation information from the user's current location to the target geographic object. The recommendation results are returned to the user terminal, and user feedback behavior data is collected for incremental updates of the heterogeneous graph structure and retraining of the optimized graph comparative learning network.

[0021] The present invention constructs a multi-source heterogeneous graph structure by integrating Beidou database positioning data, user behavior trajectories and natural language semantic features, effectively improving the ability to match user intentions with geographical contexts during map searches. With the help of a graph contrast learning network and a graph structure perturbation mechanism, the system can extract highly robust node embedding representations from the semantic structure of the graph, enhancing the adaptability of recommendation results to query semantics. At the same time, a pigeon flock optimization algorithm is introduced to dynamically tune the graph structure parameters to achieve collaborative optimization of the graph neural network structure and task objectives, significantly improving the generalization ability and accuracy of the model in different user query scenarios. In addition, the present invention constructs a central node by extracting the user's current location, extracts local subgraphs and performs high-quality embedding calculations to generate a candidate list with higher sorting accuracy and stronger spatial correlation, and performs multi-hop reasoning in combination with path connection information to generate explainable navigation recommendation results. By driving the graph structure and model retraining with feedback behavior data, the system has the ability to continuously evolve and personalize recommendations. Overall, the present invention has achieved comprehensive improvements in semantic understanding, structural modeling, recommendation interpretability and system adaptability, significantly outperforming traditional keyword retrieval methods.

[0022] In this embodiment, the Beidou database is a navigation data set that records the user's real-time location and movement path, including latitude and longitude coordinates, timestamps, speed and direction information, which is used to provide the user's spatial position and time-series trajectory information in the map, and supports graph node construction, subgraph extraction and path recommendation calculation; the cultural and tourism geographic objects include cultural and tourism-related entities with geographic location, label classification, and rating records, which are used to construct geographic entity nodes in the graph, support user interest matching, semantic association calculation and path recommendation generation; the heterogeneous graph structure includes user nodes, geographic object nodes and path nodes, and access edges, spatial adjacency edges and trajectory sequence edges are established between nodes to represent user behavior, geographic location relationships and travel trajectories, and support graph comparative learning and personalized recommendation calculation.

[0023] In this embodiment, the initial features of the user node are written into the heterogeneous graph structure, including: receiving a natural language search request input by a user through a terminal, wherein the natural language search request includes geographic preference, behavioral intention, travel purpose, and topic keywords; Perform text preprocessing on natural language search requests, including word segmentation, stop word removal, word form unification, and invalid symbol removal, to generate standardized sentence input; Perform context semantic vector extraction processing on the preprocessed natural language input text input to generate a context-dependent semantic vector representation; The extracted semantic vector representation is dimensionally normalized and feature aligned to meet the dimensional consistency and structural constraint requirements of node feature representation in heterogeneous graph structures; Bind the semantic vector representation with the user identity to generate a complete user node feature representation data structure; The generated user node features are written into the corresponding user node positions in the heterogeneous graph structure, completing the semantic embedding mapping process from natural language search requests to the graph structure.

[0024] The present invention realizes the accurate mapping of user search requests to semantic embeddings in heterogeneous graph structures by constructing a standardized natural language processing process, significantly improving the map search system's ability to understand the user's true intentions. Natural language requests often contain complex expressions and diverse query content, such as geographic preferences, behavioral intentions, and travel purposes. The present invention effectively extracts deep semantic features and regularizes them into structured embedding vectors through text preprocessing, contextual semantic vector extraction, feature alignment and normalization steps. By binding the semantic vector to the user identity and writing it into the heterogeneous graph structure, it not only ensures the consistency of the node feature dimension and the graph network structure, but also realizes the fusion of personalized semantic query and structural modeling. This technical solution solves the problem in traditional map systems that fuzzy language is difficult to parse and node features cannot directly carry natural language intentions. It enhances the expression and expansion capabilities of the graph structure for user semantics, lays a high-quality input foundation for the subsequent training of graph neural networks and the generation of recommended paths, and has the beneficial effects of strong accuracy, high generalization ability and excellent structural compatibility.

[0025] In this embodiment, the construction of a graph contrastive learning network based on a heterogeneous graph structure includes: Constructing the original graph based on heterogeneous graph structure ,in, Represents a node collection, including user nodes, geographic object nodes, and path nodes. Represents the edge set, including visited edges, spatial adjacency edges, and trajectory order edges; Introducing geographic location similarity weighting coefficient , constructing a location-aware similarity function , used to fuse the spatial relationships between nodes in the contrastive loss: ; in, Representation node and The geographical similarity of represents the cosine similarity between embedding vectors, For nodes The embedding vector representation of For nodes Embedded vector representation of ; Introducing sub-graph-aware scale control factor , used to determine the adjacency depth of the local subgraph extracted from the central node, in the first view and the second view In the construction of The neighborhood of is used as the subgraph extraction range, and node embedding extraction is performed at the subgraph level; Introducing node shading ratio adjustment factor , used to control the masking ratio of node features during the view perturbation process, when constructing the first view and the second view When, according to the ratio Mask the node feature dimensions in the original feature matrix to enhance the structural robustness of the embedding; The first view and the second view Input the graph neural network encoder to obtain the embedded representation of each node in the two sets of views 、 ; Constructing a location-aware contrastive loss function , used to optimize the consistency of embedding representation of the same node under two views: ; in, is the contrast temperature coefficient, is a logarithmic function, is an exponential function; The graph neural network is trained based on the location-aware contrastive loss function, and finally a graph contrastive learning network representation structure is obtained that integrates geographic awareness, adjustable granularity and dynamic perturbation strategy.

[0026] The present invention improves the graph contrast learning network by introducing three innovative structural parameters, significantly enhancing the robustness and representation ability of the model in terms of geographic perception, structural perturbation and semantic expression. First, a geographic location similarity weighting coefficient is introduced to construct a location-aware similarity function, so that the spatial adjacency relationship between nodes directly acts on the contrast loss term, effectively improving the model's ability to capture the spatial association between users and geographic objects. Secondly, a subgraph-aware scale control factor is introduced to dynamically adjust the neighborhood range extracted from the central node, support the construction of subgraphs with different hops, realize granularity-adjustable local structure modeling, and enhance the adaptability to multi-level semantic relationships in the graph. Thirdly, a node masking ratio adjustment factor is set to control the feature masking intensity during the graph view perturbation process, thereby improving the stability of the model in countering perturbations and unsupervised view enhancement. Through the combined effect of these structural parameters, the node embedding representation extracted by the system in the two perturbed views has stronger consistency and discriminability, and the location-aware contrast loss function further optimizes the semantic alignment effect between different views. The graph comparative learning network finally constructed has stronger geographical representation capabilities, semantic expression accuracy and structural generalization performance, providing a high-quality embedding representation foundation for subsequent recommendation ranking and path reasoning.

[0027] In this embodiment, forming an optimized graph contrast learning network includes: Set the initial population size of the pigeon swarm optimization algorithm to , initialize the pigeon flock individual set , each individual Represents the structural parameter combination of the graph contrast learning network, where , represents the weighted coefficient of geographical location similarity corresponding to the i-th individual, is the subgraph perception scale control factor corresponding to the i-th individual, is the node shielding ratio adjustment factor corresponding to the i-th individual; Randomly initialize the velocity vector for each individual , and set the individual memory vector , used to record individual historical optimal parameter configuration; For each individual parameter combination Based on the constructed graph contrastive learning network structure, node embedding calculation and contrastive learning training process are performed to obtain the corresponding contrast loss function value. Node sorting recommendation accuracy , construct the fitness function : ; in, is the weight coefficient of the contrast loss in the fitness function, which is used to control the influence of structural similarity. It is the weight coefficient of recommendation accuracy in the fitness function, which is used to control the contribution of recommendation quality; All individuals are sorted from high to low according to their fitness values, and divided into sub-populations, and the individual with the highest fitness value in each sub-population is selected as the local pilot , the remaining individuals perform local guidance updates around the local pilot body; Calculate the current iteration round Dynamic inertia factor : ; in, is the maximum inertia weight value initially set, is the maximum number of iterations, is the set minimum inertia weight value; Calculate the candidate values ​​of individual geomagnetic navigation positions respectively Candidate values ​​for landmark-guided locations : Geomagnetic navigation update: ; Landmark guidance update: ; in, is the position vector of the i-th individual at the t-th iteration, is the velocity vector of individual i in round t, is the bootstrap weight coefficient, is the individual memory weight coefficient, is the position of the individual with the best fitness in the current population, is the historical optimal memory position of the i-th individual; Introducing stage control factors , dynamically balance the two candidate positions through the fusion mechanism, perform position updates, and update the individual speeds of the next round based on the results of this round of position updates: ; in, For the i-th individual in the Position vector at the round iteration; Calculate the new round of individual positions Corresponding contrastive loss of graph contrast learning network and recommendation accuracy , update the fitness function ; If the current individual Fitness , then update the memory vector to the current solution ; Repeat until the maximum number of iterations is met Or fitness convergence threshold condition, and finally output the individual with the best fitness , as the optimal structural parameter combination of graph contrastive learning network.

[0028] To address the problem of traditional graph neural network structural parameters being unable to be automatically tuned, this paper introduces a pigeon flock optimization algorithm with a phased navigation mechanism. This algorithm achieves adaptive optimization of structural parameters within an unsupervised graph contrastive learning framework, significantly improving the model's structural expressiveness and recommendation performance. First, the algorithm sets three key structural parameters: a geographic location similarity weighting coefficient, a subgraph perception scale control factor, and a node shading ratio adjustment factor. These parameters correspond to the spatial sensitivity, subgraph extraction granularity, and feature perturbation intensity of the graph model, respectively, forming a parameter combination for each individual. A fitness function jointly evaluates each individual's performance in terms of node semantic alignment (contrast loss) and recommendation accuracy, ensuring that the optimization objective balances embedding quality and recommendation effectiveness. During the search process, the algorithm sorts the population by fitness and divides it into subpopulations. The individual with the highest fitness serves as the local leader, guiding the remaining individuals in their local updates. Furthermore, a dynamic inertia factor, geomagnetic navigation location candidates, and landmark guidance location candidate mechanisms are introduced to simulate the search behavior of a pigeon flock at different stages: emphasizing exploration capability in the early stages and convergence accuracy in the later stages. The phased control factors integrate these two guidance directions to ensure a smooth transition of the search path. Furthermore, after each iteration, individuals update their memory vectors based on their actual optimization performance, leveraging and learning from historical experience. Ultimately, through multiple rounds of optimization iterations, the system outputs a structurally optimal graph comparative learning network model. This algorithm significantly addresses the problem of traditional fixed structures being unable to adapt to diverse task characteristics. It enables adaptive optimization of graph comparative learning models across diverse data graph structures and semantic scenarios, improving recommendation accuracy, model generalization, and robustness, providing a sustainable and evolving optimization mechanism for large-scale map search and recommendation systems.

[0029] In this embodiment, generating a semantic relevance ranked list includes: Receive map search requests initiated by users on the terminal and trigger the initialization of the location-aware recommendation process; Extract the latitude and longitude coordinates of the user's current location from the Beidou database, and use the latitude and longitude coordinates of the user's current location as the location node to construct a graph node entity; The location node is used as the user's current location node and added to the currently maintained heterogeneous graph structure to expand the spatial representation capability of the graph structure; Taking the user's current location node as the central node, the associated local subgraph structure is extracted from the heterogeneous graph structure based on spatial adjacency and access relations; The optimized graph contrast learning network is called to perform embedding calculations on the extracted local subgraphs to generate semantic representations for each candidate node. Based on the semantic relevance between the user's current location node and the embedded representation of the candidate node, a ranked result list is generated and output as a candidate result set for search recommendation.

[0030] The present invention achieves a deep fusion of map search requests and current location context by introducing a subgraph embedding calculation mechanism driven by the user's current location, significantly improving the spatial matching and semantic relevance of recommendation results. After receiving a map search request, the system automatically extracts the user's current location coordinates from the Beidou database, constructs them as graph nodes and dynamically injects them into the heterogeneous graph structure, so that the user's location can participate in the graph modeling process in real time, effectively expanding the spatial perception ability of the graph structure. With the current location node as the center, the local subgraph related to its spatial adjacency and access relationship is extracted, which can accurately capture the geographic objects and path relationships that are strongly associated with the current location, avoiding the interference and computational redundancy caused by full-graph modeling. By calling the optimized graph contrast learning network, embedding calculations are performed on the local subgraph to obtain the semantic representation of the candidate node, and then combined with the embedding vector of the user's current location node, the semantic relevance is calculated and a sorted result list is generated to ensure that the recommendation results are consistent with the user's query semantics and meet the practicality of the geographic location. This mechanism realizes spatial semantic fusion recommendation based on the user's current location, and has the significant advantages of fast response speed, strong location adaptability, precise semantic matching and good recommendation interpretability. It is especially suitable for mobile smart map search and recommendation tasks in complex scenarios.

[0031] In this embodiment, the user feedback behavior data specifically includes click records, browsing dwell time, historical search paths and actual navigation behaviors, which are used to optimize the heterogeneous graph structure and the retraining process of the graph comparative learning network.

[0032] refer to Figure 2 , a cultural and tourism map search system based on deep learning and Beidou database, including the following modules: A request processing module, configured to receive a user map search request and extract a semantic representation as a user node feature; The mapping module is used to extract user location and trajectory data from the Beidou database and build a heterogeneous graph structure based on geographic object information; The graph learning optimization module is used to generate perturbation views based on heterogeneous graphs, train graph contrastive learning networks, and adjust structural parameters using the pigeon flock optimization algorithm; Subgraph calculation module, used to extract local subgraphs and generate node embedding representations; The recommendation module is used to generate ranking results based on the semantic relevance of the embedded vectors and output recommended paths and explanation information; The feedback update module is used to collect user feedback behavior data and use it for graph structure update and model retraining.

[0033] Example 1: To verify the feasibility of this invention, a city's cultural and tourism data center partnered with an artificial intelligence company to conduct a comparative experiment on map recommendation systems based on natural language user map queries. The test system deployed both a traditional keyword search solution and this invention, providing the service to users via mobile terminals and capturing their actual search and navigation behavior. Participating users, including locals and tourists, initiated map search requests in various scenarios, including daily commutes, weekend leisure activities, and cultural and museum visits. The system returned results, recording the user's click volume, average dwell time, recommendation accuracy, and actual visit paths.

[0034] In traditional systems, search requests submitted by users, such as "quiet parks suitable for walking", "cultural buildings suitable for taking photos", and "museums with exhibitions", are often returned only through keyword matching, which easily leads to recommendations that do not match the actual intention. For example, when searching for "cultural places suitable for quiet reading", the system may recommend areas with dense traffic and strong commercial attributes due to the high frequency of the "culture" keyword, while ignoring the user's preference for a quiet atmosphere, resulting in an imbalance in the recommendation ranking. In contrast, when using the system of the present invention, the search request is first converted into a semantic vector, and semantic matching is performed with the user nodes, geographic object nodes, and path nodes in the constructed heterogeneous graph structure. Then, the relevance ranking is generated by subgraph embedding in combination with the user's current location, achieving a deep understanding of the user's intention.

[0035] In the experiment, the system tested a total of ten types of locations, including cultural and historical attractions, urban park locations, exhibition halls and museums, comprehensive business districts, areas around universities and bookstores, local characteristic blocks, quiet places for humanities, night tour places, popular photography spots, and parent-child recreation areas. When faced with vague and highly descriptive natural language query requests, the control group system often failed to grasp the core intent, resulting in low recommendation clicks, inaccurate rankings, and short user stay time. After loading the user's real-time location and historical trajectory data recorded in the Beidou database, the system of the present invention constructs a heterogeneous graph structure containing spatial adjacency, access relationships, and behavioral paths, and then embeds the calculation through graph comparison learning network, effectively improving the recommendation matching degree.

[0036] Taking the "exhibition halls and museums" category as an example, the traditional system's recommendation clicks were 132 times, with an average dwell time of 61.87 seconds, and the recommendation ranking was 5th; while under the same conditions, the system of the present invention increased the clicks to 171 times, the average dwell time to 76.20 seconds, and the ranking was advanced to 1st. For another example, in the "near universities and bookstores" scenario, the traditional method had 102 clicks and the recommended position was 5th, while the system of the present invention increased it to 138 times and ranked 2nd, and the dwell time increased from 59.03 seconds to 79.98 seconds, indicating that users have a higher acceptance of the recommendation results. Through the optimized graph comparison learning model, the system can better understand the abstract semantics of "quiet atmosphere", "suitable for night tours", and "rich culture", and combined with geospatial trajectory reasoning, the recommended paths are more realistic and credible.

[0037] More importantly, in the system of the present invention, users' clicks, browsing, and path navigation behaviors are continuously collected as feedback data, continuously optimizing the edge weights and node states of the graph structure. This data is also used to retrain the graph comparative learning network, enabling the model to adapt and evolve individually. Experimental results show that, across the entire user population, the average click volume of the system of the present invention increased by approximately 31.6% compared to the traditional system, the average dwell time increased by approximately 23%, the recommendation ranking shifted significantly forward, and the actual visit rate increased from 67% to nearly 90%. User questionnaire feedback satisfaction scores increased from 3.4 points (traditional system) to 4.6 points (system of the present invention).

[0038] Table 1 Comparison of map search system recommendation effects Judging from the overall data, the system of the present invention has shown higher click-through rates, longer user dwell time, and higher recommendation rankings than traditional keyword search systems in all location types, indicating that it has obvious performance advantages in map search recommendation tasks. In the "Cultural and Historical Scenic Spots" category, the traditional system had 135 clicks, while the system of the present invention increased to 180, an increase of 33.3%; the average user dwell time increased from 56.43 seconds to 67.14 seconds, and the recommendation ranking also jumped from 5th to 1st. This shows that the system of the present invention is more capable of matching the semantic needs of "cultural history" with geographic intent.

[0039] In the "Exhibition Halls and Museums" category, the traditional system received 132 clicks, with an average dwell time of 61.87 seconds, and ranked fifth in recommendations. However, under the same query conditions, the proposed system received 171 clicks, with a dwell time of 76.20 seconds, and moved up to first place, indicating a higher degree of user acceptance of the recommended results. This trend is particularly evident in scenarios sensitive to both semantics and location, such as "urban parks," "quiet places," and "popular photography spots." The proposed system can reflect user preferences as identifiable embedded features and output more relevant candidate locations.

[0040] Notably, in the category "Parent-Child Play Area," which has a relatively clear target audience but diverse semantic descriptions, the traditional system ranked it 7th with 119 clicks. However, the proposed system, which recognized the semantic expressions "Parent-Child," "Suitable for Children," and "Safe and Open," significantly advanced its ranking to 1st, saw 164 clicks, and increased dwell time to 74.21 seconds. This result demonstrates the proposed system's high adaptability to fuzzy query intent.

[0041] From the perspective of user behavior, the increase in click volume directly reflects the improvement in the relevance of recommended content, and the increase in average dwell time reflects the close connection between recommended results and user interests, while the advance of recommendation ranking further enhances user satisfaction with the system and efficiency of use. The traditional system ranks recommendations low in many scenarios, indicating that its retrieval logic fails to understand the true intention behind the query, and the system only performs word-to-word matching, resulting in users having to click multiple times or low search efficiency. The present invention, by integrating heterogeneous graph structures with optimized graph comparison learning network models, establishes an effective connection between spatial adjacency, behavioral path information and semantic representation, making the recommendation results more reasonable in ranking and interpretability.

[0042] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A cultural and tourism map search method based on deep learning and Beidou database, characterized in that: The steps include: Obtain the user's current location data and historical movement trajectory data from the Beidou database, load a static information set containing multiple cultural and tourism geographic objects, and construct a heterogeneous graph structure; Receive a natural language search request input by the user, perform context semantic vector extraction on the natural language search request, obtain a semantic representation result, and write it into the heterogeneous graph structure as the initial feature of the user node; A graph contrast learning network is constructed based on heterogeneous graph structures. Graph structure perturbations are performed to generate two sets of views. These two sets of views are input into the graph neural network for node embedding learning. The structural parameters of the graph contrastive learning network are optimized through the pigeon flock optimization algorithm, and the optimal structure and parameter configuration of the graph contrastive learning network are finally output to form the optimized graph contrastive learning network; When a user initiates a map search request, the user's current location coordinates are extracted from the BeiDou database. A graph node is constructed based on the user's current location coordinates and added to the heterogeneous graph structure. The local subgraph is extracted from the heterogeneous graph structure as the central node. The optimized graph contrastive learning network is called to perform node embedding representation calculation on the local subgraph to generate a semantic relevance ranked list. Based on the connection relationship between the semantic relevance sorted list and the path nodes in the heterogeneous graph structure, multi-hop path reasoning is performed to generate a recommendation result containing the optimal path sequence from the user's current location to the target geographic object and recommended explanation information.

2. The cultural and tourism map search method based on deep learning and Beidou database according to claim 1 is characterized in that: The Beidou database is a navigation data set that records the user's real-time location and movement path, including latitude and longitude coordinates, timestamps, speed and direction information; the cultural and tourism geographic objects include cultural and tourism-related entities with geographic locations, label classifications, and rating records; the heterogeneous graph structure includes user nodes, geographic object nodes, and path nodes, and access edges, spatial adjacency edges, and trajectory sequence edges are established between the nodes.

3. The cultural and tourism map search method based on deep learning and Beidou database according to claim 1 is characterized in that: The initial features of the user node are written into the heterogeneous graph structure, including: receiving a natural language search request input by a user through a terminal, wherein the natural language search request includes geographic preference, behavioral intention, travel purpose, and topic keywords; Perform text preprocessing on natural language search requests, including word segmentation, stop word removal, word form unification, and invalid symbol removal, to generate standardized sentence input; Perform context semantic vector extraction processing on the preprocessed natural language input text input to generate a context-dependent semantic vector representation; The extracted semantic vector representation is dimensionally normalized and feature aligned to meet the dimensional consistency and structural constraint requirements of node feature representation in heterogeneous graph structures; Bind the semantic vector representation with the user identity to generate a complete user node feature representation data structure; The generated user node features are written into the corresponding user node positions in the heterogeneous graph structure, completing the semantic embedding mapping process from natural language search requests to the graph structure.

4. The cultural and tourism map search method based on deep learning and Beidou database according to claim 1 is characterized in that: The method of constructing a graph contrastive learning network based on a heterogeneous graph structure includes: Constructing the original graph based on heterogeneous graph structure ,in, Represents a node collection, including user nodes, geographic object nodes, and path nodes. Represents the edge set, including visited edges, spatial adjacency edges, and trajectory order edges; Introducing geographic location similarity weighting coefficient , constructing a location-aware similarity function , used to fuse the spatial relationships between nodes in the contrastive loss; Introducing sub-graph-aware scale control factor , used to determine the adjacency depth of the local subgraph extracted from the central node, in the first view and the second view In the construction of The neighborhood of is used as the subgraph extraction range, and node embedding extraction is performed at the subgraph level; Introducing node shading ratio adjustment factor , used to control the masking ratio of node features during the view perturbation process, when constructing the first view and the second view When, according to the ratio Mask the node feature dimensions in the original feature matrix; The first view and the second view Input the graph neural network encoder to obtain the embedded representation of each node in the two sets of views 、 ; Constructing a location-aware contrastive loss function , used to optimize the consistency of embedding representation of the same node in two views; The graph neural network is trained based on the location-aware contrastive loss function, and finally a graph contrastive learning network representation structure is obtained that integrates geographic awareness, adjustable granularity and dynamic perturbation strategy.

5. The cultural and tourism map search method based on deep learning and Beidou database according to claim 4 is characterized in that: The forming of the optimized graph contrast learning network includes: Set the initial population size of the pigeon swarm optimization algorithm to , initialize the pigeon flock individual set , each individual Represents the structural parameter combination of the graph contrast learning network, where , represents the weighted coefficient of geographical location similarity corresponding to the i-th individual, is the subgraph perception scale control factor corresponding to the i-th individual, is the node shielding ratio adjustment factor corresponding to the i-th individual; Randomly initialize the velocity vector for each individual , and set the individual memory vector , used to record individual historical optimal parameter configuration; For each individual parameter combination Based on the constructed graph contrastive learning network structure, node embedding calculation and contrastive learning training process are performed to obtain the corresponding contrast loss function value. Node sorting recommendation accuracy , weighted fitness function ; All individuals are sorted from high to low according to their fitness values, and divided into sub-populations, and the individual with the highest fitness value in each sub-population is selected as the local pilot , the remaining individuals perform local guidance updates around the local pilot body; Calculate the current iteration round Dynamic inertia factor ; According to the dynamic inertia factor Calculate the candidate values ​​of individual geomagnetic navigation positions respectively Candidate values ​​for landmark-guided locations : Geomagnetic navigation update: ； Landmark guidance update: ； in, is the position vector of the i-th individual at the t-th iteration, is the velocity vector of individual i in round t, is the bootstrap weight coefficient, is the individual memory weight coefficient, is the position of the individual with the best fitness in the current population, is the historical optimal memory position of the i-th individual; Introducing stage control factors ,dynamically balance the two candidate positions through the fusion mechanism, perform position update, and update the individual speed of the next round according to the position update results of this round; Calculate the new round of individual positions Corresponding contrastive loss of graph contrast learning network and recommendation accuracy , update the fitness function ; If the current individual Fitness , then update the memory vector to the current solution ; Repeat until the maximum number of iterations is met Or fitness convergence threshold condition, and finally output the individual with the best fitness , as the optimal structural parameter combination of graph contrastive learning network.

6. The cultural and tourism map search method based on deep learning and Beidou database according to claim 1 is characterized in that: Generating a semantic relevance ranked list includes: Receive map search requests initiated by users on the terminal and trigger the initialization of the location-aware recommendation process; Extract the latitude and longitude coordinates of the user's current location from the Beidou database, and use the latitude and longitude coordinates of the user's current location as the location node to construct a graph node entity; The location node is used as the user's current location node and added to the currently maintained heterogeneous graph structure to expand the spatial representation capability of the graph structure; Taking the user's current location node as the central node, the associated local subgraph structure is extracted from the heterogeneous graph structure based on spatial adjacency and access relations; The optimized graph contrast learning network is called to perform embedding calculations on the extracted local subgraphs to generate semantic representations for each candidate node. Based on the semantic relevance between the user's current location node and the embedded representation of the candidate node, a ranked result list is generated and output as a candidate result set for search recommendation.

7. The cultural and tourism map search method based on deep learning and Beidou database according to claim 1 is characterized in that: The user feedback behavior data specifically includes click records, browsing dwell time, historical search paths and actual navigation behavior, which are used to optimize the heterogeneous graph structure and the retraining process of the graph comparative learning network.

8. A cultural and tourism map search system based on deep learning and Beidou database, which executes the cultural and tourism map search method based on deep learning and Beidou database according to any one of claims 1 to 7, characterized in that: Includes the following modules: A request processing module, configured to receive a user map search request and extract a semantic representation as a user node feature; The mapping module is used to extract user location and trajectory data from the Beidou database and build a heterogeneous graph structure based on geographic object information; The graph learning optimization module is used to generate perturbation views based on heterogeneous graphs, train graph contrastive learning networks, and adjust structural parameters using the pigeon flock optimization algorithm; Subgraph calculation module, used to extract local subgraphs and generate node embedding representations; The recommendation module is used to generate ranking results based on the semantic relevance of the embedded vectors and output recommended paths and explanation information; The feedback update module is used to collect user feedback behavior data and use it for graph structure update and model retraining.

Images