A method and system for digital management and precise service of civil relief business

Abstract

The application discloses a kind of civil affairs relief business digitization management and accurate service method and system, it is related to civil affairs relief digitization management service technical field, first, the related data of relief of multiple departments such as civil affairs, public security, medical insurance, social security, housing construction, emergency management is gathered, and after standardization cleaning, difference privacy desensitization, the full-dimension portrait library of relief object is constructed;After receiving relief application, first, call rule engine to complete policy compliance preliminary check, then call pre-trained AI implicit difficulty identification model to complete recheck, after passing double-layer verification, match global relief resource margin to generate individualized relief service list, carry out whole cycle state monitoring to the object in protection, and automatically adjust relief grade according to difficulty change.The application does not need the crowd to run and issue certificate in multiple departments, greatly reduces the threshold of relief application, double-layer verification mechanism effectively reduces the problem of wrong protection and missing protection, active discovery mechanism covers special groups with weak information acquisition ability, and the adaptability of relief resources is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of digital management and service technology for civil affairs assistance, and in particular to a method and system for digital management and precise service of civil affairs assistance business. Background Technology

[0002] Civil affairs assistance is a core component of basic livelihood security, covering various scenarios such as identification of people in extreme poverty and low social status, medical assistance, temporary assistance, and special assistance. In recent years, my country has provided assistance to more than 40 million people annually, with annual assistance funds exceeding 200 billion yuan. With the advancement of digital transformation of government affairs, improving the efficiency of assistance processing and ensuring the accuracy of assistance through information technology has become a core development need of the industry.

[0003] The current mainstream civil affairs assistance information system is a single-department offline approval online system. Its working principle is that after the recipient submits paper materials offline, the staff enters the information into the system and completes online approval and fund disbursement according to the hierarchical structure. This type of system achieves electronic record-keeping of the assistance process, significantly reducing the management cost of paper materials, and has been deployed in more than 80% of district and county civil affairs departments nationwide. However, this type of system has not established cross-departmental data interfaces with civil affairs, public security, medical insurance, and social security departments. Assistance eligibility verification relies entirely on the materials submitted by the applicant and manual verification. This not only requires people to visit multiple departments to obtain certificates, but also fails to identify difficulties not reflected in the application materials, such as excessive medical expenses or hidden unemployment. The incidence of missed or incorrect coverage is high, and fund disbursement records are only stored on the civil affairs department's internal server, posing a risk of data tampering and fraud. Auditing and verification require cross-departmental material retrieval, resulting in extremely low efficiency.

[0004] AI-powered assistance identification systems, which have been gradually promoted in recent years, work by using historical assistance data from within the civil affairs department to train identification models, thus assisting staff in verifying eligibility. While these systems have improved verification efficiency to some extent, they rely solely on a single data source within the civil affairs department, resulting in insufficient feature dimensions, an accuracy rate of less than 70%, and a lack of coverage for assistance resource matching, dynamic monitoring, and end-to-end evidence preservation. This makes it impossible to achieve precise matching of supply and demand, and it also fails to address issues such as untimely dynamic adjustments to assistance eligibility and unreliable data preservation. Furthermore, these systems only support a passive assistance model where the public actively submits applications, leaving many vulnerable groups with limited access to information, such as elderly people living alone, disaster victims, and people with disabilities, unable to promptly learn about policies and receive assistance, creating blind spots in the coverage of basic social security.

[0005] Current technologies cannot simultaneously meet the requirements of cross-departmental data interoperability, accurate verification of rescue eligibility, intelligent resource matching, dynamic monitoring and adjustment, and end-to-end availability. Summary of the Invention

[0006] This invention proposes a digital management and precision service method and system for civil affairs assistance, in order to solve the problems mentioned in the prior art, such as low efficiency of cross-departmental collaboration in civil affairs assistance, insufficient accuracy of eligibility verification, blind spots in coverage, and difficulty in risk management.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for digital management and precise service of civil affairs assistance business, comprising the following steps: Connecting to the national integrated government data sharing and exchange platform, it collects multi-source, heterogeneous relief-related data from multiple departments, including civil affairs, public security, medical insurance, social security, housing and urban-rural development, disabled persons' federation, education, and emergency management. Among them, civil affairs data includes historical identification of low-income and extremely poor individuals and records of temporary assistance disbursement; public security data includes household registration, death registration, migrant population, and records of illegal and dishonest behavior; medical insurance data includes outpatient and inpatient reimbursement, identification of chronic and special diseases, and records of large medical expenditures; social security data includes records of insurance contributions, unemployment benefits, and pension disbursements; housing and urban-rural development data includes records of public rental housing rentals, dilapidated housing renovation, and property ownership; disabled persons' federation data includes records of disability level assessment, assistive device disbursement, and disability subsidies; education data includes school enrollment information and records of student aid subsidies; and emergency management data includes disaster victim registration and disaster relief records. All collected data strictly followed the "Specifications for Sharing and Exchanging Civil Affairs Business Data" to complete standardized cleaning, unified field naming, coding, and data types, uniformly mapped the unique identifiers of personnel from different departments to resident ID card numbers, used the 3σ criterion to remove outliers, used the mean of the same dimension and population or related data of the same family to complete missing value imputation, and after desensitizing sensitive information, a full-dimensional profile database of assistance recipients was constructed according to four dimensions: basic attributes, hardship attributes, demand attributes, and historical assistance attributes. A two-tiered eligibility verification process is implemented for applicants for assistance. The first tier is a hard indicator verification based on policy rules. A customizable rule engine is configured to adapt to the differentiated assistance policies of different regions. After the profile data of the applicants is input into the rule engine, the initial verification results are automatically output. Applications that do not meet the hard indicators are directly rejected and a standardized notification document is sent. Applications that meet the hard indicators proceed to the second tier of verification. The second layer is AI-based hidden difficulty identification and risk verification. It uses a pre-trained machine learning model, inputs the full-dimensional profile features of the aid recipient, and outputs the aid eligibility credibility value and risk level. Those whose credibility value reaches the set threshold pass the verification directly, while those whose credibility value does not reach the threshold are transferred to the manual review process. Those whose risk level is judged to be suspected of insurance fraud are automatically included in the cross-departmental shared risk warning list. Those who pass the verification enter the aid matching stage. Based on the demand tags automatically extracted from the profiles of those receiving assistance and the remaining information of available assistance resources across the entire region, a two-way intelligent matching is performed. On the demand side, it covers a variety of needs, including medical assistance, temporary assistance, employment assistance, age-friendly renovations, educational assistance, and home-based elderly care. On the supply side, it covers the remaining funds, service supply capacity, and service coverage of various assistance projects. During the matching process, priority is given to people in extreme poverty, those receiving minimum living allowances, those on the edge of minimum living allowances, and those facing financial difficulties due to expenses. Assistance resources corresponding to rigid needs are matched first, generating a personalized assistance service list that is simultaneously pushed to the service terminals of those receiving assistance and the work terminals of local business operators, automatically triggering the assistance processing procedure. Data shared by multiple departments is updated synchronously according to the preset monitoring cycle corresponding to different assistance projects. Among them, long-term assistance projects such as minimum living allowance and special hardship assistance are monitored quarterly, temporary assistance projects are monitored monthly, and special assistance projects such as medical assistance and education assistance are monitored according to the project execution cycle. When the monitoring expires, the re-verification of assistance eligibility and dynamic adjustment of assistance level are automatically triggered. For those who no longer meet the assistance conditions, a notice of suspension is sent in advance and an appeal window is reserved. For those whose difficulty has increased, the assistance level is automatically upgraded and the difference is paid. At the same time, service feedback from assistance recipients is collected through channels such as mini-programs and grid workers entering data at home. The feedback data is used as training samples to optimize the matching logic and AI model parameters. The entire process of providing assistance, including eligibility verification results, approval operation records, fund disbursement vouchers, dynamic adjustment records, and service feedback data, is synchronized and hashed before being stored in a consortium blockchain with nodes from civil affairs, finance, audit, and various assistance implementation departments. A multi-node consensus mechanism is used to complete data synchronization, ensuring that the entire process of data is traceable and tamper-proof, and supporting the audit department to retrieve the entire chain of records with one click to complete compliance verification.

[0008] Preferably, the specific process of multi-source heterogeneous data collection and standardized cleaning is as follows: a heterogeneous data mapping dictionary is pre-built to support the rapid access and adaptation of new data types added by different departments. After the field mapping is matched, sensitive fields such as name, ID number, mobile phone number, home address, and medical information are desensitized using the k-anonymity algorithm with differential noise. The desensitized data cannot be linked to a specific individual, but it does not affect the accuracy of group statistics, rule verification, and AI model calculation. Records with missing core fields are automatically generated as verification tasks and pushed to the local grid staff. After the grid staff completes the data entry, it is included in the profile database. All outlier removal, missing value imputation, and desensitization operations generate operation logs for storage, which are then used by data operation and maintenance personnel for retrospective verification.

[0009] Preferably, the specific process for verifying the hard indicators of the first-level policy rules is as follows: A visual drag-and-drop rule configuration engine is adopted, and civil affairs personnel can customize the identification rules for different jurisdictions and different assistance projects without programming. These rules include family per capita income thresholds, property holding limits, and personal identity identification conditions. After the rules are configured, test cases are automatically generated to verify accuracy. After they officially take effect, they are automatically adapted to all assistance applications in the corresponding jurisdiction. The notification document for initial rejection clearly lists the specific policy clauses that do not comply with, the corresponding data sources, and the channels for appealing objections. Applications that pass the initial review are automatically transferred to the second-level verification stage without manual intervention. The entire process is traceable.

[0010] Preferably, the specific process of the second-layer AI implicit difficulty identification and risk verification is as follows: an XGBoost gradient boosting tree model pre-trained with historical assistance samples is used. In the training samples, legitimate assistance recipients are positive samples, and those who have been investigated for insurance fraud are negative samples. The input features include the proportion of the assistance recipient's annual medical expenditure to family income, total education expenditure, disability level of family members, employment status, real estate and vehicle ownership, family debt scale, number of dependents, and consumption expenditure level in the past 12 months. The confidence threshold of the model output can be customized and adjusted according to the tightness of local policies. Those with confidence values ​​higher than the first threshold pass the verification directly, those with confidence values ​​lower than the first threshold but higher than the second threshold are transferred to manual review, and those with confidence values ​​lower than the second threshold are directly rejected. The list of suspected insurance fraud recipients is synchronized to all assistance business departments to prevent repeated insurance fraud across projects and regions.

[0011] Preferably, the specific process of the two-way intelligent matching is as follows: First, demand tags are automatically extracted from the full-dimensional profile of the recipients. Recipients with chronic or special diseases are automatically matched with medical assistance needs, families with school-age children are automatically matched with educational assistance needs, unemployed individuals of working age are automatically matched with employment assistance needs, and elderly people living alone are automatically matched with needs for age-friendly renovations and home-based elderly care services. Then, the remaining quotas, service supply capacity, and service coverage radius of various types of assistance resources in the corresponding locality are matched. Among them, employment assistance positions are prioritized to be matched with positions within 5 kilometers of the recipient's residence, and age-friendly renovation service providers are prioritized to be matched with compliant service providers registered in the locality. The generated personalized assistance service list specifies the subsidy amount, service content, processing time limit, and application method for each assistance project. When pushed to the recipients, the corresponding business department's processing process is automatically triggered, eliminating the need for recipients to submit materials repeatedly.

[0012] Preferably, the specific process of re-verification of eligibility for assistance and dynamic adjustment of assistance level is as follows: the monitoring cycle can be customized according to actual needs. During public emergencies and natural disasters, the monitoring cycle can be temporarily shortened to the weekly level. The eligibility re-verification is completed entirely based on shared data from multiple departments, without requiring the recipient to submit any paper materials. A notification document is sent 7 days in advance before the suspension of assistance, and an appeal window is reserved. Those who pass the appeal can continue to enjoy assistance. The difference in assistance level is automatically calculated and reissued in the first payment cycle after the adjustment. The collected service feedback includes satisfaction ratings, new needs, and objection appeals. All feedback data is automatically labeled and used as training samples to input into the AI ​​model and matching rule engine, iteratively optimizing the recognition accuracy and matching suitability.

[0013] Preferably, the consortium blockchain adopts a permissioned blockchain architecture, requiring unified authorization from the civil affairs department for node access. The consensus mechanism employs a practical Byzantine fault-tolerant algorithm, with each authorized node storing a complete copy of the ledger. Any modification to data by a single node requires consensus from more than two-thirds of the authorized nodes to take effect, completely eliminating the risk of data tampering by a single department. The on-chain data includes all operation records and vouchers for the entire relief process. Audit departments can access the entire chain record of a single relief recipient, a single relief project, or a single batch of relief funds with a single click through a dedicated authorized interface, without needing to request materials from other departments, significantly improving audit and verification efficiency.

[0014] Preferably, it also includes a proactive assistance discovery step: pre-configure rules for identifying abnormal change signals, including signals such as a single medical insurance expenditure exceeding 30% of the family's annual income, social security payments suspended for three consecutive months without receiving unemployment benefits, newly issued level one or level two disability certificates, damage to houses caused by natural disasters in the area, and new serious illness diagnoses for family members. Real-time monitoring of changes in multi-source shared data, and automatic generation of verification tasks after a signal is triggered and pushed to the corresponding local grid worker's work terminal. The task includes the basic information of the suspected person in need and details of the abnormal signal. The grid worker will automatically trigger the assistance application process if the person meets the conditions when they visit the home, without the need for the person to actively submit an application, thus realizing the transformation of assistance services from "people looking for policies" to "policies finding people".

[0015] Preferably, a digital management and precision service system for civil affairs assistance includes: The multi-source data collection module is used to connect to the government data sharing and exchange platform. It supports two data collection methods: scheduled synchronization and real-time interface calls. It has a built-in standardized cleaning rule engine and differential privacy desensitization tool, which automatically completes field mapping, outlier removal, missing value imputation, and sensitive information desensitization of multi-source heterogeneous data. It builds a full-dimensional profile library of aid recipients according to four dimensions: basic attributes, hardship attributes, demand attributes, and historical aid attributes. It supports custom tagging, full-text search, and classification statistics functions. The eligibility verification module has a built-in visual drag-and-drop policy rule engine and a pre-trained AI implicit difficulty recognition model. It supports custom configuration of rules for different regions and different relief projects, automatically completes the initial verification of policy hard indicators and AI risk verification, supports seamless connection between automatic verification and manual review processes, automatically generates standardized verification result notification documents, and automatically records the entire process. The relief resource matching module has a built-in two-way intelligent matching algorithm. It synchronizes the remaining funds, service supply capacity and service coverage of various relief projects across the entire domain in real time. It automatically extracts the demand tags of relief recipients, completes two-way matching according to relief priority and service suitability, generates a personalized relief service list, and supports automatic push of the list and automatic triggering of the relief processing process. The dynamic monitoring and adjustment module has a built-in customizable monitoring rule engine, which automatically connects to data interfaces of multiple departments to complete timed data updates, automatically triggers the process of re-verification of rescue eligibility, adjustment of rescue level, and notification push. It has two built-in feedback collection channels: a mini-program and a grid worker's work terminal, and automatically labels feedback data as training samples to iteratively optimize the AI ​​model and matching rule parameters. The blockchain evidence storage module is used to connect to consortium blockchain nodes jointly built by multiple departments. It has a built-in data hash calculation tool that automatically stores all operation records and voucher data of the entire rescue process on the blockchain. It adopts a multi-node consensus mechanism to ensure that the data is tamper-proof and supports authorized departments to retrieve on-chain data with one click to complete audit and verification.

[0016] Preferably, it also includes a proactive discovery module and a visual monitoring module. The proactive discovery module has a built-in abnormal signal recognition rule engine, which monitors changes in multi-source shared data in real time, automatically generates a list of suspected people in need and assigns grid workers to verify their information at home, and supports task tracking, feedback, and statistical functions. The visual monitoring module has a built-in multi-dimensional statistical analysis model, which automatically generates multi-dimensional visual charts showing the trend of relief fund disbursement, the geographical distribution of relief recipients, the utilization rate of relief resources, and the number of abnormal warnings. It supports drill-down queries from the provincial level to the district / county level, street level, community level, and individual relief recipient level, allowing management departments to grasp the operational status of relief operations in their jurisdiction in real time and adjust relief policies and resource allocation in a timely manner.

[0017] Compared with existing technologies, the beneficial effects of this invention are: This invention effectively addresses the shortcomings of existing technologies, such as poor cross-departmental data interoperability, high reliance on manual verification, insufficient accuracy, and high costs for the public, by combining standardized processing of multi-departmental, multi-dimensional data with a two-layer verification mechanism. It achieves automatic collection and verification of relief-related data through standardized cross-departmental data interfaces, eliminating the need for recipients to visit multiple departments to obtain certificates, significantly lowering the barrier to entry. The verification model, combining a rule engine and an AI model, ensures the uniformity of policy implementation while identifying hidden difficulties not reflected in application materials, greatly improving the efficiency and accuracy of relief eligibility verification and effectively reducing errors in coverage and omissions.

[0018] This invention effectively addresses the shortcomings of existing technologies, such as blind spots in passive assistance coverage, poor resource supply and demand matching, and untimely adjustment of eligibility, through two-way intelligent matching, full-cycle dynamic monitoring, and proactive assistance discovery mechanisms. Relying on proactive identification of abnormal signals to achieve "policy finding people," it can cover special groups with limited information access capabilities, such as elderly people living alone, people with disabilities, and disaster victims, significantly expanding the scope of assistance coverage. Intelligent matching based on demand tags and resource availability enables precise allocation of assistance resources, avoiding resource waste. Full-cycle, seamless dynamic monitoring can promptly track changes in the degree of hardship of those receiving assistance and automatically adjust the level of assistance, eliminating the need for repeated applications and greatly improving the timeliness and suitability of assistance services.

[0019] This invention achieves end-to-end data storage through a multi-departmental collaborative blockchain, effectively addressing the shortcomings of existing technologies such as easy data tampering, high risk of insurance fraud, and low audit efficiency. The hash values ​​of all relief data are stored on the blockchain, ensuring immutability and traceability. This not only meets the compliance requirements of government data remaining within its own domain but also technically prevents data falsification and fraudulent claims. The risk warning list shared across departments effectively prevents repeated insurance fraud across projects and regions. Audit departments can access the end-to-end stored data with a single click, eliminating the need for cross-departmental material requests and significantly improving audit efficiency. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall business process proposed in this invention; Figure 2 This is a flowchart of the multi-source data governance and proactive discovery process proposed in this invention; Figure 3 This is a flowchart illustrating the two-layer qualification verification logic proposed in this invention. Figure 4 This is a flowchart illustrating the two-way intelligent matching process between needs and resources proposed in this invention. Figure 5 This is a flowchart of the dynamic adjustment and blockchain-based evidence storage and supervision process proposed in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figures 1 to 5 This invention discloses a method for digital management and precise service of civil affairs assistance business. The specific implementation process, technical details, parameter configuration, and anomaly handling rules are as follows: Step 1: Multi-source data collection and construction of a comprehensive profile database of aid recipients: This step strictly follows the "National Integrated Government Service Platform Data Sharing and Exchange Interface Specification" to connect with the government data sharing and exchange platform. Before connection, government data sharing application materials must be submitted. After departmental review, the interface call key is obtained. The interface uses the SM2 national cryptographic algorithm for signing. The request header carries three verification parameters: signature string, timestamp, and caller identifier. Automatic retry after 3 seconds; if 3 consecutive retries fail, an operation and maintenance alarm is triggered; if 5 retries fail, automatic switch to backup interface is initiated to ensure data collection availability is not less than 99.95%. The full field list of the 8 types of departmental data sources is as follows: Civil Affairs Department: Low-income / Special hardship identification number, type of assistance, amount disbursed, validity period, historical appeal records, and records of penalties for fraud.

[0023] Public Security Department: Household registration address, number of cohabiting family members, death cancellation certificate, dishonest person subject to enforcement certificate, drug-related / imprisonment certificate, migrant population residence registration information, vehicle ownership information (brand, price, usage nature).

[0024] Medical insurance department: Annual total medical expenditure, amount reimbursed by the pooled fund, amount reimbursed by critical illness insurance, personal co-payment ratio, type of chronic and special disease identification, number of outpatient / inpatient visits, and medication purchase records.

[0025] Social security department: insurance status, continuous payment period, unemployment benefit validity period, work injury insurance compensation records, pension payment amount, and flexible employment registration information.

[0026] Housing and Urban-Rural Development Department: Number of properties, building area, total appraised value, validity period of public rental housing lease, records of dilapidated housing renovation, records of illegal construction, and rental registration information.

[0027] The Disabled Persons' Federation: Disability level, status of nursing subsidy receipt, rehabilitation service records, assistive device distribution records, and nursing identification for severely disabled persons.

[0028] Education Department: Records of school enrollment, receipt of student subsidies, records of nutrition meal subsidies, records of student loans, and dropout registration information.

[0029] Emergency management departments: Data collected after disaster type, disaster level, affected area, casualties, disaster relief receipt records, and building damage level undergo three levels of standardized cleaning: The first level is field mapping alignment: The TF-IDF text similarity algorithm is used to automatically match the source fields with the MZ / T 118-2018 standard fields. If the similarity is ≥92%, a mapping recommendation is automatically generated. It takes effect after confirmation by business personnel. Adding departmental data does not require modification of the core code. Only the mapping relationship needs to be configured to complete the access.

[0030] The following is a sample Python implementation snippet of the mapping logic: import jieba from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.metrics.pairwise import cosine_similarity standard_fields = ["Average monthly household income per capita", "Percentage of medical expenditure", "Disability level", "Number of properties owned"] source_field = "Average monthly income of family members" # Calculate cosine similarity after word segmentation vectorizer = TfidfVectorizer(tokenizer=jieba.lcut) vecs = vectorizer.fit_transform(standard_fields + [source_field]) similarity = cosine_similarity(vecs[-1], vecs[:-1])[0] max_idx = similarity.argmax() if similarity[max_idx]>= 0.92: print(f"The source field {source_field} is automatically matched with the standard field {standard_fields[max_idx]}, and the similarity is {similarity[max_idx]:.2f}") The second level involves outlier removal and missing value completion: The 3σ criterion is used to remove outliers from numeric fields. The calculation formula is: If... If x is an outlier, then x is considered an outlier; where This is the average value of the entire population in the corresponding region. This represents the standard deviation of the entire population in the corresponding region, such as the average monthly household income per capita in a certain district. =3216 yuan =1792 yuan. Records with income ≥3216+3*1792=8592 yuan are marked as outliers. If they are incorrectly filled in, they will be automatically returned to the data provider for verification. If they are genuine high incomes, they will be directly marked as not meeting the relief conditions.

[0031] For missing core fields (ID number, per capita family income, disability level), an automatic verification work order is generated and pushed to the local grid worker, requiring them to be entered within 24 hours. For missing non-core fields, the average value of the same dimension for the same family is used to complete the data, with an accuracy rate of ≥97%. The third level is differential privacy anonymization of sensitive information: For sensitive fields such as name, ID number, mobile phone number, home address, and medical information, the k-anonymity algorithm is used, with the k value fixed at 10, to ensure that a single record is indistinguishable from at least 9 other records. The calculation formula is the same as the Laplace differential privacy algorithm. ,in The original sensitive field value, These are the values ​​after desensitization. Global sensitivity (taking the maximum range of values ​​for a field) Scenarios for privacy budgeting and statistical analysis A value of 1.2 is used for qualification verification scenarios. The value of 0.8 balances privacy protection and data availability. The anonymized data cannot be linked to a specific individual, but it does not affect the accuracy of rule verification and model calculation, and fully complies with the compliance requirements of the Personal Information Protection Law and the Data Security Law.

[0032] After the anonymization is completed, a profile library is built according to four dimensions: basic attributes, difficulty attributes, demand attributes, and historical assistance attributes. The underlying layer uses MySQL to store structured core data, Elasticsearch to store tags that support full-text search, and HBase to store all historical data with hot and cold data separation. Historical data older than 5 years is automatically archived to object storage. Each profile contains at least 22 searchable tags, and business personnel can add custom tags.

[0033] Step 2: Two-tier eligibility verification: All assistance applications (submitted via online mini-program, offline window, grid worker on behalf of the applicant, and applications triggered by proactive discovery) will be subject to a unified two-tier verification process. The proportion of no manual intervention in the entire process is ≥90%, and the verification cycle is ≤5 minutes.

[0034] The first layer is the verification of hard indicators of policy rules: It adopts a visual drag-and-drop rule engine with five built-in components: numerical comparison, string matching, time judgment, logical AND / OR / NOT, and set inclusion. Business personnel can customize the identification rules for different regions and different relief projects without programming. The rules support three control mechanisms: version management, canary release, and automatic rollback. Version management generates a unique version number for each rule modification, recording the modifier, modification time, and modification content, and supports one-click rollback to historical versions; canary release allows for initial pilot operation in 1-3 streets for 3 days, with full rollout only if the verification pass rate fluctuation is ≤10%; automatic rollback automatically switches back to the previous version if the verification pass rate or false positive rate fluctuation exceeds the threshold after the rule goes live. Rules are stored using Groovy scripts, with a running efficiency of ≥2000 records / second. Example Groovy implementation code for a low-income assistance rule is as follows: def checkDibaoEligibility(householdPortrait) { / / Read local policy parameters and support dynamic adjustment def perIncomeThreshold = 800 / / Threshold for average monthly household income per person def propertyThreshold = 150000 / / Threshold for total family assets def luxuryCarThreshold = 150000 / / Price threshold for non-commercial vehicles / / Calculate core metrics def perMonthIncome = householdPortrait.totalAnnualIncome / 12 / householdPortrait.memberCount def totalProperty = householdPortrait.propertyTotalValue def hasLuxuryCar = householdPortrait.vehicleList.any { it.useType == "Non-Operation"&&it.price>luxuryCarThreshold} / / Output verification results if(perMonthIncome <perIncomeThreshold&& totalProperty<propertyThreshold&& !hasLuxuryCar) { return [ "pass": true, "level": "Category A Low-income assistance", "amount": perIncomeThreshold - perMonthIncome, "msg": "Meets the criteria for Category A low-income assistance" ] } else { def rejectReason = [] If (perMonthIncome>= perIncomeThreshold) rejectReason.add("The average monthly income per household exceeds the threshold of ${perMonthIncome.round(1)} yuan") If (totalProperty>= propertyThreshold) rejectReason.add("Total family assets exceed the threshold of ${totalProperty} yuan") if (hasLuxuryCar) rejectReason.add("Family owns non-commercial vehicles worth over 150,000 yuan") return [ "pass": false, "msg": "Does not meet the criteria for low-income assistance: " + rejectReason.join("、"), "appealChannel": "Bring proof of income and proof of assets to the street-level civil affairs office to file an appeal." ] } } Applications that fail the rule verification will automatically generate a PDF notification document with a national cryptographic electronic seal, listing the specific policy clauses that do not comply, the data source, and the appeal channel. This document will be simultaneously pushed to the applicant via SMS, mini-program, and door-to-door visits by grid workers. Applications that pass the verification will automatically proceed to the second-level AI verification stage.

[0035] The second layer is for AI-based implicit difficulty identification and risk verification: a pre-trained XGBoost gradient boosting tree model is used, with 1.2 million historical labeled data points as training samples. This includes 840,000 positive samples from legitimate aid recipients and 360,000 negative samples from individuals verified as fraudulent by civil affairs / auditing departments. The model hyperparameters are configured as follows: n_estimators=200, max_depth=6, learning_rate=0.1, subsample=0.8, colsample_bytree=0.8. The 10-fold cross-validation accuracy is ≥94.3%, and the false positive rate is ≤2.8%. The formula for calculating the credibility value of aid eligibility is: in Let be the weight of the i-th decision tree. For the i-th tree pair, input feature vector The output value, This is a global bias term. It is a 17-dimensional feature vector, which includes core features such as the proportion of annual medical expenditure to family income, total education expenditure, disability level of family members, employment status, number of properties owned, family debt scale, number of dependents, and consumption expenditure level in the past 12 months.

[0036] The model outputs a confidence score ranging from 0 to 100. Two custom thresholds can be set: the default first threshold is 80 and the second threshold is 50. A score ≥ 80 passes the verification directly, a score 50 ≤ Score < 80 is transferred to manual review, and a score < 50 is rejected directly and included in the cross-departmental risk warning list, prohibiting the submission of similar relief applications for one year.

[0037] To improve model interpretability, the SHAP value is used to output the feature contribution of each verification result. The example output is "This score is 92 points, of which medical expenditure contribution is +26 points, per capita family income contribution is +21 points, disability level contribution is +15 points, no high-value property contribution is +14 points, employment status contribution is +10 points, and other features contribution is +6 points". Manual reviewers can directly view the score composition without having to check the materials one by one, improving review efficiency by 80%.

[0038] The model is iterated every quarter. Before adding new samples, a fairness check is required to ensure that the accuracy difference between different genders, ages, regions, and ethnic groups is ≤2%, thus avoiding algorithmic discrimination.

[0039] Step 3: Two-way intelligent matching: For verified beneficiaries, the need tags are automatically extracted first. The tag extraction rules can be customized by business personnel. The default rules are as follows: patients with chronic and special diseases are automatically matched with the "outpatient medical assistance" tag; families with children in compulsory education are automatically matched with the "educational assistance subsidy" tag; eligible people who have been unemployed for more than 3 consecutive months are automatically matched with the "employment skills training + public welfare job recommendation" tag; elderly people over 70 years old living alone are automatically matched with the "age-friendly renovation + home-based elderly care service" tag; disaster-affected families are automatically matched with the "temporary living allowance + dilapidated house renovation subsidy" tag; and severely disabled people are automatically matched with the "nursing allowance + assistive device subsidy" tag.

[0040] Then, the real-time updated relief resource database interface is called to obtain the current available relief resources in the locality, including the remaining funds for each relief project, the remaining service capacity of service organizations, service coverage, the remaining number of positions, service provider ratings, and other data. The matching priority is then calculated using a suitability formula. in The degree of matching is defined as follows: 1 for a perfect match, 0.5 for a partial match, and 0 for a no match. The priority weights for assistance are as follows: extremely poor people = 3, those receiving minimum living allowance = 2, those on the verge of receiving minimum living allowance = 1, and those facing financial difficulties due to expenses = 0.8. Service suitability score (1 for distance <5km and service provider rating ≥4.5, 0.6 for 5-10km and rating ≥4, 0 for >10km or rating <4).

[0041] according to Sort the values ​​from high to low, select 3-6 suitable resources to generate a personalized list of assistance services, specifying the subsidy amount, service content, processing time limit and application method for each assistance project. The list for the elderly automatically enlarges the font and adds voice broadcast, while the list for the disabled adds sign language video guidance.

[0042] The system automatically triggers the processing flow of the relevant departments when the list is pushed to the recipients, eliminating the need for recipients to submit materials repeatedly. Recipients can choose whether to accept a certain type of assistance, and the resources they give up are automatically returned to the resource pool and allocated to other suitable recipients. If there are insufficient resources of a certain type of assistance in the local area, the system will automatically apply to the higher-level civil affairs department for cross-regional adjustment or match similar resources in neighboring districts and counties, with a resource matching coverage rate of ≥98%.

[0043] Step 4: Dynamic monitoring and adjustment: Set differentiated monitoring cycles for different types of assistance: long-term assistance such as minimum living allowance / extreme hardship assistance is monitored quarterly, temporary assistance is monitored monthly, and special assistance such as medical assistance / educational aid is monitored according to the project cycle. During public emergencies and natural disasters, the monitoring can be temporarily adjusted to weekly or even daily.

[0044] After the monitoring period expires, the system automatically calls multiple departmental interfaces to obtain the latest data and re-performs eligibility verification without requiring any paper materials from the recipients, achieving "seamless verification." For those who no longer meet the eligibility criteria after verification, a notification of suspension of benefits is sent 7 days in advance, with a 15-day appeal window. After an appeal is submitted, a manual review is completed within 2 working days. Those who pass the review continue to receive benefits and receive retroactive payments for the suspension period. For those whose hardship worsens, the benefit level is automatically upgraded, and the difference is paid in the first payment cycle after the adjustment, along with a notification of the adjustment.

[0045] The system is also configured with real-time anomaly monitoring rules. When the following abnormal signals appear on the aid recipients, temporary verification will be automatically triggered without waiting for the monitoring cycle: a single medical insurance expenditure exceeds 30% of the family's annual income, social security payments are suspended for 3 consecutive months, a new diagnosis of a serious disease is made, a new record of being a dishonest person subject to enforcement is made, large consumption expenditures (single transaction ≥ 50,000 yuan), and abnormal increase in family assets.

[0046] Service feedback from aid recipients is collected through two channels: a mini-program and a grid worker's work platform. Applications with a satisfaction score of ≤3 are automatically marked as failed matches. The AI ​​model and matching rules are iterated and optimized every quarter by adding to the training set. A / B testing is required before a new model is launched. It is only officially launched if the accuracy and matching fit improve by ≥1%. Otherwise, optimization continues.

[0047] Step 5: Blockchain-based end-to-end evidence storage: The eligibility verification results, approval records, fund disbursement vouchers, dynamic adjustment records, and service feedback data for the entire relief process are first calculated using SHA-256 hash values ​​and uploaded to a permissioned consortium blockchain jointly built by six core departments: civil affairs, finance, auditing, medical insurance, disabled persons' federation, and emergency management. This blockchain can be expanded to include nodes from other business departments in the future. The consortium blockchain uses a practical Byzantine fault-tolerant consensus algorithm, supporting normal operation even with up to one-third of its nodes failing. Modifications to data by a single node require consensus from more than two-thirds of the authorized nodes to take effect, completely eliminating the risk of data tampering by a single department.

[0048] Each block has the following structure: the block header contains the hash of the previous block, a timestamp, Merkle root, and node signature; the block body contains the hashes of all data uploaded to the chain this time, the operator ID, the operation type, and the data source department identifier. A block is packaged every 10 minutes, and all nodes store a complete copy of the ledger. Data is automatically backed up to an off-site disaster recovery center daily, ensuring data reliability of ≥99.999%.

[0049] Node permissions are managed hierarchically: civil affairs nodes have full data read / write permissions, finance nodes have read / write permissions for fund disbursement records, audit nodes have read-only permissions for all data, and other departmental nodes only have read / write permissions for their own relevant data. Audit departments can access the entire chain of records for a single aid recipient, a single aid project, or a single batch of aid funds with a single click via a dedicated authorization interface, inputting the recipient's ID number, project number, and batch number, without needing to request materials from other departments, improving audit efficiency by over 90%. The consortium blockchain supports cross-chain integration, enabling data exchange with aid consortium blockchains in other provinces and cities through a cross-chain gateway, facilitating cross-regional verification of aid applications from migrant populations.

[0050] In this invention, the pre-set heterogeneous data mapping dictionary in the multi-source heterogeneous data collection and standardized cleaning process supports batch import and export. The access time for new departmental data is ≤1 working day, far shorter than the 15-day access cycle of the traditional docking mode. After anonymizing sensitive fields, only the first 6 and last 4 digits of the ID number are retained, with the middle 8 digits replaced by asterisks. Only the surname and the last character of the name are retained, with the middle characters replaced by asterisks. Differential noise is also superimposed to ensure that it is impossible to identify a single individual through the anonymized data, but the identity verification can be accurately completed by comparing hash values ​​without affecting subsequent business processes. All cleaning operations (outlier removal, missing value completion, and anonymization) generate detailed operation logs, which are stored in an independent, tamper-proof log library, supporting backtracking and verification by maintenance personnel. The log retention period is ≥10 years, which meets the requirements of government data management.

[0051] In this invention, the visual drag-and-drop engine for verifying the hard indicators of the first-layer policy rules has over 120 built-in commonly used civil affairs assistance rule templates, covering the mainstream assistance policies of 31 provinces and municipalities across the country. Business personnel can directly select a template, modify parameters, and generate rules adapted to their local conditions without having to configure them from scratch. After the rules are configured, 100 test cases (50 positive samples and 50 negative samples) are automatically generated, the tests are automatically run, and a test report is generated. The rules can only be officially implemented if the test accuracy is ≥100%, avoiding verification errors caused by incorrect rule configuration. The notification document for initial rejection supports online appeals. Applicants can upload supporting materials through a mini-program, which will automatically transfer them to the manual review stage without having to go to an offline window. The appeal processing cycle is ≤3 working days.

[0052] In this invention, the feature importance ranking of the XGBoost model for the second-layer AI implicit difficulty identification and risk verification is as follows: annual medical expenditure as a percentage of household income > average monthly household income per capita > disability level > duration of continuous unemployment > total household assets. Localities can adjust the feature weights according to their local policy needs. The risk warning list is shared across departments and regions. When a recipient submits an application for assistance in other departments or regions, a risk warning is automatically triggered, requiring manual review before proceeding to the next stage. This effectively prevents repeated fraud across projects and regions. The risk warning list is valid for one year by default, and those with no abnormal records during this period can apply for early removal, which will no longer affect their assistance applications.

[0053] In this invention, the bidirectional intelligent matching rescue resource database is synchronized with the latest remaining data every 5 minutes to ensure the accuracy of the matching results. Service providers with a score of less than 4 points are automatically removed from the matching pool, and job positions with salaries lower than the local minimum wage standard are automatically excluded to ensure the quality of rescue services.

[0054] The generated personalized assistance service list supports one-click appointment. After the applicant confirms, an appointment number is automatically generated and pushed to the work terminals of the applicant and the handling personnel. When the applicant goes to the offline window for processing, he / she only needs to show his / her ID card to verify all materials, without having to bring paper materials.

[0055] For elderly people living alone with limited mobility or severely disabled people, the system automatically matches them with grid workers in the jurisdiction to provide door-to-door services. The entire process is recorded and uploaded to the blockchain for evidence storage.

[0056] For individuals facing special difficulties who still do not have corresponding assistance resources after matching, the system automatically generates a special application and pushes it to the relevant leader of the civil affairs department for approval. If the conditions are met, the reserve assistance fund can be used to provide a safety net, ensuring that all eligible individuals can receive appropriate services.

[0057] In this invention, the monitoring cycle configuration module for re-verification of aid eligibility and dynamic adjustment of aid level supports custom adjustments based on three dimensions: location, aid type, and special scenario. When public health emergencies, earthquakes, floods, and other natural disasters occur, the monitoring cycle of all aid projects in the disaster area can be temporarily adjusted to the daily level with one click, so as to keep up with the changes in the difficulties of the affected people and avoid secondary difficulties that prevent them from receiving timely aid.

[0058] The notification of suspension of aid can be pushed through multiple channels simultaneously, including SMS, WeChat official account, mini-program pop-up, and door-to-door notification by grid workers, to ensure that all aid recipients receive the notification in a timely manner. During the appeal window period, aid funds will continue to be disbursed normally to avoid misjudgment that could affect the basic living conditions of the people.

[0059] The difference in the amount due to adjustments in the level of assistance is calculated automatically, eliminating the need for manual verification. The system automatically synchronizes this information to the finance department's fund disbursement interface, ensuring timely receipt of supplementary funds. In addition to being used to iterate and optimize the AI ​​model, the collected service feedback data generates monthly optimization reports that are pushed to policy-making departments, providing data support for adjustments to assistance policies and the replenishment of assistance resources. For example, if the job matching rate for employment assistance in a certain region is less than 60%, the system automatically indicates in the optimization report that the region needs to increase the supply of public service jobs.

[0060] In this invention, permissioned nodes of the consortium blockchain must undergo unified identity verification and authorization by the provincial civil affairs department. Each node is configured with a unique public-private key pair. All operations, such as node login, data upload, and data query, must be signed with the private key. Operations that fail to sign are directly blocked to ensure node access security.

[0061] The practical Byzantine fault-tolerant consensus algorithm has an average consensus time of ≤200 milliseconds and can support at least 500 data entries per second on the blockchain, fully meeting the on-chain requirements of the daily relief business volume in all districts and counties across the country. The on-chain data only stores hash values, while the original core data is still stored in the local government data private cloud of each department. This not only meets the compliance requirements of government data not leaving the domain, but also achieves tamper-proof evidence storage of data throughout the entire process.

[0062] When the auditing department accesses on-chain data, the system automatically generates a data query log, recording the queryer, query time, query scope, and query purpose. This log is also stored on the blockchain, ensuring full traceability of the data query process and compliance with auditing requirements. For assistance applications from cross-provincial migrants, the system can connect to the assistance consortium blockchain node in the applicant's place of residence via a cross-chain gateway, retrieving historical assistance data from the applicant's place of residence with a single click. This eliminates the need for official letters of inquiry, reducing the cross-regional verification cycle from 15 days to less than 1 minute.

[0063] In this invention, the pre-set abnormal change signal identification rules in the proactive assistance discovery step allow for custom additions by operational personnel. In addition to the six default signal categories, personalized signals such as "elderly people living alone with children incarcerated," "children without guardians," and "registration for assistance to homeless and begging individuals" can be added. The system monitors incremental data pushes from various departmental data interfaces in real time. Upon triggering an abnormal signal, it first matches historical assistance records. If the individual is already receiving corresponding assistance, it is automatically filtered; otherwise, a verification task is generated and pushed to the corresponding local grid worker's workstation. The task automatically associates with all shared data for that individual. Grid workers do not need to verify other information themselves; they only need to visit the individual to confirm the authenticity of the abnormal signal and whether there are any uncovered difficulties.

[0064] Once verified and deemed eligible for assistance, the system automatically fills in all application materials. Grid workers only need to confirm online to trigger the assistance application process, eliminating the need for residents to submit any materials. This truly achieves "policy finding people," effectively covering vulnerable groups with limited access to information. If, after verification, the criteria are not met, the system automatically marks the signal as a false trigger. Subsequent similar triggers will automatically lower the priority of that individual, reducing the frequency of ineffective home visits by grid workers.

[0065] This invention discloses a digital management and precision service system for civil affairs assistance, and the technical implementation and functional details of each module are as follows: The multi-source data aggregation module adopts the Flink stream computing framework at its core, supporting two modes: batch timed synchronization of TB-level data and real-time interface calls for millisecond-level incremental data. The built-in standardized cleaning rule engine supports visual configuration of cleaning rules, and the differential privacy desensitization tool has passed the Level 3 certification of the National Cybersecurity Classified Protection 2.0, meeting the requirements for government data security.

[0066] The module builds a comprehensive profile database of aid recipients, supporting features such as custom tags, full-text search, multi-condition filtering, and categorized statistical export. The profile update delay is ≤5 minutes, accurately reflecting the latest status of aid recipients.

[0067] The dual-layer qualification verification module has a built-in visual drag-and-drop rule engine that supports rule version management, gray-scale release, and automatic testing. The pre-trained AI implicit difficulty identification model supports online iteration, interpretable output, and fairness verification. The manual review process supports online task allocation, material uploading, and review opinion annotation. The automatic verification and manual review processes are seamlessly connected. The verification results are automatically generated into a standardized notification document with a national cryptographic electronic seal. The entire process is automatically recorded and retained for ≥10 years.

[0068] The built-in two-way intelligent matching algorithm of the relief resource matching module supports custom weights, priority sorting, resource return, and cross-regional adjustment functions. It synchronizes the remaining data of relief resources across the entire region in real time with a synchronization delay of ≤5 minutes. The generated personalized relief service list supports multiple versions to suit the reading needs of four groups: the general public, the elderly, the disabled, and minors. After the list is automatically pushed, it supports online confirmation, appointment processing, and application abandonment. The operation results are synchronized to the resource pool in real time.

[0069] The dynamic monitoring and adjustment module's built-in monitoring rule engine supports customizable monitoring cycles based on location, type of assistance, and special scenarios. It automatically triggers processes such as eligibility re-verification, tier adjustment, and notification push notifications. Built-in feedback collection channels support multiple feedback formats, including text, images, voice, and video. Feedback data is automatically labeled and added to the training set. It automatically triggers iterative training and A / B testing of the AI ​​model and matching rules every quarter, requiring no manual intervention. The blockchain evidence storage module supports integration with mainstream consortium blockchain frameworks. Its built-in SHA-256 hash calculation tool has a processing efficiency of ≥10,000 records / second, automatically completing hash calculations and on-chain storage of data throughout the entire assistance process. It supports multi-dimensional on-chain data queries by ID number, project number, and batch number. Query results automatically generate verification reports with electronic signatures, which can be directly used as audit evidence.

[0070] In this invention, the system also includes an active discovery module and a visual monitoring module. The abnormal signal recognition rule engine built into the active discovery module supports real-time monitoring of incremental data, signal filtering, automatic task allocation, and tracking of grid workers' door-to-door visits. The task completion rate and verification accuracy are automatically calculated and used as a reference for the performance evaluation of grid workers.

[0071] The built-in multi-dimensional statistical analysis model of the visualization monitoring module supports custom statistical dimensions, time ranges, and drill-down levels. The automatically generated visualization charts include line charts (trends in relief fund disbursement and changes in the number of people receiving relief), bar charts (utilization rate of relief resources in each district and county, and the number of people covered by various relief projects), heat maps (geographical distribution of relief recipients and areas with high incidence of abnormal signals), and pie charts (structure of relief funds and proportion of different types of relief recipients). It supports five levels of drill-down queries from the provincial level to the district / county level, street level, community level, and individual relief recipients. All statistical data can be exported to PDF and Excel formats, allowing management departments to monitor the operational status of relief operations in their jurisdiction in real time, adjust relief policies and resource allocation in a timely manner, and improve the efficiency of relief resource utilization.

[0072] Scenario Example 1: Proactive Assistance for Seriously Ill Elderly Living Alone Scenario Description: Wang, a 78-year-old elderly person living alone in a street in a prefecture-level city in eastern China, is on the verge of receiving minimum living allowance. His children work in other places for a long time. In March 2024, he suffered a sudden cerebral hemorrhage and was hospitalized. His out-of-pocket medical expenses reached 127,000 yuan, accounting for 82% of his family's annual income. The elderly person has poor information access ability and does not know that he can apply for large medical assistance. Under the original assistance model, the elderly person cannot receive assistance in time, and his subsequent treatment and basic life will be seriously affected.

[0073] The system's proactive discovery module monitors the incremental data interface of the medical insurance department in real time. On March 21, 2024, it received data on Wang's large medical expenses, triggering an abnormal signal that "a single medical expense exceeds 30% of the family's annual income." The system automatically matched Wang's full-dimensional profile data, confirming that he was on the verge of receiving minimum living allowance, lived alone, and had no other assistance records. It automatically generated a task to be verified and pushed it to the corresponding community grid worker's work terminal. The task included all information such as Wang's medical expense records, family income records, residential address, and contact number.

[0074] Grid workers conduct door-to-door verification within 24 hours to confirm that the elderly person is seriously ill and has needs for medical assistance and home-based elderly care services. After submitting the verification results online, the system automatically fills in all application materials and enters a two-layer verification process: The first layer is rule engine verification, where the family's per capita income and assets meet the requirements of medical assistance policies, and the initial verification is passed; the second layer is AI model verification, where the system outputs a credibility score of 92 after inputting 17-dimensional features, and the verification is passed directly. The system automatically matches the remaining medical assistance funds and home-based elderly care service provider resources in the local area to generate a personalized assistance list: medical assistance subsidy of 89,000 yuan, 4 home-based elderly care services per month, and 10,000 yuan for age-friendly renovation subsidies. This list is pushed to the elderly person's family's mini-program and the street office, automatically triggering the fund disbursement and service dispatch process, without requiring the elderly person to submit any materials.

[0075] This scenario effectively addresses the shortcomings of existing technologies in passive assistance coverage, allowing assistance to be triggered without the elderly actively applying. Furthermore, relying on cross-departmental data sharing, no certificates are required, and the entire processing time is only 3 days, far shorter than the original 30-day processing time. The AI ​​verification accuracy rate reaches 100%, with no errors or omissions. All operation records are stored on the blockchain for auditing and verification at any time.

[0076] Scenario Example 2: Mass Rescue Scenario After Flood Disaster In June 2024, a county in central China suffered a severe flood disaster, with 12,000 households experiencing damage to their homes and property. Under the original relief model, residents were required to submit disaster relief certificates offline, and staff would verify each one individually. The processing time for batch relief was at least 30 days, leaving many affected residents unable to receive temporary living allowances and dilapidated housing renovation subsidies in a timely manner, making it difficult to guarantee their basic living needs. Furthermore, the batch disbursement of funds was prone to errors, omissions, and fraudulent claims, making auditing and verification extremely difficult.

[0077] After the disaster, the county's civil affairs department first imported the list of 12,000 affected households reported by the emergency management department into the system. The system automatically connected with shared data from multiple departments, including public security, housing and construction, civil affairs, and medical insurance, to create a comprehensive profile of the affected people and set temporary assistance rules: families whose houses were damaged to level C or above received a temporary living allowance of 3,000 yuan and a dilapidated house renovation subsidy of 20,000 yuan; families whose houses were damaged to level D received a temporary living allowance of 5,000 yuan and a dilapidated house renovation subsidy of 50,000 yuan; and those in extreme poverty and those receiving minimum living allowances received an additional 50% subsidy on top of the above.

[0078] The system performs batch verification at two levels, completing the verification of 12,000 applications within 2 hours. Of these, 11,200 applications with a credibility score ≥80 passed directly, while 800 applications with abnormal assets were transferred for manual review, which was completed within 2 days. The system automatically matches temporary relief funds and dilapidated housing renovation service providers, generating relief lists in batches. Subsidies are automatically synchronized to the finance department's disbursement interface, and all funds are disbursed to the social security accounts of affected residents within 3 days. Dilapidated housing renovation service providers automatically assign orders to corresponding affected families. The system stores all verification and fund disbursement records on the blockchain, allowing auditing departments to access the entire chain of records for all batches of relief with a single click, eliminating the need for individual verification of paper documents.

[0079] This scenario effectively addresses the shortcomings of existing technologies, such as low efficiency, high error rate, and unreliable evidence storage in batch rescue efforts. It improves batch verification efficiency by more than 100 times, reduces the error and omission rate by far less than the original model, ensures that the entire process is stored on the blockchain without the risk of tampering, and improves audit efficiency by 90%, ensuring that disaster-stricken people receive timely basic protection.

[0080] refer to Figure 1 This diagram illustrates the macro-level closed-loop process of digital management and precise service in civil affairs assistance. First, by connecting with multiple government service platforms, the system achieves full data collection and standardized processing, building a precise database of recipient profiles. Next, the system enters the core verification phase, employing a two-tiered approach of hard policy indicators and AI-powered identification of hidden difficulties. Upon successful verification, the system uses intelligent algorithms to achieve a two-way matching of supply and demand, generating personalized assistance lists and automatically triggering processing. During the assistance implementation process, the system dynamically monitors according to preset cycles, ensuring that the eligibility and level of assistance recipients are adjusted in real time according to actual circumstances. Finally, all data throughout the process is stored using blockchain technology, ensuring the traceability and immutability of administrative power throughout the entire process, providing underlying technical support for auditing and supervision.

[0081] refer to Figure 2This diagram details the data processing path from source to profile generation, highlighting the proactive discovery mechanism. The system uses a heterogeneous data mapping dictionary to transform heterogeneous data from multiple departments such as public security, medical insurance, and social security into a standardized format. During processing, the 3σ criterion and mean interpolation are employed to ensure data quality, while differential privacy algorithms are used to anonymize sensitive fields, ensuring information security. The process's unique proactive discovery module monitors abnormal changes in real time, such as large medical insurance expenditures or social security payment suspensions, automatically identifying potential hardship groups and generating verification tasks for grid workers. This effectively changes the previous passive situation where people had to seek policies, shifting the focus of assistance services forward.

[0082] refer to Figure 3 This diagram illustrates how the system ensures accurate eligibility for assistance through a rigorous dual-mechanism. The first layer of verification relies on a visual rule engine, transforming complex laws and regulations into machine-executable hard indicators to quickly filter out ineligible applications. The second layer of verification is the system's intelligent core, employing a pre-trained gradient boosting tree model to perform deep learning on multi-dimensional characteristics such as the recipient's family expenditure ratio and debt level, outputting a credibility value and risk level. This mechanism not only uncovers hidden difficulties behind the data but also effectively identifies suspected insurance fraud. By setting different thresholds, the system achieves a differentiated processing approach, including automatic approval, manual review, and direct rejection, significantly improving approval efficiency and fairness.

[0083] refer to Figure 4 This diagram illustrates how relief resources are allocated in a personalized and precise manner. The process begins by extracting the needs tags of those in need from profiles, such as medical care, education, and elderly care. Simultaneously, the system monitors the resource status on the supply side in real time, including the remaining amount of each special fund and the coverage radius of service organizations. The matching algorithm deeply couples demand and supply, taking into account the priority of those in need (e.g., those in extreme poverty). For example, for employment needs, the system prioritizes matching positions within a specified geographical distance. The resulting personalized service list not only includes the subsidy amount but also clearly defined processing timelines and methods, reducing the burden of repeated document submissions for the public by automatically triggering business processes.

[0084] refer to Figure 5This diagram illustrates the long-term maintenance mechanism and security system of the relief operations. The system sets differentiated monitoring cycles based on project type, enabling regular automatic re-verification of relief eligibility. This dynamic management ensures that the level of relief can be adjusted according to changes in the degree of family hardship, reflecting the flexibility of social assistance. Furthermore, the system introduces a closed-loop feedback mechanism, feeding relief evaluation and appeal data back to the AI ​​model, driving continuous algorithm evolution. Regarding data security, consortium blockchain technology is used to hash and upload key nodes such as approval, disbursement, and adjustment to the blockchain. The multi-node consensus mechanism eliminates the possibility of a single department modifying data, and the auditing department can access the complete ledger through a dedicated interface, achieving end-to-end supervision from fund allocation to end-user services.

[0085] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for digital management and precise service of civil affairs assistance, characterized in that, Includes the following steps: Collect multi-source heterogeneous assistance-related data from multiple departments such as civil affairs, public security, medical insurance, social security, housing and construction, and disabled persons' federation, and construct a comprehensive profile database of assistance recipients after standardized cleaning; A two-tiered eligibility verification process is implemented for applicants for assistance. The first tier is the verification of hard indicators based on policies and rules, and the second tier is the identification of hidden difficulties and risk verification by AI. Those who pass the verification will proceed to the assistance matching stage. Based on the needs tags of the recipients and the remaining information of available relief resources across the entire region, a two-way intelligent matching process is performed to generate a personalized relief service list that is appropriate for the recipients. Multi-source data is updated synchronously according to the monitoring cycle corresponding to the relief project, automatically triggering re-verification of relief eligibility and dynamic adjustment of relief level, and synchronously collecting service feedback from relief recipients to optimize matching logic. The eligibility verification results, operation records, and fund disbursement vouchers for the entire rescue process are all synchronously stored in a consortium blockchain node jointly participated in by multiple departments, achieving full traceability and immutability.

2. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, The specific process of collecting and standardizing heterogeneous data from multiple sources is as follows: according to the unified field coding of the national civil affairs assistance data standard, field mapping and matching are performed on heterogeneous data from different departments. After completing the intelligent imputation of missing values ​​and the automatic removal of outliers, differential privacy desensitization processing is performed on all sensitive information, and then the data is stored in the full-dimensional profile database of the assistance recipients.

3. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, The specific process for verifying the hard indicators of the first-level policy rules is as follows: various relief policies in different regions are broken down into rule engines that can be executed automatically. After inputting the profile data of the relief recipients into the rule engines, the initial verification results are automatically output. Applications that do not meet the hard indicators of the policy are directly rejected and the reasons for rejection are explained. Applications that meet the hard indicators enter the second-level verification stage.

4. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, The specific process of the second layer of AI implicit difficulty identification and risk verification is as follows: a gradient boosting tree model that integrates multiple features is adopted. The input features include the recipient's annual medical expenses, education expenses, disability level of family members, employment status, and real estate and vehicle ownership information. The output is the credibility value of the recipient's eligibility for assistance and the risk level. Those whose credibility value exceeds the set threshold pass the verification directly, while those whose credibility value is lower than the threshold are transferred to the manual review process. Those whose risk level is determined to be suspected of insurance fraud are automatically included in the risk warning list.

5. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, The specific process of the two-way intelligent matching is as follows: First, extract the demand tags from the profile of the recipient, including multiple demands such as medical assistance, temporary assistance, employment assistance, and age-friendly renovations. Then, match the remaining quota and service supply capacity of various assistance projects in the corresponding locality, prioritize matching the assistance resources corresponding to rigid demands, and simultaneously push the generated personalized assistance service list to the recipient's terminal and the local grid worker's work terminal.

6. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, The specific process for re-verifying eligibility for assistance and dynamically adjusting the level of assistance is as follows: set dynamic monitoring rules for low-income assistance on a quarterly basis, temporary assistance on a monthly basis, and special assistance on a project cycle basis. After the expiration, the latest multi-source data of the assistance recipients will be automatically synchronized. For those who no longer meet the assistance conditions, the suspension process will be automatically triggered. For those whose difficulties have increased, the level of assistance will be automatically upgraded. The collected service feedback data will be used as feature input to the matching algorithm to complete iterative optimization.

7. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, The consortium blockchain is jointly constructed by civil affairs, finance, auditing, and various relief business implementation departments as nodes. All data on the blockchain is stored using a multi-node consensus mechanism, which supports the auditing department to retrieve the entire process of relief records with one click to complete the verification, thus eliminating the risk of tampering with relief data and fraudulently obtaining relief funds.

8. The method for digital management and precise service of civil affairs assistance business according to claim 1, characterized in that, It also includes a proactive assistance discovery step: real-time monitoring of abnormal changes in multi-source data, including large medical insurance expenditures, social security suspensions, new disability certificates, and records of sudden disasters, automatically identifying suspected people in need and pushing the list to local grid workers, guiding grid workers to proactively visit people to verify their assistance needs, and realizing the transformation of assistance services from "people looking for policies" to "policies looking for people".

9. A digital management and precision service system for civil affairs assistance, used to implement the digital management and precision service method for civil affairs assistance as described in any one of claims 1-8, characterized in that, include: The multi-source data collection module is used to collect heterogeneous relief-related data from multiple departments and, after standardized cleaning, build a comprehensive profile database of relief recipients. The eligibility verification module is used to verify the hard indicators of policy rules for aid recipients and to identify and verify hidden difficulties and risks using AI. The relief resource matching module is used to perform two-way intelligent matching based on the relief recipient's need tags and the remaining relief resources to generate a personalized relief service list. The dynamic monitoring and adjustment module is used to update data synchronously according to the monitoring cycle, trigger eligibility re-verification and dynamic adjustment of assistance level, and collect service feedback to optimize matching logic; The blockchain evidence storage module is used to synchronously store the data of the entire rescue process to the consortium blockchain nodes with the participation of multiple departments, so as to achieve full traceability and immutability.

10. A digital management and precision service system for civil affairs assistance as described in claim 9, characterized in that, It also includes a proactive discovery module and a visual monitoring module. The proactive discovery module is used to identify abnormal changes in multi-source data and generate a list of suspected people in need, which is then pushed to the grid worker's terminal. The visual monitoring module is used to generate a large visual monitoring screen showing the distribution of relief funds, the distribution of relief recipients, and the use of relief resources, so that management departments can keep abreast of the operational status of relief operations in real time.

Images