Method and device for realizing progressive convergence HS classification based on large model semantic understanding, processor and computer readable storage medium thereof

Abstract

The invention relates to a method for realizing progressive convergence HS classification based on large model semantic understanding. The method comprises the following steps: determining classes in 21 large classes in an HS system; determining a specific chapter in 97 chapters in the HS system; determining specific items belonging to the confirmed chapters in the HS system; determining specific sub-items in the confirmed items belonging to the HS system; and determining a specific HS code in a confirmation subcategory belonging to the HS system. By adopting the method and the device for realizing progressive convergence HS classification based on large model semantic understanding, the processor and the computer readable storage medium thereof, the problem that classification cannot be realized due to lack of historical data based on a traditional historical data matching method is solved, the natural multi-language ability of a large model is utilized, and the classification efficiency is improved. The HS classification problem of all countries in the world is solved, the accuracy and timeliness of HS classification are greatly improved through verification of actual application scenes, and the result is quicker and more accurate than that of experienced manual classifiers especially in the scenes with classification disputes.

Description

Technical Field

[0001] This invention relates to the field of international trade, and more particularly to the field of intelligent classification. Specifically, it refers to a method, apparatus, processor, and computer-readable storage medium for progressively convergent HS classification based on large-model semantic understanding. Background Technology

[0002] In international trade exports, goods must undergo Chinese export customs declaration and import customs clearance in the destination country before reaching the port. During customs clearance, customs authorities in various countries require HS code classification and compliance with their own import requirements. The HS code, also known as the Harmonization System Code (HS-Code), was created by the World Customs Organization and is widely recognized internationally. Each HS code consists of 6 digits. The first two digits indicate the chapter to which the HS code belongs (there are 21 chapters in total, each providing a description summarizing the corresponding commodity category). The next four digits consist of the chapter title and subtitle. The 6-digit HS code is the internationally accepted code. Each country further subdivides this 6-digit HS code to form its final HS code definition. For example, China uses a 10-digit HS code; the United States also uses a 10-digit HS code. The classification methods for the last four digits of the HS code vary among countries.

[0003] With the continuous development of global trade and the increasing variety of goods, traditional manual classification methods are no longer sufficient to meet the rapidly growing trade demands. Intelligent classification technology, through automated processes, reduces repetitive manual work and improves classification efficiency. Intelligent classification systems can reduce human error, improve the accuracy of commodity classification, and help companies comply with international trade rules and tariff policies. The development of artificial intelligence and machine learning technologies provides the technological foundation for intelligent classification, enabling the system to optimize classification algorithms by learning from large amounts of commodity data. The implementation of intelligent classification systems can facilitate data sharing, improving the declaration quality and compliance management level of the entire industry. Although the overall HS coding framework is internationally universal, differences may exist in specific coding applications across different countries. Humans cannot fully understand these differences. Intelligent systems can learn from the coding rule differences of various countries and make decisions on behalf of humans to adapt to the trade environments of different countries. Furthermore, the internationally universal coding framework provides strong prior knowledge and technical feasibility for intelligent global commodity coding mapping systems.

[0004] Currently, there are several intelligent product coding and classification systems on the market. These systems can automatically match product names with corresponding product classification codes. Most of these intelligent classification models use search engines based on historical declaration data to perform keyword searches and achieve automatic code matching. Other intelligent coding and classification systems employ more advanced technical architectures, such as knowledge graphs and natural language processing (the technical details are largely similar, using Chinese word segmentation and machine learning algorithms like TextRank and BERT to calculate semantic similarity between the product name to be classified and the data in the classification history database to find semantically similar historical data to provide classification results), to improve the intelligence level of the coding process. However, these intelligent classification coding systems all rely on existing historical data. Once the HS coding system is updated (several times a year), the historical data needs to be cleaned, which is time-consuming and laborious. For newly added HS codes, due to the lack of corresponding historical data, the classification system based on search engine technology cannot map products to the new HS codes, requiring lengthy manual adjustments and labeling before the new HS codes can be classified. On the other hand, while it is technically feasible to periodically clean historical data on the mapping relationship between HS codes and commodities for domestic exports due to data accumulation, it is difficult to collect comprehensive data on the mapping relationship between HS codes and commodities for all destination countries. This is because each country has a different system after the 6-digit HS code, and the mapping relationship between HS codes and commodities for each destination country is scattered across their respective systems. Therefore, the effectiveness of HS classification systems based on traditional search engine technology for HS classification of destination countries is greatly reduced, and currently, manual classification by customs clearance professionals is often still required. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, apparatus, processor and computer-readable storage medium for progressive convergence HS classification based on large model semantic understanding that meets the requirements of high accuracy, high precision and wide applicability.

[0006] To achieve the above objectives, the present invention provides a method, apparatus, processor, and computer-readable storage medium for progressively convergent HS classification based on large-model semantic understanding, as follows: The main feature of this method for progressively convergent HS classification based on large-model semantic understanding is that the method includes the following steps: (1) Determine the categories among the 21 major categories in the HS system; (2) Identify the specific chapters among the 97 chapters in the HS system; (3) Identify the specific items that belong to the section in step (2) of the HS system; (4) Determine the specific subheading in the item identified in step (3) of the HS system; (5) Determine the specific HS code in the sub-item confirmed in step (4) of the HS system.

[0007] Preferably, step (3) specifically includes the following steps: (3.1) Query the database to confirm all item codes and item descriptions under the chapter in step (2); (3.2) Take all the item codes and item descriptions under the confirmed chapter in step (2) as input, and determine the HS item code by matching the corresponding item to which the item belongs through the large model.

[0008] Preferably, step (4) specifically includes the following steps: (4.1) Query the database to confirm all subheading codes and descriptions under the heading in step (3); (4.2) Take all the subheading codes and descriptions under the confirmed item in step (3) as input, and determine the HS subheading code by matching the corresponding subheading to which the item belongs through the large model.

[0009] Preferably, step (5) specifically includes the following steps: (5.1) Query the database to confirm all HS codes and descriptions under the subheading in step (4); (5.2) Take all the HS codes and descriptions under the confirmed sub-items in step (4) as input, and match the items to the specific HS codes through the large model.

[0010] Preferably, the HS item code in step (3.2) is a 4-digit code.

[0011] Preferably, the HS sub-item encoding in step (4.2) is a 6-bit encoding.

[0012] Preferably, the HS encoding in step (5.2) is a 10-bit encoding.

[0013] The main feature of this device for progressively convergent HS classification based on large-model semantic understanding is that the device includes: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the steps of the above-described method for progressively convergent HS classification based on large model semantic understanding.

[0014] The processor for progressive convergence HS classification based on large model semantic understanding is characterized in that the processor is configured to execute computer-executable instructions, and when the computer-executable instructions are executed by the processor, the various steps of the above-mentioned method for progressive convergence HS classification based on large model semantic understanding are implemented.

[0015] The main feature of this computer-readable storage medium is that it stores a computer program thereon, which can be executed by a processor to implement the various steps of the above-described method for progressively convergent HS classification based on large model semantic understanding.

[0016] This invention employs a method, apparatus, processor, and computer-readable storage medium for progressive convergence HS classification based on large-model semantic understanding. Leveraging the powerful semantic understanding capabilities of large-model systems, it achieves HS mapping matching, overcoming the limitations of traditional search engine-based technologies that rely on large amounts of historical data and repeated historical data cleaning to cope with changes in the HS system. For newly added HS codes, it solves the problem of traditional historical data-based matching methods being unable to classify due to a lack of historical data. The top-down, hierarchical progressive convergence matching approach avoids the problem of unsatisfactory single-match accuracy caused by an overly large HS system context, and also avoids the cost of consuming a large number of large-model tokens for a single HS system match. Utilizing the inherent multilingual capabilities of large-model systems, it solves the HS classification problem for countries worldwide, avoiding the translation preprocessing issues required for traditional NLP similarity word matching. Verified through practical applications, this method can significantly improve the accuracy and timeliness of HS classification, especially in scenarios with classification disputes, achieving faster and more accurate results than experienced human classifiers. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for progressively convergent HS classification based on large model semantic understanding according to the present invention. Detailed Implementation

[0018] To more clearly describe the technical content of the present invention, the following description is provided in conjunction with specific embodiments.

[0019] The present invention provides a method for progressively convergent HS classification based on large model semantic understanding, comprising the following steps: (1) Determine the categories among the 21 major categories in the HS system; (2) Identify the specific chapters among the 97 chapters in the HS system; (3) Identify the specific items that belong to the section in step (2) of the HS system; (4) Determine the specific subheading in the item identified in step (3) of the HS system; (5) Determine the specific HS code in the sub-item confirmed in step (4) of the HS system.

[0020] In a preferred embodiment of the present invention, step (3) specifically includes the following steps: (3.1) Query the database to confirm all item codes and item descriptions under the chapter in step (2); (3.2) Take all the item codes and item descriptions under the confirmed chapter in step (2) as input, and determine the HS item code by matching the corresponding item to which the item belongs through the large model.

[0021] In a preferred embodiment of the present invention, step (4) specifically includes the following steps: (4.1) Query the database to confirm all subheading codes and descriptions under the heading in step (3); (4.2) Take all the subheading codes and descriptions under the confirmed item in step (3) as input, and determine the HS subheading code by matching the corresponding subheading to which the item belongs through the large model.

[0022] In a preferred embodiment of the present invention, step (5) specifically includes the following steps: (5.1) Query the database to confirm all HS codes and descriptions under the subheading in step (4); (5.2) Take all the HS codes and descriptions under the confirmed sub-items in step (4) as input, and match the items to the specific HS codes through the large model.

[0023] In a preferred embodiment of the present invention, the HS item code in step (3.2) is a 4-digit code.

[0024] In a preferred embodiment of the present invention, the HS sub-item code in step (4.2) is a 6-bit code.

[0025] In a preferred embodiment of the present invention, the HS encoding in step (5.2) is a 10-bit encoding.

[0026] The apparatus for progressively convergent HS classification based on large model semantic understanding of the present invention includes: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the steps of the above-described method for progressively convergent HS classification based on large model semantic understanding.

[0027] The processor of the present invention for progressive convergence HS classification based on large model semantic understanding is configured to execute computer-executable instructions, which, when executed by the processor, implement the various steps of the above-described method for progressive convergence HS classification based on large model semantic understanding.

[0028] The computer-readable storage medium of the present invention stores a computer program that can be executed by a processor to implement the various steps of the above-described method for progressively convergent HS classification based on large model semantic understanding.

[0029] To address the problems described in the background, this invention proposes a novel asymptotic convergence method based on large-scale model semantic understanding for automatic HS code classification of any country globally. A key feature is that it does not rely on large amounts of historical data, but directly converges hierarchically from the HS tariff definitions of the target country until the final HS code is obtained. This solves the problem of HS code classification for traders during import and export processes, especially for overseas HS codes. Furthermore, the large-scale model used in this invention is not limited to a specific model; most language large-scale models can achieve semantic matching functionality.

[0030] The specific principle is as follows: Each country's HS coding system is organized hierarchically according to categories (21 categories in total), chapters (97 chapters in total), headings / subheadings (thousands), and specific HS codes (tens of thousands). This invention is based on this hierarchical structure, leveraging the semantic understanding capabilities of a large model to gradually converge from top to bottom to determine the final HS code. This differs from the traditional method that requires a large amount of historical classification data for semantic matching of product names through machine learning. It can completely "abandon" the historical database, eliminating the need to clean / build a historical database from scratch, and also avoiding the time-consuming, laborious, and error-prone data cleaning work when the HS system changes.

[0031] A key challenge of top-down hierarchical classification is that the HS description is a general overview of a group of goods, rather than a specific product name like "wooden office desk." Therefore, traditional machine learning and NLP algorithms cannot accurately map to specific HS descriptions. This is why traditional HS classification methods rely on extensive historical data for semantic matching (e.g., using historical data like "solid wood office desk" or "wooden office desk" to find the HS code for "solid wood office desk" or "wooden office desk"). However, with the increasing maturity of large-scale modeling technology, current models, which approach or even surpass human semantic understanding in some aspects, can narrow down the scope of "wooden office desk" to match it to category 9 "wood products" or category 20 "furniture," achieving high accuracy. (For ease of understanding, all 21 categories of HS descriptions are listed below; however, due to space limitations, the chapters, headings / subheadings involved are not listed further. For specific HS content, please refer to the World Customs Organization (WCO) website.) Based on the same logic, after locating the specific category within the 21 major categories using the semantic understanding capabilities of the large model, the HS descriptions of all chapters under the corresponding category and the product name to be classified can be matched to the specific chapter again using the semantic understanding capabilities of the large model. Similarly, this process is gradually refined to the item / sub-item level, finally matching the specific HS code. Specifically, by designing specific prompts, the product name to be classified and the HS code description of the target level are input into the large model, which is then required to perform matching and selection based on semantic relevance.

[0032] In specific embodiments of the present invention, the following examples illustrate how to utilize the semantic understanding capabilities of large models to achieve asymptotically convergent HS classification.

[0033] Step 1: Determine the category among the 21 major categories of the HS system; Taking the classification of "wooden office desks" in the HS coding system as an example: The HS coding system has 21 sections, which are divided according to major product categories, covering a very broad range. For example: Section I is live animals; animal products.

[0034] Section XI is textile raw materials and textile products.

[0035] Section XVI covers machines, mechanical appliances, electrical equipment and parts thereof. Wooden desks fall under the category of "furniture," which is classified under Section XX (Class 20).

[0036] Step 2: Identify the specific chapters within the 97 chapters of the HS framework; After the large model is located in the 20th category, the wooden table belongs to Chapter 94 of the 20th category (Section XX).

[0037] Step 3: Determine which heading in Chapter 94 of the HS scheme it belongs to; 3.1. Query all item codes and descriptions under Chapter 94 in the database; 3.2. Take all the item codes and descriptions under Chapter 94 as input, and let the large model use its powerful semantic understanding capabilities to match which item "wooden office desk" belongs to.

[0038] Specifically: Determine the 4-digit code for the HS item based on the given product name and HS range.

[0039] Taking wooden office desks as an example, based on the product name "wooden office desks" and the given HS range, its 4-digit HS code is 9403.

[0040] Step 4: Determine which subheading (4-digit code) under heading 9403 it belongs to; 4.1. Query all subheading codes (6-digit HS codes) and descriptions under heading 9403 in the database; 4.2. Take all subheading codes and descriptions under heading 9403 as input, and let the large model use its powerful semantic understanding capabilities to match which subheading "wooden office desk" belongs to.

[0041] Specifically: Determine the 6-digit code of the HS subheading based on the given product name and HS range.

[0042] Taking wooden office desks as an example, based on the product name "wooden office desks" and the given HS range, its 6-digit HS subheading code is 940330.

[0043] Step 5: Determine the specific 10-digit HS code belonging to subheading 940330; 5.1. Query all 10-digit HS codes and descriptions under subheading 940330 in the database; 5.2. Take all 10-digit HS codes and descriptions under subheading 940330 as input, and let the large model use its powerful semantic understanding capabilities to match "wooden office desk" to the specific 10-digit HS code.

[0044] Specifically: Determine the 10-digit code of the HS based on the given product name and HS range.

[0045] Taking a wooden office desk as an example, based on the product name "wooden office desk" and the given HS range, its correct 10-digit HS code is 9403300090.

[0046] This concludes the demonstration, through a concrete example, of how to leverage the powerful semantic understanding capabilities of large models to progressively converge at different levels to obtain the final HS encoding. The flowchart is as follows: Figure 1 As shown.

[0047] Taking the HS system, which has 21 categories, as an example, each category has a clear definition and description, as follows: 1. Category 1: Live animals; animal products (Chapters 1 to 5) This category includes live animals, meat, fish, dairy products, eggs, natural honey, and other unlisted edible and non-edible animal products.

[0048] 2. Category Two: Plant Products (Chapters 6 to 14) This category includes live plants, vegetables, edible fruits and nuts, coffee, tea, spices, grains, oilseeds, and industrial or medicinal plants.

[0049] 3. Category Three: Animal and vegetable oils and fats and their decomposition products; refined edible oils; animal and vegetable waxes (Chapter 15) This category specifically classifies fats, oils, and waxes from animal or plant sources.

[0050] 4. Class 4: Food; beverages, alcoholic beverages and vinegar; tobacco, tobacco and articles thereof (Chapters 16 to 24) This is a category of processed foods and beverages, including meat, fish, sugar products, cocoa, grain products (such as bread and pastries), processed fruits and vegetables, various beverages, alcohol, vinegar, and tobacco products.

[0051] 5. Category Five: Mineral Products (Chapters 25 to 27) This category includes salt, sulfur, soil and stone, mineral sands, slag, and mineral fuels (such as coal, oil, and natural gas).

[0052] 6. Category VI: Products of the chemical industry and related industries (Chapters 28 to 38) This is a very large category, covering inorganic chemicals, organic chemicals, pharmaceuticals, fertilizers, dyes, pigments, fragrances, cosmetics, soaps, explosives, plastics (in raw material form), etc.

[0053] 7. Class 7: Plastics and articles thereof; Rubber and articles thereof (Chapters 39 to 40) This category includes primary-shaped products of plastics and rubber, as well as their finished products, such as plastic sheets, tubes, bottles, and rubber tires, gloves, etc.

[0054] 8. Class 8: Raw hides, leather, furs and their products; saddles and harnesses; travel goods, handbags and similar containers; animal gut products (Chapters 41 to 43) This category includes raw hides, tanned leather, furs, and products made from them, such as bags and clothing.

[0055] 9. Class 9: Wood and wood products; charcoal; cork and cork products; products made of straw, stalks, needlegrass or other woven materials; baskets and wickerwork (Chapters 44 to 46) This category includes logs, timber, wood products (such as plywood, wooden parts for furniture), charcoal, and woven products such as baskets.

[0056] 10. Class 10: Wood pulp and other fibrous cellulose pulp; recycled (waste) paper or paperboard; paper, paperboard and articles thereof (Chapters 47 to 49) This category covers the entire paper industry chain, from pulp to various types of paper and cardboard, and then to printed materials such as books and newspapers.

[0057] 11. Category 11: Textile raw materials and textile products (Chapters 50 to 63) This is another very large category, which includes everything from natural fibers (silk, wool, cotton) to chemical fibers, yarns, fabrics, and finally textile products (such as clothing, bedding, and carpets).

[0058] 12. Class XII: Shoes, hats, umbrellas, canes, whips and their parts; processed feathers and their products; artificial flowers; products made from human hair (Chapters 64 to 67) This category mainly categorizes various wearable and handheld accessories, such as shoes, hats, umbrellas, and some decorative items.

[0059] 13. Class Thirteen: Stone, gypsum, cement, asbestos, mica and similar materials; ceramic products; glass and its articles (Chapters 68 to 70) This category includes stone products, ceramic products (such as bricks, tiles, and tableware), and glass products (such as flat glass, glass bottles, and glassware).

[0060] 14. Class Fourteen: Natural or cultured pearls, gemstones or semi-precious stones, precious metals, precious metal-encased products and articles thereof; imitation jewelry; coins (Chapter 71) This category specifically categorizes jewelry, precious metals (gold, silver, platinum) and their products, and coins.

[0061] 15. Class XV: Base metals and their articles (Chapters 72-83) This category covers major industrial metals such as steel, copper, nickel, aluminum, lead, zinc, and tin, as well as their products, such as metal bars, rods, plates, and pipes, and tools, knives, locks, nails, and screws.

[0062] 16. Class 16: Machines, mechanical appliances, electrical equipment and parts thereof; tape recorders and sound reproducers, television image and sound recording and reproducing equipment and parts thereof (Chapters 84 to 85).

[0063] This is the core category of industrial and consumer goods. Chapter 84 mainly covers mechanical equipment (such as engines, pumps, computers, and machine tools), and Chapter 85 mainly covers electrical equipment (such as motors, transformers, batteries, mobile phones, and televisions).

[0064] 17. Class 17: Vehicles, aircraft, ships and related transport equipment (Chapters 86 to 89) This category includes all means of transportation, such as trains, cars, airplanes, ships and their parts.

[0065] 18. Class 18: Optical, photographic, cinematographic, measuring, inspection, medical or surgical instruments and equipment; precision instruments and equipment; clocks and watches; musical instruments (Chapters 90 to 92).

[0066] This category categorizes various precision instruments, equipment, and appliances, such as eyeglasses, cameras, microscopes, medical devices, clocks, and musical instruments.

[0067] 19. Class Nineteen: Weapons, ammunition and their parts and accessories (Chapter 93) This is a specific category used to classify various weapons and ammunition.

[0068] 20. Class 20: Miscellaneous Articles (Chapters 94 to 96) This is a catch-all category that includes a variety of manufactured goods not listed above, such as furniture, lamps, toys, sporting goods, buttons, pens, etc.

[0069] 21. Category 21: Works of art, collectibles and antiquities (Chapter 97) This category is specifically used to classify items with artistic, collectible, or historical value, such as paintings, sculptures, and antiques that are over 100 years old.

[0070] For the specific implementation scheme of this embodiment, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0071] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0072] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0073] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0074] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution device. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0075] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The corresponding program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0076] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a 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 also be stored in a computer-readable storage medium.

[0077] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0078] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0079] This invention employs a method, apparatus, processor, and computer-readable storage medium for progressive convergence HS classification based on large-model semantic understanding. Leveraging the powerful semantic understanding capabilities of large-model systems, it achieves HS mapping matching, overcoming the limitations of traditional search engine-based technologies that rely on large amounts of historical data and repeated historical data cleaning to cope with changes in the HS system. For newly added HS codes, it solves the problem of traditional historical data-based matching methods being unable to classify due to a lack of historical data. The top-down, hierarchical progressive convergence matching approach avoids the problem of unsatisfactory single-match accuracy caused by an overly large HS system context, and also avoids the cost of consuming a large number of large-model tokens for a single HS system match. Utilizing the inherent multilingual capabilities of large-model systems, it solves the HS classification problem for countries worldwide, avoiding the translation preprocessing issues required for traditional NLP similarity word matching. Verified through practical applications, this method can significantly improve the accuracy and timeliness of HS classification, especially in scenarios with classification disputes, achieving faster and more accurate results than experienced human classifiers.

[0080] In this specification, the invention has been described with reference to specific embodiments thereof. However, it will be apparent that various modifications and variations can be made without departing from the spirit and scope of the invention. Therefore, the specification and drawings should be considered illustrative rather than restrictive.

Claims

1. A method for progressively convergent HS classification based on large-model semantic understanding, characterized in that, The method includes the following steps: (1) Determine the categories among the 21 major categories in the HS system; (2) Identify the specific chapters among the 97 chapters in the HS system; (3) Identify the specific items that belong to the section in step (2) of the HS system; (4) Determine the specific subheading in the item identified in step (3) of the HS system; (5) Determine the specific HS code in the sub-item confirmed in step (4) of the HS system.

2. The method for progressively convergent HS classification based on large model semantic understanding according to claim 1, characterized in that, Step (3) specifically includes the following steps: (3.1) Query the database to confirm all item codes and item descriptions under the chapter in step (2); (3.2) Take all the item codes and item descriptions under the confirmed chapter in step (2) as input, and determine the HS item code by matching the corresponding item to which the item belongs through the large model.

3. The method for progressively convergent HS classification based on large model semantic understanding according to claim 1, characterized in that, Step (4) specifically includes the following steps: (4.1) Query the database to confirm all subheading codes and descriptions under the heading in step (3); (4.2) Take all the subheading codes and descriptions under the confirmed item in step (3) as input, and determine the HS subheading code by matching the corresponding subheading to which the item belongs through the large model.

4. The method for progressively convergent HS classification based on large model semantic understanding according to claim 1, characterized in that, Step (5) specifically includes the following steps: (5.1) Query the database to confirm all HS codes and descriptions under the subheading in step (4); (5.2) Take all the HS codes and descriptions under the confirmed sub-items in step (4) as input, and match the items to the specific HS codes through the large model.

5. The method for progressively convergent HS classification based on large model semantic understanding according to claim 2, characterized in that, In step (3.2), the HS item code is a 4-digit code.

6. The method for progressively convergent HS classification based on large model semantic understanding according to claim 3, characterized in that, In step (4.2), the HS sub-item is encoded as a 6-bit code.

7. The method for progressively convergent HS classification based on large model semantic understanding according to claim 4, characterized in that, In step (5.2), the HS encoding is a 10-bit encoding.

8. A device for achieving progressively convergent HS classification based on large-model semantic understanding, characterized in that, The device includes: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the steps of the method for progressively convergent HS classification based on large model semantic understanding as described in any one of claims 1 to 7.

9. A processor for progressively convergent HS classification based on large-model semantic understanding, characterized in that, The processor is configured to execute computer-executable instructions, which, when executed by the processor, implement the steps of the method for progressively convergent HS classification based on large model semantic understanding as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that can be executed by a processor to implement the steps of the method for progressively convergent HS classification based on large model semantic understanding as described in any one of claims 1 to 7.

Images