Data splitting method and system based on natural language processing

By using a data diversion method based on natural language processing, the procurement plan is parsed to generate a set of demand attribute tags. Differentiated diversion is then performed using blockchain smart contracts, which solves the problem of the inability to identify differences in the demand attributes of the procurement plan in existing technologies, and achieves efficient and reliable procurement plan management.

CN122865131BActive Publication Date: 2026-11-10JIANGSU GUOXIN DIGITAL INTELLIGENCE SERVICE CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202611374999.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-10
Estimated Expiration
2046-09-07

AI Technical Summary

Technical Problem

The existing procurement planning management system cannot identify the differences in the demand attributes of procurement plans, which makes it impossible to allocate special needs to the appropriate professional teams, causing potential risks, and it fails to effectively distinguish between routine and demand-based procurement plans.

Method used

By using a data diversion method based on natural language processing, demand plans are analyzed, demand attribute tag sets are generated, and differentiated diversion is carried out using blockchain smart contracts. Professional teams are matched, and a diversion supply list is generated through double verification to ensure the adaptability of diversion teams and the stability of supply.

Benefits of technology

It enables accurate identification and allocation of special needs, reduces execution risks, improves procurement efficiency, ensures the accuracy and adaptability of allocation decisions, and forms a reliable data chain throughout the entire process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122865131B_ABST
    Figure CN122865131B_ABST
Patent Text Reader

Abstract

This invention discloses a data offloading method and system based on natural language processing, belonging to the field of data offloading technology. The method includes: acquiring demand data, extracting demand information, constructing a demand attribute tag set, integrating the demand information and the demand attribute tag set, and generating an initial hash value for storage; calling an offloading rule base to traverse and verify the demand attribute tag set to determine whether offloading is necessary; if the demand attribute tag set meets any condition in the offloading rule base, offloading is triggered, a differentiated offloading mechanism is initiated, and a recommended offloading list is generated; otherwise, a flow mechanism is initiated, and a flow hash value is generated; acquiring real-time status data, calculating a verification score, and performing individual and overall verification on the recommended offloading list to generate an offloading supply list; after the demand plan is completed, collecting the execution data of the demand plan and combining it with the execution evaluation system of the executing department to obtain execution evaluation indicators.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a data splitting method and system based on natural language processing, belonging to the field of data splitting technology. Background Technology

[0002] In recent years, as the economy has shifted towards innovation, traditional industries have gradually faced the challenges of slower growth and intensified market competition, leading to increasingly higher demands from enterprises for more refined and efficient internal management. As a core component of business operations, procurement management is directly related to the rationality of resource allocation and operational efficiency, and plays a crucial supporting role in the scientific nature of enterprise demand decisions and the sustainability of business development.

[0003] A patent application with publication number CN114936847A discloses a procurement planning management system. The system includes a procurement planning management module, a procurement execution planning management module, a personnel access control module, a procurement contract business ledger management module, and a performance evaluation and statistical indicator data management module. The procurement planning management module generates procurement plans and distributes them to the corresponding procurement execution departments. The procurement execution planning management module modifies and executes the procurement plans. The procurement contract business ledger management module marks the procurement plans, links them to procurement / contract management data from an e-commerce platform, and generates a contract business ledger. The performance evaluation and statistical indicator data management module analyzes the data in the contract business ledger, generates statistical indicators, and adjusts performance evaluation indicators. The personnel access control module manages the operational permissions of staff from different departments within the management system.

[0004] While existing technologies have improved the accuracy, timeliness, and convenience of procurement planning management systems, and effectively monitored the progress and key milestones of procurement projects, they have not considered the differences in the demand attributes of procurement plans and the need for targeted allocation and adaptation. Specifically, procurement platforms cannot distinguish between routine and demand-driven procurement plans, cannot identify the attributes of special needs such as critical materials and high-risk materials, and do not take into account the complexity of the supply market and the difficulty of supplier management. They also cannot allocate special procurement plans to suitable professional teams, and only process all needs in a homogenized manner, which may easily lead to potential risks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a data diversion method and system based on natural language processing. By parsing demand plans to generate a set of demand attribute tags, demand differences can be accurately identified. Data can be diverted according to rules, demand can be matched with professional teams, routine demand can be directly transferred, and the entire process can be stored on the blockchain and rely on execution iteration parameters to solve the technical problems of homogeneous processing and insufficient adaptation.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Data splitting methods based on natural language processing include:

[0008] Obtain demand data, extract demand information, construct a demand attribute tag set, integrate the demand information and the demand attribute tag set, and generate an initial hash value for evidence storage.

[0009] The system calls the traffic splitting rule library to iterate and verify the set of demand attribute tags to determine whether to split traffic. If the set of demand attribute tags meets any condition in the traffic splitting rule library, it is determined that traffic splitting is triggered, the differentiated traffic splitting mechanism is started, and a recommended traffic splitting list is generated. Otherwise, the flow mechanism is started and a flow hash value is generated.

[0010] Acquire real-time status data, calculate the verification score of each supply team in the recommended distribution list, and perform individual and overall verification on the recommended distribution list to generate a distribution supply list;

[0011] After the demand plan is completed, the execution data of the demand plan is collected and combined with the execution evaluation system of the executing department to obtain the execution evaluation indicators.

[0012] Specifically, the steps for constructing a set of requirement attribute tags include:

[0013] Obtain the requirement plan submitted by the target user, perform compatibility verification, generate supplementary prompts, and re-verify;

[0014] Natural language processing is used to extract requirement information from the requirement plan and calculate the confidence level of each field for verification. For image files, image recognition is used to parse the visualization information and generate a requirement list.

[0015] A demand identifier is constructed for the demand plan. Based on the material types in the demand list, a set of candidate suppliers is constructed and bound to the demand list to generate a demand data package.

[0016] Specifically, the steps for constructing a set of requirement attribute tags also include:

[0017] Based on the business priority of the demand plan, configure the corresponding business weights, set the substitution coefficient based on the material substitution scheme uploaded by the target user, and calculate the support coefficient.

[0018] Set a secondary material classification threshold and, in conjunction with the support coefficient, obtain the material label level;

[0019] The number of potential suppliers for the demand plan is obtained, the intention level is obtained based on the level quantification standard, a stable weight is set based on the historical cooperation data of the target user, the market complexity coefficient is calculated, a secondary complexity division threshold is set, and the supply tag level is obtained.

[0020] The number of technical parameters, the number of special processes, and the acceptance level are obtained, and a weighted calculation is performed to calculate the technical difficulty coefficient. The threshold for secondary technical classification is set, and the technical label level is obtained.

[0021] Obtain the target user's self-assessment risk value and self-assessment impact coefficient, combine them with the supply cycle, quantify the supply weight, calculate the risk coefficient, set a secondary risk threshold, and obtain the risk label level;

[0022] Construct a set of requirement attribute tags and bind them to the requirement data package, perform a hash operation, and generate an initial hash value.

[0023] Specifically, the steps of the differentiated traffic diversion mechanism include:

[0024] Call the smart contract in the blockchain and verify the validity of the smart contract;

[0025] Traverse the candidate supplier set, filter candidate supplier teams according to the smart contract, and generate a basic matching team pool;

[0026] Based on the attribute priority matching rule, obtain the tag priority of each tag level in the current demand plan;

[0027] Based on the label coding rules, assign numerical codes to each label level in the current demand plan;

[0028] The tag priority is integrated with the numerical code to generate an coded tag set;

[0029] Configure priority weights for each tag priority, and calculate the matching strength value based on the encoded tag set;

[0030] Obtain the acceptance threshold of each supply team in the basic adaptation team pool, perform positive matching, and remove supply teams whose acceptance threshold is less than the matching strength value.

[0031] Read the historical supply data of each candidate supply team, perform reverse matching based on the highest historical tag priority, filter the candidate teams, and finally generate a pool of candidate teams.

[0032] Specifically, the steps of the differentiated traffic diversion mechanism also include:

[0033] Based on the demand matching strength value and the capacity threshold, the matching ratio of each candidate team is calculated.

[0034] Based on the historical database, the historical success rate of each candidate team in handling similar needs is obtained, and the adaptation score is calculated in combination with the adaptation ratio.

[0035] Based on the fit score, the supply teams in the fit candidate team pool are sorted in descending order. If the fit scores are the same, they are sorted again by fit ratio in ascending order and historical successful execution rate in descending order to generate a list of fit teams.

[0036] The list of compatible teams is effectively verified, and all non-compliant teams are removed. If the list of compatible teams is empty after screening, the requirement fallback mechanism is activated to generate a recommended distribution list.

[0037] The recommended traffic distribution list is bound to the demand identifier, the initial hash value, and the entire filtering process data to generate a filtering data package, and a hash operation is performed to generate a matching hash value.

[0038] Specifically, the steps of the circulation mechanism include:

[0039] Retrieve a regular list pre-configured by the target user, and configure a stable priority for each supply team in the regular list. Based on the stable priority, perform targeted matching with a fixed supply team.

[0040] Conduct demand compliance verification for the fixed supply team, including verification of their qualifications and historical compliance.

[0041] If the verification fails, the replacement and verification will be carried out according to the stability priority in the regular list until a fixed supply team that passes the verification is found.

[0042] If none of the teams in the regular list pass the verification, then the backup team for regular requirements will be activated directly.

[0043] Construct flow instructions and perform integrity verification. Combine the initial hash value, demand identifier, and fixed supply team to generate a regular flow data packet and calculate the flow hash value.

[0044] Specifically, the steps for generating the distribution supply list include:

[0045] Obtain the recommended traffic distribution list and combine it with the matching hash value to construct the basic data packet;

[0046] For each supply team in the recommended distribution list, real-time status data is obtained, and data standardization transformation is performed to generate a real-time status vector.

[0047] Based on the intersection of the attribute tag levels of the demand planning and supply teams, obtain the number of intersections of attribute tag levels, and calculate the fit coefficient based on the total number of tags in the demand attribute tag set.

[0048] The attribute matching degree is calculated by combining the corresponding team's fit score with the fit coefficient.

[0049] Obtain load adaptation score, historical adaptation score, personnel adaptation score, and qualification adaptation score, and combine them with the urgency coefficient to calculate the real-time adaptation degree;

[0050] Configure attribute weights and real-time weights, and calculate the verification score.

[0051] Specifically, the steps for generating the diversion supply list also include:

[0052] Based on the demand attribute tag set and urgency coefficient, set an individual verification threshold to perform individual verification;

[0053] If the verification score is less than the individual verification threshold, the corresponding supply team is removed.

[0054] After individual verification, obtain the supply quantity of the supply team in the recommended distribution table, and set the distribution supply threshold to determine whether to activate the substitute mechanism;

[0055] Once the supply quantity is less than the diversion supply threshold, the backup mechanism is activated. Backup teams are activated sequentially based on the order of the matching team list, and individual verification is performed on the backup teams until the supply is replenished to the diversion supply threshold.

[0056] Calculate compliance score and diversity score, and use the analytic hierarchy process (AHP) to configure weights, perform weighted summation to obtain overall fit score, and set supply qualification threshold;

[0057] If the overall fit score is less than the supply qualification threshold, the corresponding supply team will be removed.

[0058] After overall verification, obtain the supply quantity of the supply team in the recommended distribution table, determine whether to activate the substitute mechanism, and generate the distribution supply list;

[0059] Based on the verification score of the diversion supply list, the allocation share of a single team is calculated based on the weighted ratio, and an upper limit for the share of a single team is set.

[0060] Based on the preset template of the smart contract, a routing instruction is generated and encrypted.

[0061] Specifically, the steps to obtain performance evaluation indicators include:

[0062] Based on demand adaptability, execution effectiveness, and compliance and security, an execution evaluation system associated with demand attribute tags is constructed;

[0063] Acquire real-time execution data, standardize and quantify individual indicators, assign weights to individual indicators, calculate comprehensive execution scores, and thus generate real-time execution evaluation indicators.

[0064] Construct a three-dimensional relationship table of team-execution-requirement, generate an execution profile of requirement attributes, and combine the requirement identifier and process evidence hash value to generate an execution data package and calculate the relationship hash value;

[0065] Perform three-dimensional feedback analysis, generate an optimization solution package based on the analysis results, and then effectively verify it.

[0066] The verified optimization solution package will be synchronously updated to the traffic distribution rule base and smart contract;

[0067] Build an optimized data packet, calculate the optimized hash value using a cryptographic hash algorithm, and write it to the blockchain.

[0068] A data offloading system based on natural language processing includes: a parsing module, an offloading module, a decision-making module, and a feedback module;

[0069] The parsing module is used to obtain demand plans, extract demand information, and screen a set of candidate suppliers with material qualifications. It also constructs a set of demand attribute tags based on support coefficients, market complexity coefficients, technical difficulty coefficients, and risk coefficients.

[0070] The traffic splitting module is used to call the traffic splitting rule library and iterate through and verify the demand attribute tag set to determine whether to split traffic. If the demand attribute tag set meets any condition in the traffic splitting rule library, the differentiated traffic splitting mechanism is activated to generate a recommended traffic splitting list. If the demand attribute tag set does not meet any of the traffic splitting rule library conditions, the flow mechanism is activated to generate a flow instruction.

[0071] The decision-making module is used to acquire real-time status data, calculate the verification score through attribute matching degree and real-time adaptability, and perform individual and overall verification on the recommended diversion list to generate a diversion supply list. It also calculates the allocation share and generates diversion instructions by using weighted ratio allocation and single team share limit.

[0072] The feedback module is used to obtain execution evaluation indicators, construct a three-dimensional association table and a requirement attribute execution profile and store them on the blockchain, perform three-dimensional feedback analysis, and generate parameter optimization schemes for updating.

[0073] The beneficial effects of this invention are:

[0074] By structurally analyzing demand plans, extracting demand information, and calculating multi-dimensional attribute indicators, a demand attribute tag set is constructed to accurately identify special demand attributes such as core materials, high-risk, and high-difficulty requirements. This overcomes the limitation of being unable to identify the essential differences in special demands. The demand information and tag set generate an initial hash value, which is then stored on the blockchain for evidence, ensuring data immutability and full-process traceability. This solves the problem of insufficient demand data credibility. A traffic diversion rule base determines whether a traffic diversion is triggered. When diversion is triggered, a recommendation list is generated based on the mapping relationship between the tag set and the professional team. If diversion is not triggered, a flow instruction is directly generated and associated with evidence storage, thus achieving differentiated processing of demand and regular demand. This approach completely changes the current situation of homogeneous processing, avoiding the execution risks caused by matching special needs with ordinary teams, while enabling the efficient flow of routine needs. By calculating verification scores based on dual verification criteria, and generating a distribution supply list through individual and overall verification, the adaptability of distribution teams and supply stability are ensured, further reducing potential risks. All process data and execution indicators are linked to the blockchain to form a complete data chain, providing a basis for the iteration of distribution rules and weight parameters, continuously improving the accuracy and adaptability of distribution decisions, and achieving an upgrade from passive matching to proactive optimization. Ultimately, this achieves a comprehensive effect of reasonable resource allocation, effective risk control, and improved procurement efficiency. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of a data splitting method based on natural language processing.

[0076] Figure 2 This is a flowchart illustrating the construction of the requirement attribute tag set in this invention;

[0077] Figure 3 This is a flowchart of the differentiated traffic splitting mechanism in this invention;

[0078] Figure 4 This is a flowchart for generating the distribution supply list in this invention. Detailed Implementation

[0079] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0080] Example 1

[0081] refer to Figures 1 to 4 As shown, this embodiment introduces a data splitting method based on natural language processing, including the following steps:

[0082] The system receives demand plans submitted by target users, performs structured parsing, extracts demand information, and calculates attribute indicators, including support coefficient, market complexity coefficient, technical difficulty coefficient, and risk coefficient, in accordance with procurement risk assessment specifications and procurement management standardization requirements. This allows for the construction of a demand attribute tag set, enabling an understanding of the essential differences between different demand plans. This achieves a leap from homogeneous processing to differentiated distribution. The system integrates demand information and demand attribute tag set, generates an initial hash value, and stores it in the enterprise consortium blockchain node for evidence storage, ensuring that the data is tamper-proof and traceable throughout the entire process.

[0083] The system invokes the built-in distribution rule library of the procurement platform to traverse and verify the demand attribute tag set to determine whether distribution should occur. If the demand attribute tag set meets any condition in the distribution rule library, distribution is triggered, and a differentiated distribution mechanism is initiated. This involves invoking distribution rules in the blockchain, and based on the mapping relationship between different combinations of demand attribute tags and professional supply teams, the system uses the demand attribute tag set as input to match teams with the distribution rules, generating a suitable recommended distribution list. If none of the demand attribute tag sets meet the distribution rule library requirements, the data is considered regular demand data, and distribution is not triggered. Instead, a transfer mechanism is initiated, generating a transfer instruction, binding the demand identifier, and generating a transfer hash value, which is then associated with the initial evidence. The distribution rule library includes: material tag level as the core, risk tag level as medium or high risk, technology tag level as high difficulty, and supply tag level as high complexity.

[0084] For each recommended target team in the recommended distribution list, real-time status data is obtained, including task load rate, historical execution scores of similar procurement projects, and availability of key personnel. Attribute matching degree and real-time adaptability are used as dual verification criteria to calculate the verification score of each recommended target team. Individual and overall verifications are performed on the recommended distribution list to generate a distribution supply list and distribution instructions.

[0085] A unique identifier for the demand plan is generated based on the diversion instruction. This identifier is then integrated with the diversion supply team, decision timestamp, and verification score to generate a new hash value. This hash value is written to the blockchain to complete the trusted notarization of the entire diversion process. After the demand plan is executed, execution data is collected and combined with the execution evaluation system of the executing department to obtain execution evaluation indicators. These indicators are then linked and bound to demand attribute tags and written to the blockchain again. This forms a complete data chain containing the demand plan, diversion decision, execution results, and execution feedback. This data serves as the basis for diversion rules and the dimensional weights of demand attribute tags. By continuously accumulating historical data, the parameters in the diversion process are iterated to continuously improve the accuracy and adaptability of diversion decisions. In this embodiment, the executing department is specifically the procurement department.

[0086] Specifically, the steps for constructing a set of requirement attribute tags include:

[0087] Obtain the demand plans submitted by target users through the procurement platform, including text files and image files, and perform compatibility verification, including verifying file readability, completeness of demand fields, and compliance of data format. If any dimension fails the verification, automatically generate supplementary prompts to be fed back to the procurement platform, and receive supplementary information for re-verification until the verification passes.

[0088] For text files, natural language processing is used to extract requirement information from the demand plan, such as material type, technical parameter requirements, supply cycle, application scenario, material purpose, and supplier pre-selection list (if any). Simultaneously, the completeness and semantic consistency of the extracted fields are assessed, and the confidence level of each field is calculated. If the confidence level of the requirement information falls below a preset standard threshold, manual review is triggered by the platform administrator, who verifies and corrects the extracted results. If image files exist, such as screenshots of technical drawings, image recognition is used to parse the visual information of the materials, such as their appearance and specifications, and supplement it into the requirement information, based on the procurement platform's internal standards. The template is used to format the requirement information and generate a structured requirement list. If there are different requirements in the same requirement plan, such as the same type of steel used for core products and office furniture, it is split into multiple independent requirement lists, each corresponding to a requirement attribute requirement. The confidence level is calculated based on the completeness and semantic consistency of the extracted fields. Samples of all requirement plans analyzed by the enterprise in the past 3 years are collected and divided into correct samples and incorrect samples. Each sample is bound to the confidence level of the corresponding field of NLP recognition output to form a full confidence dataset. The confidence level of the numerical field of all incorrect samples is extracted, and the upper quartile of the data set is taken as the standard threshold.

[0089] Based on the enterprise entity representation, demand upload timestamp, and demand type in the demand plan, a unique demand identifier for the current demand plan is generated using hash calculation. Based on the material type in the demand list, candidate suppliers with corresponding material supply qualifications are selected from the supplier resource library of the procurement platform to form a candidate supplier set. The demand identifier, demand list, and candidate supplier set are then bound together to generate a demand data package.

[0090] Based on the business priorities of the demand plan, corresponding business weights are configured. Substitution coefficients are set based on material substitution solutions uploaded by target users. Support coefficients are calculated based on the product of business weights and substitution coefficients. Secondary material classification thresholds are set. Combined with the support coefficients, material label levels are obtained, including core, regular, and general materials. Business priorities are preset by target users based on their own business needs, including core, important, and general business lines, each corresponding to different weight quantification values. All procurement needs from the past three years are obtained and divided into three sample sets according to business priority labels. The average support coefficient of the general business sample is used as the benchmark denominator. The average values ​​of the core, important, and general business samples are respectively converted to the benchmark by ratios to obtain the business weights corresponding to the three business categories. Substitution solutions include no substitution, partial substitution, and complete substitution, each with its own weight. Depending on the different substitution quantification values, the secondary material classification thresholds are pre-set by the user based on the distribution statistics of historical support coefficients for similar demands. These thresholds are set as the historical mean and the upper tertiary value. The mean represents the average level of the support coefficient for ordinary materials in the enterprise, serving as the boundary between low-end and mid-end materials, i.e., the basis for classifying general materials and conventional materials. The upper tertiary value represents the watershed of the support coefficient for scarce and irreplaceable materials in history, serving as the boundary between conventional materials and core materials. Historical material substitution records are obtained and divided into three categories of samples based on the material substitution scheme: no substitution, partial substitution, and complete substitution. Based on the ratio of the number of substitutable orders to the total number of orders, the material coverage rate of each category of samples is calculated. The material coverage rate of the no-substitutability samples is used as the benchmark denominator. The material coverage rates of the three categories of samples are compared with the benchmark to obtain the corresponding substitution coefficients.

[0091] The process involves: obtaining the number of potential suppliers for the current demand plan; determining the intention level based on a quantification standard (e.g., a quantification level of 3 for more than 5 potential suppliers); retrieving historical cooperation data of target users from the procurement platform to obtain the cooperation duration of each potential supplier; setting a stable weight based on threshold quantification rules and average cooperation duration (e.g., if the cooperation durations of different potential suppliers are unequal, the average cooperation duration is used; a stable weight of 1 is assigned for an average cooperation duration exceeding 3 years); calculating the market complexity coefficient based on the intention level and stable weight; setting a secondary complexity threshold; and obtaining the supply label level of the current demand plan, including high, medium, and low complexity. The expression is shown below:

[0092]

[0093] In the formula, For market complexity coefficient, The intended level of the current demand plan. The maximum value of the intention vectorization level. The intention level correction coefficient is configured for the target user. This is achieved by obtaining a dataset of all user intention levels and final order delivery risk data from the past three years, and performing a linear fit with intention level as the independent variable and delivery anomaly rate as the dependent variable. The coefficient of the linear term obtained from the fit is the intention level correction coefficient. To stabilize the weights, the secondary complexity threshold is preset by the user based on the quartiles of historical market complexity coefficients, namely the lower quartile and the upper quartile. The quartiles evenly divide the market complexity of all historical demands. The lower quartile represents a simple procurement scenario with stable cooperation and sufficient suppliers, and is considered low complexity if it is below the lower quartile. The upper quartile represents a complex procurement scenario with few suppliers and large fluctuations in cooperation, and is considered high complexity if it is above the upper quartile. The complete cooperation duration of all historical suppliers of the user is extracted to form a cooperation year dataset. The cooperation years are divided into three cooperation intervals: long, medium and short, according to the quartiles. The sample mean of the short cooperation interval is used as the benchmark. The sample mean of the short, medium and long intervals are respectively compared with the benchmark to generate three stable weights.

[0094] Based on the technical parameter requirements in the demand information, the number of technical parameter items and the number of special processes are obtained, as well as the acceptance level of the target user. A weighted calculation is performed on the number of technical parameter items, the number of special processes, and the acceptance level to obtain the technical difficulty coefficient. A secondary technical classification threshold is set to obtain the technical label level of the current demand plan, including high, medium, and low difficulty. The secondary technical classification threshold is preset based on the median and upper tertiary of the historical technical difficulty coefficient dataset. The median eliminates interference from extremely high / low technical requirements and represents the average difficulty of the company's routine technical procurement, serving as the boundary between low and medium difficulty. The upper tertiary filters out a small proportion of high-precision processes and multiple special parameter requirements, serving as the boundary between medium and high difficulty. A historical sample dataset containing the number of technical parameter items, the number of special processes, and acceptance is constructed. The Pearson correlation coefficients between the three indicators and the technical execution failure rate are calculated separately. The three sets of correlation coefficients are summed and normalized; the normalized values ​​are the weighted weights of the three indicators.

[0095] We obtained the target users' self-assessed risk values ​​for supply disruption risk, price fluctuation risk, and capacity risk, as well as the self-assessed impact coefficients of historical orders for the same material category after risk events. The self-assessed risk values ​​were obtained by summing the scores for each of the three risks according to a five-level quantitative scale. The scale levels were divided at equal intervals based on the frequency of similar historical demand risk events. The self-assessed impact coefficients were the average of the losses to overall project progress and costs caused by historical orders for the corresponding material category experiencing risk events. Combining the supply cycle, we quantified the supply weights and obtained a dataset of all order supply cycles from the past three years. These cycles were divided into three tiers: regular cycle, expedited cycle, and emergency cycle. The historical delivery default probability of orders in each of the three tiers was calculated, using the default probability of the regular cycle as the baseline denominator. The probability of default in the acute cycle is compared with the benchmark to obtain the corresponding supply weight. The risk coefficient is obtained by multiplying the self-assessed risk value, the self-assessed impact coefficient, and the supply weight. A secondary risk threshold is set to obtain the risk label level of the current demand plan, including high, medium, and low risk. The secondary risk threshold is preset based on the upper quartile and upper nintile of the historical risk coefficient dataset. The vast majority of procurement demand has low risk, and only a small number of orders have high risks such as supply chain disruptions, delivery defaults, and price surges. The upper quartile divides low risk into medium risk. Procurement orders with no major default records in history are defined as low risk. The upper nintile screens the top 10% of historical high-risk orders. Demands above the upper nintile have experienced delays, supply disruptions, or quality incidents and need to be assigned a professional supply team, and are defined as high risk.

[0096] The four types of attribute tag levels are integrated to generate a structured set of demand attribute tags, which are then bound to the corresponding demand data package. A hash operation is performed to generate an initial hash value, which is then synchronously written to the enterprise consortium blockchain node along with the demand identifier for initial notarization. The notarization record includes a notarization timestamp and the code of the data-owning enterprise. The blockchain node returns a notarization success receipt, which the platform binds and archives with the demand plan data as a basis for subsequent data traceability.

[0097] Specifically, the steps of the differentiated traffic diversion mechanism include:

[0098] Based on the demand identifier of the current demand plan and the enterprise code of the target user, the smart contract in the blockchain is invoked, and the validity of the smart contract is verified, including: verifying whether the contract version is the latest, whether the routing rules in the contract are consistent with the routing rule library built into the procurement platform, and whether the contract call permissions match. Among them, the smart contract is deployed to the blockchain in advance by the target user based on business needs, and stores the mapping rules of supply team-attribute type-tag level. The supply team is also associated with team responsibilities, basic qualification filing, and capacity threshold. The attribute types include materials, risks, technology, and supply.

[0099] It iterates through all the supply teams in the candidate supplier set, excluding unrelated external teams. Based on the mapping rules, it compares the four attribute tag levels of each supply team with the demand attribute tag set item by item. It eliminates supply teams whose tag levels are lower than the demand plan tag levels under the same attribute type, and only retains supply teams that can execute the demand plan as candidate suppliers. For example, if the demand attribute tag set is for regular materials, it eliminates supply teams whose attribute configuration is for general materials. It integrates all candidate supply teams that pass the comparison to generate a basic matching team pool, avoiding team files without corresponding demand capabilities from entering the subsequent deep matching process, and reducing invalid calculations and resource consumption.

[0100] Based on the attribute priority matching rules deployed in the smart contract, the tag priority of each tag level in the current demand plan is obtained. Simultaneously, based on the tag coding rules preset within the smart contract, numerical codes are assigned to each tag level in the current demand plan. The tag priorities and numerical codes are integrated, and a set of coded tags for the current demand plan is generated according to the attribute type order of materials, risks, technologies, and supply. The attribute priority matching rules store the correspondence between tag levels and tag priorities. These rules are configured based on the requirement attributes represented by the tag level itself. For example, core materials, high risk, and high difficulty are core tags, and their corresponding tag priorities are... Conventional materials, medium risk, medium difficulty, and high complexity are secondary tags, and their corresponding tag priorities are as follows: The remaining tags are regular tags, and their corresponding tag priority is... The label coding rules store the correspondence between label levels and numerical codes, including: the numerical codes corresponding to core materials, high risk, high difficulty, and high complexity are as follows: The numerical codes corresponding to conventional materials, medium risk, medium difficulty, and medium complexity are as follows: The numerical codes corresponding to the remaining labels are ;

[0101] Based on the pre-written target user information, corresponding priority weights are configured for each tag priority. Based on the coded tag set, the numerical codes and corresponding priority weights are weighted to obtain the matching strength value, which quantifies the complexity and importance of the requirements. In particular, the frequency of high, medium and low-end tags in all requirements over the past 3 years is statistically analyzed. The frequency of low-end tags is used as the benchmark value, and the frequencies of low, medium and high-end tags are inversely proportional to the benchmark value to generate corresponding priority weights.

[0102] The process involves two-way strength matching of execution demand attributes and team capabilities, including: extracting the capacity threshold of each supply team in the basic adaptation team pool and performing forward matching; removing supply teams from the basic adaptation team pool whose capacity threshold is lower than the matching strength value; and removing teams that can handle similar core materials but whose capacity threshold is lower than the current demand matching strength value. This initial narrowing of the candidate pool is then performed. Finally, reverse matching is performed on each supply team in the basic adaptation team pool, reading the historical supply data of each candidate supply team, obtaining the highest priority tag executed for each attribute type, and removing any... Candidate supply teams whose historical highest tag priority under the attribute type is lower than the tag priority of the demand are selected to ensure that the supply teams have practical experience in handling this type of demand. Finally, a pool of suitable candidate teams is generated to achieve a deep fit between the importance of the demand and the team's capacity, completing a second narrowing process. The capacity threshold is set based on the team's functional positioning and historical peak capacity. It is the upper third of the matching strength value of the historical demand that a single supply team can handle. When the demand matching strength value is not greater than the capacity threshold, it indicates that the team has the capacity and should be retained, representing the highest demand intensity level that the team can handle.

[0103] For supply teams in the candidate pool, the matching ratio of each supply team is calculated based on the ratio of demand matching intensity value to capacity threshold. At the same time, the historical database of the procurement platform is read to obtain the historical successful execution rate of each supply team in handling similar demands. The matching ratio and the historical successful execution rate are weighted to obtain the matching score of each supply team. Among them, the Pearson correlation coefficients of the historical matching ratio and historical successful execution rate with the final order delivery qualification rate are retrieved. The two sets of correlation coefficients are summed and normalized and then used as the weights of the two indicators.

[0104] Based on the fit score, the supply teams in the fit candidate team pool are sorted in descending order to make the differences between multiple teams of the same type explicit. If the fit scores are the same, they are sorted again in ascending order of fit ratio and descending order of historical successful execution rate to generate a list of fit teams.

[0105] Based on the list of compatible teams, the operational status of each candidate team is verified (whether it is operating normally and without service interruption), the validity of its basic qualification filings, and whether it is within the company's pre-selected supplier list. All non-compliant teams are eliminated. If the number of valid teams in the list of compatible teams after screening is 0, the demand fallback mechanism is activated, calling the emergency procurement team preset by the smart contract. At the same time, a demand matching alarm is generated and synchronized to the company's procurement manager to ensure that the process is not interrupted, thereby generating a recommended distribution list. The emergency procurement team is used specifically to handle the demands of teams without compatible teams.

[0106] The recommended traffic distribution list is bound to the demand identifier, the initial hash value, and the entire filtering process data to generate a filtering data package, and a hash operation is performed to generate a matching hash value.

[0107] Specifically, the steps of the circulation mechanism include:

[0108] The regular list pre-configured by the target user in the procurement platform is retrieved. All of these lists are from the locked set of candidate suppliers. At the same time, the supply teams in the regular list are configured with stable priorities. Fixed supply teams are matched based on stable priorities.

[0109] For fixed supply teams, demand compliance verification is performed, including qualification validity verification and historical compliance verification. If a verification fails, a replacement is made based on the stable priority in the regular list and verified again until a fixed supply team that passes the verification is found. If all teams in the regular list fail the verification, the regular demand backup team is directly activated to ensure the compliance of the regular demand execution team without affecting the overall workflow efficiency, effectively balancing the compliance and execution efficiency of regular demands. Qualification validity verification includes: verifying whether the team's relevant qualifications are still valid based on the existing certified basic qualification filing information, without repeatedly retrieving external data. Historical compliance verification includes: searching the procurement platform's historical database for the team's past compliance records in handling regular demands, focusing on verifying whether there are any violations. A compliance record compliance method is used for judgment, that is, if the number of violation records does not exceed the compliance threshold configured by the enterprise, the verification is considered passed. In particular, a dataset of historical violation records of all supply teams in the past 3 years is collected, and the number of violations corresponding to all orders that fail compliance verification is counted. The upper quartile of the number of violations is taken as the compliance threshold.

[0110] Extract core information from the demand data package, including demand identifier, demand list, material type, and delivery requirements, to ensure that the instructions contain all the core information required for execution. At the same time, based on the standardized flow instruction template of the procurement platform, generate structured flow instructions and perform integrity verification on the flow instructions, such as checking for missing core fields or ambiguous information descriptions, to ensure that the instruction content is complete, clear, and unambiguous. After the verification is passed, lock the instruction content to avoid subsequent modifications that may cause execution deviations.

[0111] The transfer instructions are bound to the initial hash value, demand identifier, fixed supply team, and compliance verification record to form a regular transfer data packet, ensuring the continuity and integrity of the data link. An encrypted hash algorithm is used to encrypt the regular transfer data packet to generate a transfer hash value, ensuring the immutability of the data packet. The transfer hash value, demand identifier, evidence storage timestamp, and enterprise code are synchronously written into the enterprise consortium blockchain node and associated with the initial evidence storage record. The evidence storage success receipt returned by the blockchain node is received and bound to the regular demand transfer data packet for archiving, serving as the core basis for subsequent demand traceability, execution evaluation, and dispute resolution.

[0112] Specifically, the steps for generating the distribution supply list include:

[0113] The system obtains a recommended distribution list, combines it with matching hash values ​​to construct a basic data package, and for each supply team in the recommended distribution list, it calls a blockchain trusted data source node to obtain the on-chain evidence records of each real-time data, thereby obtaining real-time status data, including task load rate, historical similar procurement project execution scores, core personnel availability, and real-time qualification validity. The system then performs data standardization transformation, converting the collected unstructured real-time status data into unified quantitative indicators, such as converting the text status of core personnel availability into corresponding quantitative values, thereby generating a real-time status vector.

[0114] A secondary verification is performed to check the matching between the demand plan and the supply team. Each supply team has corresponding matching tags, but there are cases where the level of a certain attribute tag is not clearly defined. For example, if a supply team does not involve supply tag levels, it means that this supply team has not classified the supply. Such supply teams will be assumed to be able to handle the supply situation of each tag level when screening candidate suppliers. Based on the intersection of the attribute tag levels of the demand plan and the supply team, the number of intersections of attribute tag levels is obtained. At the same time, the total number of tags in the demand attribute tag set is obtained. Based on the ratio of the number of intersections to the total number of tags, the matching coefficient is calculated. Combined with the matching score of the corresponding team, a secondary calibration is performed through product operation to obtain the attribute matching degree.

[0115] By using load rate inverse quantification, task load rate is converted into load adaptation score to avoid delivery delays after high-load teams take on requirements. Historical adaptation scores are based on the execution scores of similar procurement projects. Based on the availability quantification index of core personnel and the weight allocation of core personnel, personnel adaptation scores are calculated to ensure that the team's key execution forces are in place. Real-time qualification validity is directly quantified into a binary index, with full marks for valid qualifications and zero marks for invalid qualifications, generating qualification adaptation scores. In particular, historical samples of the on-duty status of core personnel in all orders over the past 3 years and the delivery failures caused by personnel departures are obtained. The correlation coefficients of the impact of core personnel departures in different positions on order failure rates are statistically analyzed. The correlation coefficients of multiple positions are summed and normalized, and the normalized result is the assigned weight of core personnel in each position.

[0116] The urgency level of the current demand plan is obtained from the demand information, and the corresponding urgency coefficient is configured accordingly. For example, the urgency coefficient of urgent demand is 1.2 and the urgency coefficient of regular demand is 1. The above four matching scores are weighted and summed with preset weights, and then multiplied by the urgency coefficient to calculate the real-time matching degree. In this process, the delivery cycle of all purchase orders in the past 3 years is extracted and divided into three levels: regular demand, expedited demand and urgent demand according to the ternary digits. The historical delivery delay probability of orders in the three levels is calculated for each level. The delay probability of regular demand orders is used as the benchmark, and the delay probabilities of regular, expedited and urgent demand orders are compared with the benchmark to obtain the urgency coefficient of each level.

[0117] Based on the target user's demand type, configure corresponding attribute weights and real-time weights, and calculate the verification score by weighted summation of attribute matching degree and real-time adaptability. Among them, purchase orders that have completed complete diversion and execution loop in the past 3 years are selected, and the attribute matching degree, real-time adaptability and execution score of each order are extracted. With the execution score as the target layer and attribute matching degree and real-time adaptability as the criteria layer, a pairwise comparison matrix is ​​constructed. The elements inside the matrix are the relative importance ratios of the two dimensions in the criteria layer, which are automatically generated by the Pearson correlation coefficient between each dimension and the comprehensive execution score. The matrix is ​​subjected to consistency test. After the test passes, it is normalized to obtain the attribute weights and real-time weights. If the test fails, it indicates that the sample has logical descriptions. Extreme abnormal order samples that deviate from the overall data distribution are removed, the pairwise comparison matrix is ​​reconstructed and tested again.

[0118] Based on the demand attribute tag set and urgency coefficient, the corresponding individual verification threshold is retrieved from the dynamic threshold library of the procurement platform. Each supply team in the recommended allocation list is compared with the individual verification threshold to perform individual verification. If the verification score is less than the individual verification threshold, the verification is deemed to have failed, and the supply team is removed from the recommended allocation list; otherwise, the individual verification is deemed to have passed. The dynamic threshold library is preset by the target user based on historical execution data and demand orientation, and is updated regularly through blockchain notarization to ensure the authority and adaptability of the thresholds. Specifically, historical verification score samples are obtained, and the average verification scores of all samples are calculated to obtain the individual verification threshold.

[0119] After individual verification, the supply quantity of each supply team in the recommended distribution table is obtained, and a distribution supply threshold is set as the minimum number of qualified distribution teams to avoid monopolies and ensure supply stability, in order to determine whether to activate the backup mechanism. If the supply quantity is less than the distribution supply threshold, it is determined that the supply teams in the current recommended distribution table are insufficient to meet the distribution needs of the current demand plan, and the backup mechanism is activated. Backup teams are activated sequentially based on the order of the suitable team list, and individual verification is performed on the backup teams until they are supplemented to the distribution supply threshold. Otherwise, it is determined that the recommended distribution table meets the distribution needs of the current demand plan, and the original recommended distribution table is maintained. The distribution supply threshold is a preset minimum number of supply teams. Based on the historical backup mechanism, the historical procurement distribution records that have triggered the backup mechanism in the past 3 years are traversed, and the number of backup teams that have passed individual verification before the backup is activated for each order is extracted to construct a historical reserve dataset. The minimum value of the dataset is taken as the distribution supply threshold, which represents the minimum number of qualified backup teams that the enterprise can reliably mobilize from its past supplier resources.

[0120] Compliance and diversity scores were calculated, and the Analytic Hierarchy Process (AHP) was used to extract historical order samples from the past three years showing all orders that had completed closed-loop execution. The final execution result of each order (delivery successful / execution failure) was used as the target layer of the AHP, with compliance and diversity scores serving as the two main criteria layers. A second-order pairwise comparison matrix was constructed, where the elements were generated by converting the Pearson correlation coefficients of the compliance and diversity dimensions with the order execution failure rate, respectively. The weights of the compliance and diversity scores were obtained by normalizing the sum of the matrix's largest eigenvectors. A weighted sum was then calculated to obtain the overall suitability score, and a set of parameters was set for... The overall fit score is below the supply qualification threshold. If the overall fit score is less than the supply qualification threshold, the overall verification is deemed unsuccessful, and the supply team is removed. Otherwise, the overall verification is deemed successful. The compliance score is derived from three verification dimensions extracted from the diversion compliance rule library, and is obtained through item-by-item compliance scoring. A correlation identification algorithm is used to check for monopoly risks. For example, if there is no equity relationship between teams or they share the same actual controller, full marks are awarded for compliance, and zero marks for non-compliance. This yields correlation compliance, with the fit coefficient serving as the coverage compliance. Full marks are awarded if all supply teams are on the pre-selected supplier list; otherwise, points are deducted based on the proportion of non-compliant teams (if any). The supplier pre-selection list is used to obtain the compliance score. The three scores are weighted and summed according to preset weights to obtain the compliance score. Samples of order execution failures caused by three types of issues—related violations, off-list violations, and incomplete coverage—are extracted from historical orders. The Pearson correlation coefficient between each of these three issues and the order failure rate is calculated. The three correlation coefficients are then summed and normalized; the resulting value represents the weighted weight of joint compliance, list compliance, and coverage compliance. The diversity score is based on the supplier diversity indicator library of the procurement platform and is obtained by scoring according to dimensional compliance. If the supply team covers different geographical areas, such as local + remote, a higher score is awarded. Full marks are awarded for complete coverage. Points are deducted based on the proportion of missing regions to obtain supply diversity. The core compatibility labels of the supply teams are judged to be different, such as some being good at fast delivery and others being good at cost control. Full marks are awarded for meeting the standard. Points are deducted based on the proportion of capability homogeneity to obtain service complementarity. The two scores are weighted and summed according to preset weights to obtain the diversity score. Execution failure samples corresponding to the two types of problems of regional homogeneity and capability homogeneity are extracted from historical orders. The correlation coefficients between the two types of defects and the order failure rate are calculated. After summing and normalization, the supply diversity weight and service complementarity weight are obtained. The supply qualification threshold is the median of the historical overall compatibility score.

[0121] After overall verification, the supply quantity of the supply team in the recommended distribution table is obtained, and it is determined whether to activate the substitute mechanism, and finally the distribution supply list is generated;

[0122] Based on the verification scores of the diversion supply list, the ratio of the verification score of a single supply team to the total verification scores of all candidate teams in this diversion is used as the allocation weight. The procurement share is allocated according to the allocation weight. The verification scores of each supply team in the diversion supply list are obtained, and the total verification score of all supply teams is calculated. Based on the ratio of the verification score of a single supply team to the total verification score, the weight of each supply team is calculated. Based on the product of the total demand plan share and the weight of a single team, the allocation share of a single team is calculated. At the same time, an upper limit for the share of a single team is set. If the allocation share of a team exceeds the upper limit of the share of a single team, the upper limit of the share of a single team is used as the allocation share of that team, and the remaining excess is redistributed according to the weight ratio of the remaining supply teams. This avoids any team from having an absolute dominant position and prevents the risk of monopoly from the mechanism. In particular, based on historical data, orders that have caused delivery stagnation or concentrated quality defects due to the excessive proportion of a single supply team are screened. The share proportion of the corresponding team in the faulty orders is recorded, and the lower quartile of the share proportion is taken as the upper limit of the share of a single team.

[0123] Based on the preset template of the smart contract, a structured distribution instruction is generated. The instruction includes basic demand information, team allocation information, and verification basis information. The basic demand information includes demand identifier, demand attribute tag set, and demand data package. The team allocation information includes the team identifier, allocation share and delivery node requirements of each distribution supply team. The verification basis information includes the basic data package, verification standard score and overall adaptability score of each team.

[0124] The generated distribution instructions are encrypted using the AES-256-GCM encryption algorithm to prevent the instructions from being tampered with or illegally obtained. At the same time, the distribution instructions, share allocation details, full-process verification data are bound with the demand identifier and matching hash value to generate the final decision data packet. The decision hash value is generated and written into the enterprise consortium blockchain through the encrypted hash algorithm.

[0125] The encrypted allocation instructions are pushed to the terminals of each allocation supply team through the procurement platform interface. At the same time, the allocation decision results are submitted to the user in the demand side, including the list of allocation supply teams, the allocation share of each team, and the evidence query portal, to ensure information symmetry between the demand side and the execution side.

[0126] Specifically, the steps to obtain performance evaluation indicators include:

[0127] The decision data package is obtained and structured by combining the diversion instructions, demand identifiers, diversion supply lists, decision timestamps, and verification scores from each team to form a diversion panoramic package. This ensures that the data link fully covers the entire process from demand entry to diversion decision-making and avoids data fragmentation.

[0128] An encrypted hash algorithm is used to perform encrypted calculations on the traffic distribution panoramic package to generate a process evidence hash value, which is used to identify the integrity of the entire process data. Based on the initial hash value, matching hash value, decision hash value and process evidence hash value, a hash chain generation algorithm is used to construct an associated verification hash value, forming a hash association chain of initial-filtering-verification-decision-full process to prevent data from being tampered with or the chain from being broken at any stage.

[0129] The entire process of diversion, process evidence hash value, correlation verification hash value, and evidence receipts of each stage are synchronously written into the enterprise consortium blockchain. The evidence records are indexed according to the required identifier and timestamp, and at the same time, a correlation index is established with the evidence records of the previous steps to achieve the query effect of one-click traceability of the entire process, ensuring the immutability and auditability of the diversion decision process.

[0130] After data notarization is completed, an execution evaluation system is constructed based on demand adaptability, execution effectiveness, and compliance security, and is associated with demand attribute tags. Real-time execution data is automatically acquired from the procurement platform, and execution indicators are quantified. For a single indicator, a standardized method is used to convert the real-time execution value into a standardized score within the same range. Weights are assigned to the total score of demand adaptability indicators, the total score of execution efficiency indicators, and the total score of compliance security indicators to calculate a comprehensive execution score, thereby generating real-time execution evaluation indicators. Among these, demand adaptability indicators directly correspond to… , The system uses attribute tags at the level of execution efficiency indicators to link traffic allocation and delivery node requirements, and compliance and security indicators to link compliance verification requirements, forming a highly targeted and non-redundant indicator system.

[0131] A three-dimensional relationship table is constructed for teams, execution, and requirements. A relationship table is established for each distribution supply team to record the core requirement attribute tags, screening rules, verification scores, comprehensive execution scores, and individual indicator scores corresponding to the requirement plans undertaken by the distribution supply team. Based on multiple rounds of procurement execution data of the same requirement attribute tag set, a clustering analysis algorithm is used to extract execution benchmark features, including the average comprehensive execution score and the qualification threshold of each indicator, thereby generating a requirement attribute execution profile. The three-dimensional relationship table, the requirement attribute execution profile, and the requirement identifier and process storage hash value are bound together to generate an execution data package. The association hash value is calculated through an encrypted hash algorithm and written to the blockchain. It is also linked with the previous distribution decision storage records to ensure that the relationship is tamper-proof and traceable.

[0132] A three-dimensional feedback analysis was conducted, including: in the analysis of the adaptability of the diversion rules, comparing the execution profiles of different demand attribute tag sets to determine the rationality of the mapping rule of execution team-attribute type-tag level, and clarifying which teams are more suitable to undertake a certain type of demand; in the analysis of the rationality of the weight parameters, using a multiple linear regression analysis algorithm to explore the correlation between the adaptability score weight, attribute and real-time weight, and comprehensive execution score, and to identify the correlation between weight configuration and execution improvement; in the analysis of the scientificity of the threshold standards, based on the qualified threshold of the demand attribute execution profile, verifying the rationality of the diversion trigger threshold and the individual and overall verification threshold through statistical hypothesis testing, and determining whether there are situations where the threshold is too high, causing high-quality teams to be eliminated or the threshold is too low, causing unqualified teams to be admitted.

[0133] Based on the three-dimensional analysis results, the proposed adjustments to mapping rules, updated weight configuration values, and revised threshold standards are clearly defined. The basis for each optimization is explained in detail to ensure the optimization scheme is traceable and explainable. The optimization scheme package is applied to similar historical procurement data samples to simulate the diversion process. The execution prediction scores before and after optimization are compared. An optimization scheme is considered effective only when the predicted execution improvement reaches the company's preset standard. If the standard is not met, the feedback analysis is repeated until a scientifically feasible optimization scheme is formed. Specifically, the improvement in execution prediction scores before and after simulation of all previous optimization scheme packages is extracted to construct an improvement sample set, and the lower quartile of this sample set is used as the preset standard.

[0134] The content of the verified parameter optimization solution package will be synchronously updated to the built-in diversion rule library of the procurement platform. At the same time, a rule update log will be generated to clarify the updated content, basis and scope of impact, which will facilitate subsequent auditing. The smart contract will be updated synchronously. Through the rule update interface of the blockchain smart contract, the optimized core parameters will be written into the smart contract to overwrite the original static configuration and trigger the contract version update synchronously. This will ensure that the contract rules are completely consistent with the platform rule library and avoid rule conflicts.

[0135] The parameter optimization solution package, rule update log, smart contract update record, solution verification result and requirement identifier, and associated verification hash value are bound together to generate an optimization data package. The optimization hash value is calculated using an encrypted hash algorithm and written to the blockchain.

[0136] Example 2

[0137] Another embodiment of the present invention provides a data splitting system based on natural language processing, comprising: a parsing module, a splitting module, a decision module, and a feedback module;

[0138] The parsing module receives text and image-based requirement plans submitted by target users and performs compatibility checks on file readability, requirement field completeness, and data format compliance. It extracts text requirement information through natural language processing and visualizes data through image recognition. After confidence assessment and manual review and correction, it generates a structured requirement list according to a standardized template. It splits requirements with different requirement attributes in the same plan and generates a unique requirement identifier for the requirement plan. It selects a set of candidate suppliers with material qualifications and binds them to generate a requirement data package to ensure the accuracy and completeness of requirement information. Combining procurement risk assessment specifications and management standardization requirements, it determines the four tag levels of materials, supply, technology, and risk by calculating support coefficients, market complexity coefficients, technical difficulty coefficients, and risk coefficients. It integrates and generates a structured requirement attribute tag set, binds it with the requirement data package to generate an initial hash value, and writes it to the enterprise consortium blockchain for evidence storage. This accurately identifies the differences in requirement attributes of the requirement plan, breaks the limitations of homogeneous processing, and ensures that the data is tamper-proof and traceable throughout the entire process.

[0139] The distribution module calls the procurement platform's built-in distribution rule library to iterate and verify the demand attribute tag set to determine whether distribution should occur. If the demand attribute tag set meets any condition in the distribution rule library, distribution is triggered, and a differentiated distribution mechanism is activated. This involves calling the distribution rules in the blockchain, matching the demand attribute tag set with the distribution rules based on the mapping relationship between different demand attribute tag combinations and professional supply teams, and generating a suitable recommended distribution list. If none of the demand attribute tag sets meet the distribution rule library's requirements, the data is considered regular demand data, and distribution is not triggered. Instead, a transfer mechanism is activated, generating a transfer instruction, binding the demand identifier, and generating a transfer hash value, which is then associated with the initial storage certificate. This achieves accurate division between demand and regular demand, avoiding resource mismatch and inefficiency.

[0140] The decision-making module is used to call the blockchain's trusted data source, obtain the real-time status data of the supply team and standardize and transform it. It calculates the verification score through attribute matching degree and real-time adaptability, and performs individual and overall verification on the recommended diversion list to generate a diversion supply list. It calculates the allocation share by adopting a weighted ratio allocation and a single team share limit, generates diversion instructions and binds the full-process data to generate a decision hash value on the chain to ensure that the diversion team is available in real time, compliant and stable overall, and prevents the risk of monopoly.

[0141] The feedback module is used to collect demand plan execution data. Based on the dimensions of demand adaptation, execution efficiency, and compliance and security, it constructs an execution evaluation system associated with demand attribute tags, obtains execution evaluation indicators, builds a three-dimensional association table and demand attribute execution profile, and stores it on the blockchain. Through feedback analysis of three dimensions—adaptability of diversion rules, rationality of weight parameters, and scientificity of threshold standards—it generates parameter optimization schemes, which are then verified by historical data simulation. The verified optimization parameters are synchronously updated to the procurement platform rule base and blockchain smart contract, forming a closed-loop iteration to continuously improve the accuracy and adaptability of diversion decisions.

[0142] Working principle and effects:

[0143] By analyzing the demand plans submitted by target users, extracting demand information, and calculating multi-dimensional attribute indicators, a demand attribute tag set is constructed to accurately identify the essence of special demands such as core materials and high-risk demands. This solves the pain point of traditional platforms being unable to distinguish between demand and regular demand. At the same time, an initial hash value is generated and stored on the blockchain to ensure data credibility and traceability. A diversion rule base determines whether diversion is triggered. If diversion is triggered, a recommendation list is generated based on the mapping relationship between the tag set and professional teams. After dual verification of attribute matching degree and real-time adaptability, as well as individual and overall screening, the appropriate supply team is identified. If diversion is not triggered, a flow instruction is directly generated and associated with the storage, breaking the homogeneous processing mode, allowing regular demands to flow efficiently and special demands to be matched with professional teams, reducing execution risks. All process data and execution indicators are associated with the blockchain to form a complete data chain, providing a basis for the iteration of diversion rules and parameters, continuously improving diversion accuracy, and achieving a comprehensive effect of reasonable resource allocation, effective risk control, and improved procurement efficiency.

[0144] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A data splitting method based on natural language processing, characterized in that, include: Obtain demand data, extract demand information, and construct a demand attribute tag set, which includes material tag level, supply tag level, technology tag level, and risk tag level. Integrate the demand information and the demand attribute tag set to generate an initial hash value for evidence storage. The steps for constructing the requirement attribute tag set include: Obtain the requirement plan submitted by the target user, perform compatibility verification, generate supplementary prompts, and re-verify; Natural language processing is used to extract requirement information from the requirement plan and calculate the confidence level of each field for verification. For image files, image recognition is used to parse the visualization information and generate a requirement list. Build a demand identifier for the demand plan, construct a candidate supplier set based on the material types in the demand list, and bind it to the demand list to generate a demand data package; The system calls the traffic splitting rule library to iterate and verify the set of demand attribute tags to determine whether to split traffic. If the set of demand attribute tags meets any condition in the traffic splitting rule library, it is determined that traffic splitting is triggered, the differentiated traffic splitting mechanism is started, and a recommended traffic splitting list is generated. Otherwise, the flow mechanism is started and a flow hash value is generated. The steps of the differentiated traffic diversion mechanism include: Call the smart contract in the blockchain and verify the validity of the smart contract; Traverse the candidate supplier set, filter candidate supplier teams according to the smart contract, and generate a basic matching team pool; Based on the attribute priority matching rule, obtain the tag priority of each tag level in the current demand plan; Based on the label coding rules, assign numerical codes to each label level in the current demand plan; The tag priority is integrated with the numerical code to generate an coded tag set; Configure priority weights for each tag priority, and calculate the matching strength value based on the encoded tag set; Obtain the acceptance threshold of each supply team in the basic adaptation team pool, perform positive matching, and remove supply teams whose acceptance threshold is less than the matching strength value. Read the historical supply data of each candidate supply team, perform reverse matching based on the highest historical tag priority, filter the candidate teams to be matched, and finally generate a pool of candidate teams to be matched. Acquire real-time status data, calculate the verification score of each supply team in the recommended distribution list, and perform individual and overall verification on the recommended distribution list to generate a distribution supply list; After the demand plan is completed, the execution data of the demand plan is collected and combined with the execution evaluation system of the executing department to obtain the execution evaluation indicators.

2. The data splitting method based on natural language processing according to claim 1, characterized in that, The steps for constructing a set of requirement attribute tags also include: Based on the business priority of the demand plan, configure business weights, set substitution coefficients based on the material substitution solutions uploaded by the target users, calculate support coefficients, and classify material label levels; The number of potential suppliers for the demand plan is obtained, and the intention level is obtained through level quantification. Based on the historical cooperation data of the target users, a stable weight is set, the market complexity coefficient is calculated, a secondary complexity division threshold is set, and the supply tag level is obtained. The number of technical parameters, the number of special processes, and the acceptance level are obtained, and a weighted calculation is performed to determine the technical difficulty coefficient and classify the technical label level. Obtain the target user's self-assessment risk value and self-assessment impact coefficient, combine them with the supply cycle, quantify the supply weight, calculate the risk coefficient, and classify the risk label level; Construct a set of requirement attribute tags and bind them to the requirement data package, perform a hash operation, and generate an initial hash value.

3. The data splitting method based on natural language processing according to claim 2, characterized in that, The steps of the differentiated traffic diversion mechanism also include: Based on the demand matching strength value and the capacity threshold, the matching ratio of each candidate team is calculated. Based on the historical database, the historical success rate of each candidate team in handling similar needs is obtained, and the adaptation score is calculated in combination with the adaptation ratio. Based on the fit score, the supply teams in the fit candidate team pool are sorted in descending order. If the fit scores are the same, they are sorted again by fit ratio in ascending order and historical successful execution rate in descending order to generate a list of fit teams. The list of compatible teams is effectively verified, and all non-compliant teams are removed. If the list of compatible teams is empty after screening, the requirement fallback mechanism is activated to generate a recommended distribution list. The recommended traffic distribution list is bound to the demand identifier, the initial hash value, and the entire filtering process data to generate a filtering data package, and a hash operation is performed to generate a matching hash value.

4. The data splitting method based on natural language processing according to claim 3, characterized in that, The steps of the circulation mechanism include: Retrieve a regular list pre-configured by the target user, and configure a stable priority for each supply team in the regular list. Based on the stable priority, perform targeted matching with a fixed supply team. Conduct demand compliance verification for the fixed supply team, including verification of their qualifications and historical compliance. If the verification fails, the replacement and verification will be carried out according to the stability priority in the regular list until a fixed supply team that passes the verification is found. If none of the teams in the regular list pass the verification, then the backup team for regular requirements will be activated directly. Construct flow instructions and perform integrity verification. Combine the initial hash value, demand identifier, and fixed supply team to generate a regular flow data packet and calculate the flow hash value.

5. The data splitting method based on natural language processing according to claim 4, characterized in that, The steps to generate a distribution supply list include: Obtain the recommended traffic distribution list and combine it with the matching hash value to construct the basic data packet; For each supply team in the recommended distribution list, real-time status data is obtained, and data standardization transformation is performed to generate a real-time status vector. Based on the intersection of the attribute tag levels of the demand planning and supply teams, obtain the number of intersections of attribute tag levels, and calculate the fit coefficient based on the total number of tags in the demand attribute tag set. The attribute matching degree is calculated by combining the corresponding team's fit score with the fit coefficient. Obtain load adaptation score, historical adaptation score, personnel adaptation score, and qualification adaptation score, and combine them with the urgency coefficient to calculate the real-time adaptation degree; Configure attribute weights and real-time weights, and calculate the verification score.

6. The data splitting method based on natural language processing according to claim 5, characterized in that, The steps for generating the distribution supply list also include: Based on the demand attribute tag set and urgency coefficient, set an individual verification threshold to perform individual verification; If the verification score is less than the individual verification threshold, the corresponding supply team is removed. After individual verification, the supply quantity of the supply team in the recommended distribution table is obtained. If the supply quantity is less than the distribution supply threshold, the substitute mechanism is activated and individual verification is performed on the activated substitute team. Calculate compliance score and diversity score, and use the analytic hierarchy process (AHP) to configure weights and perform a weighted summation to obtain the overall fit score. If the overall fit score is less than the supply qualification threshold, the corresponding supply team is removed. After overall verification, obtain the supply quantity of the supply team in the recommended distribution table, determine whether to activate the substitute mechanism, and generate the distribution supply list; Based on the verification score of the diversion supply list, the allocation share of a single team is calculated based on the weighted ratio allocation, a diversion instruction is generated, and then encrypted.

7. The data splitting method based on natural language processing according to claim 6, characterized in that, The steps to obtain performance evaluation metrics include: Based on demand adaptability, execution effectiveness, and compliance and security, an execution evaluation system associated with demand attribute tags is constructed; Acquire real-time execution data, standardize and quantify individual indicators, assign weights to individual indicators, calculate comprehensive execution scores, and thus generate real-time execution evaluation indicators. Construct a three-dimensional relationship table of team-execution-requirement, generate an execution profile of requirement attributes, and combine the requirement identifier and process evidence hash value to generate an execution data package and calculate the relationship hash value; Perform three-dimensional feedback analysis, generate an optimization solution package based on the analysis results, and then effectively verify it. The verified optimization solution package will be synchronously updated to the traffic distribution rule base and smart contract; Build an optimized data packet, calculate the optimized hash value using a cryptographic hash algorithm, and write it to the blockchain.

8. A data splitting system based on natural language processing, used to implement the data splitting method based on natural language processing as described in any one of claims 1-7, characterized in that, include: The module consists of an analysis module, a traffic splitting module, a decision-making module, and a feedback module. The parsing module is used to obtain demand plans, extract demand information, and screen a set of candidate suppliers with material qualifications. It also constructs a set of demand attribute tags based on support coefficients, market complexity coefficients, technical difficulty coefficients, and risk coefficients. The traffic splitting module is used to call the traffic splitting rule library and iterate through and verify the set of requirement attribute tags to determine whether to split the traffic. If the set of demand attribute tags meets any condition in the traffic splitting rule base, the differentiated traffic splitting mechanism is activated to generate a recommended traffic splitting list; If none of the required attribute tag sets meet the traffic splitting rule base, the traffic transfer mechanism is activated to generate a traffic transfer instruction; The decision-making module is used to acquire real-time status data, calculate the verification score through attribute matching degree and real-time adaptability, and perform individual and overall verification on the recommended diversion list to generate a diversion supply list. It also calculates the allocation share and generates diversion instructions by using weighted ratio allocation and single team share limit. The feedback module is used to obtain execution evaluation indicators, construct a three-dimensional association table and a requirement attribute execution profile and store them on the blockchain, perform three-dimensional feedback analysis, and generate parameter optimization schemes for updating.

Citation Information

Patent Citations

  • Procurement Planning Management System

    CN114936847A

  • Interconnection and intercommunication document editing system and method

    CN120930606A

  • Supply chain platform data management system based on intelligent Internet of Things

    CN121961433A