Supplier level-to-level management method and device based on data driving, equipment and medium

Abstract

The invention provides a supplier level-to-level management method and device based on data driving, electronic equipment and a computer readable storage medium, and relates to the technical field of supply chain management. The supplier level-to-level management method comprises the steps that comprehensive feature vectors corresponding to a plurality of suppliers are acquired, and the comprehensive feature vectors comprise terminal feature vectors and evaluation feature vectors; based on the comprehensive feature vectors corresponding to the plurality of suppliers, a distance calculation formula and a clustering algorithm, evaluating the levels of the plurality of suppliers; and based on the comprehensive feature vectors corresponding to the plurality of suppliers and a strategy gradient algorithm, generating management strategies of the suppliers under different levels. The invention aims to at least solve the problems of poor supplier management accuracy and efficiency caused by poor adaptability between a management strategy and a supplier and difficulty in adapting to self-adaptive strategy generation and optimization of the supplier in a dynamic scene in related technologies. The method is suitable for supply chain or supplier management scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain management, and in particular to a data-driven supplier hierarchical management method, device, electronic device and computer-readable storage medium. Background Art

[0002] With the rapid development of the smart home integration business of telecommunications operators, the types, brands and models of smart home terminals are increasing dramatically. There are many factors affecting the quality of equipment, response speed, brand reputation and price fluctuations provided by smart home terminal manufacturers. The decision-making basis for operators' terminal procurement is becoming more and more complicated. Users have higher and higher requirements for the use of smart home terminal equipment. They are not only limited to the performance of the equipment, but also have certain requirements and feedback on the stability, brand awareness and functional characteristics of the equipment. Operators must consider the user's perception of use and comprehensively balance the company's procurement costs. It is particularly important to collect and summarize data on the use of smart home terminals. It is necessary to use digital management methods and use AI (Artificial Intelligence) analysis capabilities to output smart home terminal supply.

[0003] The comprehensive evaluation results of the suppliers can be used for procurement guidance in the future, so as to better provide users with smart life communication services.

[0004] However, most existing technologies for automated storage, tracking, and evaluation of supplier information do not utilize the data analysis capabilities of big data modeling, but rather rely on traditional rules to score and classify suppliers, failing to cover the entire supplier lifecycle management (e.g., delivery quality, user satisfaction). Although a small number of these technologies utilize models for supplier management, they still rely on preset rules and standards for supplier classification and dynamic management strategy generation, lack the application of complex machine learning models in dynamic strategy optimization, and their intelligence needs to be improved, making it difficult to fully adapt to the complex and ever-changing market environment and supplier management needs. Furthermore, the scoring logic of existing technologies primarily relies on static formulas (e.g., linear calculations based on the lowest bid) and manually preset rules, which have limitations in supplier grading and adaptive strategy generation in dynamic scenarios, making it difficult to address the intelligent optimization needs of multi-dimensional, nonlinear relationships.

[0005] In summary, the management strategies of existing technologies are not well adapted to suppliers, and are difficult to adapt to the generation and optimization of adaptive strategies for suppliers in dynamic scenarios, resulting in poor efficiency and accuracy of supplier management. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to address the above-mentioned deficiencies in the prior art and provide a data-driven supplier hierarchical management method, device, electronic device and computer-readable storage medium, which can achieve fast and effective supplier hierarchical management.

[0007] In a first aspect, the present invention provides a data-driven supplier grading management method, comprising: obtaining comprehensive feature vectors corresponding to several suppliers, wherein the comprehensive feature vectors include terminal feature vectors and evaluation feature vectors; evaluating the grades of several suppliers based on the comprehensive feature vectors corresponding to several suppliers, a distance calculation formula, and a clustering algorithm; generating management strategies for suppliers at different grades based on the comprehensive feature vectors corresponding to several suppliers and a policy gradient algorithm.

[0008] Preferably, obtaining comprehensive feature vectors corresponding to several suppliers specifically includes: obtaining terminal information and evaluation information corresponding to several suppliers, wherein the terminal information includes terminal price, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user online time, terminal user activity, terminal user billing fees, terminal user satisfaction rate, and terminal apportionment amount, or any combination thereof; performing data preprocessing on the terminal information to obtain terminal feature vectors corresponding to several suppliers; performing text feature extraction on the evaluation information based on the RoBERTa model to obtain evaluation feature vectors corresponding to several suppliers; and concatenating the terminal feature vector and the evaluation feature vector to obtain comprehensive feature vectors corresponding to several suppliers.

[0009] Preferably, the grades of several suppliers are evaluated based on the comprehensive feature vectors corresponding to the several suppliers, the distance calculation formula and the clustering algorithm, specifically including: calculating the distance between the comprehensive feature vectors corresponding to the several suppliers based on the distance calculation formula, wherein the distance calculation formula includes any one of the following: Euclidean distance calculation formula, Manhattan distance calculation formula and cosine similarity calculation formula; dividing the several suppliers into suppliers under the target number of clusters based on the distance between the comprehensive feature vectors and the clustering algorithm, wherein the clustering algorithm includes the bottom-up aggregate hierarchical clustering AGNES algorithm; assigning grades to the suppliers under the target number of clusters to obtain the grades of the several suppliers.

[0010] Preferably, based on the distance between the comprehensive feature vectors and the clustering algorithm, several suppliers are divided into suppliers with a target number of clusters, specifically including: based on the distance between the comprehensive feature vectors and the AGNES algorithm, several suppliers are divided into suppliers with different numbers of clusters; the silhouette coefficient and Calinski-Harabasz index corresponding to different numbers of clusters are calculated respectively; the cluster number corresponding to the maximum value of the silhouette coefficient and the Calinski-Harabasz index is taken as the target number of clusters, and the suppliers with the target number of clusters are determined.

[0011] Preferably, management strategies for suppliers at different levels are generated based on the comprehensive feature vectors and policy gradient algorithm corresponding to several suppliers, specifically including: determining the terminal development status of several suppliers based on the online time and terminal user activity of several suppliers' terminal users; determining the actions to be managed of several suppliers based on the terminal prices, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user billing fees, terminal user satisfaction rate, and terminal apportionment amount of several suppliers; summarizing the terminal development status and actions to be managed of suppliers at each level respectively to obtain the state space and action space of suppliers at different levels; generating management strategies for suppliers at different levels based on the policy gradient algorithm, the state space and action space of suppliers at different levels.

[0012] Preferably, the policy gradient algorithm includes a policy function and a Monte Carlo reward function, and generates management strategies for suppliers at different levels based on the policy gradient algorithm, the state space and action space of suppliers at different levels, specifically including: using a convolutional neural network as a policy function, and based on the policy function, the state space and action space of suppliers at different levels, calculating the action probability and reward of each action to be managed of suppliers at different levels; calculating the long-term reward of each action to be managed of suppliers at different levels based on the Monte Carlo reward function and the reward of each action to be managed of suppliers at different levels; generating management strategies for suppliers at different levels based on the action probability and long-term reward of each action to be managed of suppliers at different levels.

[0013] Preferably, after generating management strategies for suppliers at different levels based on the comprehensive feature vectors and policy gradient algorithm corresponding to several suppliers, the supplier hierarchical management method also includes: obtaining management strategy execution results for suppliers at different levels; calculating assessment index data for suppliers at different levels based on the comprehensive feature vectors and management strategy execution results for suppliers at different levels; determining assessment index weights for suppliers at different levels based on the mapping relationship between assessment index weights and levels; and calculating scores for management strategies for suppliers at different levels based on the assessment index data and assessment index weights for suppliers at different levels.

[0014] In the second aspect, the present invention also provides a data-driven supplier grading management device, including a first acquisition module, an evaluation module and a generation module. The first acquisition module is used to obtain the comprehensive feature vectors of several suppliers, wherein the comprehensive feature vectors include terminal feature vectors and evaluation feature vectors. The evaluation module is connected to the first acquisition module and is used to evaluate the levels of several suppliers based on the comprehensive feature vectors, distance calculation formulas and clustering algorithms. The generation module is connected to the evaluation module and is used to generate management strategies for suppliers at different levels based on the comprehensive feature vectors and the policy gradient algorithm.

[0015] In a third aspect, the present invention further provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the data-driven supplier hierarchical management method provided in the first aspect above.

[0016] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the data-driven supplier hierarchical management method provided in the first aspect above.

[0017] The present invention provides a data-driven supplier hierarchical management method, device, electronic device, and computer-readable storage medium. By comprehensively considering terminal feature vectors and evaluation feature vectors, the supplier's grade is evaluated, enhancing the comprehensiveness and accuracy of the evaluation, thereby improving the objectivity and scientific nature of subsequent grades in subsequent supplier management decisions and reducing human subjective bias. In addition, by utilizing the reinforcement learning characteristics of the policy gradient algorithm, the supplier's management strategy is generated and continuously optimized, making supplier management decisions more adaptable to actual business dynamics, improving the scientific nature and effectiveness of decisions, and realizing intelligent, personalized, and dynamic adjustment of supplier management, significantly improving the efficiency and accuracy of supplier management and the level of supply chain operations. Therefore, the present invention can achieve rapid and effective supplier hierarchical management. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a data-driven supplier hierarchical management method according to Example 1 of the present invention;

[0019] Figure 2 This is a schematic structural diagram of a data-driven supplier tiering management system in Example 1 of the present invention;

[0020] Figure 3 This is a flowchart of a data-driven supplier hierarchical management method according to Example 2 of the present invention;

[0021] Figure 4 This is a structural diagram of a data-driven supplier hierarchical management device according to Example 3 of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0023] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0024] It is understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments may be combined with each other.

[0025] It can be understood that, for the convenience of description, the drawings of the present invention only show parts related to the present invention, while parts unrelated to the present invention are not shown in the drawings.

[0026] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0027] It will be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0028] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented using a hardware-based system that implements the specified functions, or may be implemented using a combination of hardware and computer instructions.

[0029] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0030] Example 1:

[0031] like Figure 1 As shown, this embodiment provides a data-driven supplier hierarchical management method.

[0032] In this embodiment, the data-driven supplier grading management method is applied to the data-driven supplier grading management system, such as Figure 2 As shown, the data-driven supplier tier management system includes: data base layer, service layer and presentation layer.

[0033] A data-driven supplier tier management approach, including:

[0034] S101, obtaining comprehensive feature vectors corresponding to several suppliers, wherein the comprehensive feature vector includes a terminal feature vector and an evaluation feature vector.

[0035] In this embodiment, the terminal feature vector typically refers to numerical descriptive information directly related to the supplier itself, such as the supplier's qualifications, scale, location, historical cooperation history, and supply capacity, and is typically derived from the supplier's basic attribute data. The evaluation feature vector refers to evaluation indicators of the supplier's performance or quality, such as delivery timeliness, product quality rating, service satisfaction, and price competitiveness, and is often derived from historical transaction data or third-party evaluations. The comprehensive feature vector is used to reflect the comprehensive characteristics of the supplier and is used for subsequent analysis, ranking, and decision-making.

[0036] Specifically, S101: Obtain comprehensive feature vectors corresponding to several suppliers, including steps S1011 to S1014:

[0037] S1011, obtaining terminal information and evaluation information corresponding to several suppliers, wherein the terminal information includes one or any combination of terminal price, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user online time, terminal user activity, terminal user billing fee, terminal user satisfaction rate, and terminal apportionment amount.

[0038] In this embodiment, the terminal information and evaluation information corresponding to several suppliers may be stored in different locations in the B (business support system) / O (operation support system) / M (management support system) domain due to different data types, structural differences, storage formats, or system and database design reasons. For example, the terminal information and evaluation information corresponding to the headquarters and suppliers in different provinces are stored in the headquarters database and different identity databases, respectively. The corresponding storage locations of terminal information and evaluation information with different data types and storage formats include but are not limited to: POS (Point Of Sale), SAP (Systems Applications and Products in Data Processing), supply chain system, 3GESS (3G Evolution Switching System), CBss (Converged Business Support System), RMS (Resource Management System), IOM (Integrated Operations and Maintenance System), and public customer information system.

[0039] The data base layer is used to automatically and regularly retrieve the required raw data from each data source (i.e., each storage location), classify, aggregate, and process it. Rules are pre-set based on business needs. In this embodiment, the data base layer is used as the data middle platform. The data middle platform obtains terminal information and evaluation information corresponding to several suppliers from each storage location. Evaluation information includes, but is not limited to, cooperation descriptions and user feedback.

[0040] S1012: Perform data preprocessing on the terminal information to obtain terminal feature vectors corresponding to several suppliers.

[0041] In this embodiment, the terminal information is also pre-processed through the data middle platform to screen various fields of the terminal information (i.e., terminal price, cooperating city, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user online time, terminal user activity, terminal user billing fees, terminal user satisfaction rate, and terminal apportionment amount) to generate terminal feature vectors corresponding to several suppliers shown in Table 1, wherein data pre-processing includes but is not limited to: missing value, outlier and duplicate value screening, format conversion and Min-Max normalization processing.

[0042] Table 1 Terminal feature vectors corresponding to several suppliers

[0043]

[0044] Perform data preprocessing on the terminal information, specifically including: ① Check for missing values ​​in the terminal information, calculate the missing rate of each field of the terminal information, delete fields with high missing rates (such as greater than 90%), fill fields with low missing rates (such as less than 10%) with the mode or median, assign missing values ​​in fields with qualified missing rates (between 10% and 90%) to 0, and fill in the remaining values ​​as they are. ② Check for duplicate values ​​in the terminal information and delete the duplicate values ​​directly. ③ Check for abnormal values ​​in the terminal information and treat abnormal values ​​as missing values. ④ Convert non-numeric fields of terminal information to numeric fields, for example: for binary classification fields, if the value is "yes", change it to "1", if the value is "no", change it to "0", for ordered multi-classification fields, assign values ​​to "1", "2", ..., "n" according to the order, for multi-category unordered classification fields, split them into multiple binary classification fields, and then process them according to the processing method of binary classification fields. ⑤ According to the formula Perform Min-Max normalization on the terminal information and scale all data to the interval [0, 1], where x′ represents the terminal information after Min-Max normalization, x represents the terminal information, min(·) represents the minimum function, and max(·) represents the maximum function.

[0045] S1013: Based on the RoBERTa model, text features of the evaluation information are extracted to obtain evaluation feature vectors corresponding to several suppliers.

[0046] In this example, the RoBERTa (Robustly Optimized BERT Approach) model optimizes the masked language modeling task of BERT (Bidirectional Encoder Representation from Transformers), a bidirectional language model based on Transformers, through large-scale unsupervised pre-training, enabling richer contextual semantic understanding. Therefore, this example uses the RoBERTa model as a text encoder to extract text features from the review information.

[0047] The RoBERTa model is used as a text encoder to extract text features from the review information. This includes: ① Segmenting the review information to generate a token sequence, and adding special tags ([CLS] and [SEP]). ② Inputting the token sequence into the RoBERTa model, extracting the 768-dimensional vector at the [CLS] position as the review feature vector.

[0048] It should be noted that before using the RoBERTa model as a text encoder to extract text features from the evaluation information, this embodiment also includes: obtaining historical evaluation information corresponding to several suppliers; dividing the historical evaluation information into a training set and a validation set according to a preset ratio (such as 8:2); and using a small amount of labeled data to fine-tune the RoBERTa model for different business scenarios to improve the alignment of the subsequently extracted evaluation feature vectors with the business goals.

[0049] S1014: Concatenate the terminal feature vector and the evaluation feature vector to obtain comprehensive feature vectors corresponding to the multiple suppliers.

[0050] In this embodiment, the evaluation feature vector is concatenated with the terminal feature vector to generate a 768+n-dimensional comprehensive feature vector, where n is the number of fields in the terminal feature vector. This 768+n-dimensional comprehensive feature vector is then stored in the database. This embodiment combines the evaluation feature vector, which is extracted through deep semantic features using the RoBERTa model, with the terminal feature vector after Min-Max normalization to achieve information complementarity and enhance the ability to fully express supplier characteristics. By integrating deep semantic understanding with standardized numerical features, a multi-dimensional, fine-grained description of supplier evaluations is achieved, helping to improve the accuracy of evaluations and the scientific nature of decision-making.

[0051] S102 , evaluating the levels of the multiple suppliers based on the comprehensive feature vectors, distance calculation formula, and clustering algorithm corresponding to the multiple suppliers.

[0052] In this embodiment, the service layer, based on a data base, uses big data analysis as a means, and focuses on frontline attention to provide full-process visualization, intelligent analysis, and security management functions for supplier management. Full-process visualization functions include, but are not limited to: one-code access, full lifecycle query, inventory dashboard, material statistics, variance analysis reports, and Arctic reports. Intelligent analysis functions include, but are not limited to: safety inventory model analysis, long inventory age model analysis, supplier grading, and terminal standby model analysis. Security management functions include, but are not limited to: long inventory age warning and safety inventory warning. Therefore, this embodiment primarily evaluates the levels of several suppliers through the service layer.

[0053] It should be noted that the service layer also provides system service functions, including but not limited to: user management, role management, menu management, department management, dictionary management and system settings.

[0054] Specifically, S102: Based on the comprehensive feature vectors, distance calculation formulas, and clustering algorithms corresponding to the multiple suppliers, the levels of the multiple suppliers are evaluated, including steps S1021 to S1023:

[0055] S1021, calculating the distances between the comprehensive feature vectors corresponding to the plurality of suppliers based on a distance calculation formula, wherein the distance calculation formula includes any one of the following: a Euclidean distance calculation formula, a Manhattan distance calculation formula, and a cosine similarity calculation formula.

[0056] It should be noted that, after calculating the distance between the comprehensive feature vectors corresponding to several suppliers based on the distance calculation formula, this embodiment also compares the impact of the Euclidean distance, Manhattan distance and cosine similarity between the comprehensive feature vectors corresponding to several suppliers on the clustering results to adjust the selection of the distance calculation formula.

[0057] S1022: Based on the distance between the comprehensive feature vectors and a clustering algorithm, the plurality of suppliers are divided into suppliers with a target number of clusters, wherein the clustering algorithm includes a bottom-up agglomerative hierarchical clustering (AGNES) algorithm.

[0058] Specifically, S1022: Based on the distance between the comprehensive feature vectors and the clustering algorithm, several suppliers are divided into suppliers with a target number of clusters, including: based on the distance between the comprehensive feature vectors and the AGNES algorithm, several suppliers are divided into suppliers with different numbers of clusters; the silhouette coefficient and the Calinski-Harabasz index corresponding to different numbers of clusters are calculated respectively; the number of clusters corresponding to the maximum value of the silhouette coefficient and the Calinski-Harabasz index is taken as the target number of clusters, and the suppliers with the target number of clusters are determined.

[0059] In this embodiment, a preset range of cluster numbers is determined based on business requirements, for example, [1, 5]. Taking the AGNES (Aging Clustering with Nested Series) algorithm as an example, several suppliers are divided into N clusters based on the distance between their comprehensive feature vectors and the AGNES algorithm. Specifically, the following steps are performed: ① Each supplier is initialized as an independent cluster. ② The two clusters with the closest distance between their comprehensive feature vectors are iteratively merged until all suppliers are clustered into one cluster. The merging strategy uses the Ward's method to optimize the intra-cluster variance. During the merging process, it can be seen that several suppliers can be divided into suppliers with various numbers of clusters. For example, if there are five suppliers: Supplier 1, Supplier 2, Supplier 3, Supplier 4, and Supplier 5, and the preset number of clusters is [1, 5], then the suppliers can be divided into [[Supplier 1], [Supplier 2], [Supplier 3], [Supplier 4], [Supplier 5]], [[Supplier 2 and Supplier 3], [Supplier 1, Supplier 4, and Supplier 5]], and [Supplier 1, Supplier 2, Supplier 3, Supplier 4, and Supplier 5]. If N is 1, 2, or 5, the five suppliers can be divided into one cluster, two clusters, or five clusters. This embodiment uses the AGNES algorithm to utilize the distance between comprehensive feature vectors to effectively identify similarities between suppliers and achieve natural and meaningful cluster division.

[0060] According to the formula The silhouette coefficient s corresponding to different numbers of clusters is calculated, where a(i) represents the average distance between suppliers in the i-th cluster of N clusters and other suppliers in the i-th cluster, b(i) represents the average distance between suppliers in the i-th cluster of N clusters and suppliers in the nearest cluster (e.g., the j-th cluster of N clusters), and N represents the number of clusters. In addition, this embodiment obtains the Calinski-Harabasz index corresponding to the number of clusters N by evaluating the ratio of the inter-cluster dispersion to the intra-cluster compactness. For example, five suppliers can be divided into one cluster, two clusters, or five clusters. The silhouette coefficient and Calinski-Harabasz index corresponding to a cluster number N of 1, a cluster number N of 2, and a cluster number N of 5 can be calculated, respectively.

[0061] The calculation process of the silhouette coefficient and Calinski-Harabasz index shows that the silhouette coefficient and Calinski-Harabasz index reflect the clustering effect of dividing several suppliers into N clusters, with larger values ​​indicating better clustering effects. Therefore, after dividing several suppliers into N clusters, this embodiment combines the principle of maximizing the silhouette coefficient and Calinski-Harabasz index to select the number of clusters corresponding to the maximum value of the silhouette coefficient and Calinski-Harabasz index as the target number of clusters. For example, five suppliers can be divided into one cluster, two clusters, or five clusters. If the maximum silhouette coefficient and Calinski-Harabasz index are calculated for a cluster number of two, then two clusters are determined as the target clusters, and the suppliers are clustered into two clusters. The silhouette coefficient measures the compactness within a cluster and the separation between clusters, while the Calinski-Harabasz index is based on the dispersion within and between clusters. The combination of the two can more comprehensively evaluate the clustering effect and ensure the rationality of the division results. This embodiment combines the silhouette coefficient and the Calinski-Harabasz index as two evaluation indicators to avoid subjective human specification of the number of clusters. By maximizing the indicators, the optimal number of clusters is automatically selected, improving the quality and stability of clustering. This helps to classify and manage suppliers, accurately identify risks and opportunities, and enhance the scientific nature and efficiency of procurement strategy formulation.

[0062] It should be noted that this embodiment can determine the maximum value of the silhouette coefficient and Calinski-Harabasz index by performing a weighted summation of the silhouette coefficient and the Calinski-Harabasz index and sorting the weighted summation of the silhouette coefficient and the Calinski-Harabasz index. To reduce the amount of computation and improve computational efficiency, this embodiment can also select the number of clusters corresponding to the maximum value of the silhouette coefficient or the Calinski-Harabasz index as the target number of clusters based solely on the principle of maximizing the silhouette coefficient or the Calinski-Harabasz index.

[0063] S1023 , assigning grades to suppliers under the target number of clusters to obtain grades for a number of suppliers.

[0064] In this embodiment, suppliers in the target cluster whose cluster center is closest to the operator's business objectives (such as user satisfaction and terminal quality) are marked as S-level. Suppliers in the target cluster are then assigned A / B / C / D grades in descending order based on the proximity of their cluster centers to the operator's business objectives. The grades of several suppliers are associated with their corresponding 768+n-dimensional comprehensive feature vectors to form a supplier grading data table in the database, as shown in Table 2.

[0065] Table 2 Supplier classification data table

[0066] Supplier Name Terminal price Cooperating cities .... grade Supplier 1 1200 5 .... S Supplier 2 900 3 .... B … … … .... …

[0067] S103: Based on the comprehensive feature vectors corresponding to the multiple suppliers and the policy gradient algorithm, management strategies for suppliers at different levels are generated.

[0068] In this embodiment, the intelligent analysis function of the service layer also includes management strategy generation, which follows the principle of "business-oriented, hierarchical classification", that is, using the policy gradient algorithm (Policy Gradient Methods) to determine the same management strategy for suppliers at the same level based on the supplier's comprehensive feature vector.

[0069] Specifically, S103: Based on the comprehensive feature vectors corresponding to several suppliers and the policy gradient algorithm, management strategies for suppliers at different levels are generated, including steps S1031 to S1034:

[0070] S1031: Determine the terminal development status of the plurality of suppliers based on the online time and activity of the terminal users of the plurality of suppliers.

[0071] In this embodiment, in order to facilitate calculation and comparison, before determining the terminal user online time and terminal user activity of several suppliers as the terminal development status of several suppliers, this embodiment can also normalize the terminal user online time and terminal user activity of several suppliers. For example: the terminal user online time and terminal user activity are 18 months and 80% respectively, and the normalized results are 0.60 and 0.80.

[0072] S1032, based on the terminal prices, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user billing fees, terminal user satisfaction rate, and terminal apportionment amount of the several suppliers, determine the management actions to be taken for the several suppliers.

[0073] In this embodiment, if a supplier meets at least one of the following conditions: high terminal prices, remote location, high terminal share, and high user billing fees, then the supplier's terminal costs are high, and the pending management action is "reducing terminal costs." If a supplier meets at least one of the following conditions: high terminal failure rate, high terminal refurbishment rate, and low user satisfaction rate, then the supplier's terminal equipment quality is low, and the pending management action is "improving equipment quality." Therefore, pending management actions include, but are not limited to, "reducing terminal costs" and "improving equipment quality."

[0074] S1033 , respectively summarize the terminal development status and actions to be managed of suppliers at each level to obtain the state space and action space of suppliers at different levels.

[0075] In this embodiment, the state space refers to the set of terminal development states, and the action space refers to the set of actions to be managed. For example, the terminal development states of supplier 1 and supplier 2 at level A are [0.60, 0.80] and [0.40, 0.50] respectively, and the actions to be managed of supplier 1 and supplier 2 at level A are "reduce terminal costs" and "improve equipment quality" respectively. Then the state space and action space of suppliers at level A are [[0.60, 0.80], [0.40, 0.50]] and ["reduce terminal costs", "improve equipment quality"] respectively.

[0076] S1034: Generate management strategies for suppliers at different levels based on the policy gradient algorithm, the state space and action space of suppliers at different levels.

[0077] Specifically, the policy gradient algorithm includes a policy function and a Monte Carlo reward function.

[0078] Specifically, S1034: Based on the policy gradient algorithm and the state space and action space of suppliers at different levels, management strategies for suppliers at different levels are generated, including:

[0079] Using convolutional neural network as the policy function, and based on the policy function, the state space and action space of suppliers at different levels, the action probability and reward of each action to be managed of suppliers at different levels are calculated; based on the Monte Carlo reward function and the reward of each action to be managed of suppliers at different levels, the long-term reward of each action to be managed of suppliers at different levels is calculated; based on the action probability and long-term reward of each action to be managed of suppliers at different levels, management strategies for suppliers at different levels are generated.

[0080] In this embodiment, generating management strategies for suppliers at different levels specifically includes: ① Randomly initializing the weights and biases of a convolutional neural network as the initial parameters of the management strategy, allowing the supplier to repeatedly execute different actions to be managed in the action space at different terminal development states in the state space. ② Recording the currently selected action to be managed and the terminal development state at each execution of the action to be managed, and calculating the action probability and reward of the currently selected action to be managed. The action probability of the action to be managed refers to the probability that the terminal development state will improve after executing the action to be managed. The reward is defined as 1 when the terminal development state improves, and 0 otherwise. ③ Using the action probability and reward of the currently selected action to be managed to calculate the policy gradient, that is, calculating how to adjust the parameters of the management strategy to maximize the expected return. Specifically, this includes: using the Monte Carlo method, the action probability and reward of the currently selected action to be managed, to calculate the long-term return of each action to be managed, and using the long-term return to update the parameters of the management strategy. ④ Repeating ② and ③ until the management strategy converges or reaches a preset number of repetitions. During the repetition process, the performance of improving the terminal development state will gradually improve, and the optimal management strategy can be matched to suppliers with different terminal development states. This embodiment uses convolutional neural networks (CNN) as a policy function to automatically extract complex features and spatial associations in the supplier state space, improving the policy's ability to perceive and process state information, and is particularly suitable for structured or time-series data. Management strategies are formulated based on the state space and action space of suppliers of different levels, achieving hierarchical and differentiated precise management of suppliers, and improving the targetedness and efficiency of management. By calculating the probability distribution and immediate reward of each action, the model can make reasonable trade-offs when faced with management strategy choices, balancing exploration and utilization. By integrating action probabilities and long-term returns, the management strategy is dynamically adjusted to achieve closed-loop optimization based on data and feedback, making supplier management more in line with actual operational needs and goals.

[0081] It should be noted that the parameters of the management strategy, for example: [0.6, 0.3, 0.1]# "Reducing terminal costs": 60%, "Improving equipment quality": 30%, Maintenance: 10%.

[0082] In this embodiment, management strategies for suppliers at different levels are sent to cities and managers via an office platform and text messages at preset periods, further standardizing supplier management and strengthening the compliance and rationality of cooperation. The periods include, but are not limited to, weeks and months.

[0083] Optionally, after S103: generating management strategies for suppliers at different levels based on comprehensive feature vectors corresponding to a plurality of suppliers and a policy gradient algorithm, the data-driven supplier hierarchical management method further includes:

[0084] S104, obtaining management strategy execution results of suppliers at different levels.

[0085] S105, calculating the assessment index data of suppliers at different levels based on the comprehensive feature vectors and management strategy execution results of suppliers at different levels.

[0086] In this embodiment, a supplier grading data table is obtained from a database. Based on the comprehensive feature vectors of suppliers at different levels in the supplier grading data table and the management policy execution results of suppliers at different levels, multi-dimensional assessment indicators (KPIs) for suppliers at different levels are calculated. The dimensions include, but are not limited to, user development quality, operating efficiency, task execution compliance, and sustainable development capabilities. Assessment indicators for user development quality include, but are not limited to, user number growth rate, average user billing cost, average user online time change rate, user activity fluctuation value, user satisfaction NPS (Net Promoter Score), and user complaint rate (number of complaints per thousand users). Assessment indicators for operating efficiency data include, but are not limited to, terminal commission amount achievement rate, average billing cost optimization range, and terminal life cycle cost (LCC). Assessment indicators for task execution compliance include, but are not limited to, management task completion rate, task response timeliness (e.g., SMS / platform task receipt confirmation time), abnormal event closed-loop rate (e.g., fault repair timeliness), and data reporting completeness. Assessment indicators for sustainable development capabilities include, but are not limited to, the proportion of green terminals (the proportion of equipment that meets energy efficiency standards).

[0087] S106: Based on the mapping relationship between the assessment indicator weights and the grades, the assessment indicator weights of suppliers at different grades are determined.

[0088] In this embodiment, a dynamic weight algorithm is introduced, that is, the mapping relationship between the assessment indicator weights and grades shown in Table 3, and the assessment indicator weights of suppliers at different grades (such as S / A / B / C / D) are set dynamically and differentially.

[0089] Table 3 Mapping relationship between assessment index weights and grades

[0090]

[0091] It should be noted that the assessment indicator weight is a value between 0 and 1, which is used to indicate the importance of the assessment indicator data in the comprehensive evaluation. The sum of the assessment indicator weights corresponding to all assessment indicator data of the same supplier should be 1 (i.e. ).

[0092] S107, based on the assessment indicator data and assessment indicator weights of suppliers at different levels, calculate the scores of the management strategies of suppliers at different levels.

[0093] In this embodiment, the weighted method is used to calculate the formula Perform weighted scoring on the assessment index data of suppliers at different levels and output the scores of the management strategies of suppliers at different levels, where w i Indicates the corresponding assessment indicator weight of the i-th assessment indicator data of suppliers at different levels, KPI i represents the normalized value of the i-th assessment indicator data for suppliers at different levels, and M represents the total number of assessment indicator data for suppliers at different levels. This embodiment ensures the effective implementation of management strategies for suppliers at different levels through quantified assessment indicator data and a dynamic feedback mechanism, while also providing a scientific basis for the long-term value of supplier cooperation. Subsequently, combined with the closed-loop optimization capabilities of machine learning models, it can achieve continuous adaptive improvement of management strategies, ultimately improving the overall efficiency and effectiveness of the supplier cooperation ecosystem.

[0094] It should be noted that normalization refers to the process of converting assessment indicator data of different dimensions or ranges into a unified scale (between 0 and 1) for comparison and calculation.

[0095] This embodiment may also combine managers' subjective scoring of suppliers' cooperation and coordination, and problem-solving capabilities, and calculate the scores of suppliers' management strategies at different levels based on the assessment indicator data, assessment indicator weights, and subjective scores of suppliers at different levels, where the subjective score accounts for no more than 20% of the score.

[0096] This embodiment evaluates the management strategies of suppliers at different levels based on preset cycles (i.e., obtaining the management strategy execution results of suppliers at different levels, calculating the evaluation index data of suppliers at different levels, determining the evaluation index weights of suppliers at different levels, and calculating the scores of the management strategies of suppliers at different levels). The cycles include but are not limited to: monthly, quarterly, and annual. The monthly evaluation focuses on the progress of task execution, and the annual evaluation comprehensively evaluates the long-term benefits.

[0097] The presentation layer, driven by business needs, visualizes procurement data, business data, process data, and risk data, provides multi-dimensional dashboards, empowers users of various roles with specialized capabilities, improves user work efficiency, and provides data support. After assessing the management strategies of suppliers at different levels based on a preset cycle, this embodiment automatically generates an assessment report through the presentation layer of the supplier management system, displaying the performance of suppliers at different levels through data visualization (such as heat maps and trend curves). Suppliers with continuous excellent performance are upgraded (e.g., B to A), while those with poor performance are downgraded or included in a key assistance list.

[0098] This embodiment provides a data-driven supplier grading management method, which evaluates the supplier's grade by comprehensively considering the terminal feature vector and the evaluation feature vector, thereby enhancing the comprehensiveness and accuracy of the evaluation, thereby improving the objectivity and scientific nature of subsequent grades in applying them to subsequent supplier management decisions and reducing human subjective bias. In addition, by utilizing the reinforcement learning characteristics of the policy gradient algorithm, the supplier's management strategy is generated and continuously optimized, making supplier management decisions more adaptable to actual business dynamics, improving the scientific nature and effectiveness of decisions, realizing intelligent, personalized and dynamic adjustment of supplier management, significantly improving the efficiency and accuracy of supplier management and the level of supply chain operations, and realizing fast and effective supplier grading management.

[0099] Example 2:

[0100] like Figure 3 As shown, this embodiment provides a data-driven supplier hierarchical management method. The data-driven supplier hierarchical management method includes:

[0101] S201: Obtain comprehensive feature vectors corresponding to several suppliers, wherein the comprehensive feature vector includes a terminal feature vector and an evaluation feature vector.

[0102] In this embodiment, the comprehensive feature vectors corresponding to several suppliers are Figure 2 Supplier information in .

[0103] S202 : Evaluate the levels of the multiple suppliers based on the comprehensive feature vectors, distance calculation formula, and clustering algorithm corresponding to the multiple suppliers.

[0104] S203: Generate management strategies for suppliers at different levels based on the comprehensive feature vectors corresponding to the multiple suppliers and the policy gradient algorithm.

[0105] In this embodiment, the management strategy is Figure 2 The management tasks in the above example generate management strategies for suppliers at different levels. Figure 2 Management tasks are generated in the management supplier. Figure 2 Management tasks in .

[0106] S204, obtaining the management strategy execution results of suppliers at different levels; calculating the assessment index data of suppliers at different levels based on the comprehensive feature vectors and management strategy execution results of suppliers at different levels; determining the assessment index weights of suppliers at different levels based on the mapping relationship between the assessment index weights and levels; calculating the scores of the management strategies of suppliers at different levels based on the assessment index data and assessment index weights of suppliers at different levels.

[0107] In this embodiment, the scores of the management strategies of suppliers at different levels are calculated as follows: Figure 2 Assessment and evaluation in .

[0108] This embodiment provides a data-driven supplier grading management method, which evaluates the supplier's grade by comprehensively considering the terminal feature vector and the evaluation feature vector, thereby enhancing the comprehensiveness and accuracy of the evaluation, thereby improving the objectivity and scientific nature of subsequent grades in applying them to subsequent supplier management decisions and reducing human subjective bias. In addition, by utilizing the reinforcement learning characteristics of the policy gradient algorithm, the supplier's management strategy is generated and continuously optimized, making supplier management decisions more adaptable to actual business dynamics, improving the scientific nature and effectiveness of decisions, realizing intelligent, personalized and dynamic adjustment of supplier management, significantly improving the efficiency and accuracy of supplier management and the level of supply chain operations, and realizing fast and effective supplier grading management.

[0109] Example 3: Figure 4 As shown, this embodiment provides a data-driven supplier grading management device, including a first acquisition module 31, an evaluation module 32 and a generation module 33. The first acquisition module 31 is used to obtain comprehensive feature vectors of several suppliers, wherein the comprehensive feature vector includes a terminal feature vector and an evaluation feature vector. The evaluation module 32 is connected to the first acquisition module 31 and is used to evaluate the grades of several suppliers based on the comprehensive feature vector, a distance calculation formula and a clustering algorithm.

[0110] The generation module 33 is connected to the evaluation module 32 and is used to generate management strategies for suppliers at different levels based on the comprehensive feature vector and the policy gradient algorithm.

[0111] Specifically, the first acquisition module 31 includes: an acquisition unit 311, a preprocessing unit 312, a feature extraction unit 313 and a splicing unit 314. The acquisition unit 311 is used to obtain terminal information and evaluation information corresponding to several suppliers, wherein the terminal information includes terminal price, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user online time, terminal user activity, terminal user billing fees, terminal user satisfaction rate, terminal apportionment amount or any combination thereof. The preprocessing unit 312 is used to perform data preprocessing on the terminal information to obtain terminal feature vectors corresponding to several suppliers. The feature extraction unit 313 is used to perform text feature extraction on the evaluation information based on the RoBERTa model to obtain evaluation feature vectors corresponding to several suppliers. The splicing unit 314 is used to splice the terminal feature vector and the evaluation feature vector to obtain a comprehensive feature vector corresponding to several suppliers.

[0112] Specifically, the evaluation module 32 includes: a first calculation unit 321, a clustering unit 322 and an allocation unit 323. The first calculation unit 321 is used to calculate the distance between the comprehensive feature vectors corresponding to several suppliers based on the distance calculation formula, wherein the distance calculation formula includes any one of the following: Euclidean distance calculation formula, Manhattan distance calculation formula and cosine similarity calculation formula. The clustering unit 322 is used to divide several suppliers into suppliers under the target number of clusters based on the distance between the comprehensive feature vectors and the clustering algorithm, wherein the clustering algorithm includes the bottom-up aggregate hierarchical clustering AGNES algorithm. The allocation unit 323 is used to assign grades to the suppliers under the target number of clusters to obtain the grades of several suppliers.

[0113] Specifically, the first clustering unit 322 also includes: a clustering subunit, a first calculation subunit and a determination subunit. The clustering subunit is used to divide a number of suppliers into suppliers with different numbers of clusters based on the distance between the comprehensive feature vectors and the AGNES algorithm. The first calculation subunit is used to respectively calculate the silhouette coefficient and Calinski-Harabasz index corresponding to different numbers of clusters. The determination subunit is used to take the number of clusters corresponding to the maximum value of the silhouette coefficient and the Calinski-Harabasz index as the target number of clusters, and determine the suppliers under the target number of clusters.

[0114] Specifically, the generation module 33 includes: a first determination unit 331, a second determination unit 332, a summary unit 333 and a generation unit 334. The first determination unit 331 is used to determine the terminal development status of several suppliers based on the online time and terminal user activity of several suppliers' terminal users. The second determination unit 332 is used to determine the actions to be managed of several suppliers based on the terminal prices, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user billing fees, terminal user satisfaction rate, and terminal apportionment amount of several suppliers. The summary unit 333 is used to summarize the terminal development status and actions to be managed of suppliers at each level respectively to obtain the state space and action space of suppliers at different levels. The generation unit 334 is used to generate management strategies for suppliers at different levels based on the policy gradient algorithm, the state space and action space of suppliers at different levels.

[0115] Specifically, the generation unit 334 includes: a second calculation subunit, a third calculation subunit and a generation subunit. The second calculation subunit is used to use a convolutional neural network as a policy function, and based on the policy function, the state space and action space of the suppliers at different levels, calculate the action probability and reward of each action to be managed of suppliers at different levels. The third calculation subunit is used to calculate the long-term return of each action to be managed of suppliers at different levels based on the Monte Carlo reward function and the reward of each action to be managed of suppliers at different levels. The generation subunit is used to generate management strategies for suppliers at different levels based on the action probability and long-term return of each action to be managed of suppliers at different levels.

[0116] Optionally, the data-driven supplier grading management device also includes: a second acquisition module 34, a first calculation module 35, a determination module 36 and a second calculation module 37, the second acquisition module 34 is used to obtain the management policy execution results of suppliers at different levels, the first calculation module 35 is used to calculate the assessment index data of suppliers at different levels based on the comprehensive feature vectors and management policy execution results of suppliers at different levels, the determination module 36 is used to determine the assessment index weights of suppliers at different levels based on the mapping relationship between the assessment index weights and levels, and the second calculation module 37 is used to calculate the management policy scores of suppliers at different levels based on the assessment index data and assessment index weights of suppliers at different levels.

[0117] It can be understood that the data-driven supplier grading management device provided above executes the data-driven supplier grading management method corresponding to the embodiment 1 provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects of the solution corresponding to the data-driven supplier grading management method of the embodiment 1 above, and will not be repeated here.

[0118] Example 4:

[0119] This embodiment provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the data-driven supplier hierarchical management method in the above-mentioned embodiment 1 or embodiment 2.

[0120] Example 5:

[0121] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the data-driven supplier hierarchical management method in the above-mentioned embodiment 1 or embodiment 2 is implemented.

[0122] It will be understood that the above embodiments are merely exemplary embodiments for illustrating the principles of the present invention, and the present invention is not limited thereto. Those skilled in the art will appreciate that various modifications and improvements can be made without departing from the spirit and substance of the present invention, and such modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A data-driven supplier tiered management method, characterized in that: include: Obtaining comprehensive feature vectors corresponding to several suppliers, where the comprehensive feature vector includes a terminal feature vector and an evaluation feature vector; Evaluate the levels of several suppliers based on their corresponding comprehensive feature vectors, distance calculation formulas, and clustering algorithms; Based on the comprehensive feature vectors corresponding to several suppliers and the policy gradient algorithm, management strategies for suppliers at different levels are generated.

2. The data-driven supplier hierarchical management method according to claim 1, characterized in that: The obtaining of comprehensive feature vectors corresponding to the plurality of suppliers specifically includes: Obtain terminal information and evaluation information corresponding to several suppliers, where the terminal information includes one or any combination of terminal price, cooperating cities, terminal delivery timeliness rate, terminal failure rate, terminal refurbishment rate, terminal user network time, terminal user activity, terminal user billing fees, terminal user satisfaction rate, and terminal shared amount; Perform data preprocessing on the terminal information to obtain terminal feature vectors corresponding to several suppliers; Based on the RoBERTa model, text features of the evaluation information are extracted to obtain the evaluation feature vectors corresponding to several suppliers; The terminal feature vector and the evaluation feature vector are concatenated to obtain the comprehensive feature vectors corresponding to several suppliers.

3. The data-driven supplier hierarchical management method according to claim 2, characterized in that: The evaluation of the levels of several suppliers based on the comprehensive feature vectors, distance calculation formula and clustering algorithm corresponding to the several suppliers specifically includes: Calculate the distances between the comprehensive feature vectors corresponding to the multiple suppliers based on a distance calculation formula, wherein the distance calculation formula includes any one of the following: Euclidean distance calculation formula, Manhattan distance calculation formula, and cosine similarity calculation formula; Based on the distance between the comprehensive feature vectors and the clustering algorithm, several suppliers are divided into suppliers with the target number of clusters, wherein the clustering algorithm includes the bottom-up agglomerative hierarchical clustering AGNES algorithm; Assign ranks to suppliers under the target number of clusters to obtain ranks for several suppliers.

4. The data-driven supplier hierarchical management method according to claim 3, characterized in that: Based on the distance between the comprehensive feature vectors and the clustering algorithm, several suppliers are divided into suppliers with a target number of clusters, specifically including: Based on the distance between the comprehensive feature vectors and the AGNES algorithm, several suppliers are divided into suppliers with different numbers of clusters; Calculate the silhouette coefficient and Calinski-Harabasz index corresponding to different cluster numbers respectively; The number of clusters corresponding to the maximum values ​​of the silhouette coefficient and the Calinski-Harabasz index is taken as the target number of clusters, and the suppliers under the target number of clusters are determined.

5. The data-driven supplier hierarchical management method according to claim 2, characterized in that: The management strategies for suppliers at different levels are generated based on the comprehensive feature vectors corresponding to several suppliers and the policy gradient algorithm, specifically including: The terminal development status of several suppliers is determined by the online time and activity of their terminal users; Determine the management actions for several suppliers based on their terminal prices, cooperating cities, terminal delivery timeliness, terminal failure rate, terminal refurbishment rate, terminal user billing fees, terminal user satisfaction rate, and terminal apportionment amount; Summarize the terminal development status and pending management actions of suppliers at each level to obtain the state space and action space of suppliers at different levels; Based on the policy gradient algorithm, the state space and action space of suppliers at different levels, management strategies for suppliers at different levels are generated.

6. The data-driven supplier hierarchical management method according to claim 5, characterized in that: The policy gradient algorithm includes a policy function and a Monte Carlo reward function. The management strategies for suppliers at different levels are generated based on the policy gradient algorithm, the state space and action space of suppliers at different levels, and specifically include: Using a convolutional neural network as the policy function, and based on the policy function, the state space and action space of suppliers at different levels, the action probability and reward of each action to be managed for suppliers at different levels are calculated; Based on the Monte Carlo reward function and the rewards of each management action of suppliers at different levels, the long-term returns of each management action of suppliers at different levels are calculated; Based on the action probability and long-term returns of each action to be managed for suppliers at different levels, management strategies for suppliers at different levels are generated.

7. The data-driven supplier hierarchical management method according to claim 5, characterized in that: After generating management strategies for suppliers at different levels based on the comprehensive feature vectors corresponding to the multiple suppliers and the policy gradient algorithm, the method further includes: Obtain the management strategy execution results of suppliers at different levels; Calculate the assessment index data of suppliers at different levels based on their comprehensive feature vectors and management strategy execution results; Based on the mapping relationship between assessment indicator weights and grades, determine the assessment indicator weights for suppliers at different grades; Based on the assessment indicator data and assessment indicator weights of suppliers at different levels, the scores of the management strategies of suppliers at different levels are calculated.

8. A data-driven supplier classification management device, characterized in that: It includes a first acquisition module, an evaluation module and a generation module, The first acquisition module is used to obtain comprehensive feature vectors of several suppliers, wherein the comprehensive feature vector includes a terminal feature vector and an evaluation feature vector. The evaluation module is connected to the first acquisition module and is used to evaluate the grades of several suppliers based on the comprehensive feature vector, the distance calculation formula and the clustering algorithm. The generation module is connected to the evaluation module and is used to generate management strategies for suppliers at different levels based on the comprehensive feature vector and policy gradient algorithm.

9. An electronic device, characterized in that: The system comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement a data-driven supplier hierarchical management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a data-driven supplier hierarchical management method as described in any one of claims 1 to 7.

Images