Tire system platform credit risk early warning method based on data sharing

Abstract

A tire system platform credit risk early warning method based on data sharing comprises the steps of constructing a data sharing platform, performing block chain evidence storage, multi-type feature extraction and scene type extraction on multi-dimensional data uploaded to the data sharing platform by a subject, and obtaining a feature set of the subject and a scene type of the feature set; performing dynamic index weight setting on each type of parameter in the feature set based on the scene type, and obtaining an early warning risk level of the subject according to the feature set of the subject and the dynamic index weight of each type of parameter in the feature set; and risk-credit-punishment linkage control is carried out based on the early warning risk level of the subject. According to the method, a tire industry data island is broken, the accuracy and timeliness of credit risk early warning are improved, and the transaction safety of a tire system platform and the stable operation of an industry chain are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tire system platform risk management, and particularly relates to a tire system platform credit risk early warning method based on data sharing. BACKGROUND

[0002] With the acceleration of the digital transformation of the tire industry, the tire system platform has gradually become the core carrier of the integration of industry chain resources. The credit status of platform users (mainly including tire dealers, repair shops, etc.) directly affects the safety of platform transactions, the efficiency of capital flow, and the stability of the industry chain.

[0003] The existing tire service platform generally has the following technical defects: 1. Data island problem is prominent: the service data of tire repair and replacement businesses (such as repair rate, accessory compliance), the environmental compliance data of recycling businesses (such as tire test results, flow records), and the feedback data of users are mutually isolated, which cannot form a unified risk assessment data source, resulting in one-sided risk identification.

[0004] 2. Early warning mechanism is static: most platforms only rely on historical credit scores for risk judgment, and do not dynamically update risk levels in combination with real-time data (such as sudden complaints, regulatory penalties), which has strong lagging in early warning and cannot cope with sudden risks such as "high-score business short-term violation".

[0005] 3. Insufficient risk and credit linkage: credit scores and risk early warning are disconnected, and businesses are only penalized after violating rules, without forming a closed loop of "risk triggering-early warning response-credit punishment-reform review", which allows high-risk businesses to continue taking orders through "score brushing" and harm user interests. SUMMARY

[0006] In order to solve the above technical problems, the purpose of the present application is to provide a tire system platform credit risk early warning method based on data sharing, comprising the following steps: Step s1: building a data sharing platform, performing blockchain notarization, multi-type feature extraction, and scene type extraction on the multi-dimensional data uploaded by the subject to the data sharing platform, obtaining the feature set of the subject and the scene type of the feature set; Step s2: dynamically setting the index weight of each type of parameter in the feature set based on the scene type, and obtaining the early warning risk level of the subject according to the feature set of the subject and the dynamic index weight of each type of parameter in the feature set; Step s3: risk-credit-punishment linkage control based on the early warning risk level of the subject.

[0007] Furthermore, a data sharing platform is built based on blockchain technology. The data sharing platform includes several blockchain nodes, which are interconnected to form a blockchain network. Each entity that logs into the data sharing platform communicates with a blockchain node, which is used to store the multidimensional data uploaded by the entity on the blockchain.

[0008] Furthermore, the main entities include three roles: users, service providers, and recycling companies.

[0009] Furthermore, the process of multi-type feature extraction includes: Perform outlier removal, missing value completion, and data standardization preprocessing on the multidimensional data uploaded to the blockchain node; Structured features are extracted from the preprocessed multidimensional data to obtain structured features. At the same time, the BERT model is used to perform semantic analysis on the multidimensional data to extract content keywords. Unstructured features are extracted from the content keywords to obtain unstructured features. Construct a time window, extract the historical structured features and historical unstructured features of blockchain nodes within the time window, and perform time series feature analysis on the historical structured features, historical unstructured features, and current structured features and unstructured features to obtain time series features.

[0010] Furthermore, the process of scene type extraction includes: The structured features, unstructured features, time series features, and correlation features extracted from the blockchain nodes are packaged into feature sets and stored in the blockchain nodes; It extracts the subject type, service scenario, and region of the feature set stored by the blockchain node from the multidimensional data uploaded to the blockchain node, and generates the scenario type of the feature set based on the subject type, service scenario, and region.

[0011] Furthermore, the process of obtaining the subject's early warning risk level includes: The various types of features stored in the feature set in the blockchain node are used as evaluation indicators. The dynamic indicator weights corresponding to the evaluation indicators are set according to the scenario type to which the feature set belongs. Fuzzy adaptive rules for evaluation indicators and preset warning risk levels are set for different scenario types. Based on the fuzzy adaptive rules, the membership matrix of the evaluation indicators for different warning risk levels is obtained through fuzzy comprehensive evaluation. The warning risk level corresponding to the feature set is obtained based on the membership matrix and dynamic indicator weights, and the warning risk level of the subject of the blockchain node communication link is updated based on the warning risk level corresponding to the feature set.

[0012] Furthermore, the process of setting dynamic indicator weights corresponding to evaluation indicators based on the scene type to which the feature set belongs includes: Step 1: Based on the scene type to which the historical feature sets stored by each blockchain node in the blockchain network belong, cluster all historical feature sets by scene type to generate several cluster centers. Each cluster center contains several historical feature sets belonging to the same scene type. Step 2: Remove the historical feature sets corresponding to low and medium risk levels from the cluster centers. Then, randomly select a certain type of feature from the historical feature sets included in the cluster centers. Perform high-risk association analysis on the certain type of feature in each historical feature set included in the cluster centers to obtain a significant difference threshold. Based on the significant difference threshold, perform statistical analysis on the certain type of feature in each historical feature set included in the cluster centers to obtain the probability that a certain type of feature exceeds the significant difference threshold. Repeat the above process to obtain the probability that all types of features exceed the significant difference threshold. Obtain the scene type corresponding to the cluster center. Set the dynamic indicator weights of all types of features under the scene type condition according to the probability that all types of features exceed the significant difference threshold. Step 3: Repeat Step 2 until the dynamic index weights of each type of feature under all scenario conditions are obtained, and build a weight database to store the dynamic index weights of each type of feature under all scenario conditions in the weight database. Step 4: Search and match in the weight database according to the scene type to which the feature set belongs, and obtain the dynamic indicator weights corresponding to the evaluation indicators.

[0013] Furthermore, the process of performing high-risk association analysis on a certain type of feature within each historical feature set included in the cluster center includes: The specific values ​​corresponding to a certain type of feature in each historical feature set included in the cluster center are labeled as association levels. The total number of association levels of the cluster center and the number of historical feature sets whose specific values ​​corresponding to a certain type of feature are consistent with the specific values ​​corresponding to the association levels are counted. An association histogram is constructed, and the association histogram is normalized to generate a normalized histogram. An association level proportion analysis is performed on the normalized histogram to obtain the association proportion function. Based on the normalized histogram and the association proportion function, a significance difference analysis is performed to obtain the significance difference threshold for a certain type of feature.

[0014] Furthermore, the process of implementing risk-credit-punishment linkage control includes: The system presets service score loss values ​​and warning score loss values ​​corresponding to different warning risk levels. The credit value is composed of service score and warning score. When the warning risk level of an entity is updated, the entity's credit value is updated according to the service score loss value and warning score loss value corresponding to the updated warning risk level.

[0015] Furthermore, the process of linking risk, credit, and punishment control based on the subject's early warning risk level also includes: Set a credit value range, select a threshold point within the credit value range to divide different sub-ranges of rights binding, and determine the bound rights of the subject based on the sub-range of rights binding in which the subject's credit value is located.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. Traditional tire platforms commonly suffer from data silos, data fraud, and data privacy leaks, resulting in a lack of effective basis for risk assessment. This invention constructs a data sharing platform using blockchain technology, storing multi-dimensional data from users, service providers, and recycling companies on the blockchain for evidence, while simultaneously extracting multiple types of features and scenario types, fundamentally solving the core pain points at the data level: Preventing data tampering and ensuring data authenticity: The distributed storage and immutability of blockchain nodes ensure that the multi-dimensional data uploaded by the subject (such as the repair records of service providers and the test reports of recycling providers) cannot be modified, avoiding problems such as false qualifications and forged test results, making the data source of risk assessment controllable from the source, and no longer relying on the subjective declaration of a single subject; Breaking down data silos and achieving full-dimensional data coverage: By extracting multiple types of features (structured + unstructured + time series features), it integrates unstructured data (such as negative keywords in user review texts) and time-dimensional data (such as the dynamic changes in complaints in the past 7 days) that are ignored by traditional platforms. At the same time, it associates scenario information such as "subject type + service scenario + region", so that the data is no longer fragmented and forms a full-link, multi-dimensional feature set, providing complete data support for subsequent accurate evaluation. Balancing data sharing and privacy protection: Blockchain node access control and data preprocessing (outlier removal and missing value completion) enable cross-entity data sharing (such as regulatory authorities being able to trace the flow of recycled tires) while avoiding the leakage of user privacy (such as phone numbers and location) and merchant trade secrets (such as average order value). This balances data usability and privacy security, eliminating concerns among various entities about participating in data sharing.

[0017] 2. Enable scenario-based dynamic assessment to ensure risk warnings align with actual business operations. Traditional tire platforms often employ a static evaluation model with uniform parameter weights and fixed warning rules, ignoring differences in business type (tire repair / recycling), service scenario (emergency / routine), and region (first-tier / county level). This leads to misjudging compliant businesses or overlooking high-risk businesses. This invention dynamically sets indicator weights for feature parameters based on scenario type and combines this with fuzzy adaptive rules for comprehensive evaluation, achieving scenario-based and refined risk assessment. To avoid biased assessments and improve the accuracy of early warnings: Feature sets are categorized into specific scenarios (such as "highway emergency tire repair + first-tier cities" and "county-level door-to-door recycling") by extracting scenario types. Dynamic indicator weights are then set for different scenarios (such as "order response time" having a higher weight in the highway emergency scenario and "service completion rate" having a higher weight in the county-level recycling scenario). At the same time, scenario-specific fuzzy adaptive rules are matched to ensure that risk assessments are not divorced from actual business operations. For example, the response time standard for county-level tire repair will not be used to assess highway emergency tire repair businesses, nor will the recycling rules for key environmental protection areas be used to constrain ordinary county-level recycling businesses. This effectively reduces the situation where compliant businesses are misjudged and high-risk businesses are missed. Capturing dynamic changes in risks and achieving proactive early warning: This invention extracts historical and current features by constructing a time window, performs time series feature analysis (such as fluctuations in the complaint growth rate over the past 7 days), and then combines fuzzy comprehensive evaluation to obtain a membership matrix. This allows for the keen capture of gradual trends in risks. For example, although a recycling merchant currently meets the testing pass rate standards, the pass rate has been declining over the past 7 days. The time series features can reflect this dynamic in a timely manner, and risk warnings can be triggered in advance through fuzzy rules, avoiding passive responses only when the pass rate falls below the threshold. This shifts the focus of early warning from post-event remediation to pre-event prevention. To improve the interpretability of assessments and enhance the acceptance of early warning results by various stakeholders: The design of fuzzy adaptive rules ensures that the determination of risk levels is no longer a black box output—each early warning risk level can be matched with the parameter performance in a specific scenario (e.g., a highway emergency merchant is judged to be high-risk due to a response time of more than 30 minutes and a high complaint growth rate in the past 7 days). At the same time, the setting of dynamic indicator weights can be traced back to the scenario type and historical feature analysis, avoiding the problem of inexplicable early warning results.

[0018] 3. Construct a closed loop of risk, credit, and punishment to make control more effective and guiding. Traditional tire platform risk warnings often remain at the level of simply alerting users to risks, lacking linkage with credit and penalties. This results in high-risk merchants still being able to accept orders, while compliant merchants receive no additional incentives, failing to create effective constraints and guidance. This invention, through a risk-credit-penalty linkage control system, achieves a strong binding between warnings, credit, and rights, building a healthy platform ecosystem. This invention creates rigid constraints, forcing high-risk merchants to comply with regulations: It directly links the warning risk level to the credit score (with preset service scores and warning score depletion values ​​corresponding to different warning levels). A decrease in the credit score will directly affect the rights and interests of the merchant (e.g., if the credit score is lower than the threshold, the merchant will be removed from the platform or have their order acceptance restricted). For example, if a service merchant's credit score drops significantly due to a high-risk warning, they will not only lose the priority order allocation qualification as a preferred merchant, but may also have their service privileges suspended. This forces high-risk merchants to rectify and comply with regulations from a profit perspective, avoiding the embarrassment of ineffective warnings. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a data-sharing-based tire system platform credit risk early warning method according to an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] like Figure 1 As shown, a credit risk early warning method for a tire system platform based on data sharing includes the following steps: Step s1: Build a data sharing platform, perform blockchain notarization, multi-type feature extraction, and scene type extraction on the multi-dimensional data uploaded by the subject to the data sharing platform, and obtain the subject's feature set and the scene type of the feature set; Step s2: Based on the scenario type, dynamically set the indicator weights for each type of parameter in the feature set, and obtain the warning risk level of the subject according to the subject's feature set and the dynamic indicator weights of each type of parameter in the feature set; Step s3: Implement risk-credit-punishment linkage control based on the subject's early warning risk level.

[0022] It should be further explained that, in the specific implementation process, a data sharing platform is built based on blockchain technology. The data sharing platform includes several blockchain nodes, which are interconnected to form a blockchain network. Each entity that logs into the data sharing platform communicates with a blockchain node, which is used to store the multi-dimensional data uploaded by the entity on the blockchain.

[0023] The process of blockchain-based evidence storage includes: A hash function is applied to the multidimensional data to generate a SHA-256 hash value; the SHA-256 hash value and the node signature (the private key signature of the detection point) are packaged into a transaction and broadcast to the consortium blockchain; each blockchain node in the blockchain network verifies the transaction, and after successful verification, it is written into a block (block generation time ≤ 10 seconds), and the block is added to the end of the blockchain.

[0024] It should be further explained that in the specific implementation process, the main parties include three roles: users, service providers, and recycling companies.

[0025] It should be further explained that, in the specific implementation process, the multi-type feature extraction process includes: Outlier removal is performed on the multidimensional data uploaded to the blockchain nodes: extreme data (such as "0-point reviews" due to user errors or abnormally high average order values ​​from merchants) is removed using the "3σ principle," and outliers are identified and deleted using the "box plot method." Missing value completion is performed: historical service data of newly joined merchants is filled using the K-Nearest Neighbors (KNN) algorithm; periodically missing data (such as order fluctuations during holidays) is completed using linear interpolation. Data standardization preprocessing is performed: indicators of different magnitudes (such as order volume and number of complaints) are converted into standardized values ​​in the range [0,1], with the formula: standardized value = (original value - minimum value) / (maximum value - minimum value). Structured features are extracted from the preprocessed multidimensional data to obtain structured features. At the same time, the BERT model is used to perform semantic analysis on the multidimensional data (user review text, maintenance log) to extract content keywords. Unstructured features are extracted from the content keywords to obtain unstructured features, such as extracting negative keywords such as "inferior tires", "arbitrary charges" and "illegal cutting". The "frequency ratio of negative keywords" (number of negative keywords / total number of keywords) is calculated. Construct a time window (7 days), extract the historical structured features and historical unstructured features of blockchain nodes within the time window, perform time series feature analysis on the historical structured features, historical unstructured features, and current structured features and unstructured features to obtain time series features; The structured and unstructured features, as well as the time series features, of the multidimensional data uploaded by users, service providers, and recycling businesses are shown in Table 1 below: Table 1 It should be further explained that, in the specific implementation process, the scene type extraction process includes: The structured features, unstructured features, time series features, and correlation features extracted from the blockchain nodes are packaged into feature sets and stored in the blockchain nodes; It extracts the subject type, service scenario, and region of the feature set stored by the blockchain node from the multidimensional data uploaded to the blockchain node, and generates the scenario type of the feature set based on the subject type, service scenario, and region.

[0026] The scenario types consist of subject type (user, service provider, and recycling provider) + service scenario + region, such as tire repair and replacement + highway emergency + first-tier cities, tire repair and replacement + urban daily life + county area, recycling provider + network recycling + ordinary third- and fourth-tier areas, and recycling provider + door-to-door recycling + key environmental protection areas.

[0027] It should be further explained that, in the specific implementation process, the process of obtaining the subject's early warning risk level includes: The various types of features stored in the feature set in the blockchain node are used as evaluation indicators. The dynamic indicator weights corresponding to the evaluation indicators are set according to the scenario type to which the feature set belongs. The evaluation indicators and preset warning risk levels corresponding to different scenario types are set as fuzzy adaptive rules (determined based on expert experience, with the aim of reducing uncertainty in the fuzzy comprehensive evaluation process). Based on the fuzzy adaptive rules, the membership matrix of the evaluation indicators for different warning risk levels is obtained through fuzzy comprehensive evaluation. The warning risk level corresponding to the feature set is obtained based on the membership matrix and dynamic indicator weights, and the warning risk level of the subject of the blockchain node communication link is updated based on the warning risk level corresponding to the feature set.

[0028] Among them, for setting evaluation indicators and preset early warning risk levels corresponding to different scenario types, taking the scenario types of service providers + highway emergency + first-tier cities as an example: The four parameters, "order response time, service completion rate, refund frequency, and complaint growth rate in the past 7 days," are defined as fuzzy sets, as shown in Table 2 below: Table 2 The "adaptive" nature of the rules is reflected in the fact that: evaluation indicators with higher dynamic indicator weights are given priority. The higher the weight of the dynamic indicator, the higher the priority. For example, in high-speed scenarios, the combination of "response time + complaint growth rate" is given priority. Rule 1 (High-risk trigger, priority 1): If "order response time = long" (membership ≥ 0.7) and "complaint growth rate in the past 7 days = high" (membership ≥ 0.7), then the warning risk level = high risk (level 1 warning).

[0029] Example: If a merchant's response time is 35 minutes ("long" set membership degree 0.8) and the complaint growth rate is 18% ("high" set membership degree 0.9), it meets Rule 1 and is directly judged as high risk.

[0030] Rule 2 (High-risk trigger, priority 2): If "order response time = medium" (membership ≥ 0.6), "service completion rate = low" (membership ≥ 0.7), and "refund frequency = high" (membership ≥ 0.6), then the warning risk level is high.

[0031] Example: Response time 25 minutes ("Medium" membership degree 0.7), completion rate 88% ("Low" membership degree 0.8), refunds 6 times / week ("High" membership degree 0.7), meets rule 2, and is judged as high risk.

[0032] Rule 3 (Medium Risk Trigger, Priority 3): If "Order Response Time = Medium" (Membership Degree ≥ 0.5) or "Complaint Growth Rate in the Past 7 Days = Medium" (Membership Degree ≥ 0.6) and "Service Completion Rate = Medium" (Membership Degree ≥ 0.5), then the warning risk level is medium risk (Level 2 warning).

[0033] Example: Response time 28 minutes (Medium membership degree 0.9), complaint growth rate 10% (Medium membership degree 0.7), completion rate 92% (Medium membership degree 0.6), meets rule 3, and is judged as medium risk.

[0034] Rule 4 (Low Risk Trigger, Priority 4): If "Order Response Time = Short" (Membership ≥ 0.8), "Service Completion Rate = High" (Membership ≥ 0.8), "Refund Frequency = Low" (Membership ≥ 0.8), and "Complaint Growth Rate = Low" (Membership ≥ 0.8), then the warning risk level is low (Level 3 warning).

[0035] Example: Response time 15 minutes ("Short" membership level 1.0), completion rate 98% ("High" membership level 1.0), refund once per week ("Low" membership level 1.0), complaint growth rate 3% ("Low" membership level 1.0), meets rule 4, and is judged as low risk.

[0036] It should be further explained that, in the specific implementation process, the process of obtaining the early warning risk level of the feature set based on the membership matrix and dynamic indicator weights includes: The dynamic indicator weights and membership matrix of the evaluation indicators are fused by formula to obtain the fuzzy comprehensive evaluation matrix of the evaluation indicators. The membership degree of the feature set for different warning risk levels is obtained according to the fuzzy comprehensive evaluation matrix. The warning risk level with the highest membership degree of the feature set is selected and the warning risk level with the highest membership degree of the feature set is taken as the warning risk level of the feature set. The formula is as follows: ; in, The fuzzy comprehensive evaluation matrix for the evaluation indicators. To evaluate the indicator weights, For the membership matrix, "" indicates that the elements at corresponding positions in the weight matrix and membership matrix of the evaluation index are multiplied together. The weighting parameter is used to control the balance between the weight matrix and the membership matrix in the fuzzy comprehensive evaluation matrix of the evaluation index.

[0037] It should be further explained that, in the specific implementation process, the process of setting the dynamic indicator weights corresponding to the evaluation indicators based on the scenario type to which the feature set belongs includes: Step 1: Based on the scene type to which the historical feature sets stored by each blockchain node in the blockchain network belong, cluster all historical feature sets by scene type to generate several cluster centers. Each cluster center contains several historical feature sets belonging to the same scene type. Step Two: Remove historical feature sets with low and medium risk levels from the cluster centers. Then, randomly select a certain type of feature from the historical feature sets included in the cluster centers. Perform high-risk association analysis on the certain type of feature in each historical feature set included in the cluster centers to obtain a significant difference threshold. Based on the significant difference threshold, perform statistical analysis on the certain type of feature in each historical feature set included in the cluster centers to obtain the probability that the specific value corresponding to a certain type of feature exceeds the significant difference threshold (count the total number of historical feature sets in the cluster centers and the number of historical feature sets whose specific values ​​corresponding to a certain type of feature exceed the significant difference threshold, then divide the number of historical feature sets by the threshold). The total number of historical feature sets is used to generate the probability that the specific value corresponding to a certain type of feature exceeds the significance difference threshold. The above process is repeated to obtain the probability that all types of features exceed the significance difference threshold, obtain the scene type corresponding to the cluster center, and set the dynamic indicator weight of all types of features under the scene type condition based on the probability that all types of features exceed the significance difference threshold (the higher the probability, the greater the dynamic indicator weight. For example, when the significance difference threshold of the proportion of negative keywords is 45%, when the proportion of negative keywords exceeds 45%, the proportion of merchants' warning risk level as high risk increases from 5% to 30%, then the probability that the proportion of negative keywords exceeds 45% is greater, and the dynamic indicator weight of the proportion of negative keywords is greater). Step 3: Repeat Step 2 until the dynamic index weights of each type of feature under all scenario conditions are obtained, and build a weight database to store the dynamic index weights of each type of feature under all scenario conditions in the weight database. Step 4: Search and match in the weight database according to the scene type to which the feature set belongs, and obtain the dynamic indicator weights corresponding to the evaluation indicators.

[0038] It should be further explained that, in the specific implementation process, the process of performing high-risk association analysis on a certain type of feature in each historical feature set included in the cluster center includes: The specific values ​​corresponding to a certain type of feature in each historical feature set included in the cluster center are labeled as association levels. The total number of association levels of the cluster center and the number of historical feature sets whose specific values ​​corresponding to a certain type of feature are consistent with the specific values ​​corresponding to the association levels are counted. An association histogram is constructed, and the association histogram is normalized to generate a normalized histogram. An association level proportion analysis is performed on the normalized histogram to obtain the association proportion function. Based on the normalized histogram and the association proportion function, a significance difference analysis is performed to obtain the significance difference threshold for a certain type of feature.

[0039] For example, if the total number of association levels for cluster centers is 100, the number of feature sets h for each association level (0~100) in the image is counted to obtain the association histogram. ; Normalize the correlation histogram to obtain a normalized histogram. ;in, The total set of historical features included in the cluster centers. The number of feature sets at the k-th association level; Association level proportion analysis is performed based on normalized histograms to obtain the association proportion function: the proportion of historical feature sets in the first t association levels, i.e. , ;in This represents the percentage of the cumulative historical feature sets for the first t association levels. This represents the percentage of the cumulative historical feature sets for each association level after the t-th association level. Significant difference analysis was performed based on normalized histograms and correlation proportion functions: ; (like (If the significance difference analysis of the current t is skipped); ( (If the significance difference analysis of the current t is skipped); in, The coefficient of significance is expressed as a difference coefficient. Traverse all possible Calculate the corresponding ,Pick The t value corresponding to the maximum value is used as the significance threshold.

[0040] It should be further explained that, in the specific implementation process, the risk-credit-punishment linkage control process includes: The system presets service score loss values ​​and warning score loss values ​​corresponding to different warning risk levels (low risk corresponds to a service score loss value of -4 and a warning score loss value of 0, medium risk corresponds to a service score loss value of 0 and a warning score loss value of 0, and high risk corresponds to a service score loss value of 4 and a warning score loss value of 1). The credit value consists of a service score (80 points) and a warning score (20 points). When the entity's warning risk level is updated, the entity's credit value is updated according to the service score loss value and warning score loss value corresponding to the updated warning risk level (service score minus service score loss value + warning score minus warning score loss value).

[0041] Most platforms have separate "risk warnings" and "credit scores"—merchants are only penalized with a single point deduction after being warned, without any follow-up rectification or rights linkage. This allows high-risk merchants to restore their credit through illegal means (such as buying positive reviews) and continue to accept orders.

[0042] This method splits the merchant credit score into "service score (80) + warning score (20)". The warning deduction only applies to the "warning score", and the warning score cannot be made up by the service score / warning score after it is cleared, thus avoiding violations.

[0043] It should be further explained that, in the specific implementation process, the process of linking risk, credit, and punishment based on the subject's early warning risk level also includes: Set credit range Within the credit score range, threshold points are selected to divide different sub-ranges for binding different rights. The bound rights of the main entity are determined based on the sub-range of rights binding in which the main entity's credit score is located. For example, credit score is bound to merchant rights: credit score <60 points will be automatically removed from the platform, credit score 60-70 points will be restricted from accepting orders (no emergency order permissions), and credit score ≥90 points will be listed as "preferred merchants" (priority order assignment + 10% service fee reduction).

[0044] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for early warning of credit risk in a tire system platform based on data sharing, characterized in that, Includes the following steps: Step s1: Build a data sharing platform, perform blockchain notarization, multi-type feature extraction, and scene type extraction on the multi-dimensional data uploaded by the subject to the data sharing platform, and obtain the subject's feature set and the scene type of the feature set; Step s2: Based on the scenario type, dynamically set the indicator weights for each type of parameter in the feature set, and obtain the warning risk level of the subject according to the subject's feature set and the dynamic indicator weights of each type of parameter in the feature set; Step s3: Implement risk-credit-punishment linkage control based on the subject's early warning risk level.

2. The method for credit risk early warning of a tire system platform based on data sharing according to claim 1, characterized in that, A data sharing platform is built based on blockchain technology. The data sharing platform includes several blockchain nodes, which are interconnected to form a blockchain network. Each entity that logs into the data sharing platform communicates with a blockchain node, which is used to store the multi-dimensional data uploaded by the entity on the blockchain.

3. The method for early warning of credit risk in a tire system platform based on data sharing according to claim 2, characterized in that, The main entities include three roles: users, service providers, and recycling companies.

4. The method for credit risk early warning of a tire system platform based on data sharing according to claim 3, characterized in that, The process of multi-type feature extraction includes: Perform outlier removal, missing value completion, and data standardization preprocessing on the multidimensional data uploaded to the blockchain node; Structured features are extracted from the preprocessed multidimensional data to obtain structured features. At the same time, the BERT model is used to perform semantic analysis on the multidimensional data to extract content keywords. Unstructured features are extracted from the content keywords to obtain unstructured features. Construct a time window, extract the historical structured features and historical unstructured features of blockchain nodes within the time window, and perform time series feature analysis on the historical structured features, historical unstructured features, and current structured features and unstructured features to obtain time series features.

5. The method for credit risk early warning of a tire system platform based on data sharing according to claim 4, characterized in that, The process of scene type extraction includes: The structured features, unstructured features, time series features, and correlation features extracted from the blockchain nodes are packaged into feature sets and stored in the blockchain nodes; It extracts the subject type, service scenario, and region of the feature set stored by the blockchain node from the multidimensional data uploaded to the blockchain node, and generates the scenario type of the feature set based on the subject type, service scenario, and region.

6. The method for credit risk early warning of a tire system platform based on data sharing according to claim 5, characterized in that, The process of obtaining the risk level of a subject includes: The various types of features stored in the feature set in the blockchain node are used as evaluation indicators. The dynamic indicator weights corresponding to the evaluation indicators are set according to the scenario type to which the feature set belongs. Fuzzy adaptive rules for evaluation indicators and preset warning risk levels are set for different scenario types. Based on the fuzzy adaptive rules, the membership matrix of the evaluation indicators for different warning risk levels is obtained through fuzzy comprehensive evaluation. The warning risk level corresponding to the feature set is obtained based on the membership matrix and dynamic indicator weights, and the warning risk level of the subject of the blockchain node communication link is updated based on the warning risk level corresponding to the feature set.

7. The method for credit risk early warning of a tire system platform based on data sharing according to claim 6, characterized in that, The process of setting dynamic indicator weights corresponding to evaluation indicators based on the scene type to which the feature set belongs includes: Step 1: Based on the scene type to which the historical feature sets stored by each blockchain node in the blockchain network belong, cluster all historical feature sets by scene type to generate several cluster centers. Each cluster center contains several historical feature sets belonging to the same scene type. Step 2: Remove the historical feature sets corresponding to low and medium risk levels from the cluster centers. Then, randomly select a certain type of feature from the historical feature sets included in the cluster centers. Perform high-risk association analysis on the certain type of feature in each historical feature set included in the cluster centers to obtain a significant difference threshold. Based on the significant difference threshold, perform statistical analysis on the certain type of feature in each historical feature set included in the cluster centers to obtain the probability that a certain type of feature exceeds the significant difference threshold. Repeat the above process to obtain the probability that all types of features exceed the significant difference threshold. Obtain the scene type corresponding to the cluster center. Set the dynamic indicator weights of all types of features under the scene type condition according to the probability that all types of features exceed the significant difference threshold. Step 3: Repeat Step 2 until the dynamic index weights of each type of feature under all scenario conditions are obtained, and build a weight database to store the dynamic index weights of each type of feature under all scenario conditions in the weight database. Step 4: Search and match in the weight database according to the scene type to which the feature set belongs, and obtain the dynamic indicator weights corresponding to the evaluation indicators.

8. The method for credit risk early warning of a tire system platform based on data sharing according to claim 7, characterized in that, The process of performing high-risk association analysis on a certain type of feature in each historical feature set included in the cluster centers includes: The specific values ​​corresponding to a certain type of feature in each historical feature set included in the cluster center are labeled as association levels. The total number of association levels of the cluster center and the number of historical feature sets whose specific values ​​corresponding to a certain type of feature are consistent with the specific values ​​corresponding to the association levels are counted. An association histogram is constructed, and the association histogram is normalized to generate a normalized histogram. An association level proportion analysis is performed on the normalized histogram to obtain the association proportion function. Based on the normalized histogram and the association proportion function, a significance difference analysis is performed to obtain the significance difference threshold for a certain type of feature.

9. A method for early warning of credit risk in a tire system platform based on data sharing, as described in claim 8, is characterized in that... The process of implementing risk-credit-punishment linkage control includes: The system presets service score loss values ​​and warning score loss values ​​corresponding to different warning risk levels. The credit value is composed of service score and warning score. When the warning risk level of an entity is updated, the entity's credit value is updated according to the service score loss value and warning score loss value corresponding to the updated warning risk level.

10. A method for early warning of credit risk in a tire system platform based on data sharing, as described in claim 9, is characterized in that... The process of risk-credit-punishment linkage control based on the subject's early warning risk level also includes: Set a credit value range, select a threshold point within the credit value range to divide different sub-ranges of rights binding, and determine the bound rights of the subject based on the sub-range of rights binding in which the subject's credit value is located.

Images