Method and device for constructing qualified supplier evaluation index system based on large model

Abstract

The embodiment of the application provides a qualified supplier evaluation index system construction method and device based on a large model, which comprises the following steps: generating atomic indexes and derivative indexes based on the whole life cycle business data of the qualified supplier by using a large language model, and constructing a public index library; performing semantic analysis on user evaluation requirements and qualified supplier characteristics by using the large language model, and determining structured evaluation elements; extracting adaptive indexes from the public index library according to the structured evaluation elements, generating exclusive indexes by using the large language model in combination with a retrieval enhancement generation technology, and synchronizing the exclusive indexes to the public index library; classifying the atomic indexes, the derivative indexes and the exclusive indexes, determining a composite index system, and performing redundancy inspection and logic consistency checking to determine a hierarchical composite index system; and generating individualized adjustment coefficients by using the large language model to analyze user evaluation requirements and qualified supplier characteristics, and determining the weights of indexes at different levels in the hierarchical composite index system in combination with objective index weights.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and apparatus for constructing a qualified supplier evaluation index system based on a large model. Background Technology

[0002] As the qualified supplier management department, we are responsible for the entire lifecycle management of qualified suppliers, including "entry review, process supervision, classification management, penalties, and exit." The comprehensive evaluation of qualified suppliers, as a core component of this lifecycle management, is a crucial basis for selecting high-quality partners, controlling risks, and optimizing the supply chain structure. With market expansion and the relaxation of entry barriers for qualified suppliers, we handle a large number of entry review processes annually, leading to a significant increase in the number of qualified suppliers we manage and consequently a substantial increase in the workload of risk control and evaluation. Currently, the comprehensive evaluation of qualified suppliers primarily relies on domain experts, employing a case-by-case approach to construct an evaluation indicator system. This involves determining weights and establishing an evaluation model through subjective manual assignment or traditional algorithms, extracting data from the qualified supplier management ERP system, and calculating the output results.

[0003] However, due to the large number of qualified suppliers, the wide range of product areas covered, and the specialized knowledge required, existing technical solutions have significant shortcomings. The construction and adjustment of the indicator system involves multiple stages, including needs assessment, data processing, and data mapping, consuming substantial human and material resources and resulting in a lengthy process. Furthermore, the constructed indicator systems are mostly static and fixed structures, making it difficult to comprehensively cover the differentiated characteristics of qualified suppliers in different fields and the personalized evaluation needs of qualified supplier managers. This can lead to discrepancies between evaluation results and actual conditions. In addition, existing methods lack the ability to automatically identify the personalized needs of managers and the characteristics of qualified suppliers, making it impossible to automatically generate the indicator system.

[0004] The aforementioned technical shortcomings collectively result in low efficiency, insufficient adaptability, and limited accuracy in qualified supplier evaluation, making it difficult to meet the dual challenges of enterprises' needs for both large-scale and personalized qualified supplier evaluation. Therefore, there is an urgent need to construct a qualified supplier evaluation indicator system that can dynamically adapt to and intelligently empower users' personalized needs. Summary of the Invention

[0005] This application provides a method and apparatus for constructing a qualified supplier evaluation index system based on a large model, which is used to construct a dynamically adapted, intelligently weighted qualified supplier evaluation index system that meets users' personalized needs.

[0006] In a first aspect, embodiments of this application provide a method for constructing a qualified supplier evaluation index system based on a large model, the method comprising: A public indicator library is constructed by generating atomic and derived indicators based on qualified supplier lifecycle business data using a large language model; the indicators in the public indicator library include multiple levels. Using a large language model, semantic analysis is performed on user evaluation needs and qualified supplier characteristics to determine structured evaluation elements; Based on the structured evaluation elements, appropriate indicators are extracted from the public indicator library, and exclusive indicators are generated using a large language model combined with retrieval enhancement generation technology and synchronized to the public indicator library. The atomic indicators, derived indicators, and specific indicators are categorized using a large language model to determine a composite indicator system; Redundancy checks and logical consistency verifications are performed on the composite index system to determine the hierarchical composite index system; The user evaluation needs and qualified supplier characteristics are analyzed using a large language model to generate personalized adjustment coefficients. Combined with the objective weights of the indicators, the weights of the indicators at different levels in the hierarchical composite indicator system are determined.

[0007] Secondly, embodiments of this application provide an apparatus for constructing a qualified supplier evaluation index system based on a large model, comprising: The public indicator library construction module is used to generate atomic and derived indicators based on qualified supplier full lifecycle business data using a large language model, and to construct a public indicator library; the indicators in the public indicator library include multiple levels. The structured evaluation element determination module is used to perform semantic parsing of user evaluation needs and qualified supplier characteristics using a large language model to determine the structured evaluation elements. The dedicated indicator generation module is used to extract suitable indicators from the public indicator library based on the structured evaluation elements, generate dedicated indicators using a large language model combined with retrieval enhancement generation technology, and synchronize them to the public indicator library. The composite indicator system determination module is used to classify the atomic indicators, derived indicators, and specific indicators using a large language model to determine the composite indicator system. The hierarchical composite index system determination module is used to perform redundancy checks and logical consistency checks on the composite index system to determine the hierarchical composite index system. The indicator weight configuration module is used to analyze the user evaluation needs and qualified supplier characteristics using a large language model to generate personalized adjustment coefficients, and combine them with objective indicator weights to determine the weights of indicators at different levels in the hierarchical composite indicator system.

[0008] Thirdly, embodiments of this application provide an electronic device, including: a memory, a processor, and a communication interface; wherein, the memory stores a computer program, and when the computer program is executed by the processor, the processor can at least implement the method for constructing a qualified supplier evaluation index system based on a large model as described in the first aspect.

[0009] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor of an electronic device, enables the processor to at least implement the method for constructing a qualified supplier evaluation index system based on a large model as described in the first aspect.

[0010] Fifthly, embodiments of this application provide a computer program product, including: a computer program or instructions, which, when executed by a processor of an electronic device, enable the processor to at least implement the method for constructing a qualified supplier evaluation index system based on a large model as described in the first aspect.

[0011] The solution provided in this application provides a method and apparatus for constructing a qualified supplier evaluation index system based on a large-scale model. The method includes: generating atomic and derived indicators based on qualified supplier lifecycle business data using a large-scale language model to construct a public index library; the indicators in the public index library include multiple levels; performing semantic parsing of user evaluation needs and qualified supplier characteristics using a large-scale language model to determine structured evaluation elements; extracting suitable indicators from the public index library based on the structured evaluation elements, generating exclusive indicators using a large-scale language model combined with retrieval enhancement generation technology, and synchronizing them to the public index library; classifying atomic, derived, and exclusive indicators using a large-scale language model to determine a composite index system; performing redundancy checks and logical consistency checks on the composite index system to determine a hierarchical composite index system; and using a large-scale language model to parse user evaluation needs and qualified supplier characteristics to generate personalized adjustment coefficients, and combining these with objective indicator weights to determine the weights of indicators at different levels in the hierarchical composite index system. This application's embodiments improve the efficiency of indicator design and evaluation by constructing a public indicator library. The reusability of the public indicator library significantly shortens the indicator design time. The fully automated construction and weight determination of the evaluation indicator system greatly shortens the qualified supplier evaluation cycle, effectively improving evaluation efficiency. The dynamically adjusted composite indicator system enhances the evaluation's adaptability and accuracy, accurately matching different evaluation needs and qualified supplier characteristics. The accuracy of evaluation results is significantly improved, more realistically reflecting the core capabilities of qualified suppliers and providing a reliable basis for decision-making. The public indicator library supports continuous updates, and the composite indicator generation logic can be optimized through large model prompts, adapting to the qualified supplier evaluation needs of different industries and companies of different sizes. This improves the scalability and universality of the evaluation, demonstrating significant value for widespread application. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating a method for constructing a qualified supplier evaluation index system based on a large model, as provided in this application embodiment; Figure 2 A technical framework diagram of a method for constructing a qualified supplier evaluation index system based on a large model, provided for embodiments of this application; Figure 3 A schematic diagram illustrating the semantic parsing of user evaluation requirements and qualified supplier characteristics in a method for constructing a qualified supplier evaluation index system based on a large model, provided in an embodiment of this application. Figure 4 A schematic diagram illustrating the generation of a composite index system for a method of constructing a qualified supplier evaluation index system based on a large model, provided in an embodiment of this application; Figure 5 A schematic diagram of a device for constructing a qualified supplier evaluation index system based on a large model, provided for an embodiment of this application; Figure 6 To and Figure 5 The illustrated embodiment provides a schematic diagram of the electronic device corresponding to the data processing apparatus. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] It should be noted that, in the cases involving user information in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) comply with relevant laws and standards.

[0016] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0017] The current application of qualified supplier evaluation technology has the following core shortcomings: the lack of a reusable standardized public indicator library, resulting in low efficiency in indicator design; poor adaptability of the indicator system, failing to match personalized needs and the characteristics of qualified suppliers; and strong subjectivity in determining indicator weights, lacking scientific algorithm support.

[0018] To address the technical problems of poor reusability, insufficient adaptability, and incomplete weight determination in existing qualified supplier evaluation indicators, this application provides a solution: a method for constructing a qualified supplier evaluation indicator system based on a large model.

[0019] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0020] The method for constructing a qualified supplier evaluation index system based on a large model provided in this application can be executed by an electronic device, which can be a terminal device such as a PC, laptop, or smartphone, or a server. The server can be a physical server containing an independent host, a virtual server, a cloud server, or a server cluster.

[0021] Figure 1 A flowchart illustrating a method for constructing a qualified supplier evaluation index system based on a large model, as provided in this application embodiment, is shown below. Figure 1 As shown, it may include the following steps: 101: Utilize a large language model to generate atomic and derived indicators based on qualified supplier lifecycle business data, and construct a public indicator library; the indicators in the public indicator library include multiple levels; 102: Utilize a large language model to perform semantic analysis on user evaluation needs and qualified supplier characteristics to determine structured evaluation elements; 103: Extract suitable indicators from the public indicator database based on the structured evaluation elements, generate exclusive indicators using a large language model combined with retrieval enhancement generation technology, and synchronize them to the public indicator database; 104: Use a large language model to classify atomic indicators, derived indicators, and specific indicators to determine a composite indicator system; 105: Perform redundancy checks and logical consistency verification on the composite indicator system to determine the hierarchical composite indicator system; 106: Utilize a large language model to analyze user evaluation needs and qualified supplier characteristics to generate personalized adjustment coefficients. Combine this with objective indicator weights to determine the weights of indicators at different levels in a hierarchical composite indicator system.

[0022] This application's embodiments improve the efficiency of indicator design and evaluation by constructing a public indicator library. The reusability of the public indicator library significantly shortens the indicator design time. The fully automated construction and weight determination of the evaluation indicator system greatly shortens the qualified supplier evaluation cycle, effectively improving evaluation efficiency. The dynamically adjusted composite indicator system enhances the evaluation's adaptability and accuracy, accurately matching different evaluation needs and qualified supplier characteristics. The accuracy of evaluation results is significantly improved, more realistically reflecting the core capabilities of qualified suppliers and providing a reliable basis for decision-making. The public indicator library supports continuous updates, and the composite indicator generation logic can be optimized through large model prompts, adapting to the qualified supplier evaluation needs of different industries and companies of different sizes. This improves the scalability and universality of the evaluation, demonstrating significant value for widespread application.

[0023] The existing evaluation system lacks a reusable standardized public indicator library, requiring the reconstruction of the indicator system for each assessment. This results in redundant investment of human and time resources, and each indicator design needs to be verified for supporting data. This inefficient approach prolongs the design cycle, lacks unified standards, and affects the comparability and scientific validity of the assessment results. This application's embodiment constructs a standardized public indicator library based on core data such as qualified supplier applications, contract performance, and legal compliance. This library encompasses both atomic and derived indicators, covering key dimensions and characteristics of qualified suppliers, effectively achieving standardization and reusability of indicators.

[0024] To address the problem that existing static and fixed indicator systems are difficult to adapt to different evaluation needs, qualified supplier types, and product fields, and lack refined demand analysis and qualified supplier characteristic matching methods, resulting in insufficient targeting of evaluation results and failure to truly reflect the core capability characteristics of qualified suppliers, this application's embodiment automatically and intelligently extracts suitable indicators from a public indicator library based on user evaluation needs. Combining the characteristics of qualified supplier types and their respective product fields, it adds exclusive indicators based on domain knowledge, forming a composite evaluation indicator system that meets evaluation needs and the characteristics of qualified suppliers.

[0025] To address the problem that existing weight determination methods rely on manual assignment or a single traditional algorithm, failing to consider the user's evaluation needs and priorities, and unable to dynamically adjust weights based on user needs and the characteristics of qualified suppliers, resulting in a disconnect between weight allocation and actual evaluation objectives and affecting the scientific validity of evaluation results, this application's embodiment, based on traditional indicator weight determination algorithms, uses user needs and the characteristics of qualified suppliers as inputs, employs a large model for semantic parsing and calculates personalized adjustment coefficients, automatically determining the weights of each indicator, thus balancing the objectivity of weights with the adaptability to evaluation needs.

[0026] Figure 2 A technical framework diagram of an embodiment of this application is shown, such as... Figure 2 As shown, the core technical framework of this application embodiment is to prioritize the construction of a public indicator library based on the full lifecycle data of qualified suppliers using a large model, and then dynamically generate a composite evaluation indicator system and weights by combining user evaluation needs and qualified supplier characteristics. The core steps of this application embodiment include: constructing the public indicator library, analyzing evaluation needs and qualified supplier characteristics, generating the composite indicator system, and determining the weights.

[0027] In this embodiment, a public indicator library is constructed by using a large language model to generate atomic and derived indicators based on qualified supplier lifecycle business data, including: Identify the data source tables and field names of core data items in the full lifecycle business data of qualified suppliers, and determine the standardized core data tables. The large language model is guided by a first prompt word to generate atomic metrics based on a standardized core data table; the first prompt word is a three-part prompt word that includes task instructions, constraints, and output format; atomic metrics are used to describe a single business behavior and / or event of a qualified supplier. The second prompt word guides the generation of derived indicators from the base atomic indicators of the large language model; the derived indicators are used to describe the business relationships, efficiency and / or overall performance of qualified suppliers and their association logic with the atomic indicators; Atomic and derived indicators are categorized into multiple levels to construct a public indicator library.

[0028] Specifically, the system acquires full lifecycle business data from qualified suppliers, including the data source tables and field names for core data items. Based on the classification of qualified supplier management data, the system organizes the core data items related to qualified supplier evaluation, clarifies the data source tables and field names in the qualified supplier management business system database for each data item, lays the foundation for subsequent indicator extraction and calculation, and forms a standardized core data table, which serves as a data dimension list and is stored in the database. Table 1 shows one such standardized core data table.

[0029] Table 1 Next, atomic metrics are generated. Using an API interface, the core data table is converted into text. A "large model + prompt words" approach guides the generation of standardized atomic metrics, used to describe single business behaviors or events by qualified suppliers. The prompt words employ a three-part design of "task instruction + constraint conditions + output format" to ensure the standardization and completeness of the generated metrics. The first prompt words for the atomic metric design are shown in Table 2 below: Table 2 The large model is guided to generate atomic indices using the first cue word mentioned above. Table 3 shows one example of a generated atomic indices.

[0030] Table 3 Next, derived metrics are generated. Guided by a "large model + prompt words" approach, standardized derived metrics are generated based on the atomic metric library. These metrics describe the business relationships, efficiency, or overall performance of qualified suppliers. To facilitate automatic data extraction for subsequent metric calculations, the tables and data fields in the dependent relational database are automatically supplemented during the derived metric generation process. The second prompt words also adopt a "task instruction + constraint conditions + output format" design, emphasizing the association logic with the atomic metrics. The specific settings for the second prompt words in the derived metric generation are shown in Table 4. Table 4 The large model is guided to generate derivative indicators by the second prompt word mentioned above. Table 5 shows one example of the generated derivative indicators.

[0031] Table 5 To ensure that the constructed atomic and derived indicators are clearly categorized, traceable, and compatible with subsequent automated evaluation processes, the indicator classification management adopts a "large-scale automatic classification + manual review and calibration" model. It is classified hierarchically according to multiple levels to construct a public indicator library. The public indicator library is managed hierarchically according to three levels: Level 1: "Where does the data come from, categorized by business domain"; Level 2: "What evaluation needs can it adapt to, categorized by evaluation scenario"; and Level 3: "What is the indicator type?" Table 6 shows one example of a constructed public indicator library.

[0032] Table 6 Figure 3 This illustration shows a semantic parsing diagram of user evaluation requirements and qualified supplier characteristics in an embodiment, such as... Figure 3 As shown, a large language model is used to semantically analyze user evaluation needs and qualified supplier characteristics to determine structured evaluation elements, including: The system obtains the evaluation request text input by the user, and uses third-party prompts to guide the large language model to extract the evaluation object, evaluation scenario, and core focus dimensions, generating the evaluation request parsing results. Obtain basic information about qualified suppliers to be evaluated and convert it into text information as the input context for the large language model; Using the context as input, the large language model is guided by the fourth prompt word to identify the supplier type, product field, and core business characteristics of the qualified supplier to be evaluated, and to generate the qualified supplier characteristic analysis results. By integrating the results of the evaluation requirements analysis and the results of the qualified supplier characteristics analysis, a structured evaluation element is determined, including the evaluation object, evaluation scenario, core focus dimensions, supplier type, product field, and core business characteristics.

[0033] Specifically, this embodiment uses the semantic understanding capabilities of a large model, combined with customized prompts, to analyze user evaluation needs and the core characteristics of qualified suppliers, providing a precise basis for subsequent indicator extraction and addition.

[0034] First, the evaluation request text input by the user via natural language is obtained, for example: "Conduct an annual review of Company A, focusing on its performance capabilities and legal compliance." This evaluation request text is then input into the large model. A customized third prompt word guides the extraction of core elements, identifying the evaluation object, evaluation scenario, and core focus dimensions, generating the evaluation request analysis results. Table 7 shows one example of a third prompt word design for evaluation request analysis.

[0035] Table 7 Guided by the third prompt mentioned above, the evaluation requirement parsing results are output in JSON format. Table 8 shows one example of the evaluation requirement parsing results.

[0036] Table 8 When evaluators input qualified supplier information, the data acquisition interface is automatically triggered to obtain the qualified supplier's brief introduction information and convert it into text format. This text is then used as context input for the large model. Through a customized fourth prompt word, the large model's inference is supported, identifying the characteristics of qualified suppliers, including the supplier type, product field, and core business characteristics of the supplier to be evaluated, and generating qualified supplier characteristic analysis results. Table 9 shows one design of the fourth prompt word.

[0037] Table 9 By integrating the results of the evaluation requirements analysis and the results of the qualified supplier characteristics analysis, a structured evaluation element is determined, including the evaluation object, evaluation scenario, core focus dimensions, supplier type, product domain, and core business characteristics. Table 10 shows a structured evaluation element in JSON format.

[0038] Table 10 Figure 4 This illustration shows a schematic diagram of the generation of the composite index system according to an embodiment of this application, such as... Figure 4 As shown in the embodiment, suitable indicators are extracted from the public indicator database based on structured evaluation elements. Specific indicators are generated using a large language model combined with retrieval enhancement generation technology and synchronized to the public indicator database, including: Based on structured evaluation elements, the fifth prompt word guides the large language model to perform semantic matching from the public indicator library to obtain the first candidate indicator set; The structured evaluation elements and public indicators are vectorized and the semantic similarity between them is calculated. Indicators that meet the preset similarity threshold are retained to obtain the second candidate indicator set. By merging the first and second candidate indicator sets, suitable indicators are determined. Based on the search enhancement generation technology, the industry knowledge base and historical evaluation cases are retrieved, and the large language model is guided to generate exclusive indicators through the sixth prompt word. After review, the exclusive indicators will be synchronously written into the public indicator library, and the relationship between the exclusive indicators and the applicable supplier types, product fields and evaluation scenarios will be established.

[0039] Specifically, in this application embodiment, based on the parsed evaluation requirements and qualified supplier characteristic results information, suitable indicators are intelligently extracted from the public indicator library, and exclusive indicators are added in combination with the knowledge base to form a personalized composite indicator system that meets the characteristics of qualified suppliers and evaluation requirements.

[0040] A large-scale model plus dual-validation method is adopted for indicator extraction. Through a dual validation mechanism of semantic understanding and representation similarity, accurate extraction of common indicators is achieved. The specific process consists of three stages: First, semantic understanding extraction: Based on the core elements of evaluation needs and the characteristics of qualified suppliers contained in the parsed structured evaluation elements, the semantic understanding capability of the large model is guided by the fifth prompt word to perform preliminary matching from a predefined public indicator library, forming a semantic understanding indicator set as the first candidate indicator set. Table 11 shows one design of the fifth prompt word.

[0041] Table 11 Second, similarity verification and filtering: The core elements of the parsed evaluation requirements and the characteristics of qualified suppliers are vectorized. The cosine similarity algorithm is used to calculate the semantic similarity between the requirements features and the indicator definitions in the public indicator library. Indicators with a similarity threshold > 0.7 are retained to form a similarity-verified indicator set, which serves as the second candidate indicator set. The output is still a JSON array containing indicator ID, indicator type, indicator name, indicator definition, calculation logic, indicator type, data source table, mapping fields, dependent atomic indicators (if any), and evaluation meaning fields (if any).

[0042] Third, index fusion: The semantic understanding results and similarity verification results are jointly deduplicated to output the final evaluation index. Specifically, index fusion prompts can be designed to guide the large model to jointly deduplicate the semantic understanding results and similarity verification results. Table 12 shows one example of index fusion prompt design.

[0043] Table 12 Next, custom metrics are generated. The addition of custom metrics is achieved using RAG + large model + prompt words. First, RAG technology searches the industry knowledge base and historical evaluation cases for custom metric templates that match the characteristics of qualified suppliers and evaluation needs. Second, the large model combines the search results with current requirements, using a sixth prompt word to guide the large language model to generate a draft custom metric, clearly defining the ID, metric name, metric definition, and dependent atomic metrics (if new data fields are needed, prompt the user to add them), maintaining consistency with the basic requirements of derived metrics. Finally, the feasibility and uniqueness of the custom metric are manually reviewed. Once approved, it is included in the composite metric system and simultaneously synchronized to the public metric library as a "custom metric," with notes indicating the type of qualified supplier and product it applies to, allowing for reuse by similar qualified suppliers and in similar scenarios. Table 13 shows an example of a custom metric.

[0044] Table 13 In this embodiment, a large language model is used to classify atomic indicators, derived indicators, and specific indicators to determine a composite indicator system, including: Obtain both compatible and proprietary metrics; The seventh prompt word guides the large language model to classify atomic indicators, derived indicators, and specific indicators according to preset evaluation dimensions; Based on the classification results, establish primary evaluation dimensions and corresponding secondary evaluation indicators; Based on the hierarchical relationship between primary evaluation dimensions and secondary evaluation indicators, atomic indicators, derived indicators, and specific indicators are organized to generate a composite indicator system.

[0045] Specifically, using a large model plus a seventh prompt word approach, the extracted atomic indicators, derived indicators, and newly added exclusive indicators are categorized according to a hierarchical structure of "dimension-indicator" to form a composite indicator system. Table 14 shows one design of the seventh prompt word.

[0046] Table 14 The final composite evaluation index system is shown in Table 15 below.

[0047] Table 15 In this embodiment, redundancy checks and logical consistency verifications are performed on the composite index system to determine the hierarchical composite index system, including: Obtain the evaluation data corresponding to the composite indicator system, standardize the evaluation data, and generate a standardized dataset; Based on a standardized dataset, redundant indicators in a composite indicator system are detected. Delete the detected redundant indicators and generate an optimized set of indicators; Construct the hierarchical relationships and dependencies between the optimized sets of indicators; Check whether the hierarchical relationships and dependencies meet the preset logical constraints; For indicator relationships that do not meet the preset logical constraints, the corresponding hierarchical relationships and dependencies are regenerated using a large language model to generate a hierarchical composite indicator system.

[0048] To ensure the mutual exclusivity and non-repetition of the indicator system, redundancy checks are conducted, and the secondary indicators in the resulting composite indicator system are used as the set of indicators to be tested. Collect the indicators to be tested Qualified supplier sample data Data cleaning and processing are performed to create a standardized data sample set, and Z-score standardization is applied to all indicators. For the k-th sample Original values ​​of each indicator; Let be the sample mean of the i-th indicator; The sample standard deviation of the i-th indicator; after standardization The mean is 0 and the standard deviation is 1, ensuring the stability of subsequent matrix operations. A sample matrix is ​​constructed based on the preprocessed standardized data. , dimension , in, For the first The first sample A standardized index value, .

[0049] set up For the first The indicators are in vector form, with dimension . ,Right now ;set up To eliminate After each indicator, the remaining A sample matrix of indicators, with dimensions of [missing information]. ;set up No. The optimal parameter vector (including the intercept term) of the regression model for each index has a dimension of . , For the first Each indicator prediction vector has a dimension of [value]. ; Let be the residual vector, with dimension . Characterize the errors not explained by the model.

[0050] Regarding the first Individual indicators As the dependent variable, Using the independent variable, design a matrix and regression model, as a linear regression model with an intercept matrix, to provide... Add a column A vector of all 1s is used to correspond to the coefficients of the intercept term, resulting in... 3D design matrix .

[0051] Based on the design matrix Formula for constructing a linear regression model matrix: Based on the above, we derive the optimal formula for OLS, using the minimum sum of squared residuals to guide the optimal parameter vector. .

[0052] The analytical formula for the optimal parameters is obtained through matrix differentiation. .

[0053] in, yes The transpose of the matrix, with dimension ; It is the inverse of the cross-product matrix.

[0054] The optimal parameter vector Substitute into the design matrix The first is obtained through matrix multiplication. The predicted value vector of each indicator The formula is .

[0055] The explanatory power of independent variables on dependent variables is quantified using the mutation decomposition method, and the coefficient of determination is calculated. ; Among them, the sum of squared residuals: Among them, the total sum of squares: In the above formula For the standardized first The average of the indicators, where 1 is the mean of the indicators. A vector of all 1s The value range is [0,1], and the closer it is to 1, the stronger the collinearity between the indicators.

[0056] based on Directly derive the formula for variance inflation factor Perform batch redundancy checks on n indicators. No significant collinearity, the index remains unchanged; Significantly collinear, indicators are removed to eliminate redundant information.

[0057] To ensure the consistency of the indicator system's logic, a logical consistency check is conducted, focusing on two dimensions: hierarchical logic and data logic. Based on the topological modeling of a directed acyclic graph (DAG), a conflict-free check of the indicator system is achieved through formulaic constraints, ensuring that the secondary indicators are uniquely assigned, the derived / specific indicators are effectively dependent, and the calculation logic is meaningful.

[0058] First, define the indicator set. Define the first-level indicator set: Define a set of secondary indicators: Among them, the atomic index set is: Derivative indicator set: Exclusive indicator set: .

[0059] Based on the definition of the indicator set, the DAG topology is defined, and the node set is defined as follows. It contains all first-level indicator nodes and second-level indicator nodes. The set of directed edges is defined as follows: ,in, Belonging edges represent the belonging relationship between secondary indicators and primary indicators. Dependency tables represent the dependencies between derived metrics and specific metrics. , indicates that there are no cyclic paths in the DAG.

[0060] Next, perform a hierarchical logical consistency check. Define the attribution mapping function from secondary indicators to primary indicators. The function needs to satisfy two constraints: coverage and mutual exclusion. This requires that all secondary indicators belong to a specific primary indicator without omission. Mutual exclusion constraint. The requirement is that the subsets of secondary indicators corresponding to different primary indicators have no overlap.

[0061] For any secondary indicator The out-degree of its belonging edge needs to satisfy , Indicates secondary indicators The number of edges belonging to the primary indicator, if If the hierarchical attribution is deemed abnormal (i.e., one secondary indicator corresponds to multiple primary indicators), redundant attribution edges are deleted, and only the unique attribution relationship is retained. .

[0062] like The result was determined to be an anomaly due to missing hierarchical affiliation, meaning that the secondary indicator did not correspond to the primary indicator. Therefore, a new affiliation relationship was added according to semantics. .

[0063] Finally, a data logic consistency check is performed to ensure that the dependencies of derived / dedicated metrics are valid and the calculation logic is reasonable. A dependency set is defined for any dedicated and derived metric. Its dependent atomic index set is ,if The dependency is normal. If If an anomaly is detected due to dependency issues, and the derived / dedicated metrics lack atomic metric support, the results are output to the large model via an interface as context for re-supplementation, and the relevant data in the public metric library is updated synchronously.

[0064] Design-derived / exclusive indicator calculation The calculation function is Its denominator is denoted as ,like Sample data If an error is detected, the result is output to the large model via an interface as context, the calculation logic is redefined, and the calculation logic of relevant indicators in the public indicator library is updated synchronously.

[0065] In this embodiment, a large language model is used to analyze user evaluation needs and qualified supplier characteristics to generate personalized adjustment coefficients. Combined with objective indicator weights, the weights of indicators at different levels in the hierarchical composite indicator system are determined, including: Based on user evaluation needs and structured evaluation elements, determine the degree of matching between each primary indicator and the evaluation needs; The coverage of each primary indicator is determined based on the number of secondary indicators it contains. The weights of the primary indicators are determined based on the degree of matching and coverage. Calculate the objective weight of each secondary indicator based on the data sample corresponding to each secondary indicator; The eighth prompt word is used to guide the large language model to analyze user evaluation needs and qualified supplier characteristics, and generate personalized adjustment coefficients for each secondary indicator. The objective weights are corrected based on the personalized adjustment coefficients, the weights of each secondary indicator are determined, and the weights of different levels of indicators in the hierarchical composite indicator system are determined in combination with the weights of the primary indicators.

[0066] Specifically, in determining the weights of indicators, the core principle of "user demand-driven + subjective and objective combination + personalized needs" is adopted. The weights of primary indicators are calculated based on user needs and indicator structure, while the weights of secondary indicators are determined by combining the objective discrimination of data with personalized needs.

[0067] First, determine the weights of the primary indicators. Define the set of primary indicators as follows: , No. The final weight of each primary indicator is ,satisfy Let the first The subset of secondary indicators under each primary indicator is: It includes atomic indicators, derived indicators, and specific indicators. The final weight of each secondary indicator is ,satisfy .

[0068] From the degree of matching of needs With indicator coverage Two aspects determine the primary indicators. Among them... For the first The semantic matching degree between each primary indicator and user needs, with values ​​ranging from [0,1], is output by the large model. The principle is to select the primary indicators of the core focus dimensions. Values ​​range from [0.7, 1); This is a primary indicator for non-core focus dimensions. The value is (0.1, 0.3).

[0069] For the first The coverage of a primary indicator represents the richness of secondary indicators under the primary indicator. The formula is: In the formula For the first The number of secondary indicators under each primary indicator This represents the total number of secondary indicators under all primary indicators.

[0070] set up As the weight of the degree of matching of needs, To determine the weighting coefficients for metric coverage scheduling, a priority is given to ensuring the weighting proportion of user demand. . No. The overall score of each primary indicator The final weight of the first-level indicator Next, the weights of the secondary indicators are determined. The objective weights of the secondary indicators are calculated using the entropy weight method, and are defined as follows: For the first The first-level indicator The objective weights of the second-level indicators are calculated based on the dispersion of the indicator data. First, data standardization is performed on the second-level indicators. The first primary indicator Each secondary indicator Data from a sample of qualified suppliers Normalization is performed: Positive indicators (the larger the data, the better). Contrarian indicator (the smaller the value, the better) After standardization, the index entropy value is calculated. .

[0071] Calculate objective weights based on entropy values. Based on objective weights, the large model calculates personalized adjustment coefficients. This coefficient is the first The first primary indicator The personalized adjustment coefficients for each secondary indicator are set in the range of [0.8, 1.2]. These coefficients are output by the large model based on user needs and the characteristics of qualified suppliers, according to the key secondary indicators of the core focus dimensions. Conventional secondary indicators Secondary indicators Values ​​are selected, and the combined weights of the secondary indicators are calculated based on these values. The combined weights are normalized according to the proportions of the primary indicator weights to obtain the secondary indicator weights. .

[0072] The core technology of this application's embodiments lies in constructing a full-process intelligent evaluation system of "public indicator library - dynamic composite indicator - intelligent weight," with the following key points: It realizes the construction and reuse of a standardized public indicator library. Based on core data such as qualified supplier applications, performance, and legal compliance, a public library containing atomic and derived indicators is constructed using prompt word engineering. The relationship tables and fields involved in the data required for each indicator calculation are clearly defined, enabling cross-scenario reuse, significantly improving indicator design efficiency, and facilitating rapid data extraction and indicator calculation. A dynamic generation mechanism for composite indicators of "extraction + addition" is constructed. Combining user evaluation needs and qualified supplier characteristics, a dual verification method of customized prompt words, semantic matching, and representation similarity matching guides the intelligent extraction of suitable indicators from the public library. Simultaneously, exclusive indicators are added, solving the problem of poor adaptability in traditional indicator systems and ensuring the relevance and comprehensiveness of the evaluation. An intelligent weight design of "basic anchoring - dynamic adjustment" is implemented. The weights of primary indicators are calculated by combining demand matching degree and indicator coverage. The weights of secondary indicators, based on the objective weighting method using entropy weighting, incorporate a large model to semantically analyze user demand text and qualified supplier characteristics, generating personalized adjustment coefficients for dynamic optimization. This approach ensures the objectivity of weight calculation while enhancing the adaptability of the evaluation system through dynamic mapping of demand semantics.

[0073] This application significantly improves the efficiency of indicator design and evaluation. The reuse of the public indicator library greatly shortens the indicator design time. The fully automated construction and weight determination of the evaluation indicator system significantly shortens the evaluation cycle for qualified suppliers, effectively improving evaluation efficiency. It also significantly enhances the adaptability and accuracy of the evaluation. The dynamic composite indicator system can accurately match different evaluation needs and the characteristics of qualified suppliers, greatly improving the accuracy of evaluation results and more realistically reflecting the core capabilities of qualified suppliers, providing a reliable basis for decision-making. Furthermore, it significantly improves the scalability and universality of the evaluation. The public indicator library supports continuous updates, and the composite indicator generation logic can be optimized through large model prompts, adapting to the qualified supplier evaluation needs of enterprises of different industries and sizes, demonstrating significant value for widespread application.

[0074] The following describes in detail one or more embodiments of this application an apparatus for constructing a qualified supplier evaluation index system based on a large model. Those skilled in the art will understand that these apparatuses can be configured using commercially available hardware components through the steps taught in this solution.

[0075] Figure 5 A schematic diagram of a device for constructing a qualified supplier evaluation index system based on a large model, as provided in this application embodiment, is shown below. Figure 5As shown, the device includes: a public indicator library construction module 11, a structured evaluation element determination module 12, a dedicated indicator generation module 13, a composite indicator system determination module 14, a hierarchical composite indicator system determination module 15, and an indicator weight configuration module 16.

[0076] The public indicator library construction module 11 is used to generate atomic and derived indicators based on qualified supplier full lifecycle business data using a large language model, and to construct a public indicator library; the indicators in the public indicator library include multiple levels. The structured evaluation element determination module 12 is used to perform semantic parsing of user evaluation needs and qualified supplier characteristics using a large language model to determine the structured evaluation elements. The dedicated indicator generation module 13 is used to extract suitable indicators from the public indicator library based on the structured evaluation elements, generate dedicated indicators using a large language model combined with retrieval enhancement generation technology, and synchronize them to the public indicator library. The composite indicator system determination module 14 is used to classify the atomic indicators, derived indicators and specific indicators using a large language model to determine the composite indicator system; The hierarchical composite index system determination module 15 is used to perform redundancy checks and logical consistency checks on the composite index system to determine the hierarchical composite index system. The indicator weight configuration module 16 is used to use a large language model to parse the user evaluation needs and qualified supplier characteristics to generate personalized adjustment coefficients, and combine them with objective indicator weights to determine the weights of indicators at different levels in the hierarchical composite indicator system.

[0077] Figure 5 The device shown can perform the steps described in the foregoing embodiments. For detailed execution process and technical effects, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0078] In one possible design, the above Figure 5 The structure of the data processing device shown can be implemented as an electronic device, such as... Figure 6 As shown, the electronic device may include: a memory 21, a processor 22, and a communication interface 23. The memory 21 stores a computer program, which, when executed by the processor 22, enables the processor 22 to at least implement the method for constructing a qualified supplier evaluation index system based on a large model, as provided in the foregoing embodiments.

[0079] The aforementioned memory 21 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0080] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium.

[0081] Accordingly, this application also provides a computer program product, which includes a computer program or instructions that, when executed by a processor, cause the processor to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. Furthermore, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, enabling the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to function as an apparatus for implementing the corresponding functions in the above method embodiments.

[0082] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0083] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of a necessary general-purpose hardware platform, or by a combination of hardware and software. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a computer product. This application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] Finally, it should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0085] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for constructing a qualified supplier evaluation index system based on a large model, characterized in that, include: A public indicator library is constructed by generating atomic and derived indicators based on qualified supplier lifecycle business data using a large language model; the indicators in the public indicator library include multiple levels. Using a large language model, semantic analysis is performed on user evaluation needs and qualified supplier characteristics to determine structured evaluation elements; Based on the structured evaluation elements, appropriate indicators are extracted from the public indicator library, and exclusive indicators are generated using a large language model combined with retrieval enhancement generation technology and synchronized to the public indicator library. The atomic indicators, derived indicators, and specific indicators are categorized using a large language model to determine a composite indicator system; Redundancy checks and logical consistency verifications are performed on the composite index system to determine the hierarchical composite index system; The user evaluation needs and qualified supplier characteristics are analyzed using a large language model to generate personalized adjustment coefficients. Combined with the objective weights of the indicators, the weights of the indicators at different levels in the hierarchical composite indicator system are determined.

2. The method for constructing a qualified supplier evaluation index system based on a large model according to claim 1, characterized in that, Using a large language model, atomic and derived metrics are generated based on qualified supplier lifecycle business data, and a public metric library is constructed, including: Identify the data source tables and field names of core data items in the full lifecycle business data of qualified suppliers, and determine the standardized core data tables. The large language model is guided by a first prompt word to generate atomic indicators based on the standardized core data table; the first prompt word is a three-part prompt word including task instructions, constraints and output format; the atomic indicators are used to describe a single business behavior and / or event of a qualified supplier. The second prompt word guides the generation of derived indicators from the base atomic indicators of the large language model; the derived indicators are used to describe the business relationships, efficiency and / or overall performance of qualified suppliers and their association logic with the atomic indicators; The atomic and derived indicators are categorized into multiple levels to construct a public indicator library.

3. The method for constructing a qualified supplier evaluation index system based on a large model according to claim 1, characterized in that, Using a large language model, semantic analysis of user evaluation needs and qualified supplier characteristics is performed to determine structured evaluation elements, including: The system obtains the evaluation request text input by the user, and uses third-party prompts to guide the large language model to extract the evaluation object, evaluation scenario, and core focus dimensions, generating the evaluation request parsing results. Obtain basic information about qualified suppliers to be evaluated and convert it into text information as the input context for the large language model; Using the aforementioned context as input, the large language model is guided by the fourth prompt word to identify the supplier type, product field, and core business characteristics of the qualified supplier to be evaluated, and to generate qualified supplier characteristic analysis results. By integrating the evaluation requirements analysis results and the qualified supplier characteristics analysis results, a structured evaluation element is determined, including the evaluation object, evaluation scenario, core focus dimensions, supplier type, product field, and core business characteristics.

4. The method for constructing a qualified supplier evaluation index system based on a large model according to claim 3, characterized in that, Based on the structured evaluation elements, suitable indicators are extracted from the public indicator database. Specific indicators are generated using a large language model combined with retrieval enhancement generation technology and synchronized to the public indicator database, including: Based on the structured evaluation elements, the large language model is guided by the fifth prompt word to perform semantic matching from the public indicator library to obtain the first candidate indicator set; The structured evaluation elements and public indicators are vectorized and the semantic similarity between them is calculated. Indicators that meet the preset similarity threshold are retained to obtain the second candidate indicator set. By merging the first candidate indicator set and the second candidate indicator set, the appropriate indicator is determined. Based on the search enhancement generation technology, the industry knowledge base and historical evaluation cases are retrieved, and the large language model is guided to generate exclusive indicators through the sixth prompt word. After review, the exclusive indicators will be synchronously written into the public indicator library, and the association between the exclusive indicators and the applicable supplier types, product fields and evaluation scenarios will be established.

5. The method for constructing a qualified supplier evaluation index system based on a large model according to claim 1, characterized in that, Redundancy checks and logical consistency verifications are performed on the composite index system to determine a hierarchical composite index system, including: Obtain the evaluation data corresponding to the composite indicator system, and perform standardization processing on the evaluation data to generate a standardized dataset; Based on the standardized dataset, redundant indicators in the composite indicator system are detected. Delete the detected redundant indicators and generate an optimized set of indicators; Construct the hierarchical relationships and dependencies between the optimized sets of indicators; Detect whether the hierarchical relationship and dependency relationship meet the preset logical constraints; For index relationships that do not meet the preset logical constraints, the corresponding hierarchical relationships and dependencies are regenerated using a large language model to generate a hierarchical composite index system.

6. The method for constructing a qualified supplier evaluation index system based on a large model according to claim 1, characterized in that, Using a large language model to analyze the user evaluation needs and qualified supplier characteristics to generate personalized adjustment coefficients, and combining this with objective indicator weights, the weights of indicators at different levels in the hierarchical composite indicator system are determined, including: Based on the user evaluation requirements and the structured evaluation elements, determine the degree of matching between each primary indicator and the evaluation requirements; The coverage of each primary indicator is determined based on the number of secondary indicators it contains. The weights of the primary indicators are determined based on the degree of matching and coverage. Calculate the objective weight of each secondary indicator based on the data sample corresponding to each secondary indicator; The eighth prompt word guides the large language model to analyze the user evaluation needs and qualified supplier characteristics, and generates personalized adjustment coefficients for each secondary indicator. The objective weights are corrected based on the personalized adjustment coefficients to determine the weights of each secondary indicator, and the weights of different levels of indicators in the hierarchical composite indicator system are determined in combination with the weights of the primary indicators.

7. A device for constructing a qualified supplier evaluation index system based on a large model, characterized in that, include: The public indicator library construction module is used to generate atomic and derived indicators based on qualified supplier full lifecycle business data using a large language model, and to construct a public indicator library; the indicators in the public indicator library include multiple levels. The structured evaluation element determination module is used to perform semantic parsing of user evaluation needs and qualified supplier characteristics using a large language model to determine the structured evaluation elements. The dedicated indicator generation module is used to extract suitable indicators from the public indicator library based on the structured evaluation elements, generate dedicated indicators using a large language model combined with retrieval enhancement generation technology, and synchronize them to the public indicator library. The composite indicator system determination module is used to classify the atomic indicators, derived indicators, and specific indicators using a large language model to determine the composite indicator system. The hierarchical composite index system determination module is used to perform redundancy checks and logical consistency checks on the composite index system to determine the hierarchical composite index system. The indicator weight configuration module is used to analyze the user evaluation needs and qualified supplier characteristics using a large language model to generate personalized adjustment coefficients, and combine them with objective indicator weights to determine the weights of indicators at different levels in the hierarchical composite indicator system.

8. An electronic device, characterized in that, include: The system includes a memory, a processor, and a communication interface; wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the method for constructing a qualified supplier evaluation index system based on a large model as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor of an electronic device, causes the processor to perform the method for constructing a qualified supplier evaluation index system based on a large model as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, include: A computer program or instruction, when executed by a processor of an electronic device, causes the processor to perform the method for constructing a qualified supplier evaluation index system based on a large model as described in any one of claims 1 to 6.

Images