Customs declaration information auditing method, device and equipment and storage medium

Abstract

The embodiment of the invention provides a customs declaration information auditing method, device and equipment and a storage medium, and the method comprises the steps: obtaining customs declaration information of goods, and generating prompt information according to the customs declaration information; retrieving the customs declaration information in a knowledge base to obtain retrieval information; integrating the prompt information and the retrieval information to obtain enhanced information; and generating an auditing result according to the enhanced information and the large model. According to the embodiment of the invention, the retrieval is performed in the knowledge base according to the customs declaration information, and the obtained retrieval information and the prompt information are integrated, so that the quality and accuracy of the input large model information can be improved, the data management capability and understanding capability of the large model are improved, and the user experience is improved. Therefore, the quality and the accuracy of the generated auditing result are higher.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, equipment and storage medium for customs declaration information verification. Background Technology

[0002] With the rapid increase in trade volume, the work of reviewing import and export goods has become increasingly heavy, especially in terms of verifying customs declaration data. Currently, the review work mainly relies on manual operation, which involves item-by-item comparison of customs declarations with various documents, including commercial documents such as packing lists, invoices, and trade contracts, transportation documents such as ocean freight manifests and air freight manifests, and other documents required by customs, such as licenses, certificates of origin, health certificates, and phytosanitary certificates. Manual comparison is time-consuming and prone to errors, which seriously affects customs clearance efficiency.

[0003] Given the expanding scale of imports, the increase in trading entities, and tight customs clearance times, customs faces challenges such as a shortage of human resources, low inspection efficiency, and heavy inspection pressure. Therefore, improving inspection efficiency and promoting intelligent manufacturing are crucial. Although existing research has focused on process simplification and management, it remains difficult to meet the large volume of inbound inspection demands.

[0004] Existing methods for reviewing import and export customs declaration information based on large-scale model retrieval enhancement technology aim to quickly identify potential problems and provide professional advice. This method not only improves examiners' ability to accurately assess import and export customs declaration information but also significantly reduces review costs and increases efficiency. However, large-scale model retrieval enhancement technology has limitations in data management and understanding. When the quality of the input customs declaration information is low, the quality and accuracy of the output review results are also low, making it difficult to achieve the expected results. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for reviewing customs declaration information, with the aim of improving the accuracy of the input customs declaration information, thereby improving the accuracy of the review results of the customs declaration information.

[0006] In a first aspect, embodiments of the present invention provide a method for reviewing customs declaration information, the method comprising: Obtain customs declaration information for the goods and generate notification information based on the customs declaration information; The customs declaration information is retrieved from the knowledge base to obtain the search results. The prompt information and the search information are integrated to obtain enhanced information; Based on the enhanced information and the large model, the audit results are generated.

[0007] Further, the step of retrieving the customs declaration information from the knowledge base to obtain the retrieved information includes: Extract the key information from the prompt message; Based on the key information, a search is performed in the knowledge base to obtain relevant text; The ranking quality is determined based on the hit rate and average reciprocal ranking of the relevant texts; If the sorting quality meets the preset requirements, the retrieval information is determined based on the relevant text.

[0008] Furthermore, the formula for calculating the average reciprocal rank is as follows: ; in, This indicates the average reciprocal ranking. Indicates the number of related texts. This represents the rank of the most relevant text in the i-th query, where i is a positive integer.

[0009] Furthermore, the knowledge base is obtained through the following steps: Recognize the structural features of input documents in various formats; Based on the structural features, the parsed content of the input document is obtained; A kernel-based temporal segmentation method intelligently divides the parsed content into blocks to obtain block data. Based on the word embedding model, the segmented data is transformed into segmented vectors; The knowledge base is composed of the segmented vectors.

[0010] Furthermore, the kernel-based temporal segmentation method intelligently segments the parsed content into blocks to obtain segmented data. The steps include: The parsed content is divided into multiple single sentences, and the division between single sentences is called the change point; Calculate the semantic distance between adjacent sentences using a large language model; Based on the semantic distance, the change points are filtered using a kernel-based temporal segmentation method; The segmented data is determined based on the filtered change points.

[0011] Furthermore, the objective function of the kernel-based temporal segmentation method is to minimize the variance of the semantic distance between neighboring sentences, and the formula for calculating the variance of the semantic distance between neighboring sentences is as follows: ; ; in, Let n be a positive integer, representing the number of relevant texts. This represents the i-th relevant text. This indicates a pre-defined penalty.

[0012] Furthermore, the step of generating a prompt message based on the customs declaration information includes: Based on the customs declaration information, the prompt information required for the large language model is extracted from the memory bank. The prompt information includes at least one of role description, instruction, generation constraint, operation, and example.

[0013] Secondly, embodiments of the present invention provide a customs declaration information verification device, the device comprising: The notification module is used to obtain the customs declaration information of the goods and generate notification information based on the customs declaration information; The retrieval module is used to retrieve the customs declaration information from the knowledge base to obtain the retrieved information; The enhancement module is used to integrate the prompt information and the search information to obtain enhanced information; The review module is used to generate review results based on the enhanced information and the large model.

[0014] Thirdly, embodiments of the present invention provide a computer device, the device including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of a customs declaration information verification method as provided in the first aspect.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a customs declaration information review method as provided in the first aspect.

[0016] This invention provides a method, apparatus, device, and storage medium for reviewing customs declaration information. When reviewing customs declaration information, the method first obtains the customs declaration information of the goods and generates prompt information based on this information. Then, it uses this customs declaration information to search a knowledge base, obtaining search information. The prompt information and search information are integrated to obtain enhanced information. Finally, the enhanced information is input into a large model to obtain the review result. In this invention, because the knowledge base is searched first based on the customs declaration information, and the obtained search information is integrated with the prompt information, the quality and accuracy of the information input into the large model are improved. This enhances the large model's data management and understanding capabilities, resulting in higher quality and accuracy of the generated review results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a customs declaration information review method provided in an embodiment of the present invention.

[0019] Figure 2 This is a flowchart of a large-scale model-based review process for import and export cargo customs declaration information, provided as an embodiment of the present invention.

[0020] Figure 3 This is a flowchart illustrating a process of retrieving information from a knowledge base using customs declaration information, as provided in an embodiment of the present invention.

[0021] Figure 4 A block diagram illustrating the generation of a knowledge base, as provided in an embodiment of the present invention.

[0022] Figure 5 This is a schematic diagram of a customs declaration information verification device provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] To make the description of this disclosure more detailed and complete, illustrative descriptions of embodiments and specific examples of the present invention are provided below; however, these are not the only forms of implementing or utilizing the specific embodiments of the present invention. The embodiments cover features of multiple specific embodiments and the methods, steps, and their order for constructing and operating these specific embodiments. However, other specific embodiments may also be used to achieve the same or equivalent functions and step sequences. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein.

[0026] In the description of the embodiments of the present invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The word "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more. Other quantifiers should be understood similarly. The preferred embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0027] Given the ever-expanding scale of global trade, the importance of verifying trade documents is becoming increasingly prominent. Traditional manual verification can no longer meet current review requirements, thus necessitating the search for new solutions. In the wave of Artificial Intelligence (AI), Retrieval-augmented Generation (RAG) technology has become a research hotspot and widely adopted in various applications due to its unique advantages. This technology combines the powerful generative capabilities of Large Language Models (LLMs) with efficient information retrieval systems, bringing users a completely new interactive experience. However, as the application of this technology deepens, many challenges are gradually emerging. The primary challenges of existing RAG technologies are as follows: (1) When dealing with massive amounts of data, the existing RAG system needs to balance efficiency and accuracy.

[0028] (2) Although LLM has the ability to generate fluent text, its ability to accurately grasp and recall key information is still insufficient when dealing with complex, unstructured data.

[0029] (3) In addition, the limitations of the RAG system in data management and understanding can easily lead to the problem of "garbage input, garbage output (GIGOut)," that is, if the quality of the input data is poor, the accuracy of the generated answer will also be difficult to meet expectations.

[0030] To address the aforementioned issues, this invention provides a customs declaration information review method, which is an end-to-end RAG solution. This invention aims to overcome the challenges of existing RAG technologies in data processing and answer generation by utilizing deep document understanding technology. This invention employs deep document understanding technology to effectively overcome the bottlenecks of current RAG technologies in data processing and answer generation. Furthermore, this invention not only supports multiple file formats but also possesses the ability to intelligently identify document structure and content, ensuring high-quality data input. Adhering to the principle of "high-quality input, high-quality output," this invention gains user trust in the review results by providing interpretable and controllable generated results.

[0031] Figure 1 A flowchart of a customs declaration information review method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes: S110, Obtain the customs declaration information of the goods, and generate a prompt message based on the customs declaration information; Import and export customs declaration refers to the process by which consignors and consignees of import and export goods, owners of inbound and outbound articles, or their agents complete customs formalities and related customs affairs with customs. First, customs declaration information is obtained. This information can be understood as information requiring verification, such as a clearly completed customs declaration form, commercial documents, shipping documents, and other documents deemed appropriate by customs. The clearly completed customs declaration form is the customs clearance document used to verify whether the content declared in the declaration process matches the verified documents and goods. Commercial documents include all import and export documents, such as packing lists, invoices, and trade contracts. Shipping documents include ocean freight manifests and air freight manifests. Other documents deemed appropriate by customs include official certificates such as licenses, certificates of origin, health certificates, and phytosanitary certificates. It may also include imported goods that have been approved by customs for tax reduction or exemption, such as foreign investment tax reductions or exemptions, which are submitted with relevant supporting documents along with the customs declaration. The specific content of this customs declaration information can be determined according to the actual situation; this embodiment of the invention does not impose specific limitations on this.

[0032] After extracting the customs declaration information, prompt information is generated based on the customs declaration information. The prompt information can be extracted from the memory bank. The prompt words are required by the large language model, including role descriptions, instructions, generation constraints, operations and examples.

[0033] Specifically, cue words in large language models refer to short text snippets used to guide the model to generate specific outputs. These cue words can be in the form of questions, instructions, descriptions, etc., and their purpose is to stimulate the model to generate responses related to the cue words.

[0034] Cue words can guide large language models to generate relevant content: by providing short text snippets, cue words guide the model to generate responses related to the prompts. For example, a cue word could be "Please explain what artificial intelligence is," and the large language model would generate an explanation about artificial intelligence based on this cue.

[0035] By providing clear instructions and context, prompts can help the model better understand the user's needs, thereby generating more accurate and targeted responses. For example, a user can ask the model to "explain quantum mechanics in simple language," and the large language model will generate an appropriate explanation based on this prompt.

[0036] S120, retrieve the customs declaration information from the knowledge base to obtain the retrieved information; The customs declaration information is retrieved from the knowledge base, and the data with the highest relevance to the customs declaration information is the retrieved information.

[0037] In this embodiment of the invention, the knowledge base encompasses various types of knowledge base information, including structured databases, unstructured documents, and programmatic data obtained through program interfaces. The structured database refers to a database tailored to a specific customs declaration scenario, logically expressed and implemented using a two-dimensional table structure, strictly adhering to data format and length specifications, and primarily stored and managed through relational databases. Unstructured documents refer to documents with relatively variable formats, often in various file formats. Programmatic data refers to data generated and processed through automated and algorithm-driven methods. The specific information stored in this knowledge base can be determined according to actual circumstances, and this embodiment of the invention does not impose specific limitations on this. Common retrieval methods in this embodiment of the invention include backward lookup, forward lookup, and tracing methods, which can be determined according to actual circumstances, and this embodiment of the invention does not impose specific limitations on this.

[0038] S130, integrate the prompt information and the search information to obtain enhanced information; Next, the prompt information and the search information are integrated to obtain enhanced information. This integration can be done by splicing the prompt information and the search information together, or by taking a weighted average of the prompt information and the search information. The specific method can be determined according to the actual situation, and this embodiment of the invention does not make specific limitations on this.

[0039] In this embodiment of the invention, by integrating prompt information and retrieval information, the quality and accuracy of the data input into the large model are enhanced. As the quality and accuracy of the input data are improved, the accuracy of the audit results output by the large model is also improved.

[0040] S140, Based on the enhanced information and the large model, generate the audit results.

[0041] Finally, the enhanced information is input into the large model to obtain the review result. In this embodiment of the invention, the large model can be a large language model. The large language model generates the review result based on the prompt information and the search information. During the generation process, the large language model will fully consider the retrieved information to improve the accuracy and relevance of the review result.

[0042] In this embodiment of the invention, the large language model learns and grows through continuous training and self-optimization, which is reflected in process design, decision-making, and problem-solving. The inference engine plays a crucial role in planning and decision-making, and is the core technology for executing actions. It determines how information is obtained from the environment, how future tasks are planned, how past experience is applied, and how appropriate tools are selected.

[0043] This invention provides a method for reviewing customs declaration information. When reviewing customs declaration information, the method first obtains the customs declaration information of the goods and generates prompt information based on this information. Then, it uses this customs declaration information to search a knowledge base, obtaining search information. The prompt information and search information are then integrated to obtain reinforcement information. Finally, the reinforcement information is input into a large language model to obtain the review result. In this invention, because the knowledge base is searched first based on the customs declaration information, and the obtained search information is integrated with the prompt information, the quality and accuracy of the information input into the large language model are improved. This enhances the large language model's data management and understanding capabilities, resulting in higher quality and accuracy of the generated review result.

[0044] In actual implementation, Figure 2 A flowchart illustrating the review process for import and export cargo customs declaration information based on a large model, as provided in this embodiment of the invention, is shown below. Figure 2 As shown, the user uploads customs declaration information on the user terminal, which can be a smart terminal such as a mobile phone, tablet, or computer. After receiving the customs declaration information, the user terminal generates a user query and inputs the user query into the large language model agent to generate prompt information.

[0045] The large language model agent employs the ReAct framework as its inference engine, demonstrating excellent performance. The ReAct framework, based on understanding and responding to user input, emphasizes the close relationship between reasoning and action. It provides strong support for complex reasoning and efficient action by enabling large models to automatically think and determine the next action upon receiving user queries. The ReAct framework is not limited to decision trees or fixed algorithms but is a comprehensive system integrating real-time information processing, decision-making, and action implementation. In the dynamic and uncertain process of reviewing import and export customs declaration information, effective decision-making relies on continuous learning and adaptability, as well as the ability to quickly translate reasoning into action, forming a closed loop of observation-thinking-action-re-observation.

[0046] The large language model agent retrieves the query content from the knowledge base, obtains the search information, and returns it to the large language model agent. The large language model agent integrates the prompt information and the search information to obtain reinforcement information. This reinforcement information is then input into the large language model to obtain the review result, which is returned to the user terminal through the large language model agent, allowing the user to immediately receive the review result.

[0047] In some embodiments, the step of retrieving the customs declaration information from a knowledge base to obtain retrieval information includes: Extract the key information from the prompt message; Based on the key information, a search is performed in the knowledge base to obtain relevant text; The ranking quality is determined based on the hit rate and average reciprocal ranking of the relevant texts; If the sorting quality meets the preset requirements, the retrieval information is determined based on the relevant text.

[0048] Figure 3 A flowchart illustrating a method for retrieving information from a knowledge base using customs declaration information, as provided in this embodiment of the invention, is shown below. Figure 3 As shown, the prompt information is first analyzed in depth to identify and extract key information. Through this analysis, user needs can be grasped more accurately, providing precise guidance for the search process. Then, using the key information provided by the query analyzer, relevant text can be quickly retrieved from a massive amount of documents.

[0049] Powerful search capabilities ensure that users can obtain the data they need in a timely manner. In this embodiment of the invention, response time is used to measure search efficiency, and hit rate and mean reciprocal rank (MRR) are used to evaluate ranking quality.

[0050] Here, hit rate refers to the frequency with which the expected recalled text appears in the top k relevant text search terms; that is, when recall@k, the text is expected to be found. When the ranking quality of relevant texts meets the preset requirements, it indicates a high hit rate, meaning the retrieval algorithm performs well, and the searched relevant texts can be directly used as retrieval information.

[0051] The MRR calculation method involves taking the inverse of the average ranking of all relevant documents or information generated in all queries and then averaging the results. The formula is as follows: (1) In formula (1), Indicates the number of related texts. This represents the rank of the most relevant text in the i-th query, where i is a positive integer. For example, if the correct answer in the first query of a system is ranked 4th, while the correct answer in the second query is ranked 2nd, then the MRR will be (0.25+0.5) / 2=0.375.

[0052] It should be noted that, in this embodiment of the invention, the relevant text obtained in the above steps can be further sorted and filtered to ensure that the information presented to the user is the most relevant and valuable. In this way, redundant and irrelevant data can be eliminated, improving the accuracy and usability of the information.

[0053] Finally, the integrated and enhanced information is input into the large language model to generate the final review result. Its powerful generation capability ensures the accuracy of the review result and makes the expression of the review result more natural and fluent.

[0054] In some embodiments, the knowledge base is obtained through the following steps: Recognize the structural features of input documents in various formats; Based on the structural features, the parsed content of the input document is obtained; A kernel-based temporal segmentation method intelligently divides the parsed content into blocks to obtain block data. Based on the word embedding model, the segmented data is transformed into segmented vectors; The knowledge base is composed of the segmented vectors.

[0055] Figure 4 A block diagram for knowledge base generation provided in an embodiment of the present invention, such as Figure 4 As shown, given that import and export customs declarations involve numerous documents in various formats, such as docx, excel, csv, txt, and html, this embodiment of the invention employs advanced visual information and parsing techniques to effectively and deeply interpret various documents, such as text, tables, and images. The specific process for acquiring the knowledge base is as follows: (1) Data connection: The data is rich in various formats and can be read by RAG in various formats through the loading process. The input documents include multiple document formats; (2) Transformation: Identify the structural characteristics of the input document, including basic elements such as headings, paragraphs, and line breaks. Furthermore, for higher-level elements such as images and tables, detailed layout and structural analysis are required.

[0056] For example, when processing tables, in addition to determining the table's location, it can also identify situations where each cell and its multiple internal columns of text might need to be combined into a single cell. All this information, after careful processing, will be used in conjunction with the table header information to ensure that the data is accurate and comprehensive.

[0057] Based on this, the data in the structural features are transformed into a unified format through cleaning, standardization, and organization, which facilitates subsequent analysis.

[0058] (3) Intelligent segmentation: There are many types of document segmentation, such as fixed-length segmentation, parent-child segmentation, table segmentation, segmentation by title, segmentation by sentence, and mixed text and image segmentation.

[0059] As an example, the kernel-based temporal segmentation method intelligently segments the parsed content into blocks to obtain segmented data. The steps include: The parsed content is divided into multiple single sentences, and the division between single sentences is called the change point; Calculate the semantic distance between adjacent sentences using a large language model; Based on the semantic distance, the change points are filtered using a kernel-based temporal segmentation method; The segmented data is determined based on the filtered change points.

[0060] Based on the parsed content of the input document, this embodiment of the invention uses a kernel-based temporal segmentation method to perform intelligent block segmentation, with the following steps: a. Let the vocabulary be... Given a word sequence of length n Divide it into sentences A single sentence The division between clauses is called the change point, and the lexical units of each block belong to... ; b. Calculate the semantic distance between adjacent sentences using a large language model: .

[0061] c. A kernel-based temporal segmentation method is used to filter change points. The goal is to minimize the variance of the semantic distance between adjacent sentences. The calculation formula is as follows: (2) ; In formula (2), This is a penalty to prevent the search from generating too many variables.

[0062] (4) Embedding: Using word embedding models, data is transformed into specific word embeddings, i.e., vector form; (5) Storage: Storing vector data in various storage systems, such as memory, file systems, or vector databases; (6) Retrieval: Retrieve the required data from the storage system in order to carry out subsequent operations.

[0063] In some embodiments, the step of generating a prompt message based on the customs declaration information includes: Based on the customs declaration information, the prompt information required for the large language model is extracted from the memory bank. The prompt information includes at least one of role description, instruction, generation constraint, operation, and example.

[0064] In summary, this invention presents an auxiliary inspection method based on large-scale model retrieval, designed to meet the growing demand for inbound cargo inspection. By employing this method, business professionals can effectively perform information review and risk assessment, significantly improving customs clearance efficiency.

[0065] This invention comprises a document parser, a query analyzer, a retrieval module, a reordering module, and a large-scale language model, among other components. It can efficiently process user queries, rapidly and accurately extracting the required information from massive amounts of text, and providing precise and practical answers. This invention not only significantly improves information processing efficiency but also greatly enhances the user experience.

[0066] Figure 5 This is a schematic diagram of the structure of a customs declaration information verification device provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the device includes: The prompt module 510 is used to obtain the customs declaration information of the goods and generate prompt information based on the customs declaration information; The retrieval module 520 is used to retrieve the customs declaration information from the knowledge base to obtain the retrieval information. The enhancement module 530 is used to integrate the prompt information and the search information to obtain enhancement information; The review module 540 is used to generate review results based on the enhanced information and the large model.

[0067] This embodiment is a device embodiment corresponding to the above method. Its implementation process is the same as that of the above method embodiment. For details, please refer to the above method embodiment. This device embodiment will not repeat the details.

[0068] Based on the customs declaration information review method, this embodiment of the invention also provides a computer device, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the above-mentioned customs declaration information verification method.

[0069] For other details regarding the implementation of the above technical solution by the processor in the above computer device, please refer to the description in the customs declaration information review method provided in the above embodiments of the invention, which will not be repeated here.

[0070] The processor can also be called a CPU (Central Processing Unit). A processor may be an integrated circuit chip with signal processing capabilities. A processor can also be a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0071] This invention also provides a computer-readable storage medium storing a readable computer program. The computer program can be stored in the storage medium as a software product and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks or optical disks, ROM (Read-Only Memory), RAM (Random Access Memory), or terminal devices such as computers, servers, mobile phones, and tablets.

[0072] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or modules, and may be electrical, mechanical, or other forms.

[0073] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0075] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0076] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0077] The technical solutions provided in this application have been described in detail above. Specific examples have been used in this application to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0078] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0080] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0082] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method of customs declaration information auditing, characterized by, The method comprises: obtaining customs declaration information of goods, and generating prompt information according to the customs declaration information; retrieving the customs declaration information in a knowledge base to obtain retrieval information; integrating the prompt information and the retrieval information to obtain reinforced information; generating an audit result according to the reinforced information and a large model.

2. The customs information auditing method according to claim 1, wherein, The step of retrieving the customs declaration information in the knowledge base to obtain retrieval information comprises: extracting key information in the prompt information; retrieving relevant text in the knowledge base according to the key information; determining sorting quality according to a hit rate and an average reciprocal rank of the relevant text; if the sorting quality meets a preset requirement, determining the retrieval information according to the relevant text.

3. The customs information auditing method according to claim 2, wherein, The calculation formula of the average reciprocal rank is as follows: ； wherein, denotes the average inverse rank, denotes the number of relevant texts, denotes the rank of the most relevant text in the i-th query, i being a positive integer.

4. The customs information auditing method of claim 1, wherein, The knowledge base is obtained by the following steps: identifying structural features of input documents in various formats; obtaining parsed content of the input documents according to the structural features; intelligently block the parsed content based on a kernel method time series segmentation method to obtain block data; converting the block data into block vectors based on a word embedding model; composing the block vectors into the knowledge base.

5. The customs information auditing method according to claim 4, wherein, The kernel method time series segmentation method intelligently blocks the parsed content to obtain block data, comprising: dividing the parsed content into multiple single sentences according to sentences, wherein the division between single sentences is called a change point; calculating semantic distances between adjacent single sentences using a large language model; screening the change points using a kernel method time series segmentation method based on the semantic distances; determining the block data according to the screened change points.

6. The customs information auditing method according to claim 5, wherein, The objective function of the kernel method time series segmentation method is to minimize the variance of the semantic distance between adjacent sentences, and the calculation formula of the minimization of the variance of the semantic distance between adjacent sentences is as follows: ； ； wherein, is a positive integer, n represents the number of relevant texts, represents the i-th relevant text, represents a preset penalty term.

7. The customs declaration information auditing method according to any one of claims 1 to 6, characterized in that, The step of generating prompt information according to the customs declaration information comprises: extracting the prompt information required by the large language model from the memory bank according to the customs declaration information, wherein the prompt information comprises at least one of role description, instruction, generation constraint, operation and example.

8. A customs declaration information auditing apparatus characterized by comprising: The device comprises: a prompt module configured to obtain customs declaration information of goods and generate prompt information according to the customs declaration information; a retrieval module configured to retrieve the customs declaration information in a knowledge base to obtain retrieval information; a reinforcement module configured to integrate the prompt information and the retrieval information to obtain reinforced information; an audit module configured to generate an audit result according to the reinforced information and a large model.

9. A computer device, comprising: The device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus; the memory is used to store a computer program; the processor is used to execute the program stored on the memory to implement the steps of the customs declaration information audit method in any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the customs declaration information audit method in any one of claims 1-7. The computer program is executed by the processor to implement the steps of the customs declaration information audit method in any one of claims 1-7.

Images