Automobile part classification method and equipment based on multi-model collaboration and knowledge guidance

Abstract

The invention relates to the technical field of part classification, in particular to an automobile part classification method and equipment based on multi-model collaboration and knowledge guidance. The method comprises the following steps: defining a knowledge base of automobile parts, and inputting a description text of a target part to be classified into a plurality of large language models to obtain a first candidate category and a first confidence coefficient; if the classification results of the plurality of large language models are different, obtaining a second candidate category of the target part according to the knowledge base; meanwhile, inputting the description text into the lightweight model to obtain a third candidate category and a third confidence coefficient; according to the first confidence coefficient, the matching degree of the description text in the knowledge base and the third confidence coefficient, calculating the score of each same candidate category; and according to the score of each same candidate category, reconstructing a prompt template, and according to the reconstructed prompt template, guiding a plurality of large language models to perform part classification. The problem that in the prior art, part classification is not accurate is solved.

Description

Technical Field

[0001] This application relates to the field of parts classification technology, and more specifically, to a method and device for classifying automotive parts based on multi-model collaboration and knowledge guidance. Background Technology

[0002] With the rapid development of the automotive industry, automotive parts are diverse, with inconsistent naming and complex classification systems. Traditional manual classification or rule-based systems based on keyword matching are no longer sufficient to meet the needs of efficient and accurate management. In recent years, some companies have attempted to use Large Language Models (LLMs) for zero-shot classification. However, due to issues such as "illusions," unstable output, and insufficient domain knowledge, the reliability of classification results is limited.

[0003] When the classification results of large models are unclear, manual intervention is often required, which increases labor costs.

[0004] In view of the above, this application is hereby submitted. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for classifying automotive parts based on multi-model collaboration and knowledge guidance, so as to solve the problem of inaccurate classification of parts in the prior art.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a classification method for automotive parts based on multi-model collaboration and knowledge guidance, including: Define a knowledge base for automotive parts, which includes a standard classification system, a thesaurus mapping table, and a classification rule base. The classification system includes category names, classification levels, and standard keywords. The descriptive text of the target component to be classified is input into multiple large language models to obtain the first candidate category and first confidence score output by each large language model; If the classification results of multiple large language models are different, a second candidate category for the target component is obtained based on the knowledge base; at the same time, the description text is input into a lightweight model trained from historical component classification samples to obtain a third candidate category and a third confidence level output by the lightweight model. For the same candidate category among the first, second, and third candidate categories, a score is calculated for each same candidate category based on the first confidence level, the degree of matching of the description text in the knowledge base, and the third confidence level. The prompt template is reconstructed based on the scores of each candidate category, and the reconstructed prompt template guides multiple large language models to classify parts.

[0007] Secondly, this application provides an electronic device, comprising: At least one processor, and a memory communicatively connected to at least one of the processors; The memory stores instructions that can be executed by at least one of the processors, which enable the at least one processor to execute the above-described automotive parts classification method based on multi-model collaboration and knowledge guidance.

[0008] Compared with the prior art, the beneficial effects of this application are as follows: This application's embodiments utilize multiple large language models to reduce the bias and errors of a single model, improving classification reliability through model diversity. The introduction of a knowledge base provides domain expertise, ensuring classification results conform to industry standards and avoiding misclassification due to inconsistent terminology. The inclusion of lightweight models enables rapid real-time classification, reducing computational resource consumption, making it particularly suitable for large-scale data processing. The scoring mechanism and prompt template reconstruction achieve dynamic optimization, guiding the large language models to focus on high-probability categories, thereby improving classification efficiency and consistency. Overall, this method can achieve efficient and accurate parts management in scenarios such as automobile manufacturing, supply chain management, and after-sales maintenance, reducing manual intervention and improving automation levels. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0010] Figure 1 This is a flowchart of an automotive parts classification method based on multi-model collaboration and knowledge guidance provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0011] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These 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 and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0012] The present application will be further described in detail below with reference to the embodiments.

[0013] Figure 1 This is a flowchart illustrating a multi-model collaborative and knowledge-guided method for classifying automotive parts, as provided in an embodiment of this application. This method can be executed by a computer program and integrated into an electronic device. This application is applicable to scenarios involving the automated classification of automotive parts.

[0014] like Figure 1 As shown, this embodiment provides a classification method for automotive parts based on multi-model collaboration and knowledge guidance. The core of this method lies in achieving accurate classification of automotive parts by integrating multiple large language models, a structured knowledge base, and a lightweight model, including the following steps: S110. Define a knowledge base for automotive parts, wherein the knowledge base includes a standard classification system, a thesaurus, and a classification rule base, wherein the classification system includes category names, classification levels, and standard keywords.

[0015] The classification system includes, for example, “engine system”, “braking system”, “electrical system”, “suspension system”, “transmission system”, etc., and each category is configured with a set of keywords. For example, “braking system” includes “brake pads”, “brake discs”, “calipers”, etc.

[0016] The synonym mapping table is used to process non-standard terms in descriptive text and map them to standard terms; such as "friction pad" → "brake pad", "windshield wiper" → "windshield wiper", "battery" → "storage battery", etc., supporting the mapping relationship of polysemy and polysemous terms.

[0017] The classification rule base contains matching conditions based on rule logic to assist in classification. For example, components containing "ABS" usually belong to the electrical system, while front wheel-related components are prioritized for the suspension or braking system. This knowledge base acts as the "domain brain" of the device, providing interpretable knowledge support for subsequent classification decisions and identifying implicit semantics in the input text.

[0018] S120. Input the description text of the target component to be classified into multiple large language models to obtain the first candidate category and first confidence score output by each large language model.

[0019] The descriptive text of the target component to be classified, such as "Passat front wheel friction pads," is first encapsulated into a standardized prompt by the device and simultaneously sent to multiple different large language models (such as GPT-4, Claude-3, Tongyi Qianwen, and Wenxin Yiyan) for parallel inference. Each large language model outputs a first candidate category and a corresponding first confidence score; for example, classifying it as a braking system would have a confidence score of 0.9.

[0020] S130. If the classification results of multiple large language models are different, the second candidate category of the target component is obtained according to the knowledge base; at the same time, the description text is input into a lightweight model trained from historical component classification samples to obtain the third candidate category and the third confidence level output by the lightweight model.

[0021] If the classification results of multiple large language models are inconsistent, a second candidate category will be obtained from the knowledge base at the same time, and the description text will be input into a (local) lightweight model trained by historical component classification samples to obtain a third candidate category and a third confidence score.

[0022] A lightweight model is a machine learning model trained on historical data, such as using a support vector machine (SVM) or a lightweight neural network. This model uses historical components describing text and correct category labels during training to achieve fast inference.

[0023] S140. For the same candidate category among the first, second, and third candidate categories, calculate the score for each same candidate category based on the first confidence level, the degree of matching of the description text in the knowledge base, and the third confidence level.

[0024] When calculating scores, the device identifies identical candidate categories among the first, second, and third candidate categories. For example, if multiple sources output "engine parts," a comprehensive score is calculated for that category. Score calculation can be based on a weighted formula that integrates confidence, knowledge base matching (such as keyword matching rate), and lightweight model confidence.

[0025] S150. Reconstruct the prompt template based on the scores of each candidate category, and guide multiple large language models to classify parts based on the reconstructed prompt template.

[0026] When reconstructing the prompt template, the device adjusts the prompt content according to the score, such as adding emphasis statements before high-scoring categories. Then, the reconstructed template is input into the large language model to guide the model to focus on high-probability categories and output the final classification result.

[0027] Suppose the description text of the target component to be classified is "piston rings of a car engine". First, the knowledge base is defined by a classification system, where "engine parts" is the category, and the standard keywords are "piston" and "ring". A synonym mapping table maps "piston rings" to the standard term "piston rings". The classification rule base contains rules such as "if the text contains 'piston', then the category is engine parts". Then, the description text is input into two large language models (such as GPT-4 and BERT). GPT-4 outputs "engine parts" as the first candidate category with a first confidence score of 0.9, while BERT outputs "powertrain components" as the first candidate category with a first confidence score of 0.8. Due to the different classification results, a second candidate category is obtained from the knowledge base: through the synonym mapping, "piston rings" is mapped to the standard term, and then matched in the classification system to obtain "engine parts"; through rule matching, "engine parts" is also obtained. Simultaneously, a lightweight model outputs a third candidate category based on historical training data (such as previously classified "piston" samples) as "engine parts" with a third confidence score of 0.85. At this point, the same candidate category is "engine parts". The score for this category is calculated as follows: the average first confidence level is 0.85 (from two models), indicating a high degree of knowledge base matching (keyword exact match); the third confidence level is 0.85, resulting in a high overall score. Based on the score, a suggestion template is reconstructed, such as "Based on the knowledge base and model consensus, this component is very likely to belong to the engine component category; please prioritize this category." This template is then input into the large language model, guiding the model to output the final category as "engine component."

[0028] This application's embodiments utilize multiple large language models to reduce the bias and errors of a single model, improving classification reliability through model diversity. The introduction of a knowledge base provides domain expertise, ensuring classification results conform to industry standards and avoiding misclassification due to inconsistent terminology. The inclusion of lightweight models enables rapid real-time classification, reducing computational resource consumption, making it particularly suitable for large-scale data processing. The scoring mechanism and prompt template reconstruction achieve dynamic optimization, guiding the large language models to focus on high-probability categories, thereby improving classification efficiency and consistency. Overall, this method can achieve efficient and accurate parts management in scenarios such as automobile manufacturing, supply chain management, and after-sales maintenance, reducing manual intervention and improving automation levels.

[0029] Optionally, obtaining a second candidate category for the target component based on a knowledge base includes: matching the description text in a synonym mapping table to map the description text to standard terms; matching the standard terms in a classification system to obtain a second candidate category; and matching the description text in a classification rule base to obtain a second candidate category.

[0030] The matching of synonym maps can be achieved through string similarity algorithms, such as edit distance or dictionary-based mapping, comparing words in the input descriptive text with the map, for example, mapping "engine" to "engine". Then, matching of standard terms within the classification system can be accomplished by traversing the nodes of the classification tree, for example, using keyword matching or semantic search to find the most relevant category. Matching based on the classification rule base is based on a rule engine, such as using Drools or custom logic, to perform pattern matching on the descriptive text; for example, if the text contains "brake" and "disc", the rule output "brake system component" is triggered.

[0031] Optionally, before inputting the descriptive text into a lightweight model trained from historical component classification samples to obtain the third candidate category and third confidence score output by the lightweight model, the method further includes: collecting training samples and labels; the training samples are the descriptive text of the components, and the labels are the correct categories; and training the lightweight model based on the training samples and labels.

[0032] First, training samples are collected from historical databases or manually labeled data, such as 10,000 parts description texts and their correct category labels, like "engine parts" and "brake system parts." Data preprocessing includes text cleaning, word segmentation, and vectorization, for example, using TF-IDF or word embeddings. Then, a lightweight model architecture, such as logistic regression, decision trees, or shallow neural networks, is selected, and supervised learning is performed using the training samples. Finally, the trained model is deployed for inference. The model output includes a third candidate category and a third confidence score. Fast classification based on historical data is achieved by training a lightweight model. First, the lightweight model reduces computational complexity, making it suitable for resource-constrained environments; second, the training process ensures the model adapts to domain-specific data, improving the specificity of classification; and third, by collaborating with larger models, it compensates for potential overfitting or bias issues that large language models may have on specific tasks.

[0033] Optionally, if there is only one candidate category (i.e., the large language model, knowledge base, and local lightweight model all provide the same category), then that category is selected. If there are three or more candidate categories, and there is a unique identical candidate category, then that identical candidate category is selected. If there are two or more identical candidate categories, for example, the large language model, knowledge base, and local lightweight model all provide an engine category and a braking system category, then the following operation is performed: For the same candidate category, the consensus factor for the same candidate category is calculated based on the first confidence level: ; in, is the consensus factor for the i-th identical candidate category, and N is the number of large language models. It is the weight corresponding to the historical accuracy of the m-th large language model (the number of times the classification of car parts was accurate within the historical period / the total number of classifications). This represents the first confidence score of the i-th candidate category output by the m-th large language model. Assuming the same candidate category is "engine parts," and the first confidence scores of the two large language models are 0.9 and 0.8 respectively, a consensus factor is obtained by weighted averaging of the first confidence scores (weights based on model accuracy). In this embodiment, multi-factor ensemble reduces errors from single sources, improving decision reliability; the weighted design allows for prioritization based on historical accuracy, enhancing the method's flexibility.

[0034] Based on the degree of matching between the description text and the knowledge base, calculate the knowledge base matching factor for the same candidate category: ; in, It is the knowledge base matching factor for the i-th candidate category. It is a set of keywords that describe the text. It is a category Standard keywords in the classification system It is the i-th candidate category. The exact matching category provided for the classification system. If the description text exactly matches the category name in the classification system, that is... If the matching factor is 1, then the most suitable category has been found. If the description text does not match the category name in the classification system, then it is determined according to... Calculate the knowledge base matching factor. It refers to quantity. Assuming the description text is "engine piston", the keyword set A = {"engine", "piston"}, and the standard keyword set B for the category "engine parts" = {"engine", "piston", "crankshaft"}, the knowledge base matching factor is |A∩B| / |A∪B| = 2 / 3 ≈ 0.67.

[0035] Based on the third confidence level, lightweight model factors for the same candidate category are determined. The third confidence level can be normalized and used as the lightweight model factor.

[0036] The consensus factor, knowledge base matching factor, and lightweight model factor for the same candidate category are weighted and summed to obtain a score for each candidate category. The sum of the weights of the three factors is 1. During the weighted summation, the weights can be adjusted according to the scenario; for example, the consensus factor weight could be 0.4, the knowledge base matching factor weight could be 0.4, and the lightweight model factor weight could be 0.2, to ensure balance. After calculation, each candidate category receives a score, which is used for subsequent template reconstruction.

[0037] Optionally, the prompt content can be determined based on the score difference between candidates of the same category and the number of candidates of the same category. This includes the following three scenarios: 1. If the score difference between a candidate category and other candidate categories of the same type is less than a first score threshold, and the number of other candidate categories of the same type is less than a first quantity threshold, then the matching result of the knowledge base for the candidate category and the output of the lightweight model are reinforced in the prompt content. The first score threshold is less than a second score threshold, and the first quantity threshold is less than the second quantity threshold.

[0038] 2. If the score difference between a candidate category and other candidate categories of the same type is greater than or equal to the first score threshold and less than or equal to the second score threshold, and the number of other candidate categories of the same type is greater than or equal to the first number threshold and less than or equal to the second number threshold, then the matching result of the consensus factor of the candidate category and the knowledge base is reinforced in the prompt content. 3. If the score difference between a candidate category and other candidate categories is greater than the second score threshold, and the number of other candidate categories is greater than the second quantity threshold, then the matching result of the consensus factor of the candidate category and the knowledge base is strengthened in the prompt content, while the output result of the lightweight model is weakened.

[0039] For example, the first rating threshold is 0.1, the first quantity threshold is 2, the second rating threshold is 0.4, and the second quantity threshold is 3.

[0040] First, scores for all identical candidate categories are calculated using the method provided in the above embodiments. Then, for any identical candidate category (e.g., engine system), the score difference between it and other identical candidate categories (e.g., braking system, steering system, and body) is calculated, and the highest score difference is taken, for example, 0.3. The number of identical candidate categories is 3. If the second case described above is met, the consensus factor and the matching result of the knowledge base for that identical candidate category are reinforced in the prompt content. For example, the consensus factor (i.e., the weighted average value of the first confidence of the large language model) and the matching result of the knowledge base are emphasized. The prompt template can contain content such as "Multiple large language models have reached a consensus, classifying it as engine system with a weight of 0.8, and the knowledge base rules support the category of engine system." The reconstructed template is input into the large language model, guiding the model to value collective wisdom and domain rules. In scenarios with moderate score differences and a moderate number of categories, the reliability and consistency of classification are enhanced by integrating multi-model consensus and structured knowledge. This approach reduces errors caused by the bias or noise of a single model, and is particularly suitable for situations where multiple categories compete but have relative advantages, thereby improving overall classification efficiency and robustness.

[0041] If the score difference between any identical candidate category (e.g., braking system) and other identical candidate categories, along with the number of other identical candidate categories, satisfies the conditions described in section 1 above, then the prompt content reinforces the knowledge base matching result for that identical candidate category and the output of the lightweight model. For example, the prompt content might include "The braking system category has a matching degree of 0.7 in the knowledge base, and the lightweight model's confidence level is 0.9," where the knowledge base matching result can be reinforced by listing standard keyword matching rates or rule triggering details, and the lightweight model output is emphasized by referencing its third confidence level. The reconstructed prompt content is then input into multiple large language models, guiding them to focus on this reinforcement information to output the final category. In this case, with small score differences and few competing categories, relying on domain knowledge (knowledge base) and historical data patterns (lightweight model) reduces the uncertainty and randomness of the large language model, improving the accuracy and stability of classification. This is particularly suitable for boundary cases or scenarios with ambiguous descriptive text, where even with insufficient consensus in the large language model, professional knowledge and historical experience can still provide reliable guidance.

[0042] If the score difference between any candidate category (e.g., steering system) and other candidate categories, as well as the number of other candidate categories, satisfies the third condition mentioned above, then the prompt content strengthens the consensus factor and knowledge base matching results of the candidate categories, while weakening the output of the lightweight model. For example, the prompt content could include statements such as "Multiple models and knowledge bases consistently indicate that this category is steering system, with a confidence level as high as 0.9," and "The vehicle body category output by the lightweight model may be unreliable." Specifically, the reconstructed prompt template is input into the large language model, guiding the model to rely on more reliable sources. In this complex scenario with large score differences and a large number of categories, prioritizing high consensus and domain knowledge avoids the risks that lightweight models might face due to data sparsity or overfitting, thereby improving classification accuracy and robustness. It also ensures that in highly uncertain environments, the device can focus on stable and reliable information sources, reducing the possibility of misclassification.

[0043] In addition to the three cases mentioned above, the categories, confidence levels, and reasoning processes output by the knowledge base, large language model, and lightweight model can all be filled into the prompt template.

[0044] The prompt content constructed based on each candidate category is filled into the prompt template. At the same time, the classification instructions and the description text of the target component to be classified are filled into the prompt template. The prompt template is then input into multiple large language models to obtain the final category output by the large language models.

[0045] Optionally, after guiding multiple large language models to classify components based on the reconstructed prompt template, if the outputs of the multiple large language models are inconsistent, manual classification is adopted; the lightweight model is trained based on the manual classification results, and the synonym mapping table is updated.

[0046] When the outputs of the large language model become inconsistent, a human classification interface is triggered, where an expert provides the correct category. The device then uses the human result as new training samples to update the training data of the lightweight model, retraining it to improve accuracy. Simultaneously, the mapping table is updated, for example, by adding a mapping from "motor" to "engine".

[0047] This embodiment achieves continuous improvement of equipment and continuous updating of models and knowledge base through human feedback and update mechanisms, thereby enhancing adaptability; it also reduces long-term errors and improves the level of automation through closed-loop learning.

[0048] like Figure 2 As shown, this embodiment provides an electronic device, including: At least one processor; and A memory that is communicatively connected to at least one processor; wherein, The memory stores instructions executable by at least one processor, which, when executed, enables the processor to perform the described method. Since at least one processor in this electronic device is capable of performing the described method, it possesses at least the same advantages as the described method.

[0049] Optionally, the electronic device also includes interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The components are interconnected using different buses and can be mounted on a common motherboard or otherwise installed as needed. The processor can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI (Graphical User Interface) on an external input / output device (such as a display device coupled to the interface). In other embodiments, multiple processors can be used with multiple memories, and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple electronic devices (e.g., as a server array, a group of blade servers, or a multiprocessor system) can be connected, each providing some of the necessary operations. Figure 2 Take processor 301 as an example.

[0050] The memory 302, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the automotive parts classification method based on multi-model collaboration and knowledge guidance in this embodiment. The processor 301 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 302, thereby realizing the aforementioned automotive parts classification method based on multi-model collaboration and knowledge guidance.

[0051] The memory 302 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on terminal usage. Furthermore, the memory 302 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 302 may further include memory remotely located relative to the processor 301, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0052] The electronic device may further include an input device 303 and an output device 304. The processor 301, memory 302, input device 303, and output device 304 can be connected via a bus or other means. Figure 2 Taking the example of a connection between China and Israel via a bus.

[0053] Input device 303 can receive input digital or character information, and output device 304 may include a display device, an auxiliary lighting device (e.g., an LED), and a haptic feedback device (e.g., a vibration motor). The display device may include, but is not limited to, a liquid crystal display (LCD), a light-emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touchscreen.

[0054] This embodiment provides a computer-readable storage medium storing computer instructions for causing a computer to perform the methods described above. The computer instructions on this computer-readable storage medium, used to cause a computer to perform the methods described above, thus have at least the same advantages as the methods described above.

[0055] The medium in this application may be any combination of one or more computer-readable media. The medium may be a computer-readable signal medium or a computer-readable storage medium. The medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of the medium (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, the medium may be any tangible medium containing or storing a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0056] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0057] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF (Radio Frequency), or any suitable combination thereof.

[0058] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0059] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means, such as coaxial cable, optical fiber, digital subscriber line (DSL), or wireless means, such as infrared, wireless, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium, or a semiconductor medium, etc. It is worth noting that the computer-readable storage medium mentioned in the embodiments of this application can be a non-volatile storage medium, in other words, it can be a non-transient storage medium.

[0060] 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 application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0061] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. 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 spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A classification method for automotive parts based on multi-model collaboration and knowledge guidance, characterized in that, include: Define a knowledge base for automotive parts, which includes a standard classification system, a thesaurus mapping table, and a classification rule base. The classification system includes category names, classification levels, and standard keywords. The descriptive text of the target component to be classified is input into multiple large language models to obtain the first candidate category and first confidence score output by each large language model; If the classification results of multiple large language models are different, a second candidate category for the target component is obtained based on the knowledge base; at the same time, the description text is input into a lightweight model trained from historical component classification samples to obtain a third candidate category and a third confidence level output by the lightweight model. For the same candidate category among the first, second, and third candidate categories, a score is calculated for each same candidate category based on the first confidence level, the degree of matching of the description text in the knowledge base, and the third confidence level. The prompt template is reconstructed based on the scores of each candidate category, and the reconstructed prompt template guides multiple large language models to classify parts.

2. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 1, characterized in that, The second candidate category for the target component is obtained based on the knowledge base, including: The description text is matched in a synonym mapping table, and the description text is mapped to standard terms; The standard terms are matched in the classification system to obtain the second candidate category; The description text is matched against the classification rule base to obtain the second candidate category.

3. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 1, characterized in that, Before inputting the descriptive text into a lightweight model trained from historical component classification samples to obtain the third candidate category and third confidence score output by the lightweight model, the process further includes: Collect training samples and labels; the training samples are descriptive text of the parts, and the labels are the correct categories; The lightweight model is trained based on the training samples and labels.

4. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 3, characterized in that, For the same candidate category, a score is calculated for each candidate category based on the first confidence level, the degree of matching of the description text in the knowledge base, and the third confidence level, including: Calculate the consensus factor for the same candidate category based on the first confidence level; Calculate the knowledge base matching factor for the same candidate category based on the degree of matching of the description text in the knowledge base. Based on the third confidence level, determine the lightweight model factors for the same candidate category; The consensus factor, knowledge base matching factor, and lightweight model factor for the same candidate category are weighted and summed to obtain the score for each candidate category.

5. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 4, characterized in that, The consensus factor for the same candidate category is calculated using the following formula based on the first confidence level: ； in, is the consensus factor for the i-th identical candidate category, and N is the number of large language models. These are the weights corresponding to the historical accuracy of the m-th large language model. It is the first confidence score of the i-th candidate type output by the m-th large language model.

6. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 4, characterized in that, The following formula is used to calculate the knowledge base matching factor for the same candidate category based on the degree of matching of the description text in the knowledge base: ； in, It is the knowledge base matching factor for the i-th candidate category. It is a set of keywords that describe the text. It is a category Standard keywords in the classification system It is the i-th candidate category. The exact matching category provided for the classification system.

7. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 1, characterized in that, The suggestion template is reconstructed based on the scores of each candidate category, and the reconstructed suggestion template guides multiple large language models to classify components, including: The prompt content is determined based on the score difference between the same candidate categories and the number of the same candidate categories; The prompt content constructed based on each candidate category is filled into the prompt template. At the same time, the classification instructions and the description text of the target component to be classified are filled into the prompt template. The prompt template is then input into multiple large language models to obtain the final category output by the large language models.

8. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to claim 7, characterized in that, Based on the score differences between candidates of the same category and the number of candidates of the same category, determine the prompt content, including: If the score difference between a candidate category and other candidate categories is less than a first score threshold, and the number of other candidate categories is less than a first quantity threshold, then the matching result of the knowledge base of the candidate category and the output result of the lightweight model are reinforced in the prompt content. If the score difference between a candidate category and other candidate categories of the same type is greater than or equal to the first score threshold and less than or equal to the second score threshold, and the number of other candidate categories of the same type is greater than or equal to the first number threshold and less than or equal to the second number threshold, then the matching result of the consensus factor of the candidate category and the knowledge base is reinforced in the prompt content. If the score difference between a candidate category and other candidate categories is greater than the second score threshold, and the number of other candidate categories is greater than the second number threshold, then the matching result of the consensus factor of the candidate category and the knowledge base is strengthened in the prompt content, while the output result of the lightweight model is weakened. Among them, the first scoring threshold is less than the second scoring threshold, and the first quantity threshold is less than the second quantity threshold.

9. The automotive parts classification method based on multi-model collaboration and knowledge guidance according to any one of claims 1-8, characterized in that, After guiding multiple large language models to classify components based on the reconstructed prompt template, the process also includes: If the outputs of multiple large language models are inconsistent, manual classification should be used. The lightweight model is trained based on the results of manual classification, and the thesaurus is updated.

10. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to at least one of the processors; The memory stores instructions that can be executed by at least one of the processors, which are executed by at least one of the processors to enable the at least one of the processors to perform the automotive parts classification method based on multi-model collaboration and knowledge guidance as described in any one of claims 1-9.

Images