AI-based recommendation of HS codes or other classification codes

Abstract

AI-based classification code, such as HS code, recommendation. A text prompt related to an item or service is received. Information for classifying the item or service based at least in part on the text prompt is retrieved. An input for a natural language processing model is generated based at least in part on the text prompt and the retrieved information. An output is obtained based at least in part on applying the input to the natural language processing model, the output including at least one recommended classification code for the item. The at least one recommended classification code for the item or service is output. The classification code can include an HS code related to an item.

Description

Technical Field

[0001] This disclosure relates to recommendations for Harmonized System Codes (HS Codes) or other classification codes based on Artificial Intelligence (AI). Background Technology

[0002] Classification systems are used in various fields, including trade, economics, healthcare, and information management. Typically, classification systems use classification codes, such as standardized numeric or alphanumeric codes, to organize or classify entities based on characteristics or attributes of those entities (e.g., goods, products, services, etc.).

[0003] Examples of these classification systems include the HS developed by the World Customs Organization (WCO), the Central Product Classification (CPC) system developed by the United Nations (UN), and the North American Industry Classification System (NAICS). Summary of the Invention

[0004] According to a first aspect of this disclosure, an HS-encoded recommendation system is provided, comprising: one or more processors configured to: receive textual prompts related to an item; retrieve information for classifying the item at least in part based on the textual prompts; generate input for a natural language processing model based at least in part on the textual prompts and the retrieved information; obtain an output based at least in part on applying the input to the natural language processing model; wherein the output includes at least one recommended HS code (HS code) for the item; and output the at least one recommended HS code for the item. The item is a commodity or product.

[0005] The text prompt may include questions or statements (non-questions). In one embodiment of the first aspect, the text prompt includes a description of the item. In one embodiment of the first aspect, the text prompt includes one or more keywords related to the item. In one embodiment of the first aspect, the text prompt also includes countries or regions of interest related to the recommendation.

[0006] The information used to classify the articles is retrieved from an information source. In one embodiment of the first aspect, the information source includes HS code information at least partially indexed by a trie data structure. In one embodiment of the first aspect, the information is retrieved from an information source that includes World Customs Organization (WCO) HS code information. In one embodiment of the first aspect, the information is retrieved at least partially from WCO explanatory notes. In one embodiment of the first aspect, the information is retrieved from an information source that includes one or more country- or region-specific HS code information. In one embodiment of the first aspect, the information source includes country or region tariff or customs information related to country or region-specific HS codes.

[0007] In one embodiment of the first aspect, the natural language processing model includes a large language model (LLM).

[0008] In one embodiment of the first aspect, the one or more processors are configured to: for each of the at least one recommended HS code of an article: retrieve a corresponding description of the recommended HS code of the article; and output the corresponding description of the recommended HS code of the article.

[0009] In one embodiment of the first aspect, the one or more processors are configured to: receive the text prompt based at least in part on processing an image including the item.

[0010] In one embodiment of the first aspect, the recommended HS code is a 6-bit code. The 6-bit code is the internationally standardized HS code.

[0011] In one embodiment of the first aspect, the recommended HS code is an 8-bit code. The 8-bit code is an internationally standardized HS code plus a country / region-specific classification code.

[0012] In one embodiment of the first aspect, the recommended HS code is a 10-digit code. The 10-digit code is an internationally standardized HS code plus a country / region-specific classification code.

[0013] According to a second aspect of this disclosure, an HS-coded recommendation system is provided, comprising: one or more processors configured to: receive text prompts including items or services; retrieve information for classifying the items or services at least in part based on the text prompts; generate input for a large language model based at least in part on the text prompts and the retrieved information; obtain output based at least in part on applying the input to the large language model, the output including at least one recommended classification code for the items or services; and output the at least one recommended classification code for the items or services. The recommended classification code may be a numeric code or an alphanumeric code.

[0014] The text prompt may include questions or statements (non-questions). In one embodiment of the second aspect, the text prompt includes a description of the item or service. In one embodiment of the second aspect, the text prompt includes one or more keywords related to the item or service. In one embodiment of the second aspect, the text prompt includes countries or regions of interest related to the recommendation.

[0015] In one embodiment of the second aspect, the information used to classify the items or services is retrieved from an information source.

[0016] In one embodiment of the second aspect, the information used to classify the items or services is retrieved from an information source, which includes classification code information that is at least partially based on a trie data structure index.

[0017] In one embodiment of the second aspect, the natural language processing model includes an LLM.

[0018] In one embodiment of the second aspect, the one or more processors are configured to: for each of the at least one recommended classification code for an item or service: retrieve a corresponding description of the recommended classification code for the item or service; and output the corresponding description of the recommended classification code for the item or service.

[0019] In one embodiment of the second aspect, the one or more processors are configured to: receive the text prompt based at least in part on processing an image including the item or service.

[0020] According to a third aspect of this disclosure, an HS coding recommendation method is provided, comprising: receiving a text prompt related to an item; retrieving information for classifying the item at least in part based on the text prompt; generating an input for a natural language processing model based at least in part on the text prompt and the retrieved information; obtaining an output based at least in part on applying the input to the natural language processing model; wherein the output includes at least one recommended HS code for the item; and outputting the at least one recommended HS code for the item. The item is a commodity or product.

[0021] The text prompt may include questions or statements (non-questions). In one embodiment of the third aspect, the text prompt includes a description of the item. In one embodiment of the third aspect, the text prompt includes one or more keywords related to the item. In one embodiment of the third aspect, the text prompt also includes countries or regions of interest related to the recommendation.

[0022] The information used for classifying the articles is retrieved from an information source. In one embodiment of the third aspect, the information source includes HS coding information that is at least partially indexed based on a trie data structure. In one embodiment of the third aspect, the information is retrieved from an information source that includes World Customs Organization HS coding information. In one embodiment of the third aspect, the information is retrieved at least partially from the World Customs Organization explanatory notes. In one embodiment of the third aspect, the information is retrieved from an information source that includes one or more country- or region-specific HS coding information. In one embodiment of the third aspect, the information source is retrieved from an information source that includes national or regional tariff or customs information related to country- or region-specific HS codes.

[0023] In one embodiment of the third aspect, the natural language processing model includes an LLM.

[0024] In one embodiment of the third aspect, the method includes: for each of the at least one recommended HS code of an item: retrieving a corresponding description of the recommended HS code of the item; and outputting the corresponding description of the recommended HS code of the item.

[0025] In one embodiment of the third aspect, the method includes: receiving the text prompt based at least in part on processing an image including the item.

[0026] In one embodiment of the third aspect, the recommended HS code is a 6-bit code. The 6-bit code is the internationally standardized HS code.

[0027] In one embodiment of the third aspect, the recommended HS code is an 8-bit code. The 8-bit code is an internationally standardized HS code plus a country / region-specific classification code.

[0028] In one embodiment of the third aspect, the recommended HS code is a 10-digit code. The 10-digit code is an internationally standardized HS code plus a country / region-specific classification code.

[0029] According to a fourth aspect of this disclosure, a carrier medium carrying computer-readable instructions is provided for causing or facilitating the execution of the method according to the third aspect by a computer. In one example, the carrier medium includes a computer-readable medium. In another example, the carrier medium is a non-transitory computer-readable medium storing a computer program executable by a computer. The computer program includes instructions for performing or facilitating the execution of the method according to the third aspect.

[0030] According to a fifth aspect of this disclosure, a computer program including instructions is provided, which, when executed by a computer, cause the computer to perform the method according to a third aspect.

[0031] According to a sixth aspect of this disclosure, a classification code recommendation method is provided, comprising: receiving a text prompt related to an item or service; retrieving information for classifying the item or service at least in part based on the text prompt; generating an input for a large language model based at least in part on the text prompt and the retrieved information; obtaining an output based at least in part on applying the input to the large language model, the output including at least one recommended classification code for the item or service; and outputting the at least one recommended classification code for the item or service.

[0032] The text prompt may include questions or statements (non-questions). In one embodiment of the sixth aspect, the text prompt includes a description of the item or service. In one embodiment of the sixth aspect, the text prompt includes one or more keywords related to the item or service. In one embodiment of the sixth aspect, the text prompt includes countries or regions of interest related to the recommendation.

[0033] The information used to classify the items or services is retrieved from an information source. The recommended classification code can be a numeric code or an alphanumeric code.

[0034] In one embodiment of the sixth aspect, the information used to classify the items or services is retrieved from an information source, which includes classification code information that is at least partially based on a trie data structure index.

[0035] In one embodiment of the sixth aspect, the natural language processing model includes an LLM.

[0036] In one embodiment of the sixth aspect, the method includes: for each of the at least one recommended classification code for an item or service: retrieving a corresponding description of the recommended classification code for the item or service; and outputting the corresponding description of the recommended classification code for the item or service.

[0037] In one embodiment of the sixth aspect, the method includes: receiving the text prompt based at least in part on processing an image including the item or service.

[0038] According to a seventh aspect of this disclosure, a carrier medium carrying computer-readable instructions is provided for causing or facilitating the execution of the method according to a sixth aspect. In one example, the carrier medium includes a computer-readable medium. In another example, the carrier medium is a non-transitory computer-readable medium storing a computer program executable by a computer. The computer program includes instructions for performing or facilitating the execution of the method according to a sixth aspect.

[0039] According to an eighth aspect of this disclosure, a computer program including instructions is provided, which, when executed by a computer, cause the computer to perform the method according to a sixth aspect.

[0040] Other features and aspects will become apparent from the following detailed description and accompanying drawings. Any feature described herein, as appropriate and applicable, may be combined with other features relating to any other aspect or embodiment herein. Attached Figure Description

[0041] Some specific embodiments of this disclosure will be described below by way of example with reference to the accompanying drawings.

[0042] Figure 1 This is a flowchart of an HS encoding recommendation method according to an embodiment of the present disclosure;

[0043] Figure 2A This is a schematic diagram illustrating the operation of an HS-coded recommendation system according to an embodiment of the present disclosure;

[0044] Figure 2B This is a schematic diagram illustrating the operation of an HS-coded recommendation system according to an embodiment of the present disclosure;

[0045] Figure 3 This is a flowchart of a classification code recommendation method according to an embodiment of the present disclosure;

[0046] Figure 4A This is a schematic diagram illustrating the operation of a classification code recommendation system according to an embodiment of the present disclosure;

[0047] Figure 4B A schematic diagram illustrating the operation of a classification code recommendation system according to an embodiment of the present disclosure; and

[0048] Figure 5 This is a block diagram of an information processing system according to an embodiment of the present disclosure. Detailed Implementation

[0049] Some of the embodiments disclosed herein relate to AI-based tools that can be used to recommend HS codes.

[0050] HS codes belong to a unified system, a specialized system used in over 200 countries worldwide for classifying goods (such as articles and products). HS codes are typically 6 to 10 digits long. Generally, the first 6 digits of an HS code are used internationally and standardized by the World Customs Organization. For some countries or regions, HS codes may include additional digits for more specific classification at the country / region level. Usually, to determine or find the HS code of an item, it is necessary to consult various rules, definitions, and decisions from various sources.

[0051] As is well known, manual HS code identification is challenging, especially for uneducated individuals. Furthermore, automating HS code identification has been found to be far from straightforward. In light of these limitations, some embodiments disclosed herein enable the automatic identification of HS codes for goods (HS codes may include a 6-digit number standardized by the World Customs Organization and an optional country- or region-specific suffix, e.g., an additional 4 digits for imports from China).

[0052] Manually determining an HS code, even just the first six digits, is a complex process, especially for untrained individuals. This is because assigning the correct HS code may require navigating a hierarchical classification system by identifying and applying various general and specific rules to extract the two classification digits at a time. In addition to finding relevant lists, various notes and publications may need to be consulted. These notes and publications clarify terminology, define distinctions, specify exclusions, and provide guidance on excluding similar alternatives. Some of these notes and publications are scattered across multiple sections of the World Customs Organization's official explanatory notes, which are updated regularly and may not be available online. Sometimes, other notes and documents, such as classification decisions, may need to be considered. Furthermore, for HS codes longer than six digits (e.g., 10 digits), it may be necessary to consult more difficult-to-access country- or region-specific sources, such as customs reference publications, online tax references, and customs classification decision announcements. Therefore, the cognitive load required to determine an HS code is considerable. While some existing websites or publications do provide static lists of HS codes, these resources lack higher-level rules, precautionary guidelines (such as priority guidelines), and the contextual information needed to accurately determine the HS code based on the guidelines and exceptions. Therefore, the process of manually determining or identifying HS codes is prone to errors, especially for those unfamiliar with the system or workflow.

[0053] Automating HS code identification can solve the aforementioned problems associated with manual determination. In fact, automating the HS code identification process is crucial, especially in an era of booming e-commerce and global trade. With the surge in cross-border transactions, customs authorities are struggling to effectively handle massive volumes of processing and disputes. Inaccurate HS code assignments can have serious financial, operational, and / or reputational consequences for businesses engaged in global trade. Currently, most countries report an accuracy rate of slightly above 70% in HS code determination. This means that 30% of items are misclassified, potentially leading to losses in taxable duties globally. Misclassified goods can trigger customs inspections, causing clearance delays, disrupting supply chains, impacting inventory levels, sales, and overall operations, thus risking customer dissatisfaction and reputational damage. Lost revenue stems from underpaying or overpaying customs duties and taxes, coupled with potential penalties for violations. Furthermore, HS code errors complicate compliance with trade agreements and regulations, restrict market access, and invite legal action. Resolving misclassifications can require costly audits, reassessments, and specialized consultations, exhausting administrative resources.

[0054] Developing automated HS code recognition tools is challenging for various reasons. For example, to handle the different workflow patterns involved in HS code recognition, the tool must be able to switch tasks and select appropriate reference sources, and automatically perform classification based on user queries; however, combining rule-based and classification tasks within a single tool is difficult. For instance, since some reference data may be in structured text format while others may be written as unstructured text instructions, the tool needs to be able to handle both structured and unstructured reference text sources. For example, longer 10-bit HS codes will be less accurate than shorter 6-bit HS codes because far less data is available to facilitate the determination of longer HS codes, and general applications of machine learning may not be able to address this issue. For example, the tool may need to be rapidly updatable without requiring model retraining. Specifically, as the World Customs Organization or various national or regional customs authorities may introduce changes (e.g., rapid temporary changes every five years, months, or year to adapt to changes due to sanctions, environmental factors, and product trends / evolutions), the tool needs to be able to update quickly to avoid misclassification and maintain accuracy. Ideally, the tool should be able to provide commercial-level accuracy (e.g., over 90%) for complete HS coding interpretation, depending on specific country requirements, with relevant interpretation and essentially real-time performance. Ideally, an automated HS coding identification tool should achieve high accuracy, interpretability, and high processing speed for real-world adoption.

[0055] Some of the embodiments disclosed herein relate to AI-based tools that can be used to recommend HS codes.

[0056] Figure 1 An HS encoding recommendation method 100 according to one embodiment is illustrated. Method 100 can be executed by one or more processors.

[0057] Method 100 includes, in step 102, receiving a text prompt related to an item. The item is a commodity or product. The text prompt may include a question or a statement (non-question). The text prompt may include a description of the item or one or more keywords related to the item. The text prompt may also include countries or regions of interest related to the recommendation.

[0058] Method 100 includes, in step 104, retrieving information for classifying articles, at least in part based on text prompts. Information can be retrieved from sources such as one or more databases. The database may include a carrier database. The database may be associated with information that can be used to classify goods. For example, the information source may include HS code information indexed at least in part based on a trie (prefix tree) data structure. For example, the information source may include HS information sources, which may include World Customs Organization (WCO) HS code information. For example, information may be retrieved at least in part from WCO explanatory notes. Additionally or alternatively, the information source may include one or more country- or region-specific HS code information. For example, the information source may include country or region tariff or customs information associated with country- or region-specific HS codes.

[0059] Method 100 includes, in step 106, generating input for a natural language processing model based at least in part on text prompts and retrieved information. The natural language processing model may include or may be an LLM (Local Language Management Model). The input can then be fed to the natural language processing model for processing and for generating output.

[0060] Method 100 includes, in 108, obtaining an output at least in part based on applying the input to a natural language processing model. The output includes at least one recommended HS code for the item. For example, each recommended HS code may be a 6-bit code, an 8-bit code, or a 10-bit code.

[0061] Method 100 includes: in 110, outputting at least one recommended HS code for an item, for example, for presenting at least one recommended HS code to a user.

[0062] Those skilled in the art will understand that method 100 is merely an example, and other embodiments may be provided with various modifications and / or variations to method 100. For example, method 100 may further include: for each of at least one recommended HS code of an item: retrieving a corresponding description of the recommended HS code of the item; and outputting the corresponding description of the recommended HS code of the item. The description of the recommended HS code may be output together with its corresponding recommended HS code. For example, method 100 may also include: retrieving text prompts at least in part based on processing an image including the item. For example, an image of the item may be captured, and the image may be processed using AI or computer vision techniques to determine text associated with the item in the image.

[0063] Figure 2A Operation 200A of an HS encoding recommendation system 20A according to an embodiment of the present disclosure is illustrated. The HS encoding recommendation system 20A can be implemented by at least one or more processors.

[0064] At 202A, the HS-coded recommendation system 20A receives textual prompts related to the item. This may correspond to 102 in method 100.

[0065] In 204A, the HS-encoded recommendation system 20A sends an information retrieval request to an HS-encoded information source (such as one or more databases of a vector database) based on the received text prompts.

[0066] In step 206A, the HS coding information source obtains the relevant information for classifying items and provides this information to the HS coding recommendation system 20A. In other words, the HS coding recommendation system 20A retrieves information for classifying items from the HS coding information source. This corresponds to step 104 in method 100.

[0067] The HS-encoded recommendation system 20A then generates input for a Natural Language Processing (NLP) model based on text prompts and retrieved information. This corresponds to step 106 in method 100.

[0068] At 208A, the HS-coded recommendation system 20A provides input to the natural language processing model. The natural language processing model processes the input and generates output. The output includes at least one recommended HS code for the item.

[0069] In 210A, the natural language processing model provides its output to the HS-encoded recommendation system 20A, which in turn obtains an output containing at least one recommended HS code for the item. This may correspond to 108 in method 100.

[0070] At 212A, the HS coding recommendation system 20A outputs at least one recommended HS code for the item. This may correspond to 110 in method 100.

[0071] Those skilled in the art will understand that Operation 200A is merely an example, and other embodiments may be provided for various modifications and / or variations of Operation 200A.

[0072] Figure 2B Operation 200B of an HS-encoded recommendation system 20B according to an embodiment of the present disclosure is illustrated. The HS-encoded recommendation system 20B can be implemented by at least one or more processors.

[0073] The HS-coded recommender system 20B is generally similar to the HS-coded recommender system 20A, except that the HS-coded recommender system 20B includes a natural language processing model.

[0074] At 202B, the HS-coded recommendation system 20B receives textual prompts related to the item. This may correspond to 102 in method 100.

[0075] In 204B, the HS-coded recommendation system 20B sends an information retrieval request to an HS-coded information source (such as one or more databases of a vector database) based on the received text prompts.

[0076] In step 206B, the HS coding information source obtains the relevant information for classifying items and provides this information to the HS coding recommendation system 20B. In other words, the HS coding recommendation system 20B retrieves information for classifying items from the HS coding information source. This corresponds to step 104 in method 100.

[0077] The HS-encoded recommendation system 20B then generates input for the natural language processing model based on text prompts and retrieved information. This may correspond to 106 in method 100. The HS-encoded recommendation system 20B feeds the input to the natural language processing model. The natural language processing model processes the input and generates output. The output includes at least one recommended HS code for the item. The HS-encoded recommendation system 20B obtains an output including at least one recommended HS code for the item. This may correspond to 108 in method 100.

[0078] At 208B, the HS coding recommendation system 20B outputs at least one recommended HS code for the item. This may correspond to 110 in method 100.

[0079] Those skilled in the art will understand that Operation 200B is merely an example, and other embodiments may be provided with respect to various modifications and / or variations of Operation 200B.

[0080] Some example implementations of AI-based tools for recommending HS codes involve the use of LLMs, featuring retrieval augmentation generation and trie data structures. These AI-based tool implementations leverage the versatile task-switching, rule-following, and classification capabilities of LLMs to address challenges associated with the development of automated HS code recognition tools. For example, LLMs possess rule-following capabilities. An LLM can combine task-switching, rule-following, and classification, including terminology differentiation, exclusion, and explanatory annotations that help interpret HS codes (as well as other information that helps determine the HS code). For example, combined with retrieval augmentation generation with entities and relationships, LLMs can handle unstructured text well and can respond to reference data provided at each run time to provide performance superior to traditional machine learning / LLMs can provide relevant explanations when properly programmed / hinted. In some implementations, retrieval augmentation generation in an LLM can be based on hinted knowledge rather than training operations and can quickly adapt to changes / updates in the HS coding system without requiring retraining. Near real-time integration of updated information sources and knowledge bases allows LLMs to seamlessly adjust their recommendations to conform to the latest HS coding revisions, product trends, and regulatory updates. In some implementations, AI-based tools for recommending HS codes can provide accuracy (>90%) and interpretation for 10-bit HS codes (e.g., for Chinese import data). Some implementations incorporate trie data structures as efficient text retrieval structures. For example, some implementations integrate trie (prefix tree) data structures, which are inherent in the design of the HS code hierarchy. Trie (prefix tree) data structures are particularly useful for storing and efficiently retrieving HS code rules, product descriptions, and related information at logarithmic speed in lookup time (only log(n) is needed for n matches). This data structure allows AI-based tools to have computational performance advantages over other methods and can complement the use of LLM through retrieval enhancement generation.

[0081] Some of the embodiments disclosed herein relate to AI-based tools that can be used to recommend other classification codes.

[0082] Figure 3 A classification code recommendation method 300 according to an embodiment of the present disclosure is illustrated. Method 300 can be executed by one or more processors.

[0083] Method 300 includes, in step 302, receiving a text prompt related to an item or service. The item is a commodity or product. The text prompt may include a question or a statement (non-question). The text prompt may include a description of the item or service or one or more keywords related to the item or service. The text prompt may also include countries or regions of interest related to the recommendation.

[0084] Method 300 includes, in 304, retrieving information for classifying items or services, at least in part based on text prompts. The information can be retrieved from sources such as one or more databases. The databases may include carrier databases. The information for classifying items or services can be retrieved from sources including classification code information indexed at least in part based on a trie data structure.

[0085] Method 300 includes, in 306, generating input for a natural language processing model based at least in part on text prompts and retrieved information. The natural language processing model may include or may be an LLM (Local Language Management Model). The input can then be fed to the natural language processing model for processing and for generating output.

[0086] Method 300 includes, in 308, obtaining an output at least in part based on applying the input to a natural language processing model. The output includes at least one recommended classification code for an item or service. The recommended classification code can be a numeric code or an alphanumeric code.

[0087] Method 300 includes: in 310, outputting at least one recommended category code for an item, for example, for presenting at least one recommended category code to a user.

[0088] Those skilled in the art will understand that method 300 is merely an example, and other embodiments may be provided with various modifications and / or variations to method 300. For example, method 300 may further include: for each of at least one recommended classification code for an item or service: retrieving a corresponding description of the recommended classification code for the item or service; and outputting the corresponding description of the recommended classification code for the item or service. The description of the recommended classification code may be output along with its corresponding recommended classification code. For example, method 300 may further include: retrieving text prompts at least in part based on processing an image including an item or service. For example, an image of the item or service may be captured, and the image may be processed using AI or computer vision techniques to determine text related to the item or service in the image.

[0089] Figure 4A Operation 400A of a classification code recommendation system 40A according to an embodiment of the present disclosure is illustrated. The classification code recommendation system 40A can be implemented by at least one or more processors.

[0090] In 402A, the classification code recommendation system 40A receives text prompts related to items or services. This corresponds to 302 in method 300.

[0091] In 404A, the classification code recommendation system 40A sends an information retrieval request to a classification code information source (such as one or more databases of a vector database) based on the received text prompts.

[0092] In 406A, the classification code information source obtains the relevant information for classifying items or services and provides this information to the classification code recommendation system 40A. In other words, the classification code recommendation system 40A retrieves information for classifying items or services from the classification code information source. This corresponds to 304 in method 300.

[0093] The classification code recommendation system 40A then generates input for a natural language processing model based on text prompts and retrieved information. This corresponds to 306 in method 300.

[0094] In 408A, the classification code recommendation system 40A provides input to a natural language processing model. The natural language processing model processes the input and generates output. The output includes at least one recommended classification code for an item or service. The recommended classification code can be a numeric code or an alphanumeric code.

[0095] In 410A, the natural language processing model provides its output to the classification code recommendation system 40A, which in turn obtains an output containing at least one recommended classification code for an item or service. This may correspond to 308 in method 300.

[0096] In 412A, the classification code recommendation system 40A outputs at least one recommended classification code for an item or service. This may correspond to 310 in method 300.

[0097] Those skilled in the art will understand that Operation 400A is merely an example, and other embodiments may be provided for various modifications and / or variations of Operation 400A.

[0098] Figure 4B Operation 400B of a classification code recommendation system 40B according to an embodiment of the present disclosure is illustrated. The classification code recommendation system 40B can be implemented by at least one or more processors.

[0099] Classification code recommendation system 40B is generally similar to classification code recommendation system 40A, except that classification code recommendation system 40B includes a natural language processing model.

[0100] In 402B, the classification code recommendation system 40B receives text prompts related to items or services. This corresponds to 302 in method 300.

[0101] In 404B, the classification code recommendation system 40B sends an information retrieval request to a classification code information source (such as one or more databases of a vector database) based on the received text prompts.

[0102] In 406B, the classification code information source obtains the relevant information for classifying items or services and provides this information to the classification code recommendation system 40B. In other words, the classification code recommendation system 40B retrieves information for classifying items or services from the classification code information source. This corresponds to 304 in method 300.

[0103] The classification code recommendation system 40B then generates input for the natural language processing model based on text prompts and retrieved information. This may correspond to 306 in method 300. The classification code recommendation system 40B feeds the input to the natural language processing model. The natural language processing model processes the input and generates output. The output includes at least one recommended classification code for the item or service. The classification code recommendation system 40B obtains an output including at least one recommended classification code for the item or service. This may correspond to 308 in method 300.

[0104] In 408B, the classification code recommendation system 40B outputs at least one recommended classification code for an item or service. This may correspond to 310 in method 300.

[0105] Those skilled in the art will understand that Operation 400B is merely an example, and other embodiments may be provided with respect to various modifications and / or variations of Operation 400B.

[0106] Figure 5 An example information processing system 500 that can be used to implement some embodiments is shown. For example, information processing system 500 can be used to execute method 100. For example, information processing system 500 can be used to execute method 300. For example, information processing system 500 can be used to implement recommendation systems 20A, 20B, 40A, 40B. For example, information processing system 500 can be used to operate a natural language processing model. For example, information processing system 500 can be used to implement an information source (database).

[0107] Information processing system 500 includes appropriate components required to receive, store, and execute appropriate computer instructions, commands, and / or codes. In this example, information processing system 500 includes processor 502 and memory 504. Processor 502 may include one or more of the following: one or more CPUs, one or more MCUs, one or more GPUs, one or more NPUs, one or more VPUs, one or more TPUs, one or more logic circuits, one or more Raspberry Pi chips, one or more Digital Signal Processors (DSPs), one or more Application-Specific Integrated Circuits (ASICs), one or more Field-Programmable Gate Arrays (FPGAs), and digital and / or analog circuitry (or multiple circuits) for interpreting program instructions, executing program instructions, and / or processing signals and / or information and / or data. Memory 504 may include one or more volatile memories (such as RAM, DRAM, SRAM, etc.), one or more non-volatile memories (such as ROM, PROM, EPROM, EEPROM, FRAM, MRAM, FLASH, SSD, NAND, NVDIMM, etc.), or any combination thereof. Appropriate computer instructions, commands, codes, information, and / or data are stored in memory 504. For example, computer instructions for performing steps or operations of a method embodiment may be stored in memory 504. Processor 502 and memory 504 may be integrated or may be separate (and operatively connected).

[0108] Optionally, the information processing system 500 also includes one or more input devices 506. Examples of input devices 506 include: keyboards, mice, styluses, image scanners, microphones, haptic / touch input devices (e.g., touch-sensitive screens), image / video input devices (e.g., cameras), etc. Image / video input devices can be used to capture images of objects or services.

[0109] Optionally, the data processing system 500 also includes one or more output devices 508. Examples of output devices 508 include: displays (e.g., monitors, screens, projectors, etc.), speakers, headphones, earphones, additive manufacturing machines (e.g., 3D printers), etc. The display may include an LCD display, an LED / OLED display, or other suitable displays, which may or may not be touch-sensitive.

[0110] The information processing system 500 may also include one or more disk drives 512, which may include one or more of the following: solid-state drives, hard disk drives, optical disk drives, flash memory drives, tape drives, etc. A suitable operating system may be installed in the information processing system 500, for example, on the disk drive 512 or in the memory 504. The memory 504 and the disk drive 512 may be operated by the processor 502.

[0111] Optionally, the information processing system 500 may also include a communication device 510 for establishing one or more communication links with one or more computing devices (such as servers, personal computers, terminals, tablets, telephones, watches, Internet of Things (IoT) devices, or other wireless computing devices). The communication device 510 may include one or more of the following: modems, network interface cards (NICs), integrated network interfaces, NFC transceivers, ZigBee transceivers, Wi-Fi transceivers, etc. Transceivers, radio frequency transceivers, cellular (2G, 3G, 4G, 5G, 6G, etc.) transceivers, optical ports, infrared ports, USB connections, or other wired or wireless communication interfaces. A transceiver can be implemented by one or more devices (integrated transmitters and receivers, separate transmitters and receivers, etc.). The communication link can be wired or wireless and is used to transmit commands, instructions, information, and / or data. In one example, processor 502, memory 504 (optionally, input device 506, output device 508, communication device 510, and disk drive 512, if any) can be directly or indirectly connected to each other via a bus, peripheral component interconnect (PCI) (such as PCI Express), universal serial bus (USB), optical bus, or other similar structures. In one embodiment, at least some of these components can be wirelessly connected, for example, via a network (such as the Internet, cloud computing networks). Those skilled in the art will understand that the information processing system 500 is merely an example, and in other embodiments, the information processing system 500 may have different configurations (e.g., with additional components, fewer components, alternative components, etc.).

[0112] While not strictly necessary, the embodiments described with reference to the accompanying drawings can be implemented as an application programming interface (API) or a set of libraries used by developers, or can be included in another software application, such as a terminal or personal computer operating system or a portable computing device operating system. Typically, since program modules include routines, programs, objects, components, and data files that help perform specific functions, those skilled in the art will understand that the functionality of a software application can be distributed among multiple routines, objects, and / or components to achieve the same functionality required herein.

[0113] It should also be understood that any suitable computing system architecture can be used where the methods and systems of this disclosure are implemented wholly or partially by a computing system. This will include stand-alone computers, network computers, and dedicated or non-dedicated hardware devices. When using the terms "computing system" and "computing device," these terms are intended to cover any suitable configuration of computer or information processing hardware capable of implementing the described functions.

[0114] The embodiments disclosed herein provide techniques for HS code recommendation and classification code recommendation. For example, some embodiments disclosed herein provide LLM-based classification code (e.g., HS code) recommendation systems and methods. LLM-based classification code (e.g., HS code) recommendation systems and methods can support automatic identification of accurate HS codes. In some embodiments, the system and method can support 6-bit HS codes as well as 8-bit and 10-bit country-specific HS codes, or more generally, any classification code. The embodiments disclosed herein employ retrieval-enhanced generation, which can be used to process unstructured text and update / change classification code systems (e.g., HS code systems).

[0115] Those skilled in the art will understand that variations and / or modifications can be made to the disclosed embodiments to provide other embodiments. Therefore, the disclosed embodiments should be considered illustrative rather than restrictive in all respects. While some embodiments disclosed herein specifically relate to recommendations for HS encoding, others may relate to recommendations for other types of classification codes.

Claims

1. An HS-coded recommendation system, characterized in that, include: One or more processors, which are used for: Receive text prompts related to items; Retrieve information for classifying the items, at least in part, based on the text prompts; The input for the natural language processing model is generated based at least in part on the text prompts and the retrieved information; The output is obtained at least in part by applying the input to the natural language processing model; wherein the output includes at least one recommended HS code for the item; as well as Output the at least one recommended HS code for the item.

2. The HS coding recommendation system according to claim 1, characterized in that, in, The text prompt includes: The description of the item; or One or more keywords related to the item.

3. The HS coding recommendation system according to claim 2, characterized in that, in, The text prompts also include countries or regions of interest related to the recommendations.

4. The HS coding recommendation system according to claim 3, characterized in that, in, The information used to classify the items is retrieved from an information source, which includes HS-encoded information that is at least partially based on a trie data structure index.

5. The HS coding recommendation system according to claim 1, characterized in that, in, The information was retrieved from an information source that includes World Customs Organization HS code information.

6. The HS coding recommendation system according to claim 1, characterized in that, in, The information was retrieved, at least in part, from the World Customs Organization's explanatory notes.

7. The HS coding recommendation system according to claim 1, characterized in that, in, The information is retrieved from an information source that includes one or more country- or region-specific HS coding information.

8. The HS coding recommendation system according to claim 1, characterized in that, The information source is retrieved from an information source that includes national or regional tariff or customs information related to HS codes specific to a country or region.

9. The HS coding recommendation system according to claim 1, characterized in that, in, The natural language processing model includes a large language model.

10. The HS coding recommendation system according to claim 1, characterized in that, in, The one or more processors are used for: For each of the at least one recommended HS code for the item: Retrieve the corresponding description of the recommended HS code for the item; and Output the corresponding description of the recommended HS code for the item.

11. The HS coding recommendation system according to claim 1, characterized in that, in, The one or more processors are used for: The text prompt is received based at least in part on processing an image including the item.

12. The HS coding recommendation system according to claim 1, characterized in that, in, The recommended HS code is a 6-bit code.

13. The HS coding recommendation system according to claim 1, characterized in that, in, The recommended HS encoding is an 8-bit or 10-bit code.

14. An HS-coded recommendation system, characterized in that, include: One or more processors, which are used for: Receive a text prompt that includes a description of the item or one or more keywords related to the item; Information for classifying the items based at least in part on the text prompts is retrieved from an information source that includes HS-encoded information that is at least partially indexed by a trie data structure. Input for a large language model is generated, at least in part, based on the text prompts and the retrieved information; The output is obtained at least in part by applying the input to the large language model; wherein the output includes at least one recommended HS code for the item; as well as Output the at least one recommended HS code for the item.

15. A classification code recommendation system, characterized in that, include: One or more processors, which are used for: Receive text prompts related to items or services; Retrieve information for classifying the items or services, at least in part, based on the text prompts; The input for the natural language processing model is generated based at least in part on the text prompts and the retrieved information; The output is obtained by applying the input to the natural language processing model, at least in part, and the output includes at least one recommended classification code for the item or service. as well as Output at least one recommended category code for the item or service.

16. The classification code recommendation system according to claim 15, characterized in that, in, The text prompt includes: The description of the items or services; or One or more keywords related to the item or service mentioned.

17. The classification code recommendation system according to claim 16, characterized in that, in, The text prompts also include countries or regions of interest related to the recommendations.

18. The classification code recommendation system according to claim 17, characterized in that, in, The information used to classify the items or services is retrieved from an information source, which includes classification code information that is at least partially based on a trie data structure index.

19. The classification code recommendation system according to claim 18, characterized in that, in, The natural language processing model includes a large language model.

20. The classification code recommendation system according to claim 19, characterized in that, in, The one or more processors are used for: For each of the items or services, at least one recommended classification code is required; Retrieve the description corresponding to the recommended category code of the item or service; and Output the corresponding description of the recommended category code for the item or service.

Images