Product architecture information generation method and device based on large model and electronic equipment

Abstract

The invention provides a product architecture information generation method and device based on a large model and electronic equipment, and relates to the technical field of computers, in particular to the technical fields of artificial intelligence, the large model, information retrieval, data mining and the like. The specific implementation scheme comprises the steps of obtaining a plurality of product identifiers; grouping the plurality of product identifiers to obtain an identifier group under each grouping granularity in the N grouping granularities; wherein N is greater than or equal to 2, and N is an integer; the N grouping granularities have different thickness degrees; performing identifier integration for the first identifier group by using the first large model to obtain a first identifier integration result; wherein the first identification group is an identification group under the ith grouping granularity in the N grouping granularities; 1 < = i < = N, and i is an integer; and obtaining product architecture information related to the plurality of product identifiers based on the first identifier integration result. According to the invention, the accuracy and acquisition efficiency of the product architecture information can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to the fields of artificial intelligence, big data models, information retrieval, and data mining. Specifically, it relates to a method, apparatus, and electronic device for generating product architecture information based on a big data model. Background Technology

[0002] Product architecture information is the digital hub of an enterprise's product system. It is used to systematically organize product nodes and attach digital assets such as advertising materials and landing pages to it. Its core value lies in providing an understandable and searchable framework for dynamic advertising strategies or e-commerce platforms, making it possible to accurately recall and match digital assets based on user intent, thereby driving product marketing from static configuration to intelligent matching. Summary of the Invention

[0003] This disclosure provides a method, apparatus, and electronic device for generating product architecture information based on a large model.

[0004] According to a first aspect of this disclosure, a method for generating product architecture information based on a large model is provided, comprising: Obtain multiple product identifiers; Multiple product identifiers are grouped to obtain identifier groups under each of N group granularities; where N≥2 and N is an integer; the N group granularities have different levels of coarseness; Using the first major model, we perform identifier integration for the first identifier group to obtain the first identifier integration result; where the first identifier group is the identifier group under the i-th grouping granularity among N grouping granularities; 1≤i≤N, and i is an integer; Based on the integration results of the first identifier, product architecture information related to multiple product identifiers is obtained.

[0005] According to a second aspect of this disclosure, a product architecture information generation apparatus based on a large model is provided, comprising: The identifier acquisition unit is used to acquire multiple product identifiers; The identifier grouping unit is used to group multiple product identifiers to obtain identifier groups under each of the N grouping granularities; where N≥2 and N is an integer; the N grouping granularities have different levels of coarseness; The identifier integration unit is used to perform identifier integration for the first identifier group using the large model to obtain the first identifier integration result; wherein, the first identifier group is the identifier group under the i-th grouping granularity among N grouping granularities; 1≤i≤N, and i is an integer; The information acquisition unit is used to obtain product architecture information related to multiple product identifiers based on the integration result of the first identifier.

[0006] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; Memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, which are executed by at least one processor to enable the at least one processor to perform the method provided in the first aspect of this disclosure.

[0007] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions; wherein the computer instructions are used to cause a computer to perform the method provided in the first aspect of this disclosure.

[0008] According to a fifth aspect of this disclosure, a computer program product is provided, including a computer program; wherein, when executed by a processor, the computer program is capable of implementing the method provided in the first aspect of this disclosure.

[0009] Using this disclosure can improve the accuracy and efficiency of product architecture information acquisition.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 A flowchart illustrating a method for generating product architecture information based on a large model, provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of a code group acquisition process provided in an embodiment of the present disclosure; Figure 3 A schematic diagram of the training process of a residual quantization variational autoencoder provided in an embodiment of this disclosure; Figure 4 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 1 ; Figure 5 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 2 ; Figure 6 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 3 ; Figure 7 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 4 ; Figure 8A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 5 ; Figure 9 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 6 ; Figure 10 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 7 ; Figure 11 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 8 ; Figure 12 A schematic diagram of an identifier integration process provided in this embodiment of the disclosure. Figure 9 ; Figure 13 A schematic diagram of a process for obtaining product architecture information provided in this embodiment of the disclosure. Figure 1 ; Figure 14 A schematic diagram of a process for obtaining product architecture information provided in this embodiment of the disclosure. Figure 2 ; Figure 15 This is a schematic diagram illustrating an application scenario of a product architecture information generation method based on a large model, provided in an embodiment of this disclosure. Figure 16 A schematic structural block diagram of a product architecture information generation device based on a large model provided in this disclosure embodiment; Figure 17 This is a schematic structural block diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] A target brand's official website typically provides information about its products, such as core navigation information on product detail pages, product information lists, and product hierarchy. Based on this, traditional solutions can utilize large-scale models to perform content understanding and simulated clicks on the target brand's official website. This allows for the extraction of key information from the product-related information, followed by hierarchical extraction and structuring of this key information to obtain product architecture information.

[0014] However, the inventors discovered the following problems with the above solutions: First, when performing content understanding and simulated clicks on the target brand's official website to extract key information from the relevant information of the brand's products, there are often cases of missed clicks. Therefore, some key information will be lost, thereby reducing the accuracy of product architecture information. Secondly, understanding the content and simulating clicks on the target brand's official website takes a long time, and may time out or fail for various reasons, often requiring a restart. This reduces the efficiency of obtaining product architecture information. Finally, since the key information extracted from the relevant information may be combined with the specific product description on the product details page and the entire content of the Hyper Text Markup Language (HTML) file of the product details page, the total number of the smallest text processing unit (Token) of the large model will be very large, even far exceeding the current length limit that the large model can accept. This will not only further reduce the accuracy of the product architecture information, but also further reduce the efficiency of obtaining the product architecture information.

[0015] To address at least some of the above-mentioned problems, this disclosure provides a method for generating product architecture information based on a large model, which can be applied to both service devices and terminal devices. The service device can be a server, workbench, mainframe computer, or other similar computing device; the terminal device can be a workbench, mainframe computer, conventional computer (e.g., desktop computer, laptop computer, tablet computer, etc.) or other similar computing device. The following will be combined with... Figure 1 The flowchart shown illustrates a method for generating product architecture information based on a large model, as provided in this disclosure. It should be noted that although the flowchart shows a logical order, in some cases, the steps shown or described in the flowchart may be executed in a different order.

[0016] Step S101: Obtain multiple product identifiers.

[0017] The product identifier can be the entity description information of a product under the target brand, which may include at least one of multiple subcategories of description information such as product name, product series, product model, and product configuration. Here, the target brand can be a single brand, or it can refer to each of multiple designated brands belonging to the same industry (e.g., the automotive industry, the luxury goods industry, the mobile phone industry, etc.).

[0018] In this embodiment of the disclosure, the product identifier can be obtained from the target platform. The target platform may include at least one of a search platform, a shopping platform, and a social platform.

[0019] Step S102: Group the multiple product identifiers to obtain identifier groups under each of the N grouping granularities.

[0020] Where N≥2 and N is an integer.

[0021] In this embodiment of the disclosure, the N grouping granularities have different degrees of coarseness, and the N grouping granularities can be arranged from coarse to fine.

[0022] For example, the product identifier is the entity description information of a product under the target brand, and the target brand is the designated brand XNE, which belongs to the luxury goods industry; N=4, and the first grouping granularity in the four grouping granularities is "brand", the second grouping granularity is "product line", the third grouping granularity is "category", and the fourth grouping granularity is "series". Then, we can first group multiple product identifiers according to the "brand" grouping granularity to obtain the identifier group under the "brand" grouping granularity; then group multiple product identifiers according to the "product line" grouping granularity to obtain the identifier group under the "product line" grouping granularity; next, group multiple product identifiers according to the "category" grouping granularity to obtain the identifier group under the "category" grouping granularity; finally, group multiple product identifiers according to the "series" grouping granularity to obtain the identifier group under the "series" grouping granularity.

[0023] Step S103: Using the first large model, perform identifier integration for the first identifier group to obtain the first identifier integration result.

[0024] The first type of model can be the Large Language Model (LLM). Here, an LLM can be a pre-trained neural network model (e.g., an autoregressive generative model with a Transformer architecture) that possesses general language knowledge, world knowledge, and domain-specific expertise (e.g., expertise in the field of computer technology).

[0025] In this embodiment of the disclosure, the first identifier group can be the identifier group under the i-th grouping granularity among N grouping granularities. Wherein, 1≤i≤N and i is an integer.

[0026] Furthermore, in this embodiment of the disclosure, the first identifier integration result may include at least one of an overall identifier general term and product hierarchical classification information. The overall identifier general term may be a general identifier summary result; the product hierarchical classification information may be hierarchical product classification information.

[0027] Step S104: Based on the first identifier integration result, obtain product architecture information related to multiple product identifiers.

[0028] It should be noted that in this embodiment, step S103 can be executed cyclically. That is, after taking each identifier group under the i-th group granularity in the N grouping granularities as the first identifier group, and using the first large model to perform identifier integration for the first identifier group to obtain the first identifier integration result, a "-1" operation can be performed on i, and step S103 can be executed again. This cycle continues until i=1, and when i=1, step S103 is executed again before ending. Based on this, in this embodiment, when executing step S104, all the obtained first identifier integration results can be serially and parallelly connected to obtain product architecture information related to multiple product identifiers. Among them, the product architecture information can be Standard Product Unit (SPU) information.

[0029] The product architecture information generation method based on a large model provided in this disclosure can obtain multiple product identifiers, group the multiple product identifiers to obtain identifier groups under each group granularity in N grouping granularities, and then use the first large model to perform identifier integration for the first identifier group to obtain the first identifier integration result. Based on the first identifier integration result, product architecture information related to multiple product identifiers can be obtained. In this process, on the one hand, by directly acquiring multiple product identifiers, the problems of missed clicks and timeout failures that may occur when understanding content and simulating clicks on the target brand's official website in traditional solutions can be effectively avoided. On the other hand, after acquiring multiple product identifiers, these identifiers can be hierarchically summarized according to different grouping granularities to obtain identifier groups under each of the N grouping granularities. This significantly reduces the total number of tokens input to the first main model in a single operation, avoiding exceeding the current length limit of the first main model due to excessively long input content, and ensuring the completeness and accuracy of information extraction. Based on this, by integrating the identifier groups under each of the N grouping granularities through the first main model, the corresponding first identifier integration results can be systematically and efficiently extracted for constructing product architecture information with a clear structure and complete content. In this way, the accuracy and acquisition efficiency of product architecture information are improved.

[0030] In some optional implementations, step S101, namely, "obtaining multiple product identifiers", may include: Obtain multiple user search terms from the target platform; Multiple product identifiers are obtained based on multiple user search terms.

[0031] As previously stated, in this embodiment of the disclosure, the target platform may include at least one of a search platform, a shopping platform, and a social platform; multiple user search terms may be search terms related to the target brand, and the target brand may be a single brand or each of multiple designated brands belonging to the same industry.

[0032] For example, the product identifier is the entity description information of the product under the target brand, and the target brand is the designated brand XNE, which belongs to the luxury goods industry. Multiple user search terms can be "XNE N°X", "How is XNE N°X perfume", "XNE NX perfume promotion", etc.

[0033] After obtaining multiple user search terms from the target platform, preprocessing operations (such as error correction and clarification) can be performed on these search terms. Based on these preprocessed user search terms, multiple product identifiers can be obtained. For example, it is possible to: Construct entity extraction instructions; For each user search term among multiple user search terms, the second major model is used to extract hints according to entities and obtain candidate identifiers based on the user search terms; Multiple product identifiers are obtained based on multiple candidate identifiers that correspond one-to-one with multiple user search terms.

[0034] The entity extraction indicator, serving as a prompt, can be used to instruct the second-largest model on how to obtain candidate identifiers based on the user's search terms. Here, the second-largest model can be an LLM (Limited Module Model), specifically, it can be a different LLM from the first-largest model, or it can be the same LLM as the first-largest model.

[0035] In this embodiment of the disclosure, when constructing entity extraction suggestions, entity extraction examples and industry knowledge related to multiple user search terms can be obtained, and entity extraction suggestions can be constructed based on the entity extraction examples and industry knowledge. The entity extraction examples may include: User search term "How is XNE N°X perfume?" → Candidate ID: XNE N°X perfume; User search term "XNE NX perfume promotion" → Candidate ID: XNE NX perfume; ... Industry knowledge can include: A common product hierarchy and classification system: Brand > Product Line > Category > Series > Model / Specification; The name of the product line under the target brand; Common product terminology, model naming rules, and abbreviations for the target brand; List of mainstream products currently sold by the target brand in the market.

[0036] After obtaining the entity extraction instructions, for each user search term among multiple user search terms, the second major model can be used to obtain candidate identifiers based on the user search terms according to the entity extraction prompts. Based on the multiple candidate identifiers that correspond one-to-one with the multiple user search terms, multiple product identifiers can be obtained. For example, multiple candidate identifiers can be used as multiple product identifiers, or multiple candidate identifiers can be precisely deduplicated to remove identical candidate identifiers and obtain multiple product identifiers.

[0037] Through the above methods, in this embodiment of the disclosure, multiple user search terms can be obtained from the target platform, and multiple product identifiers can be obtained based on these multiple user search terms. In other words, in this embodiment of the disclosure, multiple product identifiers are obtained using multiple user search terms obtained from the target platform as the data foundation. This bypasses the limitations of traditional solutions that require content understanding and simulated clicks on the target brand's official website. Furthermore, because user search terms naturally reflect real market concerns and product names, the coverage is broader, effectively capturing key information that the target brand's official website may not highlight or omit. This improves the richness and timeliness of the data foundation from the source, laying a solid foundation for subsequently building comprehensive and accurate product architecture information, thereby further improving the accuracy and efficiency of product architecture information acquisition. On the other hand, multiple product identifiers originating from multiple user search terms also improve the relevance between product architecture information and user search terms, thereby enhancing the interpretability of the product architecture information.

[0038] Furthermore, when multiple product identifiers are obtained based on multiple user search terms, entity extraction instructions can be constructed. For each user search term among the multiple user search terms, the second major model is used to obtain candidate identifiers based on the user search terms, following the entity extraction prompts. Then, based on the multiple candidate identifiers that correspond one-to-one with the multiple user search terms, multiple product identifiers are obtained. In other words, in this embodiment of the disclosure, intelligent and batch conversion from user search terms to product identifiers is achieved by introducing a second major model and designing entity extraction instructions. In this process, for each user search term among the multiple user search terms, the second major model can understand its natural language expression and accurately extract key entity information (e.g., brand, product line, category, series, model / specification, etc.) to form candidate identifiers. The high degree of automation not only significantly reduces the workload of manual sorting and rule writing, but also flexibly handles diverse, ambiguous, and even colloquial expressions in user search terms, effectively improving the coverage and accuracy of product identifier extraction.

[0039] Furthermore, when constructing entity extraction instructions, entity extraction examples and industry knowledge related to multiple user search terms can be obtained. Based on these examples and industry knowledge, entity extraction suggestions are constructed. In other words, this embodiment innovatively integrates entity extraction examples and specific industry knowledge when constructing entity extraction instructions, effectively providing the second model with high-quality "learning examples" and a specialized "domain dictionary" to achieve precise guidance and knowledge enhancement. This suggestion engineering construction strategy significantly improves the second model's understanding and recognition capabilities of user search terms, making the resulting candidate identifiers more standardized in format and content, thereby fundamentally improving the professionalism and structural consistency of product architecture information.

[0040] After executing step S101, step S102 will be executed, that is, the multiple product identifiers will be grouped to obtain the identifier group under each grouping granularity in N grouping granularities.

[0041] For step S102, this disclosure provides a first optional implementation method, including: Step S102-1: Obtain multiple code groups that correspond one-to-one with multiple product identifiers.

[0042] For each product identifier among multiple product identifiers, the coding group corresponding to the product identifier includes N+1 coding results. The N+1 coding results are used to quantize the product identifier from coarse to fine. That is, they are used to convert high-dimensional, discrete or unstructured product identifiers into structured and discrete N+1 coding results from coarse to fine.

[0043] In this embodiment of the disclosure, when obtaining multiple coding groups that correspond one-to-one with multiple product identifiers, each product identifier can be vectorized to obtain an identifier vector. Then, a target encoder is used to encode the identifier vector to obtain the coding group corresponding to the product identifier. "Vectorizing the product identifier to obtain the identifier vector" can mean mapping a high-dimensional, discrete, or unstructured product identifier to a low-dimensional continuous vector space to obtain the identifier vector. In one example, a quantization module can be used to vectorize the product identifier to obtain the identifier vector. Here, the quantization module can be a third major model or a specified minor model. The third major model can be, but is not limited to, the Qwen3 model; the specified minor model can be, but is not limited to, the Bidirectional Encoder Representations from Transformers (BERT) model. After vectorizing the product identifier to obtain the identifier vector, a target encoder is used to encode the identifier vector to obtain the coding group corresponding to the product identifier. The target encoder can be a Residual Quantized Variational Autoencoder (RQ-VAE), specifically a trained RQ-VAE.

[0044] Please combine Figure 2 For example, the target encoder is an RQ-VAE, which includes an encoder module and a residual quantization (RQ) module. The RQ module provides N+1 codebooks. For instance, in the luxury goods industry, where the common product hierarchy and classification system is "brand > product line > category > series > model / specification," N+1=5. The 5 codebooks can include the first codebook CB1 corresponding to "brand," the second codebook CB2 corresponding to "product line," the third codebook CB3 corresponding to "category," the fourth codebook CB4 corresponding to "series," and the fifth codebook CB5 corresponding to "model / specification."

[0045] Based on this, in the above example, when using the target encoder to encode the identifier vector to obtain the code group corresponding to the product identifier, the Encoder can be used to encode the identifier vector X to obtain the initial encoding result Z0. Then, in the RQ module: Using the first codebook CB1, the initial encoding result Z0 is quantized, that is, the first codeword EC1 that is most similar to the initial encoding result Z0 is found in the first codebook CB1, so as to characterize the encoding result for the "brand" entity directly included or implicit in the product identifier. Obtain the residual between the initial encoding result Z0 and the first codeword EC1, and use it as the first residual R1; The first residual R1 is quantized using the second codebook CB2. That is, the second codeword EC2 that is most similar to the first residual R1 is found in the second codebook CB2, so as to characterize the encoding result for the "product line" entity that is directly included or implicit in the product identifier. Obtain the residual between the first residual R1 and the second codeword EC2, and use it as the second residual R2; The second residual R2 is quantized using the third codebook CB3. That is, the third codeword EC3 that is most similar to the second residual R2 is found in the third codebook CB3 to represent the encoding result for the "category" entity that is directly included or implicit in the product identifier. Obtain the residual between the second residual R2 and the third codeword EC3, and use it as the third residual R3; The third residual R3 is quantized using the fourth codebook CB4. That is, the fourth codeword EC4 that is most similar to the third residual R3 is found in the fourth codebook CB4 to characterize the encoding result for the “series” entities that are directly included or implicit in the product identifier. Obtain the residual between the third residual R3 and the fourth codeword EC4, and use it as the fourth residual R4; The fourth residual R4 is quantized using the fifth codebook CB5. That is, the fifth codeword EC5, which is most similar to the fourth residual R4, is searched in the fifth codebook CB5 to characterize the coding result for the “model / specification” entity that is directly included or implicit in the product identification.

[0046] Among them, the first codeword EC1, the second codeword EC2, the third codeword EC3, the fourth codeword EC4, and the fifth codeword EC5 are used to form the coding group corresponding to the product identifier. That is, the coding group corresponding to the product identifier includes the first codeword EC1, the second codeword EC2, the third codeword EC3, the fourth codeword EC4, and the fifth codeword EC5.

[0047] Step S102-2: Based on multiple coding groups, group multiple product identifiers to obtain identifier groups under each grouping granularity in N grouping granularities.

[0048] Among the N grouping granularities, each grouping granularity can have at least one identifier group.

[0049] Furthermore, in this embodiment of the disclosure, "grouping multiple product identifiers based on multiple coding groups to obtain identifier groups under each of N grouping granularities" may include: From multiple coding groups, determine the M target coding groups whose first j coding results are the same; From multiple product identifiers, identify M first target product identifiers that correspond one-to-one with M target code groups; Based on M first target product identifiers, construct an identifier group under the j-th group granularity among N group granularities.

[0050] Where 1≤j≤N and j is an integer; M≥1 and M is an integer.

[0051] It should be noted that, in this embodiment of the disclosure, the specific steps of "grouping multiple product identifiers based on multiple coding groups to obtain identifier groups under each of the N grouping granularities" can be executed cyclically. That is, after "determining M target coding groups with the same first j coding results from multiple coding groups; determining M first target product identifiers that correspond one-to-one with the M target coding groups from multiple product identifiers; constructing identifier groups under the j-th grouping granularity of the N grouping granularities based on the M first target product identifiers," a "+1" operation can be performed on j, and then... The process involves repeatedly identifying M target coding groups from multiple coding groups, where the first j coding results are identical; identifying M first target product identifiers from multiple product identifiers that correspond one-to-one with the M target coding groups; and constructing identifier groups at the j-th grouping granularity among N grouping granularities. This process is repeated until j=N. At j=N, the process is repeated again, identifying M target coding groups from multiple coding groups, where the first j coding results are identical; identifying M first target product identifiers from multiple product identifiers that correspond one-to-one with the M target coding groups; and constructing identifier groups at the j-th grouping granularity among N grouping granularities. The process ends after "Constructing an identifier group under the j-th group granularity among N grouping granularities for a target product identifier"; or, after "Determining M target coding groups with the same first j coding results from multiple coding groups; determining M first target product identifiers that correspond one-to-one with the M target coding groups from multiple product identifiers; constructing an identifier group under the j-th group granularity among N grouping granularities"; or, after "Determining M target coding groups with the same first j coding results from multiple coding groups; determining M first target product identifiers that correspond one-to-one with the M target coding groups from multiple product identifiers; constructing an identifier group under the j-th group granularity among N grouping granularities based on the M first target product identifiers", a "-1" operation can be performed on j, and then "Determining M target coding groups with the same first j coding results from multiple coding groups; determining M first target product identifiers that correspond one-to-one with the M target coding groups from multiple product identifiers" can be performed again. In the identification process, M first target product identifiers are identified that correspond one-to-one with M target coding groups; based on the M first target product identifiers, an identifier group is constructed under the j-th group granularity among N grouping granularities. This process is repeated until j=1. When j=1, the process ends again by "identifying M target coding groups with the same first j coding results from multiple coding groups; identifying M first target product identifiers that correspond one-to-one with the M target coding groups from multiple product identifiers; and constructing an identifier group under the j-th group granularity among N grouping granularities".

[0052] For example, the multiple product identifiers under the target brand XNE, and the multiple identifier groups corresponding one-to-one with the multiple product identifiers, include: XNE N°X Fragrance – First coding group: 5, 8, 1, 1, 1; XNE N°X Eau de Parfum – Second coding group: 5, 8, 1, 1, 2; XNE N°X Fresh Water – Third coding group: 5, 8, 1, 1, 3; N°X Fragrance Classic - Fourth Code Group: 5, 8, 1, 1, 1; XNEYY Perfume - Fifth coding group: 5, 8, 1, 2, empty; YY Gentle Eau de Toilette – Sixth Code Group: 5, 8, 1, 2, 1; YY Vitality Eau de Toilette – Seventh Code Group: 5, 8, 1, 2, 2; Lilac No. 1 – Eighth coding group: 5, 8, 2, 1, empty; Lilac essence – Ninth coding group: 5, 8, 2, 1, 1; Lilac Face Cream – 10th Code Group: 5, 8, 2, 1, 2; XNE Lilac Eye Cream – Eleventh coding group: 5, 8, 2, 1, 3; ZZ Skin Base Eye Serum - Twelfth Code Group: 5, 8, 2, 2, 2; XNE Watch – Thirteenth Code Group: 5, 11, 1, empty, empty; XNE's JNN mechanical watch - fourteenth coding group: 5, 11, 1, 1, 1; JNN Classic Black and White Ceramic Watch - Number 15: 5, 11, 1, 1, 1; DD Original Watch - Sixteenth Code Group: 5, 11, 1, 2, 1; DD Diamond-set Watch – Number 17: 5, 11, 1, 2, 2; XNE's EE Ceramic and Diamond Ring - Code 18: 5, 11, 2, 1, 1; XNE's EE Asymmetrical Earrings - Nineteenth Code Group: 5, 11, 2, 1, 2; XNE FF Lion Head Ring - Twentieth Code Group: 5, 11, 2, 2, 1; XNE Lion Head Ring – Code Group 21: 5, 11, 2, 2, 1; XNE FF Lion Design Ring – Code Group 22: 5, 11, 2, 2, 1.

[0053] Therefore, when j=1, the 22 first target coding groups with the same first coding result will be determined from multiple coding groups, including: First coding group: 5, 8, 1, 1, 1; Second coding group: 5, 8, 1, 1, 2; Third coding group: 5, 8, 1, 1, 3; Fourth coding group: 5, 8, 1, 1, 1; Fifth coding group: 5, 8, 1, 2, empty; Sixth coding group: 5, 8, 1, 2, 1; Seventh coding group: 5, 8, 1, 2, 2; Eighth coding group: 5, 8, 2, 1, empty; Ninth coding group: 5, 8, 2, 1, 1; Tenth coding group: 5, 8, 2, 1, 2; Eleventh coding group: 5, 8, 2, 1, 3; Twelfth coding group: 5, 8, 2, 2, 2; Thirteenth coding group: 5, 11, 1, empty, empty; Fourteenth coding group: 5, 11, 1, 1, 1; The fifteenth coding group: 5, 11, 1, 1, 1; Sixteenth coding group: 5, 11, 1, 2, 1; Seventeenth coding group: 5, 11, 1, 2, 2; Eighteenth coding group: 5, 11, 2, 1, 1; Nineteenth coding group: 5, 11, 2, 1, 2; Twentieth coding group: 5, 11, 2, 2, 1; Twenty-first coding group: 5, 11, 2, 2, 1; The 22nd coding group: 5, 11, 2, 2, 1.

[0054] The 22 first target product identifiers corresponding one-to-one with the above 22 first target code groups include: XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette; Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum; XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch; XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0055] These 22 primary target product identifiers will be used to form identifier group 11 under the first group granularity (brand) of N (N=4) group granularities for the luxury goods industry.

[0056] Similarly, we can obtain the identifier group under the second grouping granularity (product line) in the 4-grouping granularity, including: Identifier Group 21 (corresponding to the beauty and personal care product line): XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette; Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum.

[0057] Logo Group 22 (corresponding to the watch and jewelry product lines): XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch; XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0058] Similarly, we can obtain the identifier group under the third grouping granularity (category) out of the four grouping granularities, including: Identifier Group 31 (corresponding to perfume category): XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette.

[0059] Identifier Group 32 (corresponding to skincare product category): Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum.

[0060] Identifier Group 33 (corresponding watch category): XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch.

[0061] Identifier group 34 (corresponding to jewelry category): XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0062] Similarly, we can obtain the identifier group under the fourth grouping granularity (series) out of the four grouping granularities, including: Identifier group 41 (corresponding to series X): XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic.

[0063] Identifier group 42 (corresponding to YY series): XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette.

[0064] Identifier Group 43 (corresponding to the Lilac series): Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream.

[0065] Identifier group 44 (corresponding to ZZ series): ZZ Skin Base Eye Serum.

[0066] Identifier group 45 (corresponding to JNN series): XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch.

[0067] Identifier group 46 (corresponding to DD series): DD Original Watches; DD diamond-set watch.

[0068] Identifier group 47 (corresponding to EE series): XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings.

[0069] Identifier group 48 (corresponding to FF series): XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0070] Through the above methods, in this embodiment of the disclosure, multiple coding groups corresponding one-to-one with multiple product identifiers can be obtained (for each product identifier among multiple product identifiers, the coding group corresponding to the product identifier includes N+1 coding results, and the N+1 coding results are used to quantify the product identifier from coarse to fine). Based on the multiple coding groups, the multiple product identifiers are grouped to obtain identifier groups under each of the N grouping granularities. In other words, in this embodiment of the disclosure, by obtaining a coding group including N+1 coding results for each product identifier, high-dimensional, discrete, or unstructured product identifiers are converted into structured, discrete N+1 coding results that can be precisely defined and processed by a computer. This process essentially endows each product identifier with a multi-dimensional "digital gene," where the coding result of each dimension corresponds to a specific product level and classification system (e.g., brand, product line, category, series, or model / specification). This will provide a unified, accurate, and efficient comparison basis for subsequent product identifier grouping, thereby improving the accuracy of the product identifier grouping results.

[0071] Furthermore, when obtaining multiple coding groups that correspond one-to-one with multiple product identifiers, each product identifier can be vectorized to obtain an identifier vector. Then, a target encoder is used to encode this identifier vector, resulting in a coding group corresponding to the product identifier. In this process, on the one hand, vectorization fully captures the semantic information of the product identifiers as identifier vectors; on the other hand, the target encoder accurately and discretely maps the abstract identifier vectors to specific product levels and classification systems. This not only avoids the rigidity and inaccuracy of hard matching strategies based on keyword rules but also intelligently understands the implicit relationships between products. This lays a reliable structured data foundation for generating clear and logically rigorous product architecture information, further improving the accuracy of the product architecture information.

[0072] Furthermore, when grouping multiple product identifiers based on multiple coding groups to obtain identifier groups under each of the N grouping granularities, M target coding groups with the same first j coding results can be determined from the multiple coding groups, and M first target product identifiers corresponding one-to-one with the M target coding groups can be determined from the multiple product identifiers. Then, based on the M first target product identifiers, the identifier group under the j-th grouping granularity of the N grouping granularities is constructed. In other words, in this embodiment of the disclosure, the identifier group under the j-th grouping granularity of the N grouping granularities is constructed by "determining M target coding groups with the same first j coding results from multiple coding groups". In this process, the inherent relationship between product hierarchy and classification system is cleverly mapped to the grouping logic to achieve the automation and hierarchical nature of product identifier grouping. That is, starting from the coarsest brand layer (j=1), it gradually advances to a finer model / specification layer (j increases); or, starting from the finest model / specification layer (j=N), it gradually advances to a coarser brand layer (j decreases). Each grouping of identifiers precisely aggregates all product identifiers that belong to the same category within the product hierarchy and classification system. This naturally and comprehensively constructs the complete hierarchical structure of the entire target brand, thereby improving the accuracy of subsequent steps and ultimately enhancing the accuracy of product architecture information.

[0073] Hereinafter, in the embodiments of this disclosure, the target encoder is... Figure 2 Using the trained RQ-VAE shown as an example, the process of training the initial encoder (the original RQ-VAE) to obtain the target encoder is explained. This process may include: Obtain product label samples; The product identification sample is vectorized to obtain the identification vector sample; The identifier vector sample is encoded using the Encoder in the initial encoder to obtain the initial encoded result sample; The initial encoded result samples are quantized using the RQ module in the initial encoder to obtain N+1 encoded result samples. Based on N+1 encoded result samples, an encoded sample corresponding to the product identification sample is obtained; The encoded sample is decoded using the Decoder in the initial encoder to obtain the decoded sample. The initial encoder is trained using the identifier vector samples, the initial encoding result samples, the encoded samples, and the decoded result samples to obtain the target encoder.

[0074] The specific functions and examples of the product identification samples can be found in the aforementioned description of product identification, and will not be repeated here. "Vectorizing the product identification samples to obtain identification vector samples" can be achieved using a quantization module to vectorize the product identification samples. Furthermore, as mentioned earlier, in this embodiment, the quantization module can be a third major model or a specified minor model. The third major model can be, but is not limited to, Qwen3; the specified minor model can be, but is not limited to, BERT.

[0075] Please combine Figure 3 In this embodiment of the disclosure, specifically, after obtaining a product identification sample and using a quantization module to vectorize the product identification sample to obtain an identification vector sample X', and then using the Encoder in the initial encoder to encode the identification vector sample X' to obtain the initial encoding result sample Z0', in the RQ module of the initial encoder: Using the first codebook CB1', the initial encoding result sample Z0' is quantized, that is, the first codeword sample EC1' that is most similar to the initial encoding result sample Z0' is found in the first codebook CB1'; Obtain the residual between the initial encoding result sample Z0' and the first codeword sample EC1', and use it as the first residual sample R1'; Using the second codebook CB2', the first residual sample R1' is quantized, that is, the second codeword sample EC2' that is most similar to the first residual sample R1' is found in the second codebook CB2'; Obtain the residual between the first residual sample R1' and the second codeword sample EC2', and use it as the second residual sample R2'; Using the third codebook CB3', the second residual sample R2' is quantized, that is, the third codeword sample EC3' that is most similar to the second residual sample R2' is found in the third codebook CB3'; Obtain the residual between the second residual sample R2' and the third codeword sample EC3', and use it as the third residual sample R3'; Using the fourth codebook sample CB4', the third residual sample difference R3' is quantized, that is, the fourth codeword sample EC4' that is most similar to the third residual sample R3' is found in the fourth codebook CB4'; Obtain the residual between the third residual sample R3' and the fourth codeword sample EC4', and use it as the fourth residual sample R4'; Using the fifth codebook CB5', the fourth residual sample R4' is quantized, that is, the fifth codeword sample EC5' that is most similar to the fourth residual sample R4' is found in the fifth codebook CB5'.

[0076] Next, based on the first codeword sample EC1', the second codeword sample EC2', the third codeword sample EC3', the fourth codeword sample EC4', and the fifth codeword sample EC5', the encoded sample ZQ corresponding to the product identification sample can be obtained. For example, the first codeword sample EC1', the second codeword sample EC2', the third codeword sample EC3', the fourth codeword sample EC4', and the fifth codeword sample EC5' can be weighted and fused to obtain the encoded sample ZQ corresponding to the product identification sample. Specifically, when performing weighted fusion on the first codeword sample EC1', the second codeword sample EC2', the third codeword sample EC3', the fourth codeword sample EC4', and the fifth codeword sample EC5', the set of weights used can be set according to application requirements. For example, the first weight M1 corresponding to the first codeword sample EC1', the second weight M2 corresponding to the second codeword sample EC2', the third weight M3 corresponding to the third codeword sample EC3', the fourth weight M4 corresponding to the fourth codeword sample EC4', and the fifth weight M5 corresponding to the fifth codeword sample EC5' can all be set to 0.20.

[0077] After obtaining the encoded sample ZQ corresponding to the product identification sample, the decoder in the initial encoder can be used to decode the encoded sample ZQ to obtain the decoded result sample X''.

[0078] Subsequently, the initial encoder can be trained using the identifier vector sample X', the initial encoded result sample Z0', the encoded sample ZQ, and the decoded result sample X'' to obtain the target encoder. For example, by minimizing the reconstruction loss between the identifier vector sample X' and the decoded result sample X'', and by introducing quantization losses (including codebook loss and commitment loss), the initial encoded result sample Z0' and the encoded sample ZQ can be made closer together, thus achieving the goal of training the initial encoder to obtain the target encoder. "Training the initial encoder to obtain the target encoder" can include adjusting the parameters of the Encoder, RQ module, and Decoder in the initial encoder. Simultaneously, the parameters of the quantization module can also be adjusted during the training process. Here, the parameters of the RQ module include the specific contents of the first, second, third, fourth, and fifth codebooks.

[0079] In the process of training the initial encoder to obtain the target encoder, on the one hand, the training data (e.g., product identification samples) can be obtained according to the target brand, so that the target encoder can be applied to the acquisition of product architecture information of the target brand, thereby improving the scalability of the product architecture information generation method based on a large model provided in this embodiment; on the other hand, the parameters of the Encoder, RQ module and Decoder are controllable and do not involve training related to the large model. Therefore, the training time is less, the training cost is low, and the training overhead can be reduced.

[0080] For step S102, this disclosure provides a second optional implementation method, including: Step S102-1: Cluster the multiple product identifiers to obtain a tree structure.

[0081] The tree structure has N clustering levels, ranging from coarse to fine.

[0082] In this embodiment of the disclosure, the value of N and the N clustering levels can be determined based on the industry to which the target brand belongs. For example, for the luxury goods industry, N=4, and the 4 clustering levels can include brand > product line > category > series > model / specification.

[0083] Furthermore, in this embodiment of the disclosure, when clustering multiple product identifiers to obtain a tree structure, a target clustering algorithm can be used to cluster the multiple product identifiers to obtain the tree structure. The target clustering algorithm can be, but is not limited to, an agglomerative hierarchical clustering algorithm, a bottom-up hierarchical clustering algorithm, or a systematic clustering method.

[0084] Step S102-2: Based on the clustering results of the kth clustering level among the N clustering levels, construct the identifier group at the kth grouping granularity among the N grouping granularities.

[0085] Where 1≤k≤N, and k is an integer.

[0086] It should be noted that in this embodiment, step S102-2 can be executed cyclically. That is, after constructing the identifier group at the kth grouping granularity among N grouping granularities based on the clustering results at the kth clustering level among N clustering levels, a "-1" operation can be performed on k, and the identifier group at the kth grouping granularity among N grouping granularities can be constructed again based on the clustering results at the kth clustering level among N clustering levels. This process is repeated until k=1, and then the identifier group at the kth grouping granularity among N grouping granularities is constructed again based on the clustering results at the kth clustering level among N clustering levels. The process ends after constructing the label group at the kth grouping granularity within N grouping levels, based on the clustering results at the kth clustering level within N clustering levels. Then, k can be incremented by 1, and the label group at the kth grouping granularity within N grouping levels can be constructed again based on the clustering results at the kth clustering level within N clustering levels. This process is repeated until k=N, and the process ends after constructing the label group at the kth grouping granularity within N grouping levels once more based on the clustering results at the kth clustering level within N clustering levels.

[0087] For example, several product identifiers under the target brand XNE include: XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette; Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum; XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch; XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0088] In this embodiment of the disclosure, multiple product identifiers are clustered to obtain a tree structure. The clustering results at the fourth clustering level (series) include: Clustering result 41 (corresponding to series X): XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic.

[0089] Clustering result 42 (corresponding to the YY series): XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette.

[0090] Clustering result 43 (corresponding to the lilac series): Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream.

[0091] Clustering result 44 (corresponding to the ZZ series): ZZ Skin Base Eye Serum.

[0092] Clustering result 45 (corresponding to JNN series): XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch.

[0093] Clustering result 46 (corresponding to DD series): DD Original Watches; DD diamond-set watch.

[0094] Clustering result 47 (corresponding to EE series): XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings.

[0095] Clustering result 48 (corresponding to the FF series): XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0096] Specifically, clustering result 41 is used to construct the label group 41 under the 4th grouping granularity among N (N=4 in this case); clustering result 42 is used to construct the label group 42 under the 4th grouping granularity among 4 grouping granularities; clustering result 43 is used to construct the label group 43 under the 4th grouping granularity among 4 grouping granularities; clustering result 44 is used to construct the label group 44 under the 4th grouping granularity among 4 grouping granularities; clustering result 45 is used to construct the label group 45 under the 4th grouping granularity among 4 grouping granularities; clustering result 46 is used to construct the label group 46 under the 4th grouping granularity among 4 grouping granularities; clustering result 47 is used to construct the label group 47 under the 4th grouping granularity among 4 grouping granularities; and clustering result 48 is used to construct the label group 48 under the 4th grouping granularity among 4 grouping granularities.

[0097] Clustering multiple product identifiers into a tree structure, the clustering results at the third clustering level (category) include: Clustering result 31 (corresponding to perfume category): XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette.

[0098] Clustering result 32 (corresponding to skincare product category): Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum.

[0099] Clustering result 33 (corresponding to watch category): XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch.

[0100] Clustering result 34 (corresponding to the jewelry category): XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0101] Among them, clustering result 31 is used to construct the label group 31 under the third grouping granularity among the four grouping granularities; clustering result 32 is used to construct the label group 32 under the third grouping granularity among the four grouping granularities; clustering result 33 is used to construct the label group 33 under the third grouping granularity among the four grouping granularities; and clustering result 34 is used to construct the label group 34 under the third grouping granularity among the four grouping granularities.

[0102] Clustering multiple product identifiers into a tree structure, the clustering results at the second clustering level (product line) include: Clustering result 21 (corresponding to beauty and personal care product lines): XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette; Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum.

[0103] Clustering result 22 (corresponding to watch and jewelry product lines): XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch; XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0104] Among them, clustering result 21 is used to construct the identifier group 21 under the second grouping granularity among the four grouping granularities; clustering result 22 is used to construct the identifier group 22 under the second grouping granularity among the four grouping granularities.

[0105] Clustering multiple product identifiers into a tree structure, the clustering results at the first cluster level (brand) include: Clustering result 11: XNE N°X fragrance; XNE N°X Eau de Parfum; XNE N°X Fresh Water; N°X perfume classic version; XNE YY perfume; YY Gentle Eau de Toilette; YY Vibrant Eau de Toilette; Lilac No. 1; Lilac extract; Lilac face cream; XNE Lilac Eye Cream; ZZ Skin Base Eye Serum; XNE watches; XNE's JNN mechanical watches; JNN Classic Black and White Ceramic Watch; DD Original Watches; DD diamond-set watch; XNE's EE ceramic and diamond ring; XNE's EE asymmetrical earrings; XNE FF Lion Head Ring; XNE Lion Head Ring; XNE FF Lion Design Ring.

[0106] Among them, clustering result 11 is used to construct the identifier group 11 under the first grouping granularity among the four grouping granularities.

[0107] Through the above methods, in this embodiment of the disclosure, multiple product identifiers can be clustered to obtain a tree structure, that is, a tree structure with N clustering levels from coarse to fine. Based on the clustering results at the j-th clustering level among the N clustering levels, an identifier group at the j-th grouping granularity among the N grouping granularities is constructed. In other words, in this embodiment of the disclosure, a tree structure can be automatically constructed based on the distribution pattern of the product identifiers themselves through clustering processing, and multiple product identifiers can be grouped accordingly to obtain grouping results, thereby improving the efficiency of obtaining grouping results.

[0108] In some optional implementations, step S103, namely, "using the first large model to perform identifier integration for the first identifier group to obtain the first identifier integration result," may include: Obtain the identifier to be integrated corresponding to the first identifier group; Build integrated identifier prompts; Using the first major model, and following the logo integration prompts, the logos to be integrated are integrated to obtain the first logo integration result.

[0109] As previously described, in this embodiment of the disclosure, the first identifier group can be the identifier group under the i-th grouping granularity among N grouping granularities. Wherein, 1≤i≤N and i is an integer.

[0110] Furthermore, in this embodiment of the disclosure, "obtaining the identifier to be integrated corresponding to the first identifier group" may include: When the granularity of the i-th group is the finest among the N group granularities, the product identifier in the first identifier group is taken as the identifier to be integrated; Alternatively, if the i-th group granularity is not the finest group granularity among the N group granularities, the identifier to be integrated is obtained based on the integration result of the second identifier.

[0111] The second identifier integration result can be obtained by integrating the identifiers of the second identifier group using the first major model; the second identifier group can be the identifier group under the (i+1)th grouping granularity in N grouping granularities, and the product identifiers in the second identifier group are included in the first identifier group.

[0112] Furthermore, it should be noted that in this embodiment of the present disclosure, when the i-th group granularity is not the finest group granularity among the N group granularities, and there are no other product identifiers in the first identifier group that are not included in the second identifier group, the identifier to be integrated can be obtained directly based on the integration result of the second identifier. When the i-th group granularity is not the finest group granularity among the N group granularities, and there are other product identifiers in the first identifier group that are not included in the second identifier group, "obtaining the identifier to be integrated based on the integration result of the second identifier" may include: obtaining a part of the identifier to be integrated based on the integration result of the second identifier, and obtaining another part of the identifier to be integrated based on other product identifiers in the first identifier group that are not included in the second identifier group. For example, other product identifiers in the first identifier group that are not included in the second identifier group may be used as another part of the identifier to be integrated.

[0113] For example, for the most detailed grouping granularity among the aforementioned four grouping granularities, such as the identifier group 41 under the fourth grouping granularity, when it is used as the first identifier group, the product identifiers in the first identifier group, namely, "XNE N°X Fragrance", "XNE N°X Eau de Parfum", "XNE N°X Fresh Water" and "N°X Fragrance Classic" can be used together as identifiers to be integrated.

[0114] For identifiers that are not the most subdivided of the four grouping granularities mentioned above, such as identifier group 34 under the third grouping granularity, when used as the first identifier group, the identifier to be integrated can be obtained based on the second identifier integration result. That is, the identifier to be integrated can be obtained based on the second identifier integration result obtained by using the first major model to integrate the identifiers of the second identifier group. Here, the second identifier group is the identifier group under the fourth grouping granularity of the four grouping granularities, and the product identifiers in the second identifier group are included in the first identifier group. That is, the second identifier group consists of identifier group 47 and identifier group 48. The second identifier integration result includes the second identifier integration result obtained by using the first major model to integrate the identifiers of identifier group 47 (hereinafter referred to as the second identifier integration result A1), and another second identifier integration result obtained by using the first major model to integrate the identifiers of identifier group 48 (hereinafter referred to as the second identifier integration result A2). Therefore, when "obtaining the identifier to be integrated based on the second identifier integration result", the identifier to be integrated can be obtained directly based on the second identifier integration result A1 and the second identifier integration result A2.

[0115] For identifier groups that are not the most subdivided among the four grouping granularities mentioned above, such as identifier group 33 under the third grouping granularity, when it is used as the first identifier group, the identifier to be integrated can be obtained based on the second identifier integration result. That is, the identifier to be integrated can be obtained based on the second identifier integration result obtained by using the first major model to perform identifier integration on the second identifier group. Here, the second identifier group is the identifier group under the fourth grouping granularity among the four grouping granularities, and the product identifiers in the second identifier group are included in the first identifier group. That is, the second identifier group is identifier group 45 and identifier group 46, and the second identifier integration result includes the second identifier integration result obtained by using the first major model to perform identifier integration on identifier group 45 (hereinafter referred to as the second identifier integration result B1), and another second identifier integration result obtained by using the first major model to perform identifier integration on identifier group 46 (hereinafter referred to as the second identifier integration result B2). Therefore, when “obtaining the identifier to be integrated based on the second identifier integration result”, a portion of the identifier to be integrated can be obtained based on the second identifier integration result B1 and the second identifier integration result B2. At the same time, other product identifiers that are not included in the second identifier group 45 and the second identifier group 46 (i.e., XNE watch) in the first identifier group 33 are used as another portion of the identifier to be integrated.

[0116] The above, "obtaining the identifier to be integrated based on the second identifier integration result" can include: If there is a general term for the overall identifier in the second identifier integration result, the general term for the overall identifier in the second identifier integration result shall be taken as the identifier to be integrated; Alternatively, if there is no overall identifier collective name in the second identifier integration result, the identifier to be integrated is obtained based on the product hierarchical classification information in the second identifier integration result.

[0117] The overall identifier can be a general identifier summary result; the product hierarchical classification information can be product classification information with a hierarchical structure. Based on this, in this embodiment of the disclosure, "obtaining the identifier to be integrated based on the product hierarchical classification information in the second identifier integration result" can be: taking the root node identifier in the product hierarchical classification information as the identifier to be integrated. For example, given first-level product classification information, including root node identifier A, and leaf node identifiers A1 and A2 below root node identifier A, root node identifier A can be used as the identifier to be integrated. As another example, given second-level product classification information, including root node identifiers B1 and B2, and leaf node identifiers B11 and B12 below root node identifier B1; since root node identifier B2 is an independent node identifier, root node identifiers B1 and B2 can be used as the identifiers to be integrated. Similarly, given third-level product classification information, including root node identifier C1, root node identifier C2, leaf node identifiers C11 and C12 below root node identifier C1, and leaf node identifiers C21 and C22 below root node identifier C2, root node identifiers C1 and C2 can be used as the identifiers to be integrated.

[0118] Furthermore, in this embodiment of the disclosure, the "build identifier integration prompt" may include: Get an example of identifier integration; Requirements for obtaining identifier integration; Based on the identifier integration examples and requirements, we construct identifier integration prompts.

[0119] Among them, the identifier integration prompt, as a Prompt message, can be used to instruct the first major model on how to integrate the identifiers to obtain the first identifier integration result.

[0120] Furthermore, it should be noted that in this embodiment, the identifier integration example can be set according to application requirements, and this embodiment does not impose any limitations on it; the identifier integration requirements may include: generating a first identifier integration result including the overall identifier general term when the identifiers to be integrated can be summarized into a general identifier term; and / or, generating a first identifier integration result including product hierarchical classification information when product hierarchical classification information can be obtained based on the identifiers to be integrated. Here, the overall identifier general term can be a general identifier summarization result. Based on this, in this embodiment, "the identifiers to be integrated can be summarized into an overall identifier general term" can be understood as: the identifiers to be integrated can be summarized into a unique overall identifier general term; the product hierarchical classification information can be product classification information with a hierarchical structure.

[0121] For example, the identifier integration prompt can be as shown in Table 1: Table 1 In this embodiment of the disclosure, after obtaining the identifier to be integrated corresponding to the first identifier group and constructing the identifier integration prompt, the first large model can be used to integrate the identifiers to be integrated according to the identifier integration prompt to obtain the first identifier integration result.

[0122] For example, when the aforementioned identifier group 41 is taken as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include: "XNE N°X Fragrance", "XNE N°X Eau de Parfum", "XNE N°X Fresh Water", and "N°X Classic Perfume". Then, after constructing the identifier integration prompts, using the first major model, the identifiers to be integrated are integrated according to the identifier integration prompts, resulting in: Figure 4 The first identifier integration result shown includes the overall identifier collectively referred to as "XNE X Perfume" and has product hierarchical classification information (where the top layer is "XNE X Perfume"; the next layer of "XNE X Perfume" is "XNE N°X Fragrance", "XNE N°X Eau de Parfum" and "XNE N°X Fresh Water"). When the aforementioned identifier group 42 is taken as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include "XNE YY Perfume", "YY Gentle Eau de Parfum" and "YY Vibrant Eau de Parfum". Therefore, after constructing the identifier integration prompts, using the first major model, according to the identifier integration prompts, the identifiers to be integrated are integrated to obtain the following result: Figure 5The first integrated logo shown includes the overall logo collectively referred to as "XNE YY Perfume" and has product hierarchical classification information (where the top layer is "XNE YY Perfume"; the next layer of "XNE YY Perfume" is "YY Gentle Eau de Toilette" and "YY Vibrant Eau de Toilette").

[0123] Subsequently, when the aforementioned identifier group 31 is used as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include "XNE No. X Perfume" and "XNE YY Perfume". Then, after constructing the identifier integration prompts, using the first major model, the identifiers to be integrated are integrated according to the identifier integration prompts, resulting in the following: Figure 6 The first integrated logo shown includes the overall logo collectively referred to as "XNE Perfume" and has product hierarchical classification information (where the top layer is "XNE Perfume"; the next layer of "XNE Perfume" is "XNE No. X Perfume" and "XNE YY Perfume").

[0124] For example, when the aforementioned identifier group 45 is used as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include: "XNE's JNN Mechanical Watch" and "JNN Classic Black and White Ceramic Watch". Since "XNE's JNN Mechanical Watch" and "JNN Classic Black and White Ceramic Watch" are different names for "XNE JNN Automatic Mechanical Watch", after constructing the identifier integration prompt, the identifiers to be integrated are integrated using the first large model according to the identifier integration prompt. The resulting first identifier integration result can include the overall identifier collective name "XNE JNN Automatic Mechanical Watch". Moreover, the returned result only includes this overall identifier collective name, that is, it does not include "XNE's JNN Mechanical Watch" and "JNN Classic Black and White Ceramic Watch", as detailed below. Figure 7 As shown; when the aforementioned identifier group 46 is taken as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include: "DD Original Watch" and "DD Diamond-set Watch". Therefore, after constructing the identifier integration prompts, using the first large model, the identifiers to be integrated are integrated according to the identifier integration prompts, resulting in the following: Figure 8 The first logo integration result shown includes the overall logo collectively referred to as "XNE DD series watches" and has product hierarchical classification information (where the top layer is "XNE DD series watches"; the next layer of "XNE DD series watches" is "DD Original Watches" and "DD Diamond-set Watches").

[0125] Subsequently, when the aforementioned identifier group 33 is used as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include "XNE JNN Automatic Mechanical Watch", "DD Series Watch", and other product identifiers in the first identifier group that are not included in the second identifier group 45 and the second identifier group 46 (i.e., "XNE Watch"). Therefore, after constructing the identifier integration prompts, using the first large model, the identifiers to be integrated are integrated according to the identifier integration prompts, resulting in the following: Figure 9 The first logo integration result shown includes the overall logo collectively referred to as "XNE Watches" and has product hierarchical classification information (where the top layer is "XNE Watches"; the next layer of "XNE Watches" is "XNE JNN Automatic Mechanical Watches" and "DD Series Watches").

[0126] For example, when the aforementioned identifier group 47 is used as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include "XNE's EE ceramic and diamond ring" and "XNE's EE asymmetrical earrings". Then, after constructing the identifier integration prompts, using the first major model, the identifiers to be integrated are integrated according to the identifier integration prompts, resulting in the following: Figure 10 The first identifier integration result shown includes the overall identifier collectively referred to as "XNE EE Series Jewelry" and has product hierarchical classification information (where the top layer is "XNE EE Series Jewelry"; the next layer of "XNE EE Series Jewelry" is "XNE's EE Ceramic and Diamond Ring" and "XNE's EE Asymmetrical Earrings"). When the aforementioned identifier group 48 is taken as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include: "XNE FF Lion Head Ring", "XNE Lion Head Ring", and "XNE FF Lion Shape Ring". Since "XNE FF Lion Head Ring", "XNE Lion Head Ring", and "XNE FF Lion Shape Ring" refer to the same product, after constructing the identifier integration prompt, the identifiers to be integrated are integrated using the first large model according to the identifier integration prompt. The resulting first identifier integration result can include the overall identifier collectively referred to as "XNE FF Lion Head Ring". Moreover, the returned result only includes this overall identifier collectively referred to as "XNE EE Series Jewelry", that is, it does not include "XNE Lion Head Ring" and "XNE EE Asymmetrical Earrings". "FF Lion Design Ring", specifically as follows Figure 11 As shown.

[0127] Subsequently, when the aforementioned identifier group 34 is used as the first identifier group, the identifiers to be integrated corresponding to the first identifier group include "XNE's EE series jewelry" and "XNE FF lion head ring". Therefore, after constructing the identifier integration prompts, using the first large model, the identifiers to be integrated are integrated according to the identifier integration prompts, resulting in the following: Figure 12The first integrated logo shown includes the overall logo collectively referred to as "XNE Jewelry" and has product hierarchical classification information (where the top layer is "XNE Jewelry"; the next layer of "XNE Jewelry" is "XNE's EE Series Jewelry" and "XNE FF Lion Head Ring").

[0128] Through the above methods, in this embodiment of the disclosure, the identifier to be integrated corresponding to the first identifier group can be obtained, and an identifier integration prompt can be constructed. Then, using the first large model, the identifier to be integrated is integrated according to the identifier integration prompt to obtain the first identifier integration result. In other words, in this embodiment of the disclosure, by clearly designing the operation chain of "obtaining the identifier to be integrated → constructing the identifier integration prompt → using the first large model to perform identifier integration", the complex semantic induction and structuring tasks are transformed into computer instructions that can be stably executed using the first large model. The general capabilities of the first large model are effectively "task-focused" and "process-encapsulated", thereby significantly improving the accuracy of the first identifier integration result and providing a reliable guarantee for the automated and batch generation of high-quality first identifier integration results.

[0129] Furthermore, when obtaining the identifier to be integrated corresponding to the first identifier group, if the granularity of the i-th group is the finest among the N group granularities, the product identifier in the first identifier group can be used as the identifier to be integrated; or, if the granularity of the i-th group is not the finest among the N group granularities, the identifier to be integrated can be obtained based on the second identifier integration result. Specifically, obtaining the identifier to be integrated based on the second identifier integration result can be: if there is a general identifier term in the second identifier integration result, the general identifier term in the second identifier integration result can be used as the identifier to be integrated; or, if there is no general identifier term in the second identifier integration result, the identifier to be integrated can be obtained based on the product hierarchical classification information in the second identifier integration result. Here, the second identifier integration result includes the identifier integration result obtained by using the first major model to integrate the identifiers for the second identifier group; the second identifier group is the identifier group under the (i+1)-th group granularity among the N group granularities, and the product identifiers in the second identifier group are included in the first identifier group. This process employs a clever bottom-up information transfer and reuse mechanism. Specifically, when processing coarser grouping granularity, it does not directly process the original product identifiers, but instead uses the second identifier integration result (overall identifier general term or product hierarchical classification information) generated at a finer granularity as input. This greatly simplifies the complexity of identifier integration, allowing the first major model to focus on the summarization and abstraction of product identifiers at the current level, avoiding information overload. At the same time, it ensures the logical coherence of product architecture information between different levels, and the identifier integration result at the lower level naturally becomes the cornerstone of identifier integration at the upper level, thereby improving the overall efficiency of product architecture information generation and the rationality of the hierarchical structure.

[0130] Furthermore, when constructing identifier integration prompts, identifier integration examples and requirements can be obtained, and identifier integration prompts can be constructed based on these examples and requirements. The identifier integration requirements may include: generating a first identifier integration result including the overall identifier term when the identifiers to be integrated can be summarized into a general identifier term; and / or, generating a first identifier integration result including product-level classification information when product-level classification information can be obtained based on the identifiers to be integrated. In other words, in this embodiment, constructing identifier integration prompts by integrating "identifier integration examples" and "identifier integration requirements" essentially involves a precise "contextual fine-tuning" and "task specification" of the first main model. The identifier integration examples provide a directly imitable template for the first main model, clarifying the ideal output format and content depth; the identifier integration requirements clearly define the task execution standards. This prompting engineering can significantly guide the reasoning direction of the first main model, effectively constrain its output, making its results more consistent with business logic and more standardized in format, greatly reducing the randomness of output caused by the first main model's "free play," and further improving the professionalism and usability of the added product information.

[0131] As described above, in this embodiment of the present disclosure, step S103 can be executed cyclically; in this embodiment of the present disclosure, when executing step S104, all the obtained first identifier integration results can be processed in series and parallel to obtain product architecture information related to multiple product identifiers. Continuing with the aforementioned example, in... Figure 4 The first identifier integration result shown and Figure 5 The first identifier integration result shown is consistent with Figure 6 The first identifier integration result shown is subjected to a series-parallel operation, that is, the... Figure 4 The first identifier integration result shown is "XNE No. X perfume" and Figure 6 The first identifier integration result shown shows that "XNE No. X perfume" can be merged into a single node identifier, thus achieving [the desired result]. Figure 4 The first identifier integration result shown is consistent with Figure 6 The serial-parallel operation of the first identifier integration result shown will similarly... Figure 5 The first identifier integration result shown includes "XNEYY perfume" and Figure 6 The "XNEYY Perfume" in the first identifier integration result shown can be merged into a single node identifier, thus achieving... Figure 5 The first identifier integration result shown is consistent with Figure 6 The serial-parallel operation of the first identifier integration result shown is used to obtain, as follows: Figure 13 The partial product architecture information shown can be used to construct product architecture information related to multiple product identifiers.

[0132] Furthermore, it should be noted that in this embodiment of the disclosure, after obtaining the local product architecture information, semantic deduplication and redundancy pruning can be performed to obtain the final local product architecture information, which can be used to construct product architecture information related to multiple product identifiers. For example, for... Figure 13 The local product architecture information shown can be semantically deduplicated and redundantly pruned to remove [deduplication]. Figure 13 The “XNE N°X Fragrance” and “N°X Classic Perfume” in the text were obtained as follows: Figure 14 The final partial product architecture information is shown.

[0133] Please see Figure 15 This is a schematic diagram illustrating an application scenario of a product architecture information generation method based on a large model, as provided in this embodiment of the disclosure.

[0134] The product architecture information generation method based on a large model provided in this disclosure is applied to electronic devices. The electronic device can be a service device or a terminal device. Here, the service device can be a server, workbench, mainframe computer, or other similar computing device; the terminal device can be a workbench, mainframe computer, conventional computer, or other similar computing device.

[0135] Electronic devices are used for: Obtain multiple product identifiers; Multiple product identifiers are grouped to obtain identifier groups under each of N group granularities; where N≥2 and N is an integer; the N group granularities have different levels of coarseness; Using the first major model, we perform identifier integration for the first identifier group to obtain the first identifier integration result; where the first identifier group is the identifier group under the i-th grouping granularity among N grouping granularities; 1≤i≤N, and i is an integer; Based on the integration results of the first identifier, product architecture information related to multiple product identifiers is obtained.

[0136] It should be noted that, in the embodiments disclosed herein, Figure 15 The application scenario diagrams shown are for illustrative purposes only and are not restrictive. Those skilled in the art can use them as a basis for their own interpretation. Figure 15 The examples may be modified in various obvious ways and / or substitutions, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of this disclosure.

[0137] To better implement the product architecture information generation method based on a large model, this disclosure also provides a product architecture information generation apparatus based on a large model, which can be integrated into an electronic device. The electronic device can be a service device or a terminal device. Here, the service device can be a server, workbench, mainframe computer, or other similar computing device; the terminal device can be a workbench, mainframe computer, conventional computer, or other similar computing device. The following will be combined with… Figure 7 The schematic block diagram shown illustrates a product architecture information generation device 1600 based on a large model provided in the disclosed embodiments.

[0138] The product architecture information generation device 1600 based on a large model includes: The identifier acquisition unit 1601 is used to acquire multiple product identifiers; The identification grouping unit 1602 is used to group multiple product identifications to obtain identification groups under each of the N grouping granularities; where N≥2 and N is an integer; the N grouping granularities have different levels of coarseness; The identifier integration unit 1603 is used to perform identifier integration for the first identifier group using a large model to obtain the first identifier integration result; wherein, the first identifier group is the identifier group under the i-th grouping granularity among N grouping granularities; 1≤i≤N, and i is an integer; The information acquisition unit 1604 is used to obtain product architecture information related to multiple product identifiers based on the integration result of the first identifier.

[0139] In some alternative implementations, the identification grouping unit 1602 is used for: Obtain multiple coding groups that correspond one-to-one with multiple product identifiers; wherein, for each product identifier among the multiple product identifiers, the coding group corresponding to the product identifier includes N+1 coding results, and the N+1 coding results are used to quantize the product identifier from coarse to fine; Based on multiple coding groups, multiple product identifiers are grouped to obtain identifier groups under each of the N grouping granularities.

[0140] In some alternative implementations, the identification grouping unit 1602 is used for: For each product identifier among multiple product identifiers, the product identifier is vectorized to obtain an identifier vector; the identifier vector is encoded using a target encoder to obtain an encoding group corresponding to the product identifier.

[0141] In some alternative implementations, the identification grouping unit 1602 is used for: From multiple coding groups, determine M target coding groups whose first j coding results are the same; where 1≤j≤N and j is an integer; M≥1 and M is an integer; From multiple product identifiers, identify M first target product identifiers that correspond one-to-one with M target code groups; Based on M first target product identifiers, construct an identifier group under the j-th group granularity among N group granularities.

[0142] In some alternative implementations, the identification grouping unit 1602 is used for: Multiple product identifiers are clustered to obtain a tree structure; the tree structure has N clustering levels from coarse to fine. Based on the clustering results at the j-th clustering level among N clustering levels, construct the identifier group at the j-th grouping granularity among N grouping granularities; where 1≤j≤N and j is an integer.

[0143] In some alternative implementations, the identification integration unit 1603 is used for: Obtain the identifier to be integrated corresponding to the first identifier group; Build integrated identifier prompts; Using the first major model, and following the logo integration prompts, the logos to be integrated are integrated to obtain the first logo integration result.

[0144] In some alternative implementations, the identification integration unit 1603 is used for: When the granularity of the i-th group is the finest among the N group granularities, the product identifier in the first identifier group is taken as the identifier to be integrated; Alternatively, if the i-th group granularity is not the finest group granularity among the N group granularities, the identifier to be integrated is obtained based on the second identifier integration result; wherein, the second identifier integration result includes the identifier integration result obtained by using the first large model to perform identifier integration for the second identifier group; the second identifier group is the identifier group under the (i+1)-th group granularity among the N group granularities, and the product identifiers in the second identifier group are included in the first identifier group.

[0145] In some alternative implementations, the identification integration unit 1603 is used for: If there is a general term for the overall identifier in the second identifier integration result, the general term for the overall identifier in the second identifier integration result shall be taken as the identifier to be integrated; Alternatively, if there is no overall identifier collective name in the second identifier integration result, the identifier to be integrated is obtained based on the product hierarchical classification information in the second identifier integration result.

[0146] In some alternative implementations, the identification integration unit 1603 is used for: Get an example of identifier integration; Requirements for obtaining identifier integration; Based on the identifier integration examples and requirements, we construct identifier integration prompts.

[0147] In some alternative implementations, the identifier integration requirements include: If the identifiers to be integrated can be summarized into a general term for the overall identifiers, generate a first identifier integration result that includes the general term for the overall identifiers. And / or, if product hierarchical classification information can be obtained based on the identifier to be integrated, a first identifier integration result including product hierarchical classification information is generated.

[0148] In some optional implementations, the identifier acquisition unit 1601 is used for: Obtain multiple user search terms from the target platform; Multiple product identifiers are obtained based on multiple user search terms.

[0149] In some optional implementations, the identifier acquisition unit 1601 is used for: Construct entity extraction prompts; For each of the multiple user search terms, the second model is used to extract hints according to the entity and obtain candidate identifiers based on the user search terms; Multiple product identifiers are obtained based on multiple candidate identifiers that correspond one-to-one with multiple user search terms.

[0150] In some optional implementations, the identifier acquisition unit 1601 is used for: Get an example of entity extraction; Acquire industry knowledge related to multiple user search terms; Based on entity extraction examples and industry knowledge, entity extraction suggestions are constructed.

[0151] In this embodiment, the specific functions and examples of each unit in the product architecture information generation device 1600 based on the large model can be found in the relevant descriptions of the corresponding steps in the aforementioned embodiment of the product architecture information generation method based on the large model applied to service equipment, and will not be repeated here.

[0152] The collection, storage, use, processing, transmission, provision, and disclosure of any type of information, such as user personal information, in this technical solution comply with relevant laws and regulations and do not violate public order and good morals.

[0153] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0154] Figure 17A schematic structural block diagram of an example electronic device 1700 that can be used to implement embodiments of the present disclosure is shown. Electronic device 1700 is intended to represent various forms of digital computers, such as in-vehicle computing devices, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 1700 can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0155] like Figure 17 As shown, the electronic device 1700 includes a computing unit 1701, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 1702 or a computer program loaded from a storage unit 1708 into a random access memory (RAM) 1703. The RAM 1703 may also store various programs and data required for the operation of the electronic device 1700. The computing unit 1701, ROM 1702, and RAM 1703 are interconnected via a bus 1704. An input / output (I / O) interface 1705 is also connected to the bus 1704.

[0156] Multiple components in electronic device 1700 are connected to I / O interface 1705, including: input unit 1706, such as keyboard, mouse, etc.; output unit 1707, such as various types of renderers, speakers, etc.; storage unit 1708, such as disk, optical disk, etc.; and communication unit 1709, such as network card, modem, wireless transceiver, etc. Communication unit 1709 allows electronic device 1700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0157] The computing unit 1701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose AI computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1701 performs the various methods and processes described above, such as a product architecture information generation method based on a large model. For example, in some embodiments, the product architecture information generation method based on a large model can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 1700 via ROM 1702 and / or communication unit 1709. When the computer program is loaded into RAM 1703 and executed by the computing unit 1701, one or more steps of the product architecture information generation method based on a large model described above can be performed. Alternatively, in other embodiments, computing unit 1701 may be configured in any other suitable manner (e.g., by means of firmware) as a product architecture information generation method based on a large model.

[0158] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0159] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data optimization device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0160] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM) or flash memory, optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0161] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a rendering device (e.g., a cathode ray tube (CRT) renderer or a liquid crystal display (LCD)) for rendering information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices are also used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0162] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0163] A computer system can include client and server components. Clients and servers are generally located far apart and typically interact via a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, a server in a distributed system, or a server incorporating blockchain technology.

[0164] This disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute a product architecture information generation method based on a large model.

[0165] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements a method for generating product architecture information based on a large model.

[0166] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure is achieved, and this is not limited herein. Furthermore, in this disclosure, relational terms such as "first," "second," and "third" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Additionally, "multiple" in this disclosure can be understood as at least two.

[0167] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for generating product architecture information based on a large model, comprising: Obtain multiple product identifiers; The multiple product identifiers are grouped to obtain identifier groups under each of N group granularities; where N≥2 and N is an integer; the N group granularities have different levels of coarseness; Using the first major model, the identifiers of the first identifier group are integrated to obtain the first identifier integration result; wherein, the first identifier group is the identifier group under the i-th grouping granularity among the N grouping granularities; 1≤i≤N, and i is an integer; Based on the first identifier integration result, product architecture information related to the multiple product identifiers is obtained.

2. The method according to claim 1, wherein, The process of grouping the multiple product identifiers to obtain identifier groups under each of the N grouping granularities includes: Obtain multiple coding groups that correspond one-to-one with the multiple product identifiers; wherein, for each product identifier among the multiple product identifiers, the coding group corresponding to the product identifier includes N+1 coding results, and the N+1 coding results are used to quantize the product identifier from coarse to fine; Based on the multiple coding groups, the multiple product identifiers are grouped to obtain the identifier group under each of the N grouping granularities.

3. The method according to claim 2, wherein, The step of obtaining multiple code groups that correspond one-to-one with the multiple product identifiers includes: For each of the plurality of product identifiers, the product identifier is vectorized to obtain an identifier vector; the identifier vector is encoded using a target encoder to obtain an encoding group corresponding to the product identifier.

4. The method according to claim 2, wherein, The step of grouping the multiple product identifiers based on the multiple coding groups to obtain identifier groups under each of the N grouping granularities includes: From the plurality of coding groups, determine M target coding groups whose first j coding results are the same; where 1≤j≤N and j is an integer; M≥1 and M is an integer; From the plurality of product identifiers, determine M first target product identifiers that correspond one-to-one with the M target code groups; Based on the M first target product identifiers, construct the identifier group under the j-th grouping granularity among the N grouping granularities.

5. The method according to claim 1, wherein, The process of grouping the multiple product identifiers to obtain identifier groups under each of the N grouping granularities includes: The multiple product identifiers are clustered to obtain a tree structure; wherein the tree structure has N clustering levels from coarse to fine. Based on the clustering results at the kth clustering level among the N clustering levels, an identifier group is constructed at the kth grouping granularity among the N grouping granularities; where 1≤k≤N and k is an integer.

6. The method according to any one of claims 1 to 5, wherein, The process of using the first major model to integrate the identifiers of the first identifier group and obtain the first identifier integration result includes: Obtain the identifier to be integrated corresponding to the first identifier group; Build integrated identifier prompts; Using the first large model, the identifiers to be integrated are integrated according to the identifier integration prompts to obtain the first identifier integration result.

7. The method according to claim 6, wherein, The step of obtaining the identifier to be integrated corresponding to the first identifier group includes: When the i-th grouping granularity is the finest grouping granularity among the N grouping granularities, the product identifier in the first identifier group is taken as the identifier to be integrated; Alternatively, if the i-th grouping granularity is not the finest grouping granularity among the N grouping granularities, the identifier to be integrated is obtained based on the second identifier integration result; wherein, the second identifier integration result is obtained by performing identifier integration for the second identifier group using the first large model; the second identifier group is the identifier group under the (i+1)-th grouping granularity among the N grouping granularities, and the product identifiers in the second identifier group are included in the first identifier group.

8. The method according to claim 7, wherein, The process of obtaining the identifier to be integrated based on the second identifier integration result includes: If there is a general term for the overall identifier in the second identifier integration result, the general term for the overall identifier in the second identifier integration result shall be taken as the identifier to be integrated; Alternatively, if there is no overall identifier collective name in the second identifier integration result, the identifier to be integrated can be obtained based on the product hierarchical classification information in the second identifier integration result.

9. The method according to claim 6, wherein, The built identifier integration prompt includes: Get an example of identifier integration; Requirements for obtaining identifier integration; Based on the aforementioned identifier integration example and the aforementioned identifier integration requirements, the identifier integration prompt is constructed.

10. The method according to claim 9, wherein, The requirements for identifier integration include: If the identifiers to be integrated can be summarized into a general term for all identifiers, a first identifier integration result including the general term for all identifiers is generated. And / or, if product hierarchical classification information can be obtained based on the identifier to be integrated, a first identifier integration result including the product hierarchical classification information is generated.

11. The method according to claim 1, wherein, The acquisition of multiple product identifiers includes: Obtain multiple user search terms from the target platform; Based on the multiple user search terms, the multiple product identifiers are obtained.

12. The method according to claim 11, wherein, The process of obtaining the multiple product identifiers based on the multiple user search terms includes: Construct entity extraction instructions; For each of the multiple user search terms, the second model is used to extract hints according to the entity and obtain candidate identifiers based on the user search terms; The multiple product identifiers are obtained based on multiple candidate identifiers that correspond one-to-one with the multiple user search terms.

13. The method according to claim 12, wherein, The entity extraction prompts include: Get an example of entity extraction; Acquire industry knowledge related to the multiple user search terms; Based on the entity extraction example and the industry knowledge, the entity extraction prompt is constructed.

14. A product architecture information generation device based on a large model, comprising: The identifier acquisition unit is used to acquire multiple product identifiers; The identification grouping unit is used to group the multiple product identifications to obtain identification groups under each of the N grouping granularities; wherein, N≥2 and N is an integer; the N grouping granularities have different levels of coarseness; The identifier integration unit is used to perform identifier integration for the first identifier group using the large model to obtain the first identifier integration result; wherein, the first identifier group is the identifier group under the i-th grouping granularity among the N grouping granularities; 1≤i≤N, and i is an integer; The information acquisition unit is used to obtain product architecture information related to the multiple product identifiers based on the first identifier integration result.

15. An electronic device comprising: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method according to any one of claims 1 to 13.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 13.

17. A computer program product comprising a computer program; wherein, When the computer program is executed by a processor, it can implement the method of any one of claims 1 to 13.

Images