Method and device for regional classification of hs code of commodity, electronic equipment and medium

Abstract

The application relates to the technical field of HS code regional classification of commodities, and discloses a HS code regional classification method and device of commodities, electronic equipment and a medium, wherein the method comprises the following steps: assigning accurate weight coefficients to each commodity feature by combining a regional preference knowledge graph; dynamically updating an initial feature association knowledge graph; obtaining the weight coefficients of the commodity features by using the updated feature association knowledge graph; and then obtaining the classification code of the to-be-classified commodity in the classification region. The application has the beneficial effects that the classification accuracy of commodities in various regions is improved, the customs clearance efficiency of enterprises is improved, the legal risks caused by improper classification are reduced, and the overall efficiency of cross-border transactions is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of regional classification of HS code of commodities, and particularly relates to a method and device for regional classification of HS code of commodities, electronic equipment and a medium. BACKGROUND

[0002] With the rapid development of global trade, the classification management of commodities in different regions is becoming more and more strict, and the regulations and preferences of different regions are becoming more and more personalized. The traditional classification method of commodities mainly relies on universal regulations and basic characteristics, ignoring the regulations, market demand and cultural differences of specific regions, which leads to high risk disputes and additional economic losses in the product declaration process of enterprises. At the same time, the classification standards of commodities are different in different regions, which leads to low clearance efficiency in cross-border trade. SUMMARY

[0003] Therefore, it is necessary to propose a method and device for regional classification of HS code of commodities, electronic equipment and a medium to solve the problem of regional classification of HS code of commodities.

[0004] A method for regional classification of HS code of commodities, the method comprising:

[0005] obtaining commodity characteristics of a plurality of preset dimensions of a commodity to be classified, and a regional preference knowledge graph of a classification region;

[0006] obtaining a weight coefficient of each commodity characteristic according to the regional preference knowledge graph;

[0007] inputting each commodity characteristic and the corresponding weight coefficient into a preset classification decision forest for coding classification to obtain an HS code of the commodity to be classified in the classification region;

[0008] Before the step of obtaining commodity characteristics of a plurality of preset dimensions of a commodity to be classified, and a regional preference knowledge graph of a classification region, the method further comprises:

[0009] obtaining a regional regulation information base and a ruling database of the classification region;

[0010] extracting entities and the association relationship between entities in the regional regulation information base and the ruling database;

[0011] updating an initial feature association knowledge graph according to the extracted entities and the association relationship between entities to obtain the regional preference knowledge graph.

[0012] Further, the step of updating the initial feature association knowledge graph according to the extracted entities and the association relationship between entities to obtain the regional preference knowledge graph comprises:

[0013] obtain a weight coefficient node corresponding to each of the entities in the initial feature correlation knowledge graph; wherein the weight coefficient node of the initial feature correlation knowledge graph stores a basic weight coefficient corresponding to a commodity feature;

[0014] calculate a node increment of the corresponding weight coefficient node according to the correlation relationship corresponding to each of the entities;

[0015] update the initial feature correlation knowledge graph according to the node increment to obtain the regional preference knowledge graph.

[0016] Further, before the step of updating the initial feature correlation knowledge graph according to the extracted entities and the correlation relationship between the entities to obtain the regional preference knowledge graph, the method further comprises:

[0017] reverse deduce the commodity features of each of the preset dimensions from a preset coding system;

[0018] obtain a set of commodity description texts and regulation text data;

[0019] extract a frequency factor of each of the commodity features based on the set of commodity description texts, and extract a regulation factor of each of the commodity features based on the regulation text data;

[0020] calculate a basic weight coefficient of each of the commodity features according to the frequency factor and the regulation factor;

[0021] construct the initial feature correlation knowledge graph based on each of the commodity features and the corresponding basic weight coefficient.

[0022] Further, in the step of obtaining the commodity features of the plurality of preset dimensions of the to-be-classified commodity and the regional preference knowledge graph of the classification region, the step of obtaining the commodity features of the plurality of preset dimensions of the to-be-classified commodity comprises:

[0023] obtain text description information of the to-be-classified commodity;

[0024] perform multi-language semantic analysis on the text description information to obtain the commodity features of the to-be-classified commodity.

[0025] Further, in the step of obtaining the commodity features of the plurality of preset dimensions of the to-be-classified commodity and the regional preference knowledge graph of the classification region, the step of obtaining the commodity features of the plurality of preset dimensions of the to-be-classified commodity comprises:

[0026] perform three-dimensional modeling on the to-be-classified commodity to obtain a point cloud three-dimensional model of the to-be-classified commodity;

[0027] modeling the point cloud three-dimensional model through a preset point cloud analysis method to identify a hidden function component of the to-be-classified commodity;

[0028] performing feature analysis on the hidden function component to obtain a commodity feature of the to-be-classified commodity.

[0029] Further, before the step of inputting each of the commodity features and the corresponding weight coefficients into a preset classification decision forest for encoding classification to obtain the HS code of the to-be-classified commodity in the classification area, the method further comprises:

[0030] obtaining regulation update information of the classification area;

[0031] updating an initial classification decision forest according to the regulation update information to obtain the preset classification decision forest.

[0032] Further, before the step of obtaining a plurality of preset dimensions of commodity features of the to-be-classified commodity and a regional preference knowledge graph of the classification area, the method further comprises:

[0033] obtaining regional regulation information of the classification area;

[0034] detecting whether the to-be-classified commodity conforms to the regulation of the regional regulation information;

[0035] If the to-be-classified commodity does not conform to the regulation of the regional regulation information, the to-be-classified commodity is identified as an out-of-compliance commodity of the classification area.

[0036] An HS code regional classification device of a commodity, the device comprising:

[0037] a commodity feature acquisition module configured to obtain a plurality of preset dimensions of commodity features of a to-be-classified commodity and a regional preference knowledge graph of a classification area;

[0038] a weight coefficient acquisition module configured to obtain a weight coefficient of each commodity feature according to the regional preference knowledge graph;

[0039] a classification code acquisition module configured to input each of the commodity features and the corresponding weight coefficients into a preset classification decision forest for encoding classification to obtain an HS code of the to-be-classified commodity in the classification area;

[0040] a ruling database acquisition module configured to obtain a regional regulation information database and a ruling database of the classification area;

[0041] an association relationship extraction module configured to extract entities and association relationships between the entities in the regional regulation information database and the ruling database;

[0042] The preference knowledge graph acquisition module is used to update the initial feature association knowledge graph based on the extracted entities and the relationships between entities, so as to obtain the regional preference knowledge graph.

[0043] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps:

[0044] Obtain product features of the products to be classified from multiple preset dimensions, as well as a regional preference knowledge graph of the classification area;

[0045] The weight coefficients of each product feature are obtained based on the regional preference knowledge graph.

[0046] Each of the product features and its corresponding weight coefficients are input into a preset classification decision forest for encoding and classification, thereby obtaining the HS code of the product to be classified in the classification region.

[0047] Before the steps of obtaining product features of multiple preset dimensions of the products to be classified and the regional preference knowledge graph of the classification area, the method further includes:

[0048] Obtain the regional regulatory information database and adjudication database for the classified area;

[0049] Extract entities and relationships between entities from the regional regulatory information database and the adjudication database;

[0050] The initial feature association knowledge graph is updated based on the extracted entities and the relationships between them to obtain the region preference knowledge graph.

[0051] A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0052] Obtain product features of the products to be classified from multiple preset dimensions, as well as a regional preference knowledge graph of the classification area;

[0053] The weight coefficients of each product feature are obtained based on the regional preference knowledge graph.

[0054] Each of the product features and its corresponding weight coefficients are input into a preset classification decision forest for encoding and classification, thereby obtaining the HS code of the product to be classified in the classification region.

[0055] Before the steps of obtaining product features of multiple preset dimensions of the products to be classified and the regional preference knowledge graph of the classification area, the method further includes:

[0056] Obtain the regional regulatory information database and adjudication database for the classified area;

[0057] Extract entities and relationships between entities from the regional regulatory information database and the adjudication database;

[0058] The initial feature association knowledge graph is updated based on the extracted entities and the relationships between them to obtain the region preference knowledge graph.

[0059] The beneficial effects of this invention are as follows: By combining a regional preference knowledge graph, precise weight coefficients are assigned to each product feature, the initial feature association knowledge graph is dynamically updated, and the weight coefficients of product features are obtained using the updated feature association knowledge graph, thereby obtaining the classification code of the product to be classified in the classification region. This improves the accuracy of product classification in various regions, thereby improving the customs clearance efficiency of enterprises, reducing legal risks caused by improper classification, and optimizing the overall efficiency of cross-border transactions. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] in:

[0062] Figure 1 This is a diagram illustrating the application environment of a regional classification method for HS codes of goods in one embodiment.

[0063] Figure 2 This is a flowchart of a regional classification method for HS codes of goods in one embodiment;

[0064] Figure 3 This is a structural block diagram of a product HS code regional classification device in one embodiment;

[0065] Figure 4 This is a structural block diagram of an electronic device in one embodiment. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Figure 1This is a diagram illustrating the application environment of regional classification of HS codes for goods in one embodiment. (Refer to...) Figure 1 The HS code regional classification method for this product is applied to the HS code regional classification system for products. This system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal; a mobile terminal can be at least one of a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers. The terminal 110 is used to acquire multiple preset dimensions of product characteristics to be classified, and the server 120 is used to generate classification codes for each classification region.

[0068] like Figure 2 As shown, in one embodiment, a method for regional classification of HS codes for goods is provided. This method can be applied to both terminals and servers; this embodiment illustrates its application to terminals. The specific steps of this method for regional classification of HS codes for goods include:

[0069] S1: Obtain product features of the products to be classified in multiple preset dimensions, as well as the regional preference knowledge graph of the classification area;

[0070] S2: Obtain the weight coefficients of each product feature based on the regional preference knowledge graph;

[0071] S3: Input each of the product features and its corresponding weight coefficients into a preset classification decision forest for encoding and classification, and obtain the HS code of the product to be classified in the classification area;

[0072] Before step S1, which involves obtaining product features of multiple preset dimensions of the product to be classified and the regional preference knowledge graph of the classification region, the method further includes:

[0073] S001: Obtain the regional regulatory information database and adjudication database for the classified area;

[0074] S002: Extract entities and relationships between entities from the regional regulatory information database and the adjudication database;

[0075] S003: Update the initial feature association knowledge graph based on the extracted entities and the relationships between them to obtain the region preference knowledge graph.

[0076] This regional product classification method can be widely applied in e-commerce platforms and supply chain management. On e-commerce platforms, merchants can accurately classify products based on regional preference knowledge graphs, according to different regional consumer preferences and regulatory requirements, thereby improving the relevance of product recommendations and sales efficiency. For example, in a specific region, consumers may prefer environmentally friendly products. By classifying and weighting product features, merchants can adjust their product strategies in a timely manner to provide products that meet local needs. Furthermore, in supply chain management, companies can optimize inventory allocation and logistics plans based on the classification results to promptly meet market demands in different regions.

[0077] As described in step S1 above, acquire product features of the product to be classified across multiple preset dimensions, as well as a regional preference knowledge graph of the classification area. The product features of each preset dimension can specifically include material composition, functional attributes, manufacturing process, size specifications, brand information, etc. For example, for an electronic product, possible product features include "battery capacity," "screen type," and "storage capacity." Simultaneously, it is also necessary to acquire a regional preference knowledge graph related to the target classification area. This is a graph containing regulations, market demands, and cultural preferences within a specific region. Regional preference knowledge graphs are typically derived from the analysis of relevant regulations, standards, and case data, and can dynamically reflect the classification preferences and approval requirements of various countries for specific products. This graph not only supports the subsequent calculation of feature weights but also serves as an important foundation for constructing more accurate classification criteria.

[0078] As described in step S2 above, the weight coefficients of each product feature are obtained based on the regional preference knowledge graph. The weight coefficients reflect the importance and applicability of each product feature within a specific classification region. For example, a particular region may place high importance on environmentally friendly materials; therefore, features related to environmental protection (such as "biodegradable plastic content") will receive higher weight coefficients. Conversely, if a region places more emphasis on other features (such as "product function" or "origin"), then the weight coefficients of these features will be increased accordingly. This process typically involves the analysis and reasoning of the regional preference knowledge graph, and also requires consideration of key terms in regulatory texts, approximations, and priority information provided by historical rulings, assigning a precise weight coefficient to each product feature to provide a quantitative basis for subsequent classification decisions.

[0079] As described in step S3 above, each commodity feature and its corresponding weight coefficient are input into a preset classification decision forest for encoding and classification, resulting in the HS code of the commodity to be classified in the classification region. A decision forest is an ensemble learning model, typically composed of multiple decision trees, each trained on a different subset of features. This method effectively reduces the risk of overfitting caused by a single decision tree and improves the overall classification accuracy. In this process, each commodity feature participates in the construction of the decision tree and classification decision according to its weight coefficient, encoding in a globally optimal manner. Finally, after processing by the classification decision forest, a classification code for the commodity to be classified in a specific region can be generated. This not only helps to accurately identify the product's classification but also facilitates subsequent compliance review and customs procedures. HS code, or Harmonized System Code, is a universal commodity classification system. The main purpose of HS code is to uniformly classify and code commodities in cross-border trade for management, statistics, trade analysis, and fee collection. HS code generally consists of six digits, where the first two digits represent the major category of goods, and the last four digits further subdivide into subcategories. For example, HS code "0102" represents live horses and mules, while "0102.10" may represent a specific breed of horse. Because different countries may expand upon HS codes by adding more digits to suit their own classification needs, some countries may have eight-digit, ten-digit, or even more digit coding systems.

[0080] As described in step S001 above, before conducting regional classification of goods, obtain the regional regulatory information database and adjudication database related to the classification area. The regional regulatory information database usually includes the regulations, standards and regulatory documents established by a specific country or region for the classification of goods. These documents can provide specific requirements on product quality, safety and environmental protection, and help to understand the legal framework faced by various goods in the region.

[0081] The adjudication database contains judicial precedents and rulings from relevant institutions. It records historical cases of goods being classified in designated regions under different circumstances. These adjudication cases provide standards and bases for approval in actual operation, helping users to clarify the classification and handling methods for specific goods in specific situations. By accessing information from these two databases, it is possible to grasp the legal dynamics and classification trends within the region in advance, laying a solid foundation for subsequent feature extraction and the construction of preference knowledge graphs.

[0082] As described in step S002 above, after obtaining the regulatory information database and the ruling database, the next step is to extract relevant entities and their relationships. This process involves using Natural Language Processing (NLP) technology to achieve efficient analysis and information extraction from regulatory texts and ruling cases. Specifically, the first step is to identify key entities from the legal texts, such as characteristic terms, restrictions, compliance requirements, and responsible parties in the regulations. For the ruling database, the structure of goods, characteristics, legal clauses, and corresponding rulings needs to be identified. These are key elements that need to be focused on during the product classification process. Named entity recognition technology is used to automatically extract these key elements. Secondly, it is necessary to clarify the relationships between these entities. For example, a specific regulatory clause may correspond to multiple product characteristics, or a certain characteristic may involve the requirements of multiple regulatory agencies; there may also be a correlation between ruling cases from different periods. This relationship mapping not only enhances the understanding of specific regulations but also lays the foundation for subsequent entity-based knowledge reasoning. Specifically, entity relationships can be extracted using a large language model. Regional regulatory information databases contain laws and regulations within specific regions, and these legal texts usually have complex structures and rich legal terminology. The adjudication database stores legal documents such as court judgments and arbitration awards, reflecting the application of law and the logic of adjudication. When using LLM for entity extraction, the model can identify legal entities in the text, such as legal clauses, case numbers, court names, parties involved (e.g., institutions, individuals), time, and place. In addition to extracting individual entities, LLM can also identify the relationships between entities, such as the application relationship between a legal clause and a specific case, or the mutual influence between different entities.

[0083] As described in step S003 above, after completing entity extraction and mapping of relationships, the initial feature association knowledge graph is updated based on these extracted entities and relationships to form a regional preference knowledge graph. First, existing feature nodes in the initial feature knowledge graph need to be identified. These nodes represent key product features such as materials, functions, and uses. Then, based on entities and their relationships extracted from the regulatory information database and adjudication database, the attributes and weights of these feature nodes are updated one by one. This includes assigning new weight coefficients to specific features, such as setting a higher compliance priority for a certain material, or marking a characteristic as compliant with specific regulatory requirements. Simultaneously, new feature nodes and relevant legal provisions should be dynamically added to ensure the completeness and usability of the knowledge graph. By continuously updating these features and their interrelationships, a knowledge graph reflecting regional regulatory requirements and market preferences (regional preference knowledge graph) is established. This graph not only provides a basis for subsequent product feature analysis and classification but also provides effective support for enterprises to formulate market strategies and compliance decisions in actual business activities.

[0084] In one embodiment, step S003, which updates the initial feature association knowledge graph based on the extracted entities and the relationships between entities to obtain the region preference knowledge graph, includes:

[0085] S0031: Obtain the weight coefficient nodes in the initial feature association knowledge graph corresponding to each entity; wherein, the weight coefficient nodes in the initial feature association knowledge graph store the basic weight coefficients of the corresponding product features;

[0086] S0032: Calculate the node increment of the corresponding weight coefficient node according to the association relationship of each entity;

[0087] S0033: Update the initial feature association knowledge graph according to the node increment to obtain the region preference knowledge graph.

[0088] As described in step S0031 above, the weight coefficient nodes in the initial feature association knowledge graph corresponding to each entity are obtained. Using the extracted entities, the weight coefficient nodes in the initial feature association knowledge graph corresponding to these entities are searched. First, the entities obtained from the regional regulatory information database and adjudication database typically include product characteristics, compliance requirements, and relevant legal clauses. Each entity has a corresponding feature node in the initial feature association knowledge graph. These nodes store basic weight coefficients, representing the importance of the feature in the globally accepted classification system of relevant institutions. Specifically, the initial feature association knowledge graph is queried to find the feature node matching each entity. In the query operation, techniques such as keyword matching and similarity calculation can be used to ensure that the corresponding feature can be accurately found. For example, if an extracted entity involves the feature "biodegradable materials," the system will retrieve the relevant feature nodes in the initial feature association knowledge graph and extract the basic weight coefficient of that node, usually represented by wbase. This lays the foundation for subsequent weight calculations, ensuring that the updated regional preference knowledge graph truly reflects regulatory requirements.

[0089] As described in step S0032 above, the node increment of the corresponding weight coefficient node is calculated based on the association relationships of each entity. The increment of the weight coefficient node corresponding to each extracted entity is calculated, and the extracted entities and their association relationships are used to quantify the specific impact of regulations or rulings on feature weights. Each entity typically contains information such as model, compliance requirements, risk assessment, and market trends; its association relationships may clarify the change in the importance of the feature within a specific region. By comparing the entity information with the weight coefficients in the initial feature association knowledge graph, the system will analyze the association relationships of each entity. For example, if a feature is emphasized in green environmental protection regulations, the increment Δwk of each weight coefficient node can be obtained based on its importance and the stipulated compliance coefficients, providing a specific quantitative basis for subsequent weight updates.

[0090] As described in step S0033 above, the initial feature association knowledge graph is updated based on the node increments to obtain the regional preference knowledge graph. These increments are then used to update the initial feature association knowledge graph, thus forming a new regional preference knowledge graph. This process involves dynamically adjusting the weight nodes of each feature to ensure the graph reflects the latest regulatory requirements and market demands. First, the system traverses all corresponding feature weight nodes and applies the previously calculated increments to the initial weights. During this process, the system also performs constraint checks to ensure that the updated weights do not exceed a predetermined effective range (e.g., between 0 and 1). If the updated weight of a feature exceeds the range, appropriate measures should be taken to adjust it. The updated weight coefficients are stored in the new regional preference knowledge graph to support subsequent product classification. This update ensures the real-time and dynamic nature of the knowledge graph, enabling it to provide accurate product classification criteria in practice and helping enterprises conduct business compliantly in different regions. In this way, the regional preference knowledge graph provides enterprises with classification standards based on specific regulations, greatly improving the adaptability and compliance of products in the market.

[0091] In one embodiment, before step S003, which updates the initial feature association knowledge graph based on the extracted entities and the relationships between entities to obtain the region preference knowledge graph, the method further includes:

[0092] S0021: Derive product features of each preset dimension from the preset coding system;

[0093] S0022: Obtain product description text set and regulatory text data;

[0094] S0023: Extract frequency factors of each product feature based on the product description text set, and extract regulatory factors of each product feature based on the regulatory text data;

[0095] S0024: Calculate the basic weight coefficients for each commodity feature based on the frequency factor and regulatory factor;

[0096] S0025: Construct the initial feature association knowledge graph based on each of the product features and its corresponding basic weight coefficients.

[0097] As described in step S0021 above, the HS coding system, which derives product features for each preset dimension from a pre-defined coding system, categorizes products into several types. Different codes can identify specific product features, such as material type, purpose, and appearance. The system first identifies the pre-defined coding system within a specified area. This information comes from various areas and can be obtained by accessing the corresponding database. Specifically, the HS codes and their corresponding descriptive information are extracted to identify various dimensions related to product features. For example, a product coded 8424 typically involves "machinery for a specific purpose" as a feature dimension. During the derivation process, corresponding coding mapping rules can be used to extract key feature items such as "power," "weight," and "size." A feature set F={f1,f2,...,fn} is established, which will lay the foundation for subsequent data analysis and weight calculation.

[0098] As described in step S0022 above, product description text sets and regulatory text data are acquired. A comprehensive information foundation is established to support subsequent product feature parsing and weight calculation. Product description text sets typically contain product introductions from e-commerce platforms, manufacturer websites, and industry catalogs, covering various product specifications, materials, uses, and other information. This text data provides rich context for feature extraction. Meanwhile, regulatory text data originates from the laws, regulations, and industry standards of various countries. This data can be obtained by accessing standard databases and digital archives of legal documents, or by using APIs to retrieve real-time regulatory information. By comparing these text data, the system can ensure that the extracted product features not only conform to market description standards but also comply with all legal and regulatory requirements.

[0099] As described in step S0023 above, frequency factors of each product feature are extracted from the product description text set, and regulatory factors of each product feature are extracted from the regulatory text data. After completing the data collection of product text and regulatory text, frequency factors of product features are extracted by analyzing the product description text set, and regulatory factors are extracted from the regulatory text. The frequency factors reflect the frequency of occurrence of product features in the description text, which can help identify which features receive widespread attention in the market. For example, by using natural language processing (NLP) techniques, such as the term frequency-inverse document frequency (TF-IDF) algorithm, the frequency of a certain feature appearing in the product description text can be determined. In addition, regulatory factors are used to quantify the importance of specific product features in the regulatory text, involving the analysis of the frequency and importance of mentioning a certain feature in the regulatory text. For example, if a regulation emphasizes "environmentally friendly materials," then regulatory factors related to this feature will receive higher weights. In this way, not only is the market demand status (frequency factors) of product features obtained, but the compliance requirements of different product features in various regulatory documents can also be understood.

[0100] As described in step S0024 above, the basic weight coefficients of each product feature are calculated based on the frequency factor and regulatory factor extracted earlier. This process combines market demand (frequency factor) and compliance requirements (regulatory factor) to arrive at an effective weight allocation scheme. The basic weight coefficients are typically calculated using a weighted formula. In this way, product features with high frequency and high regulatory impact will receive higher basic weight coefficients, while features that are less frequently mentioned or have lower compliance requirements will have their weights reduced accordingly. This calculation ensures that the resulting feature weights not only conform to market trends but also meet the requirements of local laws and regulations, providing a solid foundation for the construction of a regional preference knowledge graph. Specifically, the specific method for calculating the basic weight coefficients of each product feature can be... ,in, Based on the weighting coefficient, For frequency factors, As a regulatory factor, , This is the adjustment coefficient.

[0101] As described in step S0025 above, an initial feature association knowledge graph is constructed using the previously calculated basic weight coefficients and various product features. Specifically, the system needs to create nodes for each product feature and assign basic weight coefficients to them. These nodes will be stored in the graph database of the knowledge graph, forming a feature set with a clear structure. Relationships between these nodes are established, and edges are constructed using the similarity and relevance of features. For example, if two features frequently appear in the same product description, weighted edges can be established based on their frequency of co-occurrence. Such edges can represent the mutual influence between features, enhancing the expressive power of the knowledge graph. Ultimately, the resulting initial feature association knowledge graph can lay the foundation for updating the regional preference knowledge graph and provide strong data support for subsequent regional classification methods. This integrated information structure will also become a key tool for further analyzing local markets and legal environments, helping companies understand the adaptability and compliance of their products in specific markets.

[0102] In one embodiment, the step S1 of obtaining multiple preset dimensions of product features of the product to be classified and the regional preference knowledge graph of the classification region includes:

[0103] S101: Obtain the text description information of the product to be classified;

[0104] S102: Perform multilingual semantic parsing on the text description information to obtain the product characteristics of the product to be classified.

[0105] As described in step S101 above, the text description information of the product to be classified is obtained. Relevant data about the product is acquired from multiple sources, typically including e-commerce platforms, product manuals, manufacturer websites, industry directories, and user-generated content (such as reviews and ratings). The text description information should be as comprehensive as possible, covering details of the product's materials, functions, specifications, uses, brand, manufacturing process, and other aspects. During implementation, multiple information sources must first be identified and accessed. This may require calling APIs or using web crawlers to extract information. Automated data collection tools can periodically extract product information from specified web pages or databases to ensure the timeliness and accuracy of the information. When acquiring product text information, the system must ensure that the extracted content conforms to a unified format for subsequent processing. After acquiring the relevant text information, this data is cleaned and standardized. The text may contain inconsistencies in fields, duplicate information, spelling errors, etc., so deduplication and formatting are performed. After this process, the system will obtain a clean, structured set of product description information for subsequent analysis.

[0106] As described in step S102 above, multilingual semantic parsing technology is used to analyze the previously acquired text description information to extract the features of the goods. Since goods to be classified may be described in different languages ​​in global trade, multilingual processing technology is needed to ensure accurate identification and extraction of product features. Specifically, firstly, the system preprocesses the text description, including removing stop words, punctuation marks, special characters, etc., and then performs word segmentation. Advanced Natural Language Processing (NLP) models (such as BERT, Transformer, etc.) are used for semantic understanding and feature extraction. These models can identify key features in the product description, such as "environmentally friendly materials," "battery capacity," and "usage method," and map them to the corresponding feature sets. In a multilingual environment, semantic parsing also needs to consider cultural and grammatical differences between languages. Multilingual models are trained to handle various local dialects and industry terms to improve the accuracy and robustness of feature extraction. After this process, the final product feature set will be formatted as structured data, providing necessary input information for subsequent classification decisions and the construction of regional preference knowledge graphs. In one specific embodiment, the multilingual semantic parsing technique is only used for the BERT multilingual model or the Transformer model, without any other processing.

[0107] In one embodiment, the step S1 of obtaining multiple preset dimensions of product features of the product to be classified and the regional preference knowledge graph of the classification region includes:

[0108] S111: Perform 3D modeling on the product to be classified to obtain a point cloud 3D model of the product to be classified;

[0109] S112: Model the three-dimensional model of the point cloud using a preset point cloud analysis method to identify the hidden functional components of the product to be classified;

[0110] S113: Perform feature analysis on the hidden functional components to obtain the product features of the product to be classified.

[0111] As described in step S111 above, a 3D model of the product to be classified is performed to obtain a point cloud 3D model of the product. This process typically employs advanced computer vision and image processing technologies to obtain the 3D point cloud model of the product. First, multiple angle images of the product are captured using multi-view image acquisition devices (such as high-definition cameras or laser scanners). These images are used to generate a dataset containing a large number of coordinate points, i.e., the so-called point cloud. During this process, the system uses methods such as structured light scanning, stereo vision, or laser scanning to acquire the required data. The point cloud is acquired by a range sensor (such as LiDAR) or a depth camera, and it represents the geometric data of the object's surface. This modeling technology can provide information about the product's size, shape, and structure, surpassing traditional 2D image analysis. By applying image processing technology, the acquired multiple images are transformed into a 3D spatial model, ultimately forming a point cloud representation.

[0112] As described in step S112 above, the three-dimensional model of the point cloud is modeled using a preset point cloud analysis method to identify the hidden functional components of the product to be classified. The point cloud analysis process typically includes multiple techniques such as feature extraction and geometric analysis. For example, the system can apply algorithms based on normal estimation and clustering to identify high-density regions in the point cloud, which may correspond to hidden component functions. The preset point cloud analysis method can be PointNet, clustering algorithms, etc. Machine learning models, such as Support Vector Machines (SVM) or Convolutional Neural Networks (CNNs) in deep learning, can be used to train the point cloud data, teaching the model how to identify function-related features from the point cloud. Furthermore, by using geometric shape comparison and model matching techniques, the system can compare against a known design model database to further confirm the existence of hidden components.

[0113] As described in step S113 above, feature analysis is performed on the identified hidden functional components to obtain a complete feature set for the product to be classified. The focus of this process is on in-depth interpretation and feature extraction of the hidden components, ensuring that all relevant information about the physical product can be mapped to classifiable product features. First, the system receives the geometric and physical data of the hidden functional components, such as size, location, and material type. This information helps the system better understand the functionality and market value of the product. Next, by applying feature extraction algorithms, the system can analyze the functional attributes of the components, such as electrical connections, compatibility (e.g., USB interface), and functional attributes (e.g., power switch location). To further enrich the product features, the analysis results also include information on the relevance and compliance requirements of the product design. For example, a certain feature may require certification according to relevant regulations, such as the RoHS standard (Restriction of Hazardous Substances Directive) required in Western markets. Finally, through these steps, the system obtains a comprehensive product feature set, which not only provides sufficient basis for subsequent product classification but also reflects the functional requirements and market positioning of the product. This feature data will be stored, analyzed, and used to update the product classification knowledge base, making future product classification more efficient and accurate.

[0114] In one specific embodiment, text is used preferentially to identify product features, and 3D modeling is enabled to identify product features if the text is missing or blurry.

[0115] In one embodiment, before step S3, which involves inputting each of the product features and their corresponding weight coefficients into a preset classification decision forest for encoding and classification to obtain the HS code of the product to be classified in the classification region, the method further includes:

[0116] S201: Obtain the regulatory update information for the classification area;

[0117] S202: Update the initial classification decision forest according to the updated regulatory information to obtain the preset classification decision forest.

[0118] As described in step S201 above, regulatory update information for the classification region is obtained. Because commodity regulations in different regions around the world are constantly evolving, the relevant standards, quality control, and safety certification requirements also change accordingly. Therefore, continuously tracking and obtaining local regulatory updates is crucial to ensuring the compliance of commodity classification. First, the system needs to access dedicated data sources, such as official websites of various countries, industry associations, standardization organizations, or even real-time updated regulatory databases. These resources can provide the latest regulatory texts, newly released policies, and new requirements for specific commodity categories. The system needs to automatically capture this information periodically (e.g., daily or weekly) to ensure the timeliness and accuracy of updates. Next, the collected regulatory update information needs to be structured. Using natural language processing technology, the regulatory text can be parsed to extract key information such as key clauses, time points, and scope of application. By establishing a regulatory update log, the system can clearly record when and what regulations have changed and identify key points with significant impact. This information provides the necessary basic data for subsequent updates to the classification decision forest, ensuring that it can reflect actual regulatory requirements in real time and improve the compliance and market adaptability of commodities.

[0119] As described in step S202 above, the initial classification decision forest is updated according to the updated regulatory information to obtain the preset classification decision forest. During the decision forest update process, the updated regulatory information needs to be matched with the existing decision tree. Specifically, it's necessary to identify which features or nodes are directly affected by the new regulations. This may involve analyzing each node in the decision tree to ensure the update targets specific features. For example, if a new regulation requires increasing the proportion of recycled materials in a certain type of product, the system needs to update the branching conditions of nodes related to that material feature. Based on the updated information, the system will readjust the weights or partitioning thresholds in the decision forest. The update operation may involve merging, adding new tree structures, or modifying existing nodes. At this time, consistency in the update logic should be ensured to avoid unnecessary complexity or errors caused by over-adjustment of the decision tree. At the end of the update process, the system needs to create a new version of the decision forest and retain a backup of the original version for restoration if necessary. Finally, the resulting new preset classification decision forest will serve as the basis for subsequent product feature classification. It can accurately capture the latest regulatory dynamics and ensure the compliance and adaptability of products in the target market.

[0120] In one embodiment, before step S1 of obtaining product features of multiple preset dimensions of the product to be classified and the regional preference knowledge graph of the classification region, the following steps are included:

[0121] S011: Obtain the regional specification information of the classification area;

[0122] S012: Detect whether the goods to be classified comply with the requirements of the area regulations;

[0123] S013: If the goods to be classified do not comply with the requirements of the regional regulations, the goods to be classified shall be identified as non-compliant goods of the classification region.

[0124] As described in step S011 above, regional regulatory information for the classified region is obtained. This information includes factors such as stability, security, and economic policies within a country or region, which directly impact the compliance and market access of goods. With the increasing complexity of cross-border trade, the acceptance of foreign goods by various countries is not only restricted by laws and regulations but also influenced by regional regulatory situations. First, the system needs to collect regional regulatory information from multiple reliable data sources, including global reports, documents published in specific regions, or analysis reports provided by market research companies. These data sources can comprehensively present information related to the current sales environment, credit risk, trade restrictions, and sanctions lists. Second, the collected regional regulatory information is structured and analyzed to extract specific content related to the product classification. This may involve identifying specific policies, market access barriers, or reflecting the preferences or restrictions of specific countries and regions on a particular type of product. By integrating this information into the system, support can be provided for subsequent product compliance checks, thereby helping companies conduct risk assessments and responses, ensuring that business activities in specific regions comply with regional regulations through risk prediction and avoidance strategies.

[0125] As described in step S012 above, it is checked whether the goods to be classified comply with the regulations of the region. With changes in the relationship between regions, some goods may be restricted or prohibited due to regional regulations, making compliance testing crucial. Various characteristic information of the goods to be classified, such as material, purpose, country of origin, and manufacturer, are extracted. This information is then compared with the current regional regulations to determine its compliance.

[0126] As described in step S013 above, if the goods to be classified do not comply with the regulations of the region, then the goods to be classified are identified as non-compliant goods in the classification region. If the goods to be classified do not comply with the relevant regional regulations, then the goods need to be identified as non-compliant goods and processed accordingly. Furthermore, the non-compliance status of the goods can be recorded, and a detailed compliance report can be generated. The report will list in detail the specific reasons for non-compliance, such as country of origin, materials, supply chain risks, or technological limitations.

[0127] Reference Figure 3 The present invention also provides a regional classification device for HS codes of commodities, the device comprising:

[0128] The product feature acquisition module 902 is used to acquire product features of multiple preset dimensions of the product to be classified, as well as the regional preference knowledge graph of the classification area;

[0129] The weight coefficient acquisition module 904 is used to acquire the weight coefficients of each product feature based on the regional preference knowledge graph.

[0130] The classification coding acquisition module 906 is used to input each of the product features and their corresponding weight coefficients into a preset classification decision forest for coding and classification, so as to obtain the HS code of the product to be classified in the classification area.

[0131] The adjudication database acquisition module 908 is used to acquire the regional regulatory information database and adjudication database of the classification area;

[0132] The association extraction module 910 is used to extract entities and the association relationships between entities from the regional legal information database and the adjudication database.

[0133] The preference knowledge graph acquisition module 912 is used to update the initial feature association knowledge graph based on the extracted entities and the relationships between entities, so as to obtain the regional preference knowledge graph.

[0134] In one embodiment, the preference knowledge graph acquisition module 912 includes:

[0135] The weight coefficient node acquisition submodule is used to acquire the weight coefficient nodes in the initial feature association knowledge graph corresponding to each entity; wherein, the weight coefficient nodes in the initial feature association knowledge graph store the basic weight coefficients of the corresponding product features;

[0136] The node increment calculation submodule is used to calculate the node increment of the corresponding weight coefficient node according to the association relationship of each entity.

[0137] The preference knowledge graph acquisition submodule is used to update the initial feature association knowledge graph according to the node increment to obtain the region preference knowledge graph.

[0138] In one embodiment, the HS code regional classification device for goods further includes:

[0139] The derivation module is used to reverse-engineer product features of various preset dimensions from a preset coding system;

[0140] The regulatory text data acquisition module is used to acquire product description text sets and regulatory text data;

[0141] The frequency factor extraction module is used to extract frequency factors of each product feature based on the product description text set, and to extract regulatory factors of each product feature based on the regulatory text data.

[0142] The basic weight coefficient calculation module is used to calculate the basic weight coefficient of each commodity feature based on the frequency factor and regulatory factor.

[0143] A construction module is used to construct the initial feature association knowledge graph based on each of the product features and its corresponding basic weight coefficients.

[0144] In one embodiment, the product feature acquisition module 902 includes:

[0145] The text description information submodule is used to obtain the text description information of the goods to be classified;

[0146] The parsing submodule is used to perform multilingual semantic parsing on the text description information to obtain the product features of the product to be classified.

[0147] In one embodiment, the product feature acquisition module 902 includes:

[0148] The point cloud 3D model submodule is used to perform 3D modeling of the product to be classified, and obtain the point cloud 3D model of the product to be classified.

[0149] The identification submodule is used to model the three-dimensional model of the point cloud using a preset point cloud analysis method in order to identify the hidden functional components of the product to be classified.

[0150] The product feature acquisition submodule is used to perform feature analysis on the hidden functional components to obtain the product features of the product to be classified.

[0151] In one embodiment, the HS code regional classification device for goods further includes:

[0152] The regulatory update information acquisition module is used to acquire regulatory update information for the classification area;

[0153] The preset classification decision forest acquisition module is used to update the initial classification decision forest according to the updated regulatory information to obtain the preset classification decision forest.

[0154] In one embodiment, the HS code regional classification device for goods further includes:

[0155] The regional specification information acquisition module is used to acquire the regional specification information of the classified regions;

[0156] The detection module is used to detect whether the goods to be classified conform to the requirements of the area regulations.

[0157] The non-compliant product identification module is used to identify the product to be classified as a non-compliant product of the classification area if the product to be classified does not comply with the regulations of the area.

[0158] Figure 4 An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a regional classification method for HS codes of goods. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement a regional classification method for HS codes of goods. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0159] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps:

[0160] Obtain product features of the products to be classified from multiple preset dimensions, as well as a regional preference knowledge graph of the classification area;

[0161] The weight coefficients of each product feature are obtained based on the regional preference knowledge graph.

[0162] Each of the product features and its corresponding weight coefficients are input into a preset classification decision forest for encoding and classification, thereby obtaining the HS code of the product to be classified in the classification region.

[0163] Before the steps of obtaining product features of multiple preset dimensions of the products to be classified and the regional preference knowledge graph of the classification area, the method further includes:

[0164] Obtain the regional regulatory information database and adjudication database for the classified area;

[0165] Extract entities and relationships between entities from the regional regulatory information database and the adjudication database;

[0166] The initial feature association knowledge graph is updated based on the extracted entities and the relationships between them to obtain the region preference knowledge graph.

[0167] By combining regional preference knowledge graphs, precise weight coefficients are assigned to each product feature, and the initial feature association knowledge graph is dynamically updated. The updated feature association knowledge graph is then used to obtain the weight coefficients of the product features, thereby obtaining the classification code of the product in the classification region. This improves the accuracy of product classification in various regions, thereby increasing customs clearance efficiency for enterprises, reducing legal risks caused by improper classification, and optimizing the overall efficiency of cross-border transactions.

[0168] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following steps:

[0169] Obtain product features of the products to be classified from multiple preset dimensions, as well as a regional preference knowledge graph of the classification area;

[0170] The weight coefficients of each product feature are obtained based on the regional preference knowledge graph.

[0171] Each of the product features and its corresponding weight coefficients are input into a preset classification decision forest for encoding and classification, thereby obtaining the HS code of the product to be classified in the classification region.

[0172] Before the steps of obtaining product features of multiple preset dimensions of the products to be classified and the regional preference knowledge graph of the classification area, the method further includes:

[0173] Obtain the regional regulatory information database and adjudication database for the classified area;

[0174] Extract entities and relationships between entities from the regional regulatory information database and the adjudication database;

[0175] The initial feature association knowledge graph is updated based on the extracted entities and the relationships between them to obtain the region preference knowledge graph.

[0176] By combining regional preference knowledge graphs, precise weight coefficients are assigned to each product feature, and the initial feature association knowledge graph is dynamically updated. The updated feature association knowledge graph is then used to obtain the weight coefficients of the product features, thereby obtaining the classification code of the product in the classification region. This improves the accuracy of product classification in various regions, thereby increasing customs clearance efficiency for enterprises, reducing legal risks caused by improper classification, and optimizing the overall efficiency of cross-border transactions.

[0177] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, 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. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0178] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0179] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for regional classification of HS codes for commodities, characterized in that, The method includes: Obtain product features of the products to be classified from multiple preset dimensions, as well as a regional preference knowledge graph of the classification area; The weight coefficients of each product feature are obtained based on the regional preference knowledge graph. Each of the product features and its corresponding weight coefficients are input into a preset classification decision forest for encoding and classification, thereby obtaining the HS code of the product to be classified in the classification region. Before the steps of obtaining product features of multiple preset dimensions of the products to be classified and the regional preference knowledge graph of the classification area, the method further includes: Obtain the regional regulatory information database and adjudication database for the classified area; Extract entities and relationships between entities from the regional regulatory information database and the adjudication database; The initial feature association knowledge graph is updated based on the extracted entities and the relationships between them to obtain the region preference knowledge graph.

2. The method for regional classification of HS codes for commodities according to claim 1, characterized in that, The step of updating the initial feature association knowledge graph based on the extracted entities and the relationships between entities to obtain the region preference knowledge graph includes: Obtain the weight coefficient nodes in the initial feature association knowledge graph corresponding to each entity; wherein, the weight coefficient nodes in the initial feature association knowledge graph store the basic weight coefficients of the corresponding product features; Calculate the node increment of the corresponding weight coefficient node based on the association relationship of each entity; The initial feature association knowledge graph is updated based on the node increments to obtain the region preference knowledge graph.

3. The method for regional classification of goods using HS codes according to claim 1, characterized in that, Before the step of updating the initial feature association knowledge graph based on the extracted entities and the relationships between entities to obtain the region preference knowledge graph, the method further includes: The product characteristics of each preset dimension are derived from the preset coding system. Obtain product description text sets and regulatory text data; Frequency factors of each product feature are extracted from the product description text set, and regulatory factors of each product feature are extracted from the regulatory text data. Calculate the basic weight coefficients for each commodity feature based on the frequency factor and regulatory factor; The initial feature association knowledge graph is constructed based on each of the product features and its corresponding basic weight coefficients.

4. The method for regional classification of goods using HS codes according to claim 1, characterized in that, The step of obtaining multiple preset dimensions of product features for the products to be classified, and the regional preference knowledge graph of the classification region, includes the following steps: Obtain the text description information of the product to be classified; Multilingual semantic parsing is performed on the text description information to obtain the product characteristics of the product to be classified.

5. The method for regional classification of commodities using HS codes according to claim 1, characterized in that, The step of obtaining multiple preset dimensions of product features for the products to be classified, and the regional preference knowledge graph of the classification region, includes the following steps: The product to be classified is modeled in three dimensions to obtain a point cloud three-dimensional model of the product to be classified. The point cloud 3D model is modeled using a preset point cloud analysis method to identify the hidden functional components of the goods to be classified. Feature analysis is performed on the hidden functional components to obtain the product features of the product to be classified.

6. The method for regional classification of HS codes for commodities according to claim 1, characterized in that, Before the step of inputting each of the product features and their corresponding weight coefficients into a preset classification decision forest for encoding and classification to obtain the HS code of the product to be classified in the classification region, the method further includes: Obtain the updated regulatory information for the classification area; The initial classification decision forest is updated based on the updated regulatory information to obtain the preset classification decision forest.

7. The method for regional classification of HS codes for commodities according to claim 1, characterized in that, Before the steps of obtaining product features of multiple preset dimensions of the products to be classified and the regional preference knowledge graph of the classification area, the method further includes: Obtain the regional definition information of the classification area; Detect whether the goods to be classified comply with the regulations of the area; If the goods to be classified do not comply with the requirements of the area regulations, the goods to be classified will be identified as non-compliant goods in the classification area.

8. A regional classification device for HS codes of commodities, characterized in that, The device includes: The product feature acquisition module is used to acquire product features of the products to be classified in multiple preset dimensions, as well as the regional preference knowledge graph of the classification area; The weight coefficient acquisition module is used to acquire the weight coefficients of each product feature based on the regional preference knowledge graph. The classification coding acquisition module is used to input each of the product features and their corresponding weight coefficients into a preset classification decision forest for coding and classification, so as to obtain the HS code of the product to be classified in the classification area. The adjudication database acquisition module is used to acquire the regional regulatory information database and adjudication database of the classified area; The relationship extraction module is used to extract entities and the relationships between entities from the regional legal information database and the adjudication database. The preference knowledge graph acquisition module is used to update the initial feature association knowledge graph based on the extracted entities and the relationships between entities, so as to obtain the regional preference knowledge graph.

9. A computer-readable storage medium, characterized in that, The computer program is stored thereon, which, when executed by a processor, causes the processor to perform the steps of the HS coding regional classification method for the goods as claimed in any one of claims 1 to 7.

10. An electronic device, characterized in that, The device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the HS coding regional classification method for the goods as claimed in any one of claims 1 to 7.

Images