Cross-border trade process optimization method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of cross-border trade process optimization, and discloses a cross-border trade process optimization method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining cargo information and customs clearance information of a to-be-cross-border trade; generating a plurality of preset dimensions based on the cargo information and the customs clearance information; collecting a dimension value corresponding to the preset dimension from a preset knowledge graph to form a state space; wherein the state space is a space corresponding to all preset dimensions; and performing strategy optimization on the state space through a preset learning algorithm to obtain an optimized customs clearance process. The method has the advantages that the cross-border trade process is more flexible and higher in adaptability, the overall operation efficiency can be improved, the customs clearance time can be shortened, an enterprise can optimize resource configuration, the operation cost is reduced, and finally the overall market response ability and customer satisfaction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cross-border trade process optimization, and particularly relates to a cross-border trade process optimization method and device, computer equipment and a storage medium. BACKGROUND

[0002] In recent years, although information technology and automation means have been continuously introduced into the industry, most systems still cannot efficiently integrate multi-source data and achieve dynamic and intelligent decision support. Currently, a large number of complex factors such as policies and regulations, cargo information and path planning are involved in cross-border trade, and traditional systems are difficult to cope with this complexity. At the same time, the existing process decomposition method often lacks flexibility and cannot adapt to policy changes and real-time market demand. SUMMARY

[0003] Therefore, it is necessary to propose a cross-border trade process optimization method for the existing cross-border trade process optimization problem.

[0004] A cross-border trade process optimization method, the method comprising:

[0005] obtaining cargo information and customs clearance information of goods to be traded across borders;

[0006] generating a plurality of preset dimensions based on the cargo information and the customs clearance information;

[0007] collecting dimension values corresponding to the preset dimensions in a preset knowledge graph to form a state space; wherein the state space is a space corresponding to all preset dimensions;

[0008] optimizing the state space through a preset learning algorithm to obtain an optimized customs clearance process.

[0009] Further, before the step of collecting dimension values corresponding to the preset dimensions in a preset knowledge graph to form a state space, the method further comprises:

[0010] obtaining historical customs clearance data in cross-border trade;

[0011] dividing the historical customs clearance data into structured data and unstructured data according to data types;

[0012] using a dual-flow transformer architecture, using a pattern perception encoder to extract field-level semantic features for the structured data, and using a pre-trained language model to perform entity recognition for the unstructured data to obtain each target entity;

[0013] aligning logistics track coordinates and target entities through a cross-modal attention mechanism based on the field-level semantic features to establish a space-semantic association matrix;

[0014] Import each of the target entities and the space-semantic correlation matrix into a graph database with preset values to obtain the preset knowledge graph.

[0015] Further, the step of obtaining historical clearance data in cross-border trade further comprises:

[0016] Detecting the clearance language of the historical clearance data;

[0017] Obtaining a corresponding translation model based on the clearance language;

[0018] Translating the historical clearance data based on the translation model to obtain historical clearance data in a specified language.

[0019] Further, the step of importing each of the target entities and the space-semantic correlation matrix into a graph database with preset values to obtain the preset knowledge graph further comprises:

[0020] Real-time monitoring of the historical clearance data for changes by a preset crawler;

[0021] If changes have occurred, obtaining a first target text after the changes and a second target text before the changes;

[0022] Calculating the similarity between the first target text and the second target text;

[0023] Determining whether the similarity is greater than a threshold value;

[0024] If the similarity is greater than the threshold value, constructing a GNN network to simulate the conduction effect, obtaining a predicted entity affected by the prediction, and updating the space-semantic correlation matrix in the graph database to obtain an updated knowledge graph.

[0025] Further, the step of generating a plurality of preset dimensions based on the cargo information and the clearance information comprises:

[0026] Using a preset multi-modal artificial intelligence model to process the cargo information and the clearance information to generate a flow tree structure;

[0027] Using a preset deep learning model to identify nodes in the flow tree structure to obtain a plurality of preset dimensions.

[0028] Further, the step of optimizing the state space by a preset learning algorithm to obtain an optimized clearance process comprises:

[0029] Obtaining a related target model based on the clearance information;

[0030] The state space is input into the target model, and a preset value function is set in the target model, so as to obtain an optimized customs clearance process.

[0031] Further, before the step of obtaining a relevant target simulator based on the customs clearance information, the method further comprises:

[0032] obtaining a customs clearance category of various customs clearance information;

[0033] generating a category data set of each customs clearance category using a simulator; wherein the category data set comprises customs clearance category data of multiple same customs clearance categories;

[0034] inputting each category data set into a preset model for training to obtain a target model corresponding to each customs clearance category.

[0035] A cross-border trade process optimization device, the device comprising:

[0036] an acquisition module configured to acquire goods information and customs clearance information of a cross-border trade;

[0037] a generation module configured to generate multiple preset dimensions based on the goods information and the customs clearance information;

[0038] a collection module configured to collect dimension values corresponding to the preset dimensions in a preset knowledge graph to form a state space; wherein the state space is a space corresponding to all preset dimensions;

[0039] an optimization module configured to perform policy optimization on the state space through a preset learning algorithm to obtain an optimized customs clearance process.

[0040] A computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to cause the processor to perform the following steps:

[0041] acquiring goods information and customs clearance information of a cross-border trade;

[0042] generating multiple preset dimensions based on the goods information and the customs clearance information;

[0043] collecting dimension values corresponding to the preset dimensions in a preset knowledge graph to form a state space; wherein the state space is a space corresponding to all preset dimensions;

[0044] performing policy optimization on the state space through a preset learning algorithm to obtain an optimized customs clearance process.

[0045] A computer readable storage medium stores a computer program, the computer program is executed by a processor, so that the processor executes the following steps:

[0046] Obtain goods information and customs clearance information to be traded across borders;

[0047] Generate a plurality of preset dimensions based on the goods information and the customs clearance information;

[0048] Collect dimension values corresponding to the preset dimensions in a preset knowledge graph to form a state space; wherein the state space is a space corresponding to all preset dimensions;

[0049] Optimize the state space through a preset learning algorithm to obtain an optimized customs clearance process.

[0050] The beneficial effects of the present application are: not only make the cross-border trade process more flexible, more adaptable, but also improve the overall operation efficiency, shorten the customs clearance time. In addition, the realization of dynamic decision support enhances the intelligence of the system, ensures that enterprises have a competitive advantage in the rapidly changing international trade environment. Through this innovative method, enterprises can optimize resource allocation, reduce operating costs, and ultimately improve overall market responsiveness and customer satisfaction. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0052] Among them:

[0053] Figure 1 It is an application environment diagram of the cross-border trade process optimization method in one embodiment;

[0054] Figure 2 It is a flowchart of the cross-border trade process optimization method in one embodiment;

[0055] Figure 3 It is a structure block diagram of the cross-border trade process optimization device in one embodiment;

[0056] Figure 4 It is a structure block diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0058] Figure 1 This is a diagram of an application environment for cross-border trade process optimization in an embodiment. Figure 1 This cross-border trade process optimization method is applied to a cross-border trade process optimization system. This cross-border trade process optimization system includes a terminal 110 and a server 120. Terminal 110 and server 120 are connected via a network. Terminal 110 can be a desktop terminal or a mobile terminal. The mobile terminal can be at least one of a mobile phone, tablet computer, and laptop computer. Server 120 can be implemented as a standalone server or a server cluster consisting of multiple servers. Terminal 110 is used to collect data, and server 120 is used to optimize cross-border trade processes.

[0059] like Figure 2 As shown, in one embodiment, a method for optimizing cross-border trade processes is provided. This method can be applied to both a terminal and a server. This embodiment uses the server as an example for illustration. The cross-border trade process optimization method specifically includes the following steps:

[0060] S1: Obtain information on goods to be traded across borders and customs clearance information;

[0061] S2: Generate multiple preset dimensions based on the cargo information and the customs clearance information;

[0062] S3: collecting dimension values ​​corresponding to the preset dimensions from a preset knowledge graph to form a state space; wherein the state space is the space corresponding to all preset dimensions;

[0063] S4: Optimizing the state space strategy through a preset learning algorithm to obtain an optimized customs clearance process.

[0064] As described in step S1, obtain the goods information and customs clearance information of the cross-border trade. The smooth progress of cross-border trade first requires accurate and comprehensive goods information and customs clearance information. Relevant data needs to be collected from multiple sources, including but not limited to the basic characteristics of the goods (such as the name of the goods, HS code, quantity, value, etc.), and the customs clearance requirements and policies of each country (such as customs standards, required documents, inspection process, etc.). Information can be obtained through interface integration, data crawling, enterprise system (such as ERP, WMS, etc.) static data or real-time data transmission, etc. For example, enterprises can obtain the latest customs and regulations through API interface connected with customs system, and use the built-in information stored in the database and logistics system to automatically aggregate all data, ensuring the real-time and accuracy of the data. The collected information can lay the foundation for the entire optimization process and help subsequent decision-making and analysis work.

[0065] As described in step S2, generate a plurality of preset dimensions based on the goods information and the customs clearance information. After the collection of information is completed, a plurality of preset dimensions are generated based on the goods information and the customs clearance information. The preset dimensions are the classification and feature extraction of the information, which usually include commodity classification, destination country, transportation mode, declaration status, declaration progress, inspection probability, etc. Complex and raw data are converted into multi-dimensional structures that can be quantitatively analyzed, facilitating subsequent processing and optimization work. For example, when analyzing a certain type of commodity, the dimensions may include the impact of different countries' regulations on the commodity, the time cost and economic comparison of different transportation modes, and the different customs clearance documents required in each case. In actual operation, machine learning or data mining techniques can be used to automatically identify and extract useful features, and data mining tools and statistical analysis can be used to establish a dimension model suitable for business needs. Through accurate preset dimension generation, subsequent process analysis and decision-making will become more efficient and accurate. It should be noted that the preset dimensions here are the dimensions that need to be obtained, and the specific dimension information has not been obtained.

[0066] As described in step S3 above, the dimension values corresponding to the preset dimensions are collected in the preset knowledge graph to form a state space. After generating the preset dimensions, the corresponding dimension information is obtained and converted into a state space. In this step, the collected data of the preset dimensions needs to be mapped to form a complete state space. The state space is a collection of all possible states that the entire system can be in, which provides a foundation for the subsequent policy optimization. The knowledge graph is a knowledge network represented by a graph, which can associate different information. In this process, different dimension information will become nodes in the knowledge graph, and the relationship between nodes is represented by edges. For example, the HS code of a commodity can be associated with the customs clearance requirements of the destination country, and a more complex graph can be formed. In this state space, different states reflect various situations and scenarios that may exist under different conditions, and the system can effectively identify potential risks and opportunities through the state space. Through this method, different data can be integrated into a unified model, facilitating subsequent data acquisition.

[0067] As described in S4 above, the state space is optimized by a preset learning algorithm to obtain an optimized customs clearance process. After establishing the state space, the space is optimized by a preset learning algorithm to refine a more efficient customs clearance process. The preset learning algorithm can be reinforcement learning, supervised learning, etc., aiming to explore and utilize the state space, constantly learning and optimizing decisions. For example, the PPO (Proximal Policy Optimization) reinforcement learning algorithm can be used. PPO is a reinforcement learning algorithm proposed by OpenAI and widely used in deep reinforcement learning. It aims to optimize the policy while limiting the update amplitude to improve the stability of training. The core of this algorithm is to prevent the policy from updating too drastically through proximal optimization, making the training process more stable and reliable. The strategy of reinforcement learning can be applied to define a reward function to drive the model to preferentially select the optimal customs clearance path and method. Through training the model to simulate and adjust multiple times in the state space, the best operation method for each link can be found. The optimized customs clearance process will be reflected in the optimal strategy set found, which can not only improve the speed of customs clearance, but also improve compliance and reduce costs. In this process, accurate state representation, real-time data feedback, and learning algorithm cooperation are crucial. Through continuous learning and iteration, the process is optimized, enabling enterprises to more efficiently cope with complex international environments and policy changes in cross-border trade.

[0068] In one embodiment, before the step S3 of collecting the dimension values corresponding to the preset dimensions in the preset knowledge graph to form a state space, it further includes:

[0069] S201: Obtain historical clearance data in cross-border trade;

[0070] S202: Divide the historical clearance data into structured data and unstructured data according to data types;

[0071] S203: Use a dual-flow transformer architecture to extract field-level semantic features using a schema-aware encoder for structured data, and perform entity recognition using a pre-trained language model for unstructured data to obtain each target entity;

[0072] S204: Align the logistics track coordinates and target entities based on the field-level semantic features through a cross-modal attention mechanism to establish a space-semantic association matrix;

[0073] S205: Import each target entity and the space-semantic association matrix into a graph database with a preset value to obtain the preset knowledge graph.

[0074] As described in step S201 above, historical clearance data related to cross-border trade is collected from multiple channels. These data may cover a wide range of content, including declaration documents, encountered customs policies, inspection and release status, and specific requirements of different countries' customs, etc. The source of historical clearance data can be obtained from the API interface of the customs system, the ERP system of the enterprise, the document management system or the online trade platform. During the information collection process, it is necessary to ensure the diversity and real-time nature of the data to avoid information lag or inaccuracy in subsequent steps. For example, real-time tracking of the latest customs policies and import regulations, identifying various clearance documents required. After obtaining the clearance data, the data needs to be preliminarily sorted to facilitate subsequent analysis and processing.

[0075] As described in step S202 above, after completing the acquisition of clearance data, the collected clearance data is classified and divided into structured data and unstructured data. This step is very important because different types of data require different processing techniques and models. Structured data generally refers to data with fixed formats, such as tables in databases, CSV files, etc., containing standardized information such as quantities, product types, HS codes, etc. This type of data is easy to automate and analyze, and can be processed using pattern-aware encoders. Unstructured data comes from text, documents, emails, etc., such as clearance application letters, contracts, email communications, PDF documents, etc. This part of data is often without specific format, and is more complex to process, it needs to use natural language processing technology to extract information. For example, using natural language processing (NLP) methods for text analysis and information extraction. After completing this classification, the data processing team can develop appropriate processing strategies to ensure that subsequent feature extraction and data analysis can proceed smoothly. By clearly defining and classifying data in the initial stage, it helps to improve the accuracy and efficiency of the subsequent process.

[0076] As described in step S203 above, a dual-stream transformer architecture is used to extract field-level semantic features using a pattern-aware encoder for structured data and to perform entity recognition using a pre-trained language model for unstructured data to obtain each target entity. The dual-stream transformer architecture is used to process structured and unstructured data, which consists of two independent processing paths for structured data and unstructured data. The purpose of this architecture is to take full advantage of the characteristics of each type of data to extract more informative features. For structured data, a pattern-aware encoder is used to extract field-level semantic features. This encoder can understand the relationship between fields in a data table and extract key information based on field patterns. For example, extract information such as the name, quantity and tax rate of the goods, and convert these features into high-dimensional vector representations for subsequent processing. For unstructured data, a pre-trained language model (such as BERT, RoBERTa, etc.) is used for entity recognition. This step can identify key entities in the text through NLP technology, such as extracting clearance-related terms, product names and shipping locations, etc., to convert scattered information into structured data. This provides rich content support for subsequent knowledge graph construction.

[0077] As described in step S204 above, based on the field-level semantic features, the logistics trajectory coordinates and the target entity are aligned through the cross-modal attention mechanism to establish a spatial-semantic association matrix. Effectively combine data from different sources (i.e., logistics trajectory and customs clearance targets) to improve the correlation and consistency between data. The spatial-semantic association matrix provides an abstract expression of spatial relationships for the data. Specifically, the information of the logistics trajectory coordinates provides information about the geographical location during the transportation of goods, while the target entity refers to the specific items and requirements related to the customs clearance process. Through the cross-modal attention mechanism, the relationship between the two types of information can be analyzed, such as determining the transportation route of a product within a specific time and its corresponding customs clearance requirements. For example, by comparing the coordinate points in the logistics trajectory with the position of the target entity, the system can identify the customs clearance operations that need to be performed at each transportation point and establish connections between the various tasks. The resulting spatial-semantic association matrix not only provides data support for subsequent decision-making, but also helps detect potential problems and risks, laying the foundation for the optimization of cross-border trade processes.

[0078] As described in step S205 above, each target entity and the spatial-semantic association matrix are imported into a preset graph database to obtain the preset knowledge graph. The extracted target entities and the established spatial-semantic association matrix are imported into the preset graph database to form a complete preset knowledge graph. A knowledge graph is an information repository represented by a graph structure, allowing information to be stored, queried, and analyzed in a more flexible and dynamic manner. By representing information as nodes and connections, a network is formed, revealing relationships and associations between data. When importing the target entities and association matrix into the graph database, each target entity is treated as a node, and the spatial-semantic association matrix is ​​used to define edges between nodes, representing the relationships and interactions between different nodes. Using a graph database (such as Neo4j) enables efficient relationship querying and data management, improving the efficiency and convenience of data access. By constructing a knowledge graph, relevant information related to cross-border trade is systematized, enabling rapid querying, reasoning, and decision-making, thereby providing strong support for subsequent process optimization. Ultimately, the constructed knowledge graph will lay the foundation for subsequent state-space-based decision optimization and strategy generation, improving the overall efficiency of the cross-border trade process.

[0079] In one embodiment, after step S201 of obtaining historical customs clearance data in cross-border trade, the method further includes:

[0080] S2021: Detecting the customs clearance language of the historical customs clearance data;

[0081] S2022: Obtaining a corresponding translation model based on the customs clearance language;

[0082] S2023: Translate the historical clearance data based on the translation model to obtain historical clearance data in a specified language.

[0083] As described in steps S2021-S2023 above, the language used in the clearance data is identified and determined, and the accurate detection of the language provides a basis for subsequent processing. Since cross-border trade involves different countries and regions, clearance data may be presented in multiple languages, so it is crucial to identify the language of the data. The process of detecting the language usually uses language recognition technology, which can be implemented through natural language processing (NLP) tools or libraries such as LangDetect, Google Cloud Natural Language API, etc. The system will analyze the input clearance text and use statistical features or pre-trained models to determine the specific language used in the text. According to language features such as vocabulary distribution, syntactic structure, etc., the system can quickly determine the language type of the text. This process not only improves the quality of subsequent translation, thereby avoiding inaccurate translation due to incorrect language recognition, but also provides input for the selection and setting of related parameters for the subsequent translation model. Accurate language recognition is a prerequisite for effective processing and utilization of historical clearance data. In one embodiment, if it is a voice input, the NVIDIA Riva framework can also be integrated, and an acoustic model can be customized for cross-border professional terms (such as INCOTERMS2020), supporting end-to-end recognition of Southeast Asian dialects, so that voice input can be converted into text, and then the text can be analyzed.

[0084] After detecting the language of the clearance data, the corresponding translation model is obtained based on the clearance language. Select the appropriate translation tool to translate according to the original language of the historical clearance data to ensure the effectiveness and accuracy of the translation. The acquisition of the translation model usually involves the following aspects: first, determine the translation model required for different languages, which may need to integrate multiple translation systems. For example, for foreign language clearance data, the translation models that may be used include Google Translate, Microsoft Translator, or open source translation frameworks such as OpenNMT, Fairseq, etc. Second, consider the scope of the translation model to ensure that the selected model can effectively handle specific industry terms and clearance terms, which usually requires Specialized Vocabulary Tuning. Finally, when introducing some deep learning-based translation models, the model can be trained and fine-tuned to enhance the model's performance in specific fields. Specifically, a multilingual BERT variant (XLM-RoBERTa) can be pre-built, and the semantic space of 107 languages of the tariff clauses can be aligned through contrastive learning, and then a T5-3B model can be used for zero-shot translation, combined with a domain memory bank to realize the context-aware generation of templates such as "Tax Refund Application-Spanish".

[0085] After determining the translation model, the final step is to translate the historical customs clearance data based on the translation model, generating the data in the specified language. It should be noted that the translation process selects the translation model corresponding to the original language of the customs clearance data and translates the text according to pre-set translation rules and parameters. The translation model processes the input customs clearance data sentence by sentence or paragraph by paragraph, generating the corresponding target language content. This process goes beyond simple word-by-word translation and takes into account context, grammatical structure, and specialized terminology to ensure that the translation meets customs clearance requirements in terms of both semantics and purpose.

[0086] Furthermore, a post-processing phase can be incorporated into the translation process to optimize the translation results. This can include correcting potential grammatical errors, improper punctuation, or further verifying industry terminology to ensure that the final output meets customs clearance requirements. This step, resulting in historical customs clearance data in a specific language, will facilitate subsequent decision-making, process optimization, and the efficient conduct of cross-border trade activities, enabling seamless data transmission and exchange, and providing support for a company's international operations.

[0087] In one embodiment, after step S204 of importing each of the target entities and the spatial-semantic association matrix into a preset graph database to obtain the preset knowledge graph, the method further includes:

[0088] S2051: Monitor in real time whether the historical customs clearance data has changed using a preset crawler;

[0089] S2052: If a change occurs, obtaining the first target text after the change and the second target text before the change;

[0090] S2053: Calculating the similarity between the first target text and the second target text;

[0091] S2054: Determine whether the similarity is greater than a threshold;

[0092] S2055: If the similarity is greater than a threshold, a GNN network is constructed to simulate the conduction effect, the affected predicted entities are obtained, and the spatial-semantic association matrix is ​​updated in the graph database to obtain an updated knowledge graph.

[0093] As described in steps S2051-S2055 above, real-time monitoring of historical clearance data is performed using a pre-set crawler technique to determine whether the data has changed. A web crawler is an automated software program that extracts information from specific web pages or data sources. In the clearance process, relevant data may frequently change due to policy changes, regulatory updates, or changes in market conditions, so a real-time monitoring mechanism is established for monitoring. The pre-set crawler usually accesses specified data sources such as customs official websites, trade-related databases, and various industry reports, etc. by obtaining HTML pages, CSV files or API interfaces to focus on changes in relevant data. The monitoring content may include changes in tariff policies, updates to laws and regulations, and requirements for clearance documents, etc. When determining whether the data has changed, the crawler compares the current data obtained with the previously stored data. If differences are found, further processing of the changes is required to ensure that future decision-making and process optimization can reflect the latest real situation. In this way, the timeliness and accuracy of the knowledge graph are maintained, and the response capability of the cross-border trade process to market changes is improved. After monitoring the data changes, the first target text after the changes and the second target text before the changes need to be obtained. This stage corresponds to the target text data at two different time points, namely the latest crawled and updated data (first target text) and the previously stored data (second target text). First, the new data (serial data) extracted by the pre-set crawler is used as the first target text, which usually covers the latest clearance policies, information of goods, and related laws and regulations, etc. Second, the original, unchanged clearance information is retained as the second target text for difference analysis. By directly comparing the two texts, specific change information such as the addition, modification or deletion of clause content can be effectively identified. This detailed information acquisition process provides the necessary basis for subsequent similarity calculation and entity update, ensuring that the system can adapt to new changes in a timely manner, thereby providing important support for the optimization and maintenance of cross-border trade processes. After obtaining the first target text after the changes and the second target text before the changes, the next step is to calculate the similarity between the first target text and the second target text. Similarity calculation is an important process to compare the similarity of two texts, and its purpose is to determine the degree of change between the updated data and the historical data. Similarity calculation usually uses text similarity algorithms in natural language processing, such as cosine similarity, Jaccard coefficient, word embedding (such as Word2Vec, BERT), etc. Specifically, by converting the two texts into vector representations, mathematical formulas are used to calculate the similarity score between them. For example, when using cosine similarity, the two texts are considered as vectors in a high-dimensional space, and the cosine value of the included angle is calculated to obtain a similarity score ranging from 0 to 1. The higher the value, the more similar the texts.The calculation of similarity can help quickly identify the differences in key data during the text update process, and then determine whether the current knowledge graph needs to be updated to ensure its real-time and accuracy.

[0094] After calculating the similarity between the first target text and the second target text, it is necessary to determine whether the similarity is greater than a threshold. This step is used to determine whether the changes between historical data and new data are significant, i.e., whether the similarity score exceeds the pre-set threshold. The purpose of setting the threshold is to avoid invalid updates for minor changes, and to ensure that the system will only update and adjust the knowledge graph in the case of real and important changes. If the similarity result is higher than the threshold, it means that the changes between the two texts are not significant, and it can be considered as content fine-tuning or format correction, and the graph database can not be updated. However, if the similarity is lower than the threshold, it means that the difference between the texts is large, which may mean significant changes in policies, significant updates in clearance requirements, etc. At this time, further action needs to be taken to ensure that the system can dynamically respond and timely adjust the relevant data. This judgment process is a key link to ensure the effectiveness and accuracy of the knowledge graph, and helps to improve the adaptability and effectiveness of the system in the cross-border trade process.

[0095] After determining whether the similarity is greater than the set threshold, the last step is to construct a GNN (Graph Neural Network) to simulate the conduction effect if the similarity is greater than the threshold, to obtain the predicted entities affected, and to update the space-semantic association matrix in the graph database to obtain the updated knowledge graph. If the similarity exceeds the threshold, it means that there are substantial changes between the historical data and the new data, therefore, the graph neural network (GNN) is needed to further analyze the impact of these changes on the entire knowledge graph. The design of GNN can handle graph structure data, and is very suitable for simulating the link relationship between entities and dynamic information update. Based on the current knowledge graph and the newly obtained change data, the GNN model will analyze how the change content affects the associated nodes, and use the structure in the graph to mine the potential impact entities. This step can effectively identify the clearance policies, process nodes or related news entities that may be affected due to changes. After completing the prediction, the space-semantic association matrix in the graph database is updated according to these affected entities. This update ensures the timeliness and accuracy of the data, so that the knowledge graph can always reflect the latest state of cross-border trade related information. Through this comprehensive and systematic method, it ensures that cross-border trade can quickly respond when facing policy and regulation changes, optimize the process, and improve efficiency. Finally, the updated knowledge graph generated will provide solid data support and decision basis for subsequent applications.

[0096] In one embodiment, the step S2 of generating a plurality of preset dimensions based on the cargo information and the clearance information comprises:

[0097] S211: processing the cargo information and the clearance information using a preset multi-modal artificial intelligence model to generate a process tree structure;

[0098] S212: identifying nodes in the process tree structure using a preset deep learning model to obtain a plurality of preset dimensions.

[0099] As described in steps S211-S212 above, the collected cargo information and clearance information are comprehensively processed using a preset multi-modal artificial intelligence model, and finally a JSON format process tree structure is generated. The advantage of the multi-modal artificial intelligence model is that it can process multiple types of data at the same time, such as text, image and audio, etc., which is suitable for extracting and analyzing information in complex business situations. Specifically, when the cargo information (such as commodity classification, HS code, quantity, shipper, etc.) and clearance information (such as clearance requirements, customs declaration documents, etc.) are input into the multi-modal artificial intelligence model, the model will perform feature extraction and representation learning through a multi-layer neural network architecture. This process includes data preprocessing, feature selection and feature fusion. In addition to extracting keywords and sentences from the text, the model may also combine relevant visual data (such as product images) to enhance the depth of understanding. The extracted information is converted into a process tree structure. This structure displays each link and its relationship in a hierarchical manner, starting from the root node and gradually revealing each sub-task. For example, the root node may represent the entire clearance process, and the sub-nodes will show each specific step or condition. By generating a process tree structure, the complex clearance process can be clearly captured and expressed, making subsequent analysis and decision-making more concise and efficient. After the generation of the process tree structure, a preset deep learning model is used to identify each node in the process tree to extract a plurality of preset dimensions. The purpose of this process is to simplify the complex structure built by the multi-modal model into specific dimensions that can effectively support subsequent decision-making and analysis. Deep learning models, especially those using convolutional neural networks (CNN) or recurrent neural networks (RNN) architectures, can exhibit excellent capabilities in identifying specific node information. These models are trained on large amounts of data to automatically identify and classify the types and characteristics of nodes. The model will classify each node and convert it into specific preset dimensions. These dimensions may include multiple key factors such as the tax rate of the product, the required documents for clearance, the declaration time and the legal framework involved. By clearly defining and extracting these dimensions, conditions are created for subsequent optimization and analysis, and a foundation is laid for efficient management of the entire cross-border trade process.

[0100] In one embodiment, the step S4 of optimizing the state space through a preset learning algorithm to obtain an optimized clearance process includes:

[0101] S401: obtaining a target model related to the clearance information based on the clearance information;

[0102] S402: inputting the state space into the target model and setting a preset value function in the target model, so as to obtain an optimized clearance process.

[0103] As described in steps S401-S402 above, it is necessary to extract the target model related to the optimization of the clearance process from the clearance information. The target model usually refers to a deep learning model. Based on historical data training, it has captured the clearance process under certain conditions in the past and many variables that affect the process output. Specifically, it can use deep learning, reinforcement learning and other technologies to comprehensively consider different variables (such as cargo type, regulations of the destination country, freight, time limit, etc.) and simulate their impact on clearance efficiency. Through such training, the model can provide more accurate predictions for different strategies and their potential results. After the target model is determined, the selected state space is input into the target model, and a preset value function is set in the model, so as to obtain an optimized clearance process. Specifically, various features (such as cargo information, clearance policy, logistics information, etc.) carried by the state space will be input into the target model in the form of an input vector. The model will receive these state information and run simulation algorithms to evaluate potential clearance decisions. The value function is used to evaluate the utility of executing a specific action in each state. In a specific embodiment, the time limit (R1), the cost (R2) and the compliance (R3) can be weighted and summed to evaluate the three indicators. By setting this value function, the target model can reflect the expected returns of different decision paths, so as to select the strategy that is most beneficial to optimizing the clearance process. The optimization process can use a policy optimization algorithm in reinforcement learning (such as PPO, Q-learning, etc.) to compare the returns of different strategies repeatedly, adjust the strategy selection, and finally determine the best clearance process.

[0104] In one embodiment, before the step S401 of obtaining a target model related to the clearance information based on the clearance information, the method further comprises:

[0105] S4001: obtaining various clearance categories of clearance information;

[0106] S4002: generating a category data set for each clearance category using a simulator; wherein the category data set includes clearance category data of the same clearance category;

[0107] S4003: inputting each category data set into a preset model for training to obtain a target model corresponding to each clearance category.

[0108] As described in steps S4001-S4003 above, the clearance category of various clearance information needs to be obtained. This process involves classifying the collected clearance information to identify different types of clearance procedures and related requirements. For example, the clearance category can include import clearance, export clearance, temporary import and export, bonded zone clearance, and other different forms. Each type of clearance procedure has its specific policies, regulations and operational requirements, therefore, confirming the clearance category is the basis for subsequent data processing and model training. Obtaining the clearance category usually involves comparison and analysis of regulatory documents related to customs regulations, industry standards, etc. Data analysis tools or techniques such as text mining, natural language processing can be relied on to identify and classify keywords and information in the documents.

[0109] After obtaining the clearance category of various clearance information, the simulator generates a category data set for each clearance category. A corresponding training data set is created for each clearance category for subsequent training of the target model.

[0110] The generation of the category dataset using the simulator is based on the understanding of real customs clearance data and processes, which is transformed into simulated data. The simulator can generate legitimate and specific category characteristic compliant customs clearance data by setting rules and scenarios. For example, for the import customs clearance category, the simulator can generate corresponding data with real customs document requirements, cargo properties, and different shipment origins. The generated dataset should contain multiple samples, each representing the customs clearance requirements of the same category while covering a variety of cargo types and customs clearance situations to improve the model's generalization ability. The number and diversity of category datasets directly affect the training effect and accuracy of the target model, so it is very important to form a sufficient and representative training set. This can help the model better understand the standards and processes of different customs clearance categories in subsequent steps, improving its application effect in real scenarios. Ultimately, the generated category dataset will provide rich data support for the training of the target model for customs clearance categories, ensuring that the model can effectively handle future customs clearance scenarios. After generating the category dataset for each customs clearance category, each category dataset is input into the pre-set model for training to obtain the target model corresponding to each customs clearance category. This process is the core of the entire customs clearance process optimization, and its goal is to build intelligent models that can effectively handle different customs clearance categories. The process of inputting the category dataset into the pre-set model usually includes the following aspects: First, select a model architecture suitable for the current task, such as a deep learning model (e.g., convolutional neural networks, graph neural networks, or recurrent neural networks, etc.) to capture complex patterns in the data. Then, set appropriate hyperparameters (such as learning rate, batch size, etc.) to ensure that the model can effectively learn during training. During training, the model will continuously adjust its internal parameters to minimize the loss function, thereby improving its ability to accurately identify and predict customs clearance categories. The training algorithm used can be stochastic gradient descent (SGD) or its variant optimization algorithms such as the Adam optimizer. Finally, the trained target model will correspond to a specific customs clearance category and be able to accurately identify, predict, and handle future customs clearance requests. This targeted model enables the entire cross-border trade process to dynamically adapt to different customs clearance requirements, achieving optimization in actual applications and improving the efficiency of the process and reducing potential operational risks. Through this series of operations, enterprises can establish a stronger response capability in customs clearance, ensuring efficient and compliant cross-border trade process operations.

[0111] Referring to Figure 3 The application also provides a cross-border trade process optimization device, which comprises:

[0112] The acquisition module 902 is configured to acquire cargo information and customs clearance information to be traded across borders.

[0113] The generation module 904 is configured to generate a plurality of preset dimensions based on the cargo information and the customs clearance information.

[0114] The collection module 906 is configured to collect dimension values corresponding to the preset dimensions in the preset knowledge graph to form a state space; the state space is a space corresponding to all the preset dimensions;

[0115] The optimization module 908 is configured to perform policy optimization on the state space by using a preset learning algorithm to obtain an optimized clearance process.

[0116] In an embodiment, the cross-border trade process optimization apparatus further comprises:

[0117] The historical clearance data acquisition module is configured to acquire historical clearance data in cross-border trade;

[0118] The historical clearance data division module is configured to divide the historical clearance data into structured data and unstructured data according to data types;

[0119] The feature recognition module is configured to use a dual-flow transformer architecture, use a pattern perception encoder to extract field-level semantic features from the structured data, and use a pre-trained language model to perform entity recognition on the unstructured data to obtain target entities;

[0120] The matrix establishment module is configured to align logistics track coordinates and target entities by using a cross-modal attention mechanism based on the field-level semantic features to establish a space-semantic association matrix;

[0121] The matrix import module is configured to import each target entity and the space-semantic association matrix into a graph database with preset values to obtain the preset knowledge graph.

[0122] In an embodiment, the cross-border trade process optimization apparatus further comprises:

[0123] The clearance language detection module is configured to detect a clearance language of the historical clearance data;

[0124] The translation model acquisition module is configured to acquire a corresponding translation model based on the clearance language;

[0125] The translation module is configured to translate the historical clearance data based on the translation model to obtain historical clearance data in a specified language.

[0126] In an embodiment, the cross-border trade process optimization apparatus further comprises:

[0127] The crawler monitoring module is configured to monitor whether the historical clearance data changes in real time by using a preset crawler;

[0128] The target text acquisition module is configured to acquire the first target text after the change and the second target text before the change if the change occurs.

[0129] The similarity calculation module is configured to calculate a similarity between the first target text and the second target text.

[0130] The similarity judgment module is configured to judge whether the similarity is greater than a threshold.

[0131] The matrix update module is configured to construct a GNN network to simulate a conduction effect, obtain a predicted entity affected by prediction, and update a space-semantic association matrix in the graph database to obtain an updated knowledge graph if the similarity is greater than the threshold.

[0132] In one embodiment, the generation module 904 includes:

[0133] The flow tree structure generation submodule is configured to process the cargo information and the clearance information using a preset multi-modal artificial intelligence model to generate a flow tree structure.

[0134] The flow tree structure identification submodule is configured to identify nodes in the flow tree structure using a preset deep learning model to obtain a plurality of preset dimensions.

[0135] In one embodiment, the optimization module 908 includes:

[0136] The target model acquisition submodule is configured to acquire a relevant target model based on the clearance information.

[0137] The state space input submodule is configured to input the state space into the target model and set a preset value function in the target model to obtain an optimized clearance process.

[0138] In one embodiment, the optimization module 908 further includes:

[0139] The clearance category acquisition submodule is configured to acquire clearance categories of various clearance information.

[0140] The category dataset generation submodule is configured to generate a category dataset of each clearance category using a simulator. The category dataset includes clearance category data of a plurality of same clearance categories.

[0141] The training submodule is configured to input each category dataset into a preset model for training to obtain a target model corresponding to each clearance category.

[0142] Figure 4 An internal structure diagram of a computer device in one embodiment is shown. The computer device can be a terminal or a server. As shown in FIG. 1, the computer device includes a processor 1001, memory 1002, and a bus 1003.Figure 4 As shown in the figure, the computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and can also store a computer program, which, when executed by the processor, can enable the processor to implement the cross-border trade process optimization method. The computer program can also be stored in the internal memory, and the computer program, when executed by the processor, can enable the processor to execute the cross-border trade process optimization method. Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0143] In one embodiment, a computer device is proposed, comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to enable the processor to perform the following steps:

[0144] Obtaining cargo information and customs clearance information to be traded across borders;

[0145] Generating a plurality of preset dimensions based on the cargo information and the customs clearance information;

[0146] Collecting dimension values corresponding to the preset dimensions in a preset knowledge graph to form a state space; wherein the state space is a space corresponding to all preset dimensions;

[0147] Optimizing the state space through a preset learning algorithm to obtain an optimized customs clearance process.

[0148] Not only makes the cross-border trade process more flexible and more adaptable, but also can improve the overall operation efficiency and shorten the customs clearance time. In addition, the implementation of dynamic decision support enhances the intelligence of the system, ensuring that enterprises have a competitive advantage in the rapidly changing international trade environment. Through this innovative method, enterprises can optimize resource allocation, reduce operating costs, and ultimately improve overall market responsiveness and customer satisfaction.

[0149] In one embodiment, a computer readable storage medium is proposed, storing a computer program, the computer program being executed by a processor to enable the processor to perform the following steps:

[0150] Obtaining cargo information and customs clearance information to be traded across borders;

[0151] Generating a plurality of preset dimensions based on the cargo information and the customs clearance information;

[0152] Collecting dimension values ​​corresponding to the preset dimensions from a preset knowledge graph to form a state space; wherein the state space is the space corresponding to all preset dimensions;

[0153] The state space is strategically optimized using a preset learning algorithm to obtain an optimized customs clearance process.

[0154] This not only makes cross-border trade processes more flexible and adaptable, but also improves overall operational efficiency and shortens customs clearance times. Furthermore, the implementation of dynamic decision support enhances the system's intelligence, ensuring that companies maintain a competitive advantage in the ever-changing international trade environment. This innovative approach enables companies to optimize resource allocation, reduce operating costs, and ultimately enhance overall market responsiveness and customer satisfaction.

[0155] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchl ink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0156] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0157] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific manner, but should not be construed as limiting the scope of the patent of the present application. It should be noted that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A cross-border trade process optimization method, characterized in that: The method comprises: Obtain information on goods to be traded across borders and customs clearance information; generating a plurality of preset dimensions based on the cargo information and the customs clearance information; Collecting dimension values ​​corresponding to the preset dimensions from a preset knowledge graph to form a state space; wherein the state space is the space corresponding to all preset dimensions; The state space is strategically optimized using a preset learning algorithm to obtain an optimized customs clearance process.

2. The cross-border trade process optimization method according to claim 1, characterized in that: Before the step of collecting dimension values ​​corresponding to the preset dimensions from the preset knowledge graph to form a state space, the method further includes: Obtain historical customs clearance data in cross-border trade; Classifying the historical customs clearance data into structured data and unstructured data according to data type; A dual-stream transformer architecture is adopted to extract field-level semantic features from the structured data using a pattern-aware encoder, and to perform entity recognition on the unstructured data using a pre-trained language model to obtain target entities. Based on the field-level semantic features, the logistics trajectory coordinates and the target entity are aligned through the cross-modal attention mechanism to establish a spatial-semantic association matrix; Import each of the target entities and the spatial-semantic association matrix into a preset graph database to obtain the preset knowledge graph.

3. The cross-border trade process optimization method according to claim 1, characterized in that: After the step of obtaining historical customs clearance data in cross-border trade, the following steps are also included: Detecting the customs clearance language of the historical customs clearance data; Obtaining a corresponding translation model based on the customs clearance language; The historical customs clearance data is translated based on the translation model to obtain historical customs clearance data in a specified language.

4. The cross-border trade process optimization method according to claim 2, characterized in that: After the step of importing each of the target entities and the spatial-semantic association matrix into a preset graph database to obtain the preset knowledge graph, the method further includes: Monitor in real time whether the historical customs clearance data has changed through a preset crawler; If a change occurs, obtaining the first target text after the change and the second target text before the change; Calculating the similarity between the first target text and the second target text; Determining whether the similarity is greater than a threshold; If the similarity is greater than a threshold, a GNN network is constructed to simulate the conduction effect, the affected predicted entities are obtained, and the spatial-semantic association matrix is ​​updated in the graph database to obtain an updated knowledge graph.

5. The cross-border trade process optimization method according to claim 1, characterized in that: The step of generating a plurality of preset dimensions based on the cargo information and the customs clearance information includes: Processing the cargo information and the customs clearance information using a preset multimodal artificial intelligence model to generate a process tree structure; A preset deep learning model is used to identify nodes in the process tree structure to obtain multiple preset dimensions.

6. The cross-border trade process optimization method according to claim 1, characterized in that: The step of performing strategy optimization on the state space by using a preset learning algorithm to obtain an optimized customs clearance process includes: Obtaining a relevant target model based on the customs clearance information; The state space is input into the target model, and a preset value function is set in the target model to obtain an optimized customs clearance process.

7. The cross-border trade process optimization method according to claim 6, characterized in that: Before the step of obtaining the relevant target simulator based on the customs clearance information, the method further includes: Obtain customs clearance categories for various customs clearance information; Using a simulator to generate a category data set for each customs clearance category; wherein the category data set includes customs clearance category data for a plurality of the same customs clearance categories; Input each category data set into the preset model for training to obtain the target model corresponding to each customs clearance category.

8. A cross-border trade process optimization device, characterized in that: The device comprises: The acquisition module is used to obtain information about goods to be traded across borders and customs clearance information; A generating module, configured to generate a plurality of preset dimensions based on the cargo information and the customs clearance information; A collection module, configured to collect dimension values ​​corresponding to the preset dimensions from a preset knowledge graph to form a state space; wherein the state space is the space corresponding to all preset dimensions; The optimization module is used to perform strategy optimization on the state space through a preset learning algorithm to obtain an optimized customs clearance process.

9. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the cross-border trade process optimization method as described in any one of claims 1 to 7.

10. A computer device, characterized in that: The device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the cross-border trade process optimization method according to any one of claims 1 to 7.

Images