Service work order identification and distribution method and system based on tourist spatiotemporal behavior graph

Abstract

The present invention discloses a method and system for identifying and distributing service work orders based on a tourist spatiotemporal behavior graph, which solves the problem that existing scenic spot service work orders contain multiple modal information such as voice, text, and pictures, and are difficult to accurately understand the user's true intention due to the ambiguity and diversity of language and the existence of scenic spot-specific terms. The method includes pre-constructing a tourist spatiotemporal behavior graph based on historical correlation data, and the spatiotemporal behavior graph performs semantic understanding on tourist demand data, and distributes personalized service work orders to the corresponding optimal service provider based on the degree of interactivity; the present invention pre-constructs a tourist spatiotemporal behavior graph containing an entity extraction model based on historical correlation data, which can more comprehensively capture the spatiotemporal behavior characteristics of tourists and service providers, so that the spatiotemporal behavior graph can accurately understand the tourist demand data, ensuring that personalized service work orders are accurately distributed to the optimal service provider with the highest correlation, thereby improving service efficiency and tourist satisfaction.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and in particular relates to a method and system for identifying and distributing service work orders based on a tourist's spatiotemporal behavior graph. Background Art

[0002] As people's living standards improve, more and more people choose to travel during holidays, leading to a booming tourism industry. Predicting tourism demand and providing services based on that demand is a key issue in the tourism industry. Accurately predicting trends and patterns can improve infrastructure and resource scheduling, providing high-quality services for tourists. Knowledge graph technology is widely used in various fields. By building knowledge connections, it enables efficient management and deep mining of complex information. The tourist spatiotemporal behavior graph is a knowledge graph constructed based on the temporal and spatial behavior data of tourists during their travels. It integrates information such as their location, time, stopover points, and activity sequence to present their behavioral patterns and preferences in a graph structure.

[0003] A scenic spot service work order is a work record used in scenic spot management to record, process, and track visitor needs, issues, or requests. It plays a vital role in the daily operations of a scenic spot and is a key tool for providing efficient service and resolving problems.

[0004] Currently, traditional scenic spot service ticket processing relies on manual dispatch, which is inefficient and prone to errors. However, combining tourist spatiotemporal behavior graphs with knowledge graph technology can enable automated classification and intelligent distribution of service tickets. This intelligent processing approach not only reduces manual intervention, but scenic spot service tickets may contain multimodal information such as voice, text, and images. Accurately understanding the user's true intent can be difficult due to the ambiguity and diversity of language and the presence of scenic spot-specific terminology. Effectively integrating this multimodal information and accurately understanding its semantics is a challenge. To address these issues, we propose a service ticket identification and distribution method and system based on tourist spatiotemporal behavior graphs. Summary of the Invention

[0005] The purpose of the present invention is to address the shortcomings of the existing technology and provide a service ticket identification and distribution method and system based on the spatiotemporal behavior graph of tourists. This solves the problem that existing scenic spot service tickets may contain multiple modal information such as voice, text, and pictures, and it is difficult to accurately understand the user's true intention due to the ambiguity and diversity of language and the existence of scenic spot-specific terminology.

[0006] The present invention is implemented as follows: a service ticket identification and distribution method based on a tourist spatiotemporal behavior graph, the method comprising:

[0007] Crawling historical correlation data related to tourists' spatiotemporal behavior, and pre-building a tourist spatiotemporal behavior graph based on the historical correlation data;

[0008] Real-time collection of multi-source heterogeneous tourist demand data. The spatiotemporal behavior graph uses advanced statistical functions to perform semantic understanding of tourist demand data, triggering service ticket generation instructions. In response to the generation instructions, personalized service tickets are generated.

[0009] Load personalized service tickets. Based on the pre-trained entity extraction model, extract ticket priority, user tags, and ticket information features and perform weighted fusion to obtain a fused feature set. The entity extraction model traverses the spatiotemporal behavior graph to obtain the user-server interaction matrix, user-association interaction matrix, and service-association interaction matrix based on the fused feature set. The interaction matrices are weighted and combined to obtain the in-degree interactivity matrix.

[0010] The in-degree interactivity matrix is ​​input into the entity extraction model. Based on the ant colony optimization algorithm and fuzzy clustering analysis, the in-degree interactivity matrix corresponds to the interaction degree between personalized service tickets and service providers. Based on the interaction degree, the personalized service tickets are distributed to the corresponding optimal service providers.

[0011] Preferably, the method of pre-constructing a spatiotemporal behavior graph by combining historical correlation data includes:

[0012] The pattern layer of the graph is defined based on the four dimensions of user spatiotemporal behavior, service provider spatiotemporal behavior, personnel interaction behavior, and personnel prediction behavior. These four dimensions serve as the pattern layer ontology to guide the construction of the graph data layer.

[0013] Use the Scrapy crawler framework to crawl multi-source heterogeneous historical related data, extract historical related data entities, entity attributes, and entity relationships, and pre-process the historical related data. The historical related data includes user location data, time data, behavior data, external data, user name, gender, age, education level, home address, historical demand data, and destination information.

[0014] Pre-build an entity extraction model, use historical association data to iteratively train the entity extraction model, and output a converged entity extraction model;

[0015] Load the pre-processed historical association data, extract the entity feature attributes of the historical association data based on the entity extraction model, and call the CEA integration algorithm to assign weights to the entity feature attributes from the perspectives of user spatiotemporal behavior, service provider spatiotemporal behavior, and human interaction behavior;

[0016] Associating the entity feature values ​​of historical association data with the attribute weight values ​​to obtain the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix of the entity feature attributes;

[0017] Calculate the predicted behavior influence weights of entity feature attributes in the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the human interaction behavior weight matrix based on the Tanimoto coefficient;

[0018] Preset a prediction weight threshold, extract entity feature values ​​and attribute weight values ​​of the spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix based on the prediction weight threshold, and obtain a personnel prediction behavior matrix based on the user spatiotemporal behavior, the service provider spatiotemporal behavior, and the personnel interaction behavior;

[0019] Load the user spatiotemporal behavior weight matrix, service provider spatiotemporal behavior weight matrix, personnel interaction behavior weight matrix, and personnel prediction behavior matrix. Use the LOAD CSV command to convert the data in the matrix into triples in the knowledge graph data layer. Store the data in the form of a graph and visualize it using Neo4j to complete the construction of the spatiotemporal behavior graph.

[0020] Preferably, the entity extraction model takes the BP neural network model as the basic model, and introduces the Transformer architecture after the input layer of the basic model. The Transformer architecture consists of an encoder and a decoder. The FreeLB adversarial training algorithm is introduced in the encoder, and the spatiotemporal federated learning framework is used to replace the hidden layer of the basic model. Advanced statistical functions are introduced in the spatiotemporal federated learning framework. A CRF annotation sequence layer embedded with a Softmax layer is added between the spatiotemporal federated learning framework and the output layer. The sigmoid activation function is used to activate the output of the Transformer architecture. The entity extraction model is iteratively trained based on the FreeLB adversarial training algorithm combined with historical association data, and the hyperparameters of the entity extraction model are adjusted through the Adma optimizer to minimize the cross-entropy loss function of the entity extraction model.

[0021] Preferably, the predicted behavior impact weight is calculated using the following formula:

[0022]

[0023] Among them, q d Represents the predicted behavior influence weight of entity feature attributes, W s a ,W s b ,W s c are the entity feature values ​​of entity feature attribute s in user spatiotemporal behavior weight matrix A, service provider spatiotemporal behavior weight matrix B, and personnel interaction behavior weight matrix C, respectively. a ,q b ,q c are the attribute weight values ​​of entity feature attributes s, Sim TTanimoto coefficients representing the user spatiotemporal behavior weight matrix, the server spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix, φ a,b,c represents the embedding vector of entity feature attribute s in user spatiotemporal behavior weight matrix A, service provider spatiotemporal behavior weight matrix B, and personnel interaction behavior weight matrix C. T represents the number of entity features, |KU i +KU j | represents the interaction value between the user and the server, |A i +B i +C i |,|A j +B j +C j | represents the number of entity features of interacting user i and server j in the user spatiotemporal behavior weight matrix A, the server spatiotemporal behavior weight matrix B, and the personnel interaction behavior weight matrix C, respectively. |A∩B∩C| represents the matrix intersection, and |A∪B∪C| represents the matrix union.

[0024] Preferably, the spatiotemporal behavior graph performs semantic understanding of tourist demand data based on advanced statistical functions, including:

[0025] Loading tourist demand data, preprocessing the tourist demand data, identifying the preprocessed tourist demand data, and determining the data type;

[0026] If the data type is non-text data, the entity extraction model input layer extracts the non-text data signal features and denoises the signal features based on inverse fast Fourier transform;

[0027] If the data type is text data, the encoder of the Transformer architecture of the entity extraction model serializes and encodes the features of the text data to obtain a text encoding set;

[0028] After loading the signal reconstruction features and text encoding set, the spatiotemporal federated learning framework combines advanced statistical functions to perform statistical fusion on the signal reconstruction features and text encoding set to obtain a semantic output set. The CRF annotation sequence layer then annotates the semantic output set with user labels, determines priorities, and outputs personalized semantic information.

[0029] Using personalized semantic information as an index, the spatiotemporal behavior graph is traversed based on retrieval enhancement generation technology. The spatiotemporal behavior graph outputs the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix corresponding to the personalized semantic information based on the response to the personalized semantic information.

[0030] Load the user spatiotemporal behavior weight matrix, service provider spatiotemporal behavior weight matrix, and personnel interaction behavior weight matrix corresponding to the personalized semantic information, and generate the personnel prediction behavior matrix corresponding to the personalized semantic information from the spatiotemporal behavior graph;

[0031] Extract the triple entities and entity relationships corresponding to the personnel prediction behavior matrix in the knowledge graph data layer, load the triple entities and entity relationships in the knowledge graph data layer into the preset work order template, and generate personalized service work orders.

[0032] Preferably, when performing denoising on the signal features:

[0033] Load the original signal and identify the original amplitude and phase corresponding to the original signal frequency, and construct a screening window with adjustable length and positive and negative directions;

[0034] The screening window identifies the regular amplitude, discrete amplitude, and high-energy amplitude in the original amplitude based on the cluster spectrum algorithm, and suppresses the discrete amplitude;

[0035] The cluster spectrum algorithm is used to calculate the Euclidean distance between the high-energy amplitude and the conventional amplitude, and to determine whether the Euclidean distance between the cluster centers of the high-energy amplitude and the conventional amplitude exceeds the conventional amplitude threshold interval;

[0036] If the Euclidean distance between the high-energy amplitude and the cluster center of the conventional amplitude exceeds the conventional amplitude threshold interval, the high-energy amplitude is suppressed; if the Euclidean distance between the high-energy amplitude and the cluster center of the conventional amplitude does not exceed the conventional amplitude threshold interval, the current high-energy amplitude is retained;

[0037] The original signal frequency after signal feature denoising is loaded, and the original signal is reconstructed based on the inverse fast Fourier transform to obtain the signal reconstruction features of the non-text data.

[0038] Preferably, the method for determining the degree of interaction between personalized service tickets and service providers based on the in-degree interactivity matrix using the ant colony optimization algorithm and fuzzy clustering analysis includes:

[0039] Load the in-degree interactivity matrix and the associated personalized service work order, identify the predicted spatiotemporal point information in the personnel prediction behavior matrix based on the personalized service work order, and use the ant colony optimization algorithm to find the path from the current position point l1 to the predicted spatiotemporal point l n l paths, the ant colony optimization algorithm iteratively searches the l paths to find the optimal path l best ;

[0040] Based on the optimal path best Extract service providers within the spatiotemporal behavior, use the Pearson correlation coefficient to calculate the initial interaction between personalized service tickets and service providers, and generate an initial interaction sequence;

[0041] The fuzzy clustering analysis algorithm uses Minkowski distance to characterize the weighted interaction between personalized service tickets and service providers, and obtains a weighted interaction degree sequence.

[0042] The initial interaction degree sequence between personalized service work orders and service providers and the weighted interaction degree sequence between personalized service work orders and service providers are loaded, and the time axis and spatial axis of the time series objects personalized service work orders and service providers are locally compressed. The weighted interaction degree sequence is optimally mapped to the initial interaction degree sequence to obtain the interaction degree between personalized service work orders and service providers. The service providers are sorted based on the interaction degree between personalized service work orders and service providers to obtain the ranking position of each service provider. The ant colony optimization algorithm performs the initial weighted update of pheromones based on the ranking of the service providers to obtain the optimal service provider.

[0043] The interaction degree between the personalized service ticket and the service provider is calculated using the following formula:

[0044]

[0045] Among them, J best , J0, They represent interaction degree, initial interaction degree, and weighted interaction degree respectively, and δ is the optimal path l best The spatiotemporal expectation of the server in , G, G represents the in-degree interactivity matrix and the mean of the in-degree interactivity matrix, k is the number of servers in the spatiotemporal behavior, AG i , AG j They represent the weight matrix of the in-degree interactivity matrix corresponding to user i and server j, They represent the in-degree interaction values ​​between user i and server j in the in-degree interactivity matrix.

[0046] On the other hand, the present invention also provides a service ticket identification and distribution system based on a tourist spatiotemporal behavior graph, the service ticket identification and distribution system based on a tourist spatiotemporal behavior graph comprising:

[0047] The knowledge graph module is used to crawl historical data related to tourists' spatiotemporal behavior and pre-build a tourist spatiotemporal behavior graph based on the historical data;

[0048] The demand identification module is used to collect multi-source heterogeneous tourist demand data in real time. The spatiotemporal behavior graph uses advanced statistical functions to perform semantic understanding of tourist demand data, trigger service ticket generation instructions, and generate personalized service tickets in response to the generation instructions.

[0049] The work order indexing module loads personalized service work orders and extracts work order priority, user tags, and work order information features based on a pre-trained entity extraction model. These features are weighted and fused to obtain a fused feature set. The entity extraction model traverses the spatiotemporal behavior graph to obtain a user-service interaction matrix, a user-associate interaction matrix, and a service-associate interaction matrix based on the fused feature set. The interaction matrices are weighted and combined to obtain an in-degree interactivity matrix.

[0050] The work order distribution module is used to input the in-degree interactivity matrix into the entity extraction model, and based on the ant colony optimization algorithm and fuzzy clustering analysis, the in-degree interactivity matrix corresponds to the interaction degree between personalized service work orders and service providers, and distributes the personalized service work orders to the corresponding optimal service provider based on the interaction degree.

[0051] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0052] In an embodiment of the present invention, a tourist spatiotemporal behavior graph containing an entity extraction model is pre-constructed based on historical correlation data. The entity extraction model introduces an entity extraction model based on a BP neural network and a Transformer architecture, and combines it with the FreeLB adversarial training algorithm to significantly improve the model's processing capability and robustness for complex data. Through the spatiotemporal federated learning framework and advanced statistical functions, the model's ability to model spatiotemporal data is further enhanced, enabling a more comprehensive capture of the spatiotemporal behavior characteristics of tourists and service providers, ensuring the accuracy and efficiency of feature extraction. This enables the spatiotemporal behavior graph to accurately understand tourist demand data and generate personalized semantic information and a predicted behavior matrix based on tourist demand data. This dynamically evaluates the matching degree between service providers and work orders, ensuring that personalized service orders are accurately distributed to the most optimal service providers with the highest degree of relevance, thereby improving service efficiency and tourist satisfaction. This overcomes the problem that existing scenic spot service orders may contain multiple modal information such as voice, text, and images, making it difficult to accurately understand the user's true intentions due to the ambiguity and diversity of language and the presence of scenic spot-specific terminology.

[0053] In an embodiment of the present invention, the entity extraction model significantly improves the accuracy, robustness, and generalization of entity extraction by combining the nonlinear modeling capabilities of the BP neural network and the long-distance dependency capture capabilities of the Transformer architecture. At the same time, the Transformer architecture can effectively capture long-distance dependencies through the self-attention mechanism, making it particularly suitable for processing long sequence data. Its parallel computing capability significantly improves training and inference efficiency. The introduction of the FreeLB adversarial training algorithm can enhance the robustness of the model, enabling it to more accurately identify semantic entities when faced with modal information with strong ambiguity and diversity. Adversarial training helps the model learn more stable feature representations by simulating adversarial samples, thereby improving generalization capabilities. The spatiotemporal federated learning framework replaces the hidden layer of the BP neural network, which can better handle the complexity of spatiotemporal data. This framework further improves the model's modeling capabilities for spatiotemporal features by introducing advanced statistical functions. Adding a CRF annotation sequence layer with an embedded Softmax layer between the spatiotemporal federated learning framework and the output layer can effectively handle the annotation dependency problem in sequence annotation tasks. The Softmax layer is used to activate the output of the Transformer architecture, further optimizing the model's output results.

[0054] The present invention provides a method for semantically understanding visitor demand data using spatiotemporal behavior graphs based on advanced statistical functions. This method achieves precise semantic understanding of visitor demand data by preprocessing the data, denoising signal features using an inverse fast Fourier transform, integrating statistics using a spatiotemporal federated learning framework, and generating a predictive behavior matrix corresponding to personalized semantic information from the spatiotemporal behavior graph. This method not only processes multimodal data but also dynamically adjusts service content and generates personalized service tickets, significantly improving the quality and efficiency of scenic spot services.

[0055] In this embodiment of the present invention, the processed signal features are reconstructed using an inverse fast Fourier transform (iFFT), fully restoring the signal's time-domain characteristics. This method not only removes noise but also preserves the signal's original structure and key features, ensuring the usability of the reconstructed signal. The clustering spectrum algorithm and dynamic threshold adjustment mechanism make the entire signal reconstruction process highly flexible and scalable. It can be adjusted to suit different application scenarios and signal characteristics, adapting to a variety of complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the implementation process of the service work order identification and distribution method based on the tourist spatiotemporal behavior graph provided by the present invention.

[0057] Figure 2 A schematic diagram of the implementation process of the method for pre-constructing a spatiotemporal behavior graph by combining historical correlation data is shown.

[0058] Figure 3 The figure shows a flowchart of the implementation of a method for semantically understanding tourist demand data based on advanced statistical functions using a spatiotemporal behavior graph.

[0059] Figure 4 The schematic diagram of the implementation process of the method of corresponding personalized service work order and service provider interaction method based on ant colony optimization algorithm and fuzzy cluster analysis in-degree interactivity matrix is ​​shown.

[0060] Figure 5 The schematic diagram shows the structure of the service ticket identification and distribution system based on the tourist spatiotemporal behavior graph. DETAILED DESCRIPTION

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0062] Existing scenic spot service work orders may contain multiple modal information such as voice, text, and pictures. Due to the ambiguity and diversity of language and the existence of scenic spot-specific terminology, it is difficult to accurately understand the user's true intention. To address the above problems, we propose a service work order identification and distribution method and system based on the tourist spatiotemporal behavior graph. In short, when implementing the method, the tourist spatiotemporal behavior graph is first pre-constructed in combination with historical correlation data. The spatiotemporal behavior graph performs semantic understanding of tourist demand data based on advanced statistical functions to generate personalized service work orders. The work order priority, user label, and work order information features are extracted based on the pre-trained entity extraction model and weighted fused to obtain a fused feature set. Based on the fused feature set, the user-server interaction matrix, user-associated interaction matrix, and service-associated interaction matrix are obtained. The interaction matrices are weightedly combined to obtain the in-degree interactivity matrix. Based on the ant colony optimization algorithm and fuzzy clustering analysis, the in-degree interactivity matrix corresponds to the personalized service work order and the interaction degree between the service provider. Based on the interaction degree, the personalized service work order is distributed to the corresponding optimal service provider. In an embodiment of the present invention, a tourist spatiotemporal behavior graph containing an entity extraction model is pre-constructed based on historical correlation data. The entity extraction model introduces an entity extraction model based on a BP neural network and a Transformer architecture, and combines it with the FreeLB adversarial training algorithm to significantly improve the model's processing capability and robustness for complex data. Through the spatiotemporal federated learning framework and advanced statistical functions, the model's ability to model spatiotemporal data is further enhanced, enabling a more comprehensive capture of the spatiotemporal behavior characteristics of tourists and service providers, ensuring the accuracy and efficiency of feature extraction. This enables the spatiotemporal behavior graph to accurately understand tourist demand data and generate personalized semantic information and a predicted behavior matrix based on tourist demand data. This dynamically evaluates the matching degree between service providers and work orders, ensuring that personalized service orders are accurately distributed to the most optimal service providers with the highest degree of relevance, thereby improving service efficiency and tourist satisfaction. This overcomes the problem that existing scenic spot service orders may contain multiple modal information such as voice, text, and images, making it difficult to accurately understand the user's true intentions due to the ambiguity and diversity of language and the presence of scenic spot-specific terminology.

[0063] The embodiment of the present invention provides a service ticket identification and distribution method based on the tourist spatiotemporal behavior graph. Figure 1 The following is a schematic diagram of the implementation process of a service ticket identification and distribution method based on a tourist spatiotemporal behavior graph. The service ticket identification and distribution method based on a tourist spatiotemporal behavior graph specifically includes:

[0064] Step S10: crawling historical correlation data associated with tourists' spatiotemporal behavior, and pre-building a tourist spatiotemporal behavior graph based on the historical correlation data;

[0065] In this embodiment, by crawling heterogeneous historical data from multiple sources, we can comprehensively collect rich information related to tourists' spatiotemporal behavior, providing a solid data foundation for graph construction. Compared with traditional methods that rely on limited data sources or manual data compilation, this method can more comprehensively reflect tourist behavior patterns and preferences.

[0066] Step S20: Real-time collection of multi-source heterogeneous tourist demand data. The spatiotemporal behavior graph performs semantic understanding of the tourist demand data based on advanced statistical functions, triggers a service ticket generation instruction, and generates a personalized service ticket in response to the generation instruction.

[0067] It should be noted that tourist demand data includes but is not limited to travel decision-making behavior data, tourist activity trajectories in scenic areas, tourist tour plans, tour time, willingness to visit scenic spots, whether to participate in scenic area activities, whether to communicate with staff, tourist consumption behavior data during the tour, online booking data, tourist name, contact information, entry time, location, specific description of the problems or needs encountered by tourists, work orders submitted when tourists need additional services, work orders submitted when tourists are dissatisfied with the services, facilities or management of the scenic area, and work orders submitted when tourists encounter emergencies (such as injury, getting lost, bad weather, etc.).

[0068] By collecting multi-source, heterogeneous tourist demand data in real time, including voice, text, images, and other modal information, and fusing them together, the spatiotemporal behavior graph fully leverages the advantages of multimodal data. Compared to single-modal data processing, it can more comprehensively capture tourist demand information and improve the accuracy of work order generation. By combining the spatiotemporal behavior graph with advanced statistical functions to perform semantic understanding of tourist demand data, it can more accurately understand tourist intent and accurately identify even in the face of ambiguity and diversity in language. This solves the problem of incorrect or incomplete work order generation caused by inaccurate semantic understanding in traditional methods, thereby improving the quality and relevance of work orders.

[0069] In step S30, the personalized service ticket is loaded. Based on the pre-trained entity extraction model, the ticket priority, user tags, and ticket information features are extracted and weighted and fused to obtain a fused feature set. This deep feature extraction and fusion method can more comprehensively characterize the characteristics of the ticket compared to simple feature extraction methods, providing a more accurate basis for subsequent ticket distribution. The entity extraction model traverses the spatiotemporal behavior graph to obtain a user-server interaction matrix, a user-relationship interaction matrix, and a service-relationship interaction matrix based on the fused feature set. Weighted combination of these interaction matrices yields an in-degree interactivity matrix. The entity extraction model traverses the spatiotemporal behavior graph to obtain various interaction matrices. These matrices reflect the complex relationships between users and service providers, users and relationships, and services and relationships. By weighted combination of these interaction matrices, an in-degree interactivity matrix is ​​obtained, which further quantifies these relationships and provides an important basis for subsequent service ticket distribution. For example, on a travel service platform, the interaction matrix between users and hotel staff can reflect user satisfaction with hotel services and the frequency of requests; the interaction matrix between users and travel agency staff can reflect user preferences and feedback on travel products. The in-degree interactivity matrix integrates this information to help the system distribute service tickets to the most appropriate service providers more accurately.

[0070] It should be noted that the user-server interaction matrix represents the relationship between the current user and the server. The rows of this matrix represent users, and the columns represent servers. Each element in the matrix represents the interaction strength or degree of association between the user and the server. For example, the values ​​in the matrix could be the user's rating of the server, the number of services provided, or other quantitative indicators. The user-associate interaction matrix represents the relationship between the current user and other users in the same spatiotemporal context, while the service-associate interaction matrix represents the relationship between other users and the server in the same spatiotemporal context. Through weighted combination, the relationship between users and servers, users and other users, and server and other users can be comprehensively considered, thereby more comprehensively reflecting the interactive relationships between different entities.

[0071] Step S40: input the in-degree interactivity matrix into the entity extraction model, analyze the interaction degree between the in-degree interactivity matrix and the personalized service ticket and the service provider based on the ant colony optimization algorithm and fuzzy clustering, and distribute the personalized service ticket to the corresponding optimal service provider based on the interaction degree.

[0072] In this embodiment, an ant colony optimization algorithm and fuzzy clustering analysis of the in-degree interactivity matrix intelligently assess the degree of interaction between personalized service tickets and service providers, and accordingly distribute the tickets to the optimal service provider. Compared to traditional manual distribution or simple rule-based distribution methods, this intelligent optimization method can more efficiently allocate resources and improve service efficiency and quality.

[0073] In an embodiment of the present invention, a tourist spatiotemporal behavior graph containing an entity extraction model is pre-constructed based on historical correlation data. The entity extraction model introduces an entity extraction model based on a BP neural network and a Transformer architecture, and combines it with the FreeLB adversarial training algorithm to significantly improve the model's processing capability and robustness for complex data. Through the spatiotemporal federated learning framework and advanced statistical functions, the model's ability to model spatiotemporal data is further enhanced, enabling a more comprehensive capture of the spatiotemporal behavior characteristics of tourists and service providers, ensuring the accuracy and efficiency of feature extraction. This enables the spatiotemporal behavior graph to accurately understand tourist demand data and generate personalized semantic information and a predicted behavior matrix based on tourist demand data. This dynamically evaluates the matching degree between service providers and work orders, ensuring that personalized service orders are accurately distributed to the most optimal service providers with the highest degree of relevance, thereby improving service efficiency and tourist satisfaction. This overcomes the problem that existing scenic spot service orders may contain multiple modal information such as voice, text, and images, making it difficult to accurately understand the user's true intentions due to the ambiguity and diversity of language and the presence of scenic spot-specific terminology.

[0074] The embodiment of the present invention provides a method for pre-constructing a spatiotemporal behavior graph based on historical correlation data. In short, the spatiotemporal behavior graph is constructed through multi-dimensional data collection and pre-processing, entity extraction and feature weighting, weight matrix construction and predictive behavior analysis, combined with graph database storage and visualization, to construct a tourist spatiotemporal behavior graph. Figure 2 A schematic diagram of the implementation process of a method for pre-constructing a spatiotemporal behavior graph based on historical correlation data is shown. The method for pre-constructing a spatiotemporal behavior graph based on historical correlation data specifically includes:

[0075] Step S101: Define the graph's pattern layer based on four dimensions: user spatiotemporal behavior, service provider spatiotemporal behavior, personnel interaction behavior, and personnel prediction behavior. Use these four dimensions as the pattern layer ontology to guide the construction of the graph data layer.

[0076] In this embodiment, the model layer of the graph is defined from multiple key dimensions, encompassing the behaviors of users and service providers, as well as interactions and predicted behaviors between people. This provides a comprehensive framework for constructing the entire spatiotemporal behavior graph. This multidimensional consideration can more systematically reflect the interrelationships between various factors in real-world scenarios, making the graph more comprehensive and practical.

[0077] Step S102: Using the Scrapy crawler framework to crawl multi-source heterogeneous historical related data, extracting historical related data entities, entity attributes, and entity relationships, and pre-processing the historical related data. The historical related data includes user location data, time data, behavior data, external data, user name, gender, age, education level, home address, historical demand data, and destination information.

[0078] It should be noted that the Scrapy crawler framework is used to collect multi-source heterogeneous historical correlation data, covering user location, time, behavior, external data, and other aspects of information, providing a rich data foundation for graph construction. The preprocessing step ensures data quality, reduces noise and redundancy, and improves subsequent processing efficiency.

[0079] Step S103: pre-build an entity extraction model, iteratively train the entity extraction model using historical association data, and output a converged entity extraction model;

[0080] In an embodiment of the present invention, the entity extraction model can not only extract entities from historical association data, but also perform entity recognition on tourist demand data. It can solve the problem that existing graphs are difficult to accurately understand semantic entities due to the presence of ambiguous multi-modal information and scenic spot-specific terms, and the existing graph recognition has the problem of fuzzy entity boundaries and weak generalization ability. The entity extraction model uses the BP neural network model as the basic model, introduces the Transformer architecture after the input layer of the basic model, and the Transformer architecture consists of an encoder and a decoder. The FreeLB adversarial training algorithm is introduced in the encoder, and the hidden layer of the basic model is replaced by the spatiotemporal federated learning framework. Advanced statistical functions are introduced in the spatiotemporal federated learning framework. A CRF annotation sequence layer embedded with a Softmax layer is added between the spatiotemporal federated learning framework and the output layer. The sigmoid activation function is used to activate the output of the Transformer architecture. The entity extraction model is iteratively trained based on the FreeLB adversarial training algorithm combined with historical association data. The hyperparameters of the entity extraction model are adjusted by the Adma optimizer to minimize the cross-entropy loss function of the entity extraction model. The performance of the model is regularly evaluated during the training process, and the accuracy and generalization ability of the model are checked using a validation set. Adjust the training strategy and hyperparameters based on the evaluation results. After training, use an independent test set to conduct a final test on the model to ensure its effectiveness and reliability in real-world application scenarios.

[0081] In an embodiment of the present invention, the entity extraction model significantly improves the accuracy, robustness, and generalization of entity extraction by combining the nonlinear modeling capabilities of the BP neural network and the long-distance dependency capture capabilities of the Transformer architecture. At the same time, the Transformer architecture can effectively capture long-distance dependencies through the self-attention mechanism, making it particularly suitable for processing long sequence data. Its parallel computing capability significantly improves training and inference efficiency. The introduction of the FreeLB adversarial training algorithm can enhance the robustness of the model, enabling it to more accurately identify semantic entities when faced with modal information with strong ambiguity and diversity. Adversarial training helps the model learn more stable feature representations by simulating adversarial samples, thereby improving generalization capabilities. The spatiotemporal federated learning framework replaces the hidden layer of the BP neural network, which can better handle the complexity of spatiotemporal data. This framework further improves the model's modeling capabilities for spatiotemporal features by introducing advanced statistical functions. Adding a CRF annotation sequence layer with an embedded Softmax layer between the spatiotemporal federated learning framework and the output layer can effectively handle the annotation dependency problem in sequence annotation tasks. The Softmax layer is used to activate the output of the Transformer architecture, further optimizing the model's output results.

[0082] Step S104: Load the pre-processed historical association data, extract entity feature attributes of the historical association data based on the entity extraction model, and call the CEA integration algorithm to assign weights to the entity feature attributes from the perspectives of user spatiotemporal behavior, service provider spatiotemporal behavior, and human interaction behavior;

[0083] In this embodiment of the present invention, entity feature attributes are extracted based on an entity extraction model, enabling rapid location and extraction of key information related to spatiotemporal behavior from massive amounts of data. The CEA integrated algorithm weights entity feature attributes from different perspectives, further clarifying the importance of each feature, helping to highlight key factors and providing a more targeted basis for subsequent analysis and decision-making.

[0084] Step S105: Associating the entity feature values ​​of the historical association data with the attribute weight values ​​to obtain the user spatiotemporal behavior weight matrix, the server spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix of the entity feature attributes;

[0085] By associating entity feature values ​​with attribute weights, an intuitive weight matrix is ​​formed, enabling a quantitative representation of the complex relationships between user spatiotemporal behavior, service provider spatiotemporal behavior, and human interaction. This quantification facilitates computer processing and analysis, facilitating subsequent calculations and predictions. For example, in constructing a spatiotemporal behavior map for a tourism service platform, establishing a user spatiotemporal behavior weight matrix allows for a clear visualization of the travel behavior preferences of different user groups over different time periods and regions, providing data support for the design and promotion of tourism products.

[0086] Step S106: Calculate the predicted behavior influence weights of entity feature attributes in the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the human interaction behavior weight matrix based on the Tanimoto coefficient. Using the Tanimoto coefficient to calculate the predicted behavior influence weights can scientifically assess the degree of influence of each entity feature attribute on the predicted behavior. This method considers the similarities and differences between features, making the calculated weights more reasonable and accurate, helping to improve the reliability and effectiveness of the prediction.

[0087] In this embodiment of the present invention, the predicted behavior impact weight is calculated using the following formula:

[0088]

[0089] Among them, q d Represents the predicted behavior influence weight of entity feature attributes, W s a ,W s b ,W s c are the entity feature values ​​of entity feature attribute s in user spatiotemporal behavior weight matrix A, service provider spatiotemporal behavior weight matrix B, and personnel interaction behavior weight matrix C, respectively. a ,q b ,q c are the attribute weight values ​​of entity feature attributes s, Sim T Tanimoto coefficients representing the user spatiotemporal behavior weight matrix, the server spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix, φ a,b,c represents the embedding vector of entity feature attribute s in user spatiotemporal behavior weight matrix A, service provider spatiotemporal behavior weight matrix B, and personnel interaction behavior weight matrix C. T represents the number of entity features, |KU i +KU j | represents the interaction value between the user and the server, |A i +B i +C i |,|A j +B j +C j | represents the number of entity features of interacting user i and server j in the user spatiotemporal behavior weight matrix A, the server spatiotemporal behavior weight matrix B, and the personnel interaction behavior weight matrix C, respectively. |A∩B∩C| represents the matrix intersection, and |A∪B∪C| represents the matrix union.

[0090] In this embodiment of the present invention, a calculation method and formula for predicting the impact weight of behavior are defined. By combining a weight matrix of user spatiotemporal behavior, service provider spatiotemporal behavior, and human interaction behavior, this formula can comprehensively consider information from multiple dimensions. This multi-dimensional integration makes predictions more comprehensive and accurate. Using embedding vectors and Tanimoto coefficients to calculate weights, the weights of different entity attributes can be dynamically adjusted based on their importance. This improves the model's adaptability and flexibility to different scenarios.

[0091] Step S107: Preset a prediction weight threshold, which can be set to 0.15-0.3. Based on the prediction weight threshold, extract entity feature values ​​and attribute weight values ​​of the spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix to obtain a personnel prediction behavior matrix based on the user spatiotemporal behavior, the service provider spatiotemporal behavior, and the personnel interaction behavior.

[0092] Step S108, load the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, the personnel interaction behavior weight matrix, and the personnel prediction behavior matrix, use the LOAD CSV command to convert the data in the matrix into a triple form of the knowledge graph data layer, store the data in the form of a graph and use Neo4j to visualize it, and complete the construction of the spatiotemporal behavior graph. For example, in a city's tourism planning, by constructing a spatiotemporal behavior graph and displaying it in a visual way, city planners and tourism companies can intuitively understand the behavior patterns of tourists, the distribution of tourism resources, and the trends of market demand, so as to formulate more scientific and reasonable tourism development plans and marketing strategies.

[0093] In this embodiment of the present invention, by presetting prediction weight thresholds, we can filter out features and weights that significantly influence predicted behavior, remove noise and redundant information, and improve prediction accuracy. The resulting predicted behavior matrix directly reflects the combined impact of different behavioral factors on a person's future behavior, providing strong support for personalized recommendations and service optimization.

[0094] In this embodiment of the present invention, by integrating four dimensions—user spatiotemporal behavior, service provider spatiotemporal behavior, human interaction behavior, and human predicted behavior—a comprehensive depiction of tourist behavior patterns and preferences is achieved. The spatiotemporal behavior graph provides rich contextual information, helping the entity extraction model better understand ambiguous and diverse modal information and improve the accuracy of entity recognition. By leveraging the spatiotemporal correlation information in the graph, the model can more clearly define entity boundaries and reduce misidentifications. Combined with this graph information, the entity extraction model can more accurately identify tourist needs and generate service tickets that better meet their actual needs.

[0095] The embodiment of the present invention provides a method for semantic understanding of tourist demand data based on advanced statistical functions using spatiotemporal behavior graphs. This method achieves accurate semantic understanding of tourist demand data by pre-processing tourist demand data, combining inverse fast Fourier transform to denoise signal features, integrating statistics within a spatiotemporal federated learning framework, and generating a personnel prediction behavior matrix corresponding to personalized semantic information using spatiotemporal behavior graphs. This method can not only process multimodal data, but also dynamically adjust service content and generate personalized service work orders, significantly improving the quality and efficiency of scenic area services. Figure 3 The following is a schematic diagram showing the implementation process of a method for semantically understanding tourist demand data using a spatiotemporal behavior graph based on advanced statistical functions. The method specifically includes:

[0096] Step S201, loading tourist demand data, preprocessing the tourist demand data, and identifying the preprocessed tourist demand data;

[0097] In this embodiment of the present invention, preprocessing of visitor demand data includes, but is not limited to, outlier cleaning and deletion, removing noise and redundant information from the data and improving data quality. By identifying preprocessed data, a clear and clean data foundation is provided for subsequent semantic understanding, ensuring the accuracy of subsequent processing.

[0098] Step S202: Determine the data type. The data type is determined based on whether it is text data, so that appropriate processing methods can be selected in a targeted manner, avoiding a one-size-fits-all approach. This classification process can better adapt to the characteristics of different modal data and improve the efficiency and accuracy of semantic understanding.

[0099] Step S203: If the data type is non-text data, the entity extraction model input layer extracts non-text data signal features and denoises the signal features using an inverse fast Fourier transform (iFFT). For non-text data (such as speech and images), extracting signal features and performing denoising using an inverse fast Fourier transform (iFFT) can effectively remove noise interference and retain key information. This step significantly improves the usability of non-text data and provides high-quality input for subsequent semantic understanding.

[0100] Step S204: If the data type is text data, the encoder of the Transformer architecture of the entity extraction model serializes and encodes the features of the text data to obtain a text encoding set;

[0101] Step S205: Load the signal reconstruction features and text encoding set. The spatiotemporal federated learning framework combines the signal reconstruction features and text encoding set with advanced statistical functions to obtain a semantic output set. The CRF annotation sequence layer annotates the semantic output set with user labels, determines the priority, and outputs personalized semantic information.

[0102] In this embodiment of the present invention, the spatiotemporal federated learning framework combines advanced statistical functions to perform fusion statistics on signal reconstruction features and text encoding sets, enabling comprehensive processing of multimodal data to produce a more comprehensive set of semantic outputs. The CRF annotation sequence layer further annotates and prioritizes the semantic output sets with user tags, providing clear semantic information for personalized services.

[0103] Step S206: Using the personalized semantic information as an index, the spatiotemporal behavior graph is traversed based on the retrieval enhancement generation technology. The spatiotemporal behavior graph outputs the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix corresponding to the personalized semantic information in response to the personalized semantic information.

[0104] Step S207: Load the user spatiotemporal behavior weight matrix, the server spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix corresponding to the personalized semantic information, and generate a personnel prediction behavior matrix corresponding to the personalized semantic information from the spatiotemporal behavior graph;

[0105] Step S208: extract the triple entities and entity relationships corresponding to the personnel prediction behavior matrix in the knowledge graph data layer, load the triple entities and entity relationships in the knowledge graph data layer into the preset work order template, and generate a personalized service work order.

[0106] In this embodiment of the present invention, triple entities and relationships are extracted from the knowledge graph data layer and loaded into a preset work order template to generate personalized service work orders. This method not only improves the automation level of work order generation, but also ensures the accuracy and personalization of work order content, thereby improving visitor satisfaction and scenic spot management efficiency.

[0107] In this embodiment, the signal feature denoising process is as follows: the original signal is loaded and an adjustable screening window is constructed, a cluster spectrum algorithm is used to identify and suppress discrete amplitudes, and then the Euclidean distance between the high-energy amplitude and the normal amplitude is calculated. A threshold is used to determine whether to suppress the high-energy amplitude, and an inverse fast Fourier transform (iFFT) is performed on the processed signal features to reconstruct the signal. Specifically, it includes:

[0108] Step S2031, loading the original signal and identifying the original amplitude and phase corresponding to the original signal frequency, and constructing a screening window with adjustable length and positive and negative directions;

[0109] Step S2032: The screening window identifies the regular amplitude, discrete amplitude, and high-energy amplitude in the original amplitude based on the cluster spectrum algorithm, and suppresses the discrete amplitude;

[0110] By using a clustered spectrum algorithm to identify discrete and high-energy amplitudes, it is possible to accurately distinguish between noise components and useful information in a signal. Compared to traditional fixed threshold methods, this method can more flexibly adapt to the characteristics of different signals and avoid false suppression of useful information.

[0111] Step S2033: Calculate the Euclidean distance between the high-energy amplitude and the normal amplitude using a cluster spectrum algorithm, and determine whether the Euclidean distance between the cluster centers of the high-energy amplitude and the normal amplitude exceeds a normal amplitude threshold interval;

[0112] By calculating the Euclidean distance between the high-energy amplitude and the normal amplitude and comparing it with a dynamic threshold range, the embodiment of the present invention can dynamically adjust the suppression strategy based on the actual characteristics of the signal. This method avoids the problem of over-suppression or under-suppression caused by fixed thresholds.

[0113] Step S2034: If the Euclidean distance between the high-energy amplitude and the center of the regular amplitude cluster exceeds the regular amplitude threshold interval, the high-energy amplitude is suppressed; if the Euclidean distance between the high-energy amplitude and the center of the regular amplitude cluster does not exceed the regular amplitude threshold interval, the current high-energy amplitude is retained;

[0114] Step S2035 , loading the original signal frequency after the signal feature denoising process, reconstructing the original signal based on the inverse fast Fourier transform, and obtaining the signal reconstruction feature of the non-text data.

[0115] In this embodiment of the present invention, the processed signal features are reconstructed using an inverse fast Fourier transform (iFFT), fully restoring the signal's time-domain characteristics. This method not only removes noise but also preserves the signal's original structure and key features, ensuring the usability of the reconstructed signal. The clustering spectrum algorithm and dynamic threshold adjustment mechanism make the entire signal reconstruction process highly flexible and scalable. It can be adjusted to suit different application scenarios and signal characteristics, adapting to a variety of complex environments.

[0116] The embodiment of the present invention provides a method for determining the interaction degree between personalized service work orders and service providers based on an ant colony optimization algorithm and a fuzzy clustering analysis in-degree interactivity matrix. In short, the implementation method is to find the optimal path through the ant colony optimization algorithm, extract the service provider information on the path, calculate the initial interaction degree between the personalized service work order and the service provider, use fuzzy clustering analysis to calculate the weighted interaction degree, map the weighted interaction degree sequence to the initial interaction degree sequence to obtain the final interaction degree, and rank the service providers. According to the ranking, the pheromone weight is updated to determine the optimal service provider. Figure 4 The following is a schematic diagram of the implementation process of a method for determining the degree of interaction between personalized service work orders and service providers based on an ant colony optimization algorithm and a fuzzy clustering analysis in-degree interactivity matrix. The method specifically includes:

[0117] Step S301: load the in-degree interactivity matrix and the associated personalized service work order, identify the predicted spatiotemporal point information in the personnel prediction behavior matrix based on the personalized service work order, and use the ant colony optimization algorithm to find the space-time point from the current position point l1 to the predicted spatiotemporal point l n l paths, the ant colony optimization algorithm iteratively searches the l paths to find the optimal path l best ;

[0118] In this embodiment of the present invention, the ant colony optimization algorithm, by simulating the food-seeking behavior of ants, can efficiently find the optimal path in complex environments. In this embodiment, the algorithm is used to search for a path from the current location to a predicted spatiotemporal point. It can dynamically adapt to environmental changes and update the path in real time. This approach not only improves the efficiency of path search but also ensures the optimality of the path, providing a scientific basis for subsequent service provider selection.

[0119] Step S302: Based on the optimal path l best Extract service providers within the spatiotemporal behavior, use the Pearson correlation coefficient to calculate the initial interaction between personalized service tickets and service providers, and generate an initial interaction sequence;

[0120] By calculating the initial engagement between personalized service tickets and service providers using the Pearson correlation coefficient, we can quantify the correlation between the two. This method provides basic data for subsequent engagement optimization and ensures the scientific and accurate calculation of engagement.

[0121] In step S303, the fuzzy cluster analysis algorithm uses the Minkowski distance to characterize the weighted interaction between the personalized service ticket and the service provider, obtaining a weighted interaction degree sequence. In this embodiment, calculating the weighted interaction degree using the Minkowski distance can more accurately reflect the interaction intensity between the personalized service ticket and the service provider. This method not only improves the accuracy of the interaction degree calculation but also enhances the robustness of the system.

[0122] Step S304: load the initial interaction degree sequence between personalized service work orders and service providers, and the weighted interaction degree sequence between personalized service work orders and service providers, perform local compression on the time axis and space axis of the time series objects personalized service work orders and service providers, optimally map the weighted interaction degree sequence to the initial interaction degree sequence, obtain the interaction degree between personalized service work orders and service providers, sort the service providers based on the interaction degree between personalized service work orders and service providers, obtain the ranking position of each service provider, and use the ant colony optimization algorithm to perform the initial weighted update of pheromones based on the ranking of the service providers to obtain the optimal service provider.

[0123] In this embodiment, the interaction degree between the personalized service ticket and the service provider is calculated using the following formula:

[0124]

[0125] Among them, J best , J0, They represent interaction degree, initial interaction degree, and weighted interaction degree respectively, and δ is the optimal path l best The spatiotemporal expectation of the server in this embodiment can be 0.2-0.8, G, represents the in-degree interactivity matrix and the mean of the in-degree interactivity matrix, k is the number of service providers in the spatiotemporal behavior, AG i , AG j They represent the weight matrix of the in-degree interactivity matrix corresponding to user i and server j, They represent the in-degree interaction values ​​between user i and server j in the in-degree interactivity matrix, which can be 0.2-5.

[0126] In this embodiment of the present invention, by optimally mapping the weighted interaction degree sequence to the initial interaction degree sequence, a more reasonable interaction degree between personalized service tickets and service providers can be obtained by comprehensively considering the initial and weighted interaction degrees. Service providers are ranked based on this interaction degree, and pheromone weights are updated using an ant colony optimization algorithm. This allows for dynamic adjustment of service provider priorities to ensure the selection of the optimal service provider.

[0127] The embodiment of the present invention provides a service ticket identification and distribution system based on the tourist spatiotemporal behavior graph. Figure 5 The schematic diagram of the structure of the service ticket identification and distribution system based on the tourist spatiotemporal behavior graph is shown. The service ticket identification and distribution system based on the tourist spatiotemporal behavior graph specifically includes:

[0128] The knowledge graph module 100 is used to crawl historical correlation data related to tourists' spatiotemporal behavior and pre-build a tourist spatiotemporal behavior graph based on the historical correlation data;

[0129] Demand identification module 200 is used to collect multi-source heterogeneous tourist demand data in real time, perform semantic understanding of tourist demand data based on advanced statistical functions using spatiotemporal behavior graphs, trigger service ticket generation instructions, and generate personalized service tickets in response to the generation instructions;

[0130] The work order indexing module 300 loads personalized service work orders and extracts work order priority, user tags, and work order information features based on a pre-trained entity extraction model. These features are weighted and fused to obtain a fused feature set. The entity extraction model traverses the spatiotemporal behavior graph to obtain a user-server interaction matrix, a user-associate interaction matrix, and a service-associate interaction matrix based on the fused feature set. The interaction matrices are weighted and combined to obtain an in-degree interactivity matrix.

[0131] The work order distribution module 400 is used to input the in-degree interactivity matrix into the entity extraction model, analyze the interaction degree between the in-degree interactivity matrix and the personalized service work order and the service provider based on the ant colony optimization algorithm and fuzzy clustering, and distribute the personalized service work order to the corresponding optimal service provider based on the interaction degree.

[0132] In this embodiment, the knowledge graph module 100, the demand identification module 200, the work order index module 300, the work order distribution module 400 and the service work order identification and distribution system based on the tourist spatiotemporal behavior graph correspond to the above-mentioned service work order identification and distribution method based on the tourist spatiotemporal behavior graph. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the service work order identification and distribution method based on the tourist spatiotemporal behavior graph, and will not be repeated here.

[0133] In summary, the present invention provides a method and system for identifying and distributing service tickets based on a tourist spatiotemporal behavior graph. In an embodiment of the present invention, a tourist spatiotemporal behavior graph containing an entity extraction model is pre-constructed based on historical correlation data. The entity extraction model introduces an entity extraction model based on a BP neural network and a Transformer architecture. Combined with the FreeLB adversarial training algorithm, this significantly improves the model's ability to process complex data and its robustness. Through a spatiotemporal federated learning framework and advanced statistical functions, the model's ability to model spatiotemporal data is further enhanced, enabling a more comprehensive capture of the spatiotemporal behavior characteristics of tourists and service providers, ensuring the accuracy and efficiency of feature extraction. This enables the spatiotemporal behavior graph to accurately understand tourist demand data and generate personalized semantic information and a predicted behavior matrix based on the tourist demand data. This allows for dynamic evaluation of the matching degree between service providers and ticket requests, ensuring that personalized service tickets are accurately distributed to the optimal service provider with the highest correlation, thereby improving service efficiency and tourist satisfaction. This overcomes the problem that existing scenic spot service tickets, which may contain multiple modal information such as voice, text, and images, can be difficult to accurately understand the user's true intent due to the ambiguity and diversity of language and the presence of scenic spot-specific terminology.

[0134] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A service ticket identification and distribution method based on tourist spatiotemporal behavior graph, characterized in that: include: Crawling historical correlation data related to tourists' spatiotemporal behavior, and pre-building a tourist spatiotemporal behavior graph based on the historical correlation data; Real-time collection of multi-source heterogeneous tourist demand data. The spatiotemporal behavior graph uses advanced statistical functions to perform semantic understanding of tourist demand data, triggering service ticket generation instructions. In response to the generation instructions, personalized service tickets are generated. Load personalized service tickets. Based on the pre-trained entity extraction model, extract ticket priority, user tags, and ticket information features and perform weighted fusion to obtain a fused feature set. The entity extraction model traverses the spatiotemporal behavior graph to obtain the user-server interaction matrix, user-association interaction matrix, and service-association interaction matrix based on the fused feature set. The interaction matrices are weighted and combined to obtain the in-degree interactivity matrix. The in-degree interactivity matrix is ​​input into the entity extraction model. Based on the ant colony optimization algorithm and fuzzy clustering analysis, the in-degree interactivity matrix corresponds to the interaction degree between personalized service tickets and service providers. Based on the interaction degree, the personalized service tickets are distributed to the corresponding optimal service providers. The method for pre-constructing a spatiotemporal behavior graph by combining historical correlation data includes: The pattern layer of the graph is defined based on the four dimensions of user spatiotemporal behavior, service provider spatiotemporal behavior, personnel interaction behavior, and personnel prediction behavior. These four dimensions serve as the pattern layer ontology to guide the construction of the graph data layer. Use the Scrapy crawler framework to crawl multi-source heterogeneous historical related data, extract historical related data entities, entity attributes, and entity relationships, and pre-process the historical related data. The historical related data includes user location data, time data, behavior data, external data, user name, gender, age, education level, home address, historical demand data, and destination information. Pre-build an entity extraction model, use historical association data to iteratively train the entity extraction model, and output a converged entity extraction model; Load the pre-processed historical association data, extract the entity feature attributes of the historical association data based on the entity extraction model, and call the CEA integration algorithm to assign weights to the entity feature attributes from the perspectives of user spatiotemporal behavior, service provider spatiotemporal behavior, and human interaction behavior; Associating the entity feature values ​​of historical association data with the attribute weight values ​​to obtain the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix of the entity feature attributes; Calculate the predicted behavior influence weights of entity feature attributes in the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the human interaction behavior weight matrix based on the Tanimoto coefficient; Preset a prediction weight threshold, extract entity feature values ​​and attribute weight values ​​of the spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix based on the prediction weight threshold, and obtain a personnel prediction behavior matrix based on the user spatiotemporal behavior, the service provider spatiotemporal behavior, and the personnel interaction behavior; Load the user spatiotemporal behavior weight matrix, service provider spatiotemporal behavior weight matrix, personnel interaction behavior weight matrix, and personnel prediction behavior matrix. Use the LOAD CSV command to convert the data in the matrix into triples in the knowledge graph data layer. Store the data in the form of a graph and visualize it using Neo4j to complete the construction of the spatiotemporal behavior graph. The predicted behavior impact weight is calculated using the following formula: （1） （2） （3） （4） in, Represents the predicted behavior influence weight of entity feature attributes, Entity feature attributes In the user spatiotemporal behavior weight matrix , server spatiotemporal behavior weight matrix , Personnel Interaction Behavior Weight Matrix The entity eigenvalues ​​in , Entity feature attributes The attribute weight value of The Tanimoto coefficients representing the user spatiotemporal behavior weight matrix, the server spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix, Represents entity feature attributes In the user spatiotemporal behavior weight matrix , server spatiotemporal behavior weight matrix , Personnel Interaction Behavior Weight Matrix The embedding vector in , Indicates the number of entity features, Indicates the interaction value between users and service providers. , Represent the user's spatiotemporal behavior weight matrix , server spatiotemporal behavior weight matrix , Personnel Interaction Behavior Weight Matrix Interactive users 、Service Provider The number of entity features, represents matrix intersection, Represents a matrix union.

2. The service ticket identification and distribution method based on the tourist spatiotemporal behavior graph according to claim 1 is characterized by: The entity extraction model is based on the BP neural network model. A Transformer architecture is introduced after the input layer of the basic model. The Transformer architecture consists of an encoder and a decoder. The FreeLB adversarial training algorithm is introduced in the encoder. The spatiotemporal federated learning framework is used to replace the hidden layer of the basic model. Advanced statistical functions are introduced in the spatiotemporal federated learning framework. A CRF annotation sequence layer embedded with a Softmax layer is added between the spatiotemporal federated learning framework and the output layer. The sigmoid activation function is used to activate the output of the Transformer architecture. The entity extraction model is iteratively trained based on the FreeLB adversarial training algorithm combined with historical association data. The hyperparameters of the entity extraction model are adjusted through the Adma optimizer to minimize the cross-entropy loss function of the entity extraction model.

3. The service ticket identification and distribution method based on the tourist spatiotemporal behavior graph according to claim 2 is characterized by: The spatiotemporal behavior graph performs semantic understanding of tourist demand data based on advanced statistical functions, including: Loading tourist demand data, preprocessing the tourist demand data, identifying the preprocessed tourist demand data, and determining the data type; If the data type is non-text data, the entity extraction model input layer extracts the non-text data signal features and denoises the signal features based on inverse fast Fourier transform; If the data type is text data, the encoder of the Transformer architecture of the entity extraction model serializes and encodes the features of the text data to obtain a text encoding set; After loading the signal reconstruction features and text encoding set, the spatiotemporal federated learning framework combines advanced statistical functions to perform statistical fusion on the signal reconstruction features and text encoding set to obtain a semantic output set. The CRF annotation sequence layer then annotates the semantic output set with user labels, determines priorities, and outputs personalized semantic information. Using personalized semantic information as an index, the spatiotemporal behavior graph is traversed based on retrieval enhancement generation technology. The spatiotemporal behavior graph outputs the user spatiotemporal behavior weight matrix, the service provider spatiotemporal behavior weight matrix, and the personnel interaction behavior weight matrix corresponding to the personalized semantic information based on the response to the personalized semantic information. Load the user spatiotemporal behavior weight matrix, service provider spatiotemporal behavior weight matrix, and personnel interaction behavior weight matrix corresponding to the personalized semantic information, and generate the personnel prediction behavior matrix corresponding to the personalized semantic information from the spatiotemporal behavior graph; Extract the triple entities and entity relationships corresponding to the personnel prediction behavior matrix in the knowledge graph data layer, load the triple entities and entity relationships in the knowledge graph data layer into the preset work order template, and generate personalized service work orders.

4. The service ticket identification and distribution method based on the tourist spatiotemporal behavior graph according to claim 3 is characterized by: When denoising signal features: Load the original signal and identify the original amplitude and phase corresponding to the original signal frequency, and construct a screening window with adjustable length and positive and negative directions; The screening window identifies the regular amplitude, discrete amplitude, and high-energy amplitude in the original amplitude based on the cluster spectrum algorithm, and suppresses the discrete amplitude; The cluster spectrum algorithm is used to calculate the Euclidean distance between the high-energy amplitude and the conventional amplitude, and to determine whether the Euclidean distance between the cluster centers of the high-energy amplitude and the conventional amplitude exceeds the conventional amplitude threshold interval; If the Euclidean distance between the high-energy amplitude and the cluster center of the conventional amplitude exceeds the conventional amplitude threshold interval, the high-energy amplitude is suppressed; if the Euclidean distance between the high-energy amplitude and the cluster center of the conventional amplitude does not exceed the conventional amplitude threshold interval, the current high-energy amplitude is retained; The original signal frequency after signal feature denoising is loaded, and the original signal is reconstructed based on the inverse fast Fourier transform to obtain the signal reconstruction features of the non-text data.

5. The service ticket identification and distribution method based on the tourist spatiotemporal behavior graph according to claim 3 is characterized by: The method for determining the degree of interaction between personalized service orders and service providers based on the ant colony optimization algorithm and fuzzy cluster analysis in-degree interactivity matrix includes: Load the in-degree interactivity matrix and the associated personalized service work order, identify the personnel prediction behavior matrix based on the personalized service work order, and use the ant colony optimization algorithm to find the time-space point information from the current position point. To the predicted time and space point of Paths, ant colony optimization algorithm Iterative search of paths to find the optimal path ; Based on the optimal path Extract service providers within the spatiotemporal behavior, use the Pearson correlation coefficient to calculate the initial interaction between personalized service tickets and service providers, and generate an initial interaction sequence; The fuzzy clustering analysis algorithm uses Minkowski distance to characterize the weighted interaction between personalized service tickets and service providers, and obtains a weighted interaction degree sequence. The initial interaction degree sequence between personalized service work orders and service providers and the weighted interaction degree sequence between personalized service work orders and service providers are loaded, and the time axis and spatial axis of the time series objects personalized service work orders and service providers are locally compressed. The weighted interaction degree sequence is optimally mapped to the initial interaction degree sequence to obtain the interaction degree between personalized service work orders and service providers. The service providers are sorted based on the interaction degree between personalized service work orders and service providers to obtain the ranking position of each service provider. The ant colony optimization algorithm performs the initial weighted update of pheromones based on the ranking of the service providers to obtain the optimal service provider.

6. The service ticket identification and distribution method based on the tourist spatiotemporal behavior graph according to claim 5 is characterized by: The interaction degree between the personalized service ticket and the service provider is calculated using the following formula: （5） （6） （7） in, Represents interaction degree, initial interaction degree, and weighted interaction degree respectively. The optimal path The time and space expectations of the service providers, represents the in-degree interactivity matrix and the mean of the in-degree interactivity matrix, is the number of service providers in the spatiotemporal behavior, They represent the in-degree interactivity matrix corresponding to the user With the server The weight matrix, Represent the users in the in-degree interactivity matrix With the server The in-degree interaction value of .

7. A service ticket identification and distribution system based on a tourist spatiotemporal behavior graph, for implementing the service ticket identification and distribution method based on a tourist spatiotemporal behavior graph according to any one of claims 1 to 6, characterized in that: The service ticket identification and distribution system based on the tourist spatiotemporal behavior graph includes: The knowledge graph module is used to crawl historical data related to tourists' spatiotemporal behavior and pre-build a tourist spatiotemporal behavior graph based on the historical data; The demand identification module is used to collect multi-source heterogeneous tourist demand data in real time. The spatiotemporal behavior graph uses advanced statistical functions to perform semantic understanding of tourist demand data, trigger service ticket generation instructions, and generate personalized service tickets in response to the generation instructions. The work order indexing module loads personalized service work orders and extracts work order priority, user tags, and work order information features based on a pre-trained entity extraction model. These features are weighted and fused to obtain a fused feature set. The entity extraction model traverses the spatiotemporal behavior graph to obtain a user-service interaction matrix, a user-associate interaction matrix, and a service-associate interaction matrix based on the fused feature set. The interaction matrices are weighted and combined to obtain an in-degree interactivity matrix. The work order distribution module is used to input the in-degree interactivity matrix into the entity extraction model, and based on the ant colony optimization algorithm and fuzzy clustering analysis, the in-degree interactivity matrix corresponds to the interaction degree between personalized service work orders and service providers, and distributes the personalized service work orders to the corresponding optimal service provider based on the interaction degree.

Images