Crowdsourcing human platform payment security system and method based on multi-dimensional risk control

Abstract

The invention belongs to the technical field of big data, and particularly relates to a crowdsourcing human platform payment security system based on multi-dimensional risk control, and the system comprises a packet receiving party collection verification module which is used for carrying out identity verification and dynamic transaction verification; the hierarchical transaction restriction engine comprises an evaluation sub-module and a risk control threshold module, and the evaluation sub-module performs evaluation by using a multi-dimensional credit scoring model to obtain a credit score; the risk control threshold submodule is used for performing dynamic threshold calculation in the aspect of risk control from multiple aspects of industry, region and holiday type, and performing credit rating and limit control based on the credit rating by comparing a credit score with the dynamic threshold; the account transfer monitoring system is used for monitoring account transfer behaviors; the system also comprises a geographic position and IP verification module, and a system for verifying the consistency between a packet receiving party and a payee. According to the invention, accurate risk identification and dynamic management and control are realized through technology fusion.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of big data, and particularly relates to a system and method for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control. BACKGROUND

[0002] In the prior art, with the development of big data technology, transaction full-process risk interception is achieved through a multi-dimensional verification mechanism based on big data, which has become an important means to ensure payment security. However, in the prior art, there are still three major risks in the payment link of the crowdsourcing human resource platform: Identity impersonation risk: stolen or false registration of the account of the service receiver, leading to the flow of funds to illegal subjects; Transaction abnormality risk: large amount of transfer during non-working hours, cross-regional transactions and other behaviors may involve money laundering or fraud; Operation compliance risk: inconsistency between the payee and the contract subject may cause tax disputes or legal liability.

[0003] The existing risk control system mostly uses a single rule engine, which has the defects of high misjudgment rate, poor real-time performance and inability to adapt to dynamic scenarios. For example, in the transaction abnormality risk, due to factors such as holidays and shopping festivals, transaction volume and transaction frequency change over time, many human resource risk control systems use static holiday threshold settings, that is, the threshold remains fixed regardless of changes in business volume, personnel flow and other factors. However, the business characteristics and risk status of different holidays, different regions and different industries may differ greatly. For example, during the Spring Festival, the construction industry has the behavior of paying wages at the end of the year, which leads to a significant increase in transaction amount. At this time, using a static threshold for monitoring cannot detect potential risks. SUMMARY

[0004] The purpose of the present application is to provide a system for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control, to solve the technical problems of high misjudgment rate and poor real-time performance caused by the use of a single rule engine and static threshold settings that cannot adapt to dynamic changes in the prior art.

[0005] The system for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control comprises: a service receiver payment verification module for identity verification and dynamic transaction verification; a hierarchical transaction restriction engine including an evaluation submodule and a risk control threshold submodule, the evaluation submodule uses a multi-dimensional credit scoring model for evaluation to obtain a credit score, and the risk control threshold submodule calculates dynamic thresholds from the aspects of industry, region and holiday type, and performs credit level classification and limit control based on the credit level by comparing the credit score with the dynamic threshold; a transfer monitoring system for monitoring transfer behavior; And its geographical location and IP verification module, the package and the payee consistency verification system.

[0006] Preferably, the evaluation submodule: adopts a multi-dimensional credit scoring model, based on each standardized characteristic value, adopts a weighted summation algorithm, calculates the credit score, the calculation formula is: , wherein, is the weight of the ith dimension, is the standardized characteristic value; the data dimension specifically includes a basic data dimension and an external data dimension, the basic data dimension includes a task completion rate, a dispute rate and an evaluation score, and the external data dimension includes central bank credit investigation and industrial and commercial information.

[0007] Preferably, the risk control threshold submodule calculates the dynamic threshold based on industry, region, holiday type and other dimensions, and the risk control threshold submodule includes: a dimension setting unit for setting each dimension; a dynamic threshold calculation engine for calculating a dynamic threshold as a risk control threshold; a security guarantee module: security is guaranteed by delaying the account and manual review.

[0008] Preferably, each dimension includes an industry classification dimension, a regional economic attribute, a holiday nature classification, a holiday cycle characteristic and a holiday influence level; the basic formula of the dynamic threshold calculation engine is: risk control threshold T = basic threshold T0 x industry adjustment coefficient I x regional adjustment coefficient R x holiday adjustment coefficient H x risk correction coefficient V; wherein, the basic threshold T0 refers to the industry general standard or the past historical data of the enterprise, and sets the initial basic threshold for different business scenarios; the risk correction coefficient V is used to cope with sudden risks, policy changes and other special situations, and the initial value is set to 1 and is dynamically adjusted according to the actual situation.

[0009] Preferably, the industry classification dimension includes cash flow characteristics, risk preference and settlement period, the regional economic attribute includes first-tier cities and city groups, new first-tier cities and provincial capitals, and small and medium-sized cities and counties, the holiday nature classification includes traditional consumption type holidays, statutory commemorative type holidays, religious folk holidays and e-commerce artificial holidays, the holiday cycle characteristic includes long holidays and short holidays, and the holiday influence level includes national major holidays and regional characteristic holidays; the calculation formula of the industry adjustment coefficient I is: industry adjustment coefficient I = cash flow characteristic coefficient I xj x risk preference coefficient I fx x settlement period coefficient I zq ; the calculation formula of the regional adjustment coefficient R is: regional adjustment coefficient R = geographical economic attribute coefficient R dl x special area difference coefficient R sp ; the calculation formula of the holiday adjustment coefficient H is: holiday adjustment coefficient H = holiday nature coefficient H nHoliday cycle coefficient H c Holiday influence coefficient H i .

[0010] Preferably, the transfer monitoring system comprises the following functions: Time-amount matrix: the monitoring amount threshold is adjusted according to the time period to form a time-amount matrix; Behavioral anomaly detection: including transaction frequency and fund flow direction anomaly detection; Transaction frequency early warning: statistics the user's daily transaction frequency and total amount, and calculates the transaction frequency index; set the historical baseline of 3 times the standard deviation of the transaction frequency index as the threshold for triggering high-frequency transaction early warning; Fund flow tracking: correlation analysis is performed on the payee of large amount transfer, if the payee is a newly registered account or has no historical transaction record, the blocking is triggered.

[0011] The application also provides a method for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control, which adopts the system for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control as described above, and comprises a preparation process before threshold calculation, a threshold calculation process and a management process after threshold calculation; wherein the preparation process before threshold calculation comprises: a first step of data acquisition and preprocessing, and a second step of parameter setting and model initialization; the threshold calculation process comprises: a first step of preliminary calculation of a calculation model, a second step of calculation of various adjustment coefficients, and a third step of final threshold calculation; the management process after threshold calculation comprises: monitoring and dynamic adjustment, special situation response and threshold correction, and effect evaluation and optimization iteration.

[0012] Preferably, in the threshold calculation process, the second step comprises static interval calculation and dynamic coefficient calculation, which comprises: Holiday nature coefficient H n : the holiday nature is classified, and the static interval is obtained through the benchmark interval calculation algorithm, but the corresponding data source and dynamic adjustment formula are selected according to the holiday nature; Holiday cycle coefficient H c : the holiday cycle is determined according to the holiday name within a certain time range, and the holiday cycle is classified into a long holiday model and a short holiday model, and the date is used to determine which stage the current time belongs to, and the static interval of different stages of different holiday nature types is calculated based on historical data and the benchmark interval calculation algorithm a imin , a imax ], i represents the i th stage, the holiday cycle coefficient H c的 Dynamic coefficient calculation is divided into two cases, in the growth state, the holiday cycle coefficient Hc = a imax × Period Effective Coefficient, in the reduced state, holiday period coefficient H c = a imin × Period Effective Coefficient; Period Effective Coefficient = T*+ Lambda* Delta t , T* is a static reference value, Lambda is a correction coefficient, Delta t is the difference between the current time in the corresponding stage and the earliest time threshold, and an adjustment amplitude constraint is also introduced, setting the upper limit of the adjustment amplitude of the static interval and the lower limit of the adjustment amplitude.

[0013] Preferably, in the threshold calculation process, the second step further comprises: influence level coefficient H i : according to the influence factors of the holiday on the influence level, the influence level of the holiday is determined, the corresponding static interval is calculated based on the type through the reference interval calculation algorithm, and the corresponding static basic coefficient is further calculated based on the lower limit of the static interval. The basic coefficient is required to be upper limited, that is, not greater than the upper limit constraint of the static interval, and finally, the dynamic coefficient calculation of the influence level coefficient: H i = Static Basic Coefficient * Dynamic Factor; the dynamic factor is obtained according to historical data analysis.

[0014] The advantages of the present application lie in that the present application constructs a five-dimensional verification system (receiver payment verification, hierarchical transaction restriction, non-working time large amount transfer monitoring, geographic location and IP verification, and payee consistency verification), and realizes precise risk identification and dynamic control through technology fusion.

[0015] In the hierarchical transaction restriction, in view of the problem that the static threshold value is difficult to adapt to changes, the present application adopts targeted dynamic threshold adjustment, especially for holidays. Since the risk may differ greatly due to different holidays, different regions, and different business characteristics and risk conditions of different industries, the present application analyzes from three aspects of holiday nature classification, holiday period, and holiday influence, and sets corresponding classification, static interval calculation based on classification, and dynamic adjustment formula to dynamically calculate the threshold value. In combination with the dynamic coefficient calculation of industry adjustment coefficient, regional adjustment coefficient, etc., the final threshold value finally calculated in the present application can be dynamically adjusted and calculated based on changes in multiple factors, has good flexibility and adaptability, and can effectively play a risk early warning role. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The basic flowchart of the system and method for payment security of the crowdsourcing human resource platform based on multi-dimensional risk control in the present application.

[0017] Figure 2Flow chart for the preparation process before the holiday threshold calculation in the present application.

[0018] Figure 3 Flow chart for the preparation process before the holiday threshold calculation in the present application.

[0019] Figure 4 Flow chart for the management process after the holiday threshold calculation in the present application. DETAILED DESCRIPTION

[0020] The specific embodiments of the present application will be further described in detail below with reference to the accompanying drawings, and the description of the embodiments will help the skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solutions of the present application.

[0021] As shown in Figures 1-4 The present application provides a crowdsourcing human resource platform payment security system based on multi-dimensional risk control, which includes: The order taker payment verification module is used for identity verification and dynamic transaction verification.

[0022] The hierarchical transaction restriction engine uses a multi-dimensional credit scoring model for evaluation and restricts the amount based on dynamic threshold.

[0023] The transfer monitoring system monitors the transfer behavior.

[0024] And other verification modules, such as the geographic location and IP verification module for verifying the geographic location and IP; the order taker and payee consistency verification system for a series of consistency verification of the order taker and payee.

[0025] I. The order taker payment verification module includes the following functions: (1) Dual identity binding: 1.1 Biological feature binding: complete bank card four elements + face liveness detection (3D structured light technology anti-fake) during registration, establish strong association of biological features-payment account-bank card.

[0026] 1.2 Device fingerprint binding: collect device IMEI, MAC address, SIM card information, generate unique device identifier, and prohibit cross-device login.

[0027] (2) Dynamic transaction verification: forcedly start SMS verification for high-risk transactions (such as single transaction over 50,000 yuan).

[0028] II. Grading transaction restriction engine: among the above modules, the grading transaction restriction engine is the core part of the system, including the evaluation sub-module and the risk control threshold module. The evaluation sub-module uses a multi-dimensional credit scoring model for evaluation to obtain a credit score. The risk control threshold sub-module calculates the dynamic threshold value from the aspects of industry, region, and holiday type. The credit level is classified by comparing the credit score with the dynamic threshold value, and the limit control is based on the credit level.

[0029] (1) Evaluation sub-module: a multi-dimensional credit scoring model is used, and a weighted summation algorithm is used based on the standardized characteristic values to calculate the credit score. The calculation formula is: wherein, is the weight of the ith dimension, is the standardized characteristic value (with a value range of 0-1).

[0030] The relevant data dimensions include: 1.1 Basic data dimension (65%): including: Task completion rate (30%): calculated by historical task delivery rate, project acceptance rate, etc.

[0031] Dispute rate (25%): statistics of the proportion of disputed orders in the user's historical transactions, using a sliding window algorithm (such as the last 3 months) to smooth abnormal fluctuations.

[0032] Evaluation score (10%): comprehensive platform user rating, customer satisfaction survey data, etc., using NLP to extract evaluation keywords sentiment.

[0033] 1.2 External data dimension (35%): including: Central bank credit information (20%): access to overdue records, debt ratio, etc. in personal credit reports, using Logistic regression model to quantify risk.

[0034] Business information (15%): enterprise subcontractors need to verify business registration status, shareholder structure changes, etc., using graph algorithm to detect associated risks.

[0035] (2) Risk control threshold sub-module: dynamic threshold processing and further determination of the corresponding relationship between credit level and limit control through dynamic limit strategy.

[0036] In dynamic threshold processing, the current human crowd sourcing platform risk control system has the following problems: 1) Lack of flexibility in threshold setting: Static thresholds are difficult to adapt to changes: Many human risk control systems use static holiday threshold settings, which remain fixed regardless of changes in business volume, personnel flow, and other factors. However, business characteristics and risk conditions may vary greatly during different holidays, in different regions, and in different industries. For example, during the Spring Festival, the construction industry experiences a significant increase in transaction volume due to year-end wage payments. Using static thresholds for monitoring during this period cannot detect potential risks.

[0037] 2) Data accuracy and timeliness issues: Data delays affect judgment: Data updates for human risk control systems may be delayed, especially during holidays, due to adjustments in business processes or the absence of data processing personnel. As a result, data may not accurately reflect actual conditions in a timely manner.

[0038] 3) Limited reference to historical data: When setting holiday thresholds, historical data is often referenced, but historical data may not fully cover the special circumstances of the current holiday. For example, an enterprise may have launched a new business this year, which has different human resource needs and risk characteristics during the holiday than in the past. Setting thresholds based solely on historical data may not effectively monitor risks associated with the new business. 4) Lack of consideration for special circumstances: Inadequate response to emergencies: Various emergencies may occur during holidays, such as natural disasters and public health events, which can significantly impact an enterprise's human resources and render normal thresholds inapplicable. For example, during the pandemic, an enterprise may face challenges such as employees unable to report to work and a sharp decline in business volume. If the holiday threshold of the risk control system does not take into account such special circumstances, it will be difficult to effectively perform risk early warning.

[0039] To address the above issues, the risk control threshold submodule in the present scheme performs detailed dynamic threshold calculation based on industry, region, holiday type, and other dimensions. The risk control threshold submodule includes the following content.

[0040] 2.1 Dimension setting unit, each dimension is set as follows.

[0041] (I) Industry classification dimension: Cash flow characteristics: including high turnover (retail / logistics), project-based (construction / design), and seasonality (agriculture / tourism). Risk preference: including high fraud rate (digital content crowdsourcing) and low risk (physical services). Settlement period: including T+0 (immediate service) and monthly settlement (long-term projects).

[0042] (II) Regional economic attributes: Tier-1 cities and urban clusters: such as Beijing-Tianjin-Hebei, Yangtze River Delta, Guangdong-Hong Kong-Macao Greater Bay Area, with highly developed economy and high industrial concentration, high labor cost and large business scale. Enterprises have strong liquidity, high payment frequency, large amount, and threshold setting needs to reflect higher upper limit and more flexible fluctuation range. For example, financial enterprises in Beijing and Shanghai can increase the payment threshold by 30%-50% during holidays to meet the demand of high-frequency large transactions. New first-tier cities and provincial capitals: economic development is active, emerging industries and traditional industries coexist, and business stability is relatively strong, but the impact of regional economic cycle on capital fluctuation. Threshold setting is 70%-80% of that in first-tier cities, and industry characteristics weight is increased, such as Hangzhou (Internet industry is prominent) to digital economy enterprises threshold gives 10%-15% elastic adjustment. Small and medium-sized cities and counties: local consumption and basic industries are the main, small scale of capital flow, low frequency, and relatively low risk sensitivity. Threshold benchmark value is set to 40%-60% of that in first-tier cities, focusing on protecting basic operating capital needs, such as county retail enterprises holiday payment threshold pays more attention to consumer goods procurement funds.

[0043] (Three) Holiday nature classification: Traditional consumption holidays: Spring Festival, Mid-Autumn Festival, Double Eleven, etc., consumption demand explosion, retail, e-commerce, logistics industry business surge. Threshold is set at 120%-150% of the historical peak value, and dynamic monitoring is started 1-2 months in advance, and real-time adjustment is made according to pre-sale data and consumption trends. Statutory commemorative holidays: National Day, Labor Day, etc., some industries stop work or business volume decreases, but government projects and public service enterprises need to ensure basic operation. Threshold is set at 80%-100% of the regular working day, focusing on emergency fund protection. Religious or folk holidays: such as Tibetan New Year, etc., regional influence is significant. In areas inhabited by ethnic minorities, the threshold of related enterprises is adjusted according to holiday arrangements and business characteristics, such as tourism enterprises threshold can be increased by 20%-30%.

[0044] (Four) Holiday cycle characteristics: Long holiday (7 days and above): Spring Festival, National Day, etc., enterprises need to reserve funds in advance to cope with the holiday and the business peak before and after the holiday. Threshold starts pre-adjustment 15 days before the holiday, set at 130% of the historical average, and gradually returns to normal after the holiday. Short holiday (3 days): Qingming, Dragon Boat Festival, etc., business fluctuation is relatively small, threshold adjustment range is controlled within ±10%, focusing on monitoring liquidity risk.

[0045] (Five) Holiday influence level: National major holidays: The impact covers the whole country, and all industries need to assess the risk comprehensively. The threshold adjustment uses a comprehensive calculation of "base value + industry coefficient + regional coefficient", for example, the threshold of national retail enterprises = base value x 1.2 (consumption holiday coefficient) x regional adjustment coefficient. Regional characteristic holidays: Only have an impact on specific regions, such as the impact of Water-Splashing Festival on Yunnan tourism. Threshold adjustment is only for the affected regions and industries, and the adjustment range is determined according to historical data of the holiday and market forecast this year.

[0046] 2.2 Dynamic threshold calculation engine for calculating dynamic thresholds as risk control thresholds.

[0047] The basic formula of the dynamic threshold calculation engine is: risk control threshold T = base threshold T0 x industry adjustment coefficient I x regional adjustment coefficient R x holiday adjustment coefficient H x risk correction coefficient V. Among them, the base threshold T0 is set according to the industry general standard. Further specific description is as follows.

[0048] 2.2.1 Risk correction coefficient V is used to deal with sudden risks, policy changes and other special circumstances, and the initial value is set to 1, which is dynamically adjusted according to actual conditions.

[0049] 2.2.2 The calculation formula of industry adjustment coefficient I is: industry adjustment coefficient I = cash flow characteristic coefficient I xj x risk preference coefficient I fx x settlement period coefficient I zq .

[0050] The above coefficients for calculation are determined according to the type respectively, and the specific setting method is as follows: The type of cash flow characteristic coefficient (I xj ) includes high turnover industry (retail / logistics), project-based industry (construction / design), and seasonal industry (agriculture / tourism), among which seasonal industry has different value ranges according to off-season and peak season.

[0051] The type of risk preference coefficient (I fx ) includes high fraud rate industry (digital content crowdsourcing) and low risk industry (physical service).

[0052] The type of settlement period coefficient (I zq ) includes T+0 (immediate service) and monthly settlement (long-term project).

[0053] 2.2.3 The calculation formula of regional adjustment coefficient R is: regional adjustment coefficient R = geographical economic attribute coefficient R dl x special area difference coefficient R sp .

[0054] The coefficients used in the above calculation are determined according to the type respectively, and the specific setting mode is as follows: Geoeconomic attribute coefficient (R dl ) types include first-tier cities and urban agglomerations, new first-tier cities and provincial capitals, and small and medium-sized cities and counties.

[0055] Special area difference coefficient (R sp ) types include policy experimental areas (free trade zones and national new areas) and ordinary administrative areas.

[0056] The holiday adjustment coefficient H calculation formula in Section 2.2.4 is: Holiday adjustment coefficient H = Holiday nature coefficient (H n ) x Holiday cycle coefficient (H c ) x Holiday influence coefficient (H i ). The coefficients used in the above calculation are determined according to the type respectively, and the specific setting mode is as follows: Holiday nature coefficient (H n ) types include traditional consumption holidays (Spring Festival, Mid-Autumn Festival, Double Eleven), statutory commemorative holidays (National Day, Labor Day), and religious / folk holidays (Tibetan New Year), among which the coefficient of religious / folk holidays is only adjusted in the affected area during the corresponding holiday.

[0057] Holiday cycle coefficient (H c ) types include long holidays (7 days and above) and short holidays (3 days).

[0058] Holiday influence coefficient (H i ) types include national major holidays and regional characteristic holidays, and regional characteristic holidays are only adjusted in the affected area and industry.

[0059] Three, the transfer monitoring system includes the following functions: (1) Time-amount matrix: The monitoring amount threshold is adjusted according to the time period, for example, define 22:00-6:00 as non-working hours and belong to high-risk period, the monitoring amount threshold is 50,000, that is, single transfer over 50,000 yuan triggers an early warning. Define weekends and holidays as special periods, and the monitoring amount threshold is 30,000.

[0060] (2) Behavior anomaly detection: including transaction frequency and fund flow direction two aspects of anomaly detection. Specifically as follows.

[0061] Transaction frequency warning: the window is N days (such as 7 days), the daily transaction frequency and the total amount of the user are counted, the transaction frequency index is calculated, and the calculation formula is as follows: , Threshold setting: set the historical baseline 3 times standard deviation as the threshold, if the daily trading frequency index exceeds the historical baseline 3 times standard deviation, trigger high-frequency trading warning.

[0062] Fund flow tracking: correlation analysis of large amount of transfer recipients, if the recipient is a newly registered account or has no historical transaction record, trigger blocking.

[0063] (3) Security module: security is ensured by delaying the account and manual review. For high-risk transactions, the monitoring system sets a 24-hour delay in the account, which can be initiated through the APP during the period. For transactions exceeding the threshold, the monitoring system will push them to the risk control officers for manual review (need to upload contracts, invoices and other documents).

[0064] Four, other verification modules: (1) Geolocation and IP verification module includes multi-source positioning fusion, anti-proxy attack and device / IP association analysis functions.

[0065] Multi-source positioning fusion: through GPS positioning, base station positioning and other multi-source positioning technologies combined with IP address library to analyze geographical location, realize positioning data fusion and determine geographical location.

[0066] ① GPS positioning: real-time latitude and longitude are obtained through the APP, with an accuracy of ±5 meters.

[0067] ② Base station positioning: combined with operator base station information, GPS blind area is corrected.

[0068] ③ IP address library: call third-party IP library (such as MaxMind) to analyze geographical location and calculate confidence.

[0069] Anti-proxy attack: including IP reputation library dynamic update and proxy penetration detection technology.

[0070] ① IP reputation library dynamic update: access third-party threat intelligence (such as FireholIP blacklist) to update high-risk IP list in real time; ② Proxy penetration detection: identify proxy servers through HTTP headers (such as X-Forwarded-For), DNS resolution delay, TCP connection fingerprint and other multi-dimensional.

[0071] Device / IP association analysis: check if the transaction is consistent with the user's common device and IP address. If the device fingerprint and IP address of a transaction have no intersection with historical records, or are located in high-risk areas (such as VPN, proxy server), device / IP risk verification is triggered.

[0072] (2) Consistency verification system of package recipient and payee: four elements comparison, intelligent contract analysis and exception scenario management to determine the consistency of package recipient and payee.

[0073] Four key elements of strong comparison: In the payment link, the name, ID number, bank card number, and bank of the payee are checked for consistency with the contract subject. For inconsistent transactions, automatic interception and push to the risk control officer for manual review.

[0074] (2) Exception scenario management: including relatives' collection, enterprise sub-account and cross-border payment Relatives' collection: relationship proof documents (household register, notarization) need to be uploaded and manually reviewed; Enterprise sub-account: sub-account information needs to be recorded on the platform, and a white list mechanism needs to be established; Cross-border payment: the platform does not allow cross-border payment at present.

[0075] The present application also discloses a method for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control, which is realized based on the above-mentioned system for payment security of a crowdsourcing human resource platform based on multi-dimensional risk control.

[0076] The method comprises the following steps: after the contractor sends a payment request to the crowdsourcing human resource platform, the crowdsourcing human resource platform mechanically performs risk control verification, and then performs the following steps in sequence: payee collection verification, hierarchical transaction restriction, non-working time large amount transfer monitoring, geographic location and IP verification, and consistency verification between the receiver and the payee. If all the verification steps are passed, the payment is automatically made immediately; if there is a verification step that fails, the whole process of fee payment is realized by combining delayed account and manual review.

[0077] In the threshold calculation process of the hierarchical transaction restriction in the above-mentioned method, the process comprises a preparation process before threshold calculation, a threshold calculation process, and a management process after threshold calculation.

[0078] The preparation process before threshold calculation comprises the following steps: First step: data collection and preprocessing.

[0079] (1) Multi-source data acquisition: through external data sources such as the official website of the statistics bureau, industry association platform, and policy and regulation release website, external data such as regional economic indicators, industry risk data, and holiday-related policies are obtained to build a comprehensive data base.

[0080] (2) Data cleaning and integration: use data cleaning algorithms to process the collected data, identify and eliminate duplicate data, error data (such as negative payment amount) and abnormal data beyond the normal range of enterprise business; according to the unified data standard and format, integrate the multi-source data into the data warehouse to ensure the accuracy, integrity and standardization of the data.

[0081] Second step: parameter setting and model initialization.

[0082] (I) Basic threshold determination: reference industry general standards or enterprise past historical data, set the initial basic threshold (T0) for different business scenarios. For example, according to the industry average level, set a benchmark limit for the payment risk of a certain type of enterprise as the basic threshold.

[0083] (II) Initial configuration of coefficients: According to the classification rules of industry, region, and holiday type, give initial values to the adjustment coefficients of each dimension. For example, set the cash flow characteristic coefficient (I) of high turnover industry in the range of 1.2-1.5, and set the geographical and economic attribute coefficient (R) of first-tier cities and urban agglomerations in the range of 1.2-1.5, and clearly define the initial value of risk correction coefficient V as 1.

[0084] (III) Deployment of calculation model: deploy the basic formula of risk control threshold and the sub-computation formula of each dimension coefficient to the dynamic threshold calculation engine to complete model initialization. The basic formula is as follows: risk control threshold T = basic threshold T0 x industry adjustment coefficient I x regional adjustment coefficient R x holiday adjustment coefficient H x risk correction coefficient V.

[0085] The threshold calculation process includes the following steps: First step: preliminary calculation of calculation model: static benchmark interval calculation, through the analysis of industry historical data, using kernel density estimation method to determine the safe transaction frequency range. The core code is as follows: def calc_static_range(data): # Kernel density estimation kde = gaussian_kde(data) x = np.linspace(min(data), max(data), 1000) pdf = kde(x) # Find 95% confidence interval cdf = np.cumsum(pdf) / np.sum(pdf) lower = x[np.where(cdf>=0.025)[0][0]] upper = x[np.where(cdf>=0.975)[0][0]] return (lower, upper)。

[0086] Second step: calculation of each type of adjustment coefficient: including static interval calculation and dynamic coefficient calculation, the following will explain each adjustment coefficient.

[0087] (1) Cash flow characteristic coefficient I xjStatic interval [ is derived by the benchmark interval calculation algorithm a min , a max ].

[0088] Dynamic coefficient calculation of cash flow characteristic coefficient: I xj = clamp (base coefficient x dynamic adjustment factor, a min , a max ), the clamp function limits the value within the specified range, the base coefficient= a min + 0.03 x (actual transaction frequency-50), and the dynamic adjustment factor=1-0.5x transaction volatility.

[0089] (2) Risk preference coefficient I fx : Static interval [ is derived by the benchmark interval calculation algorithm a min , a max ]. The data used is the industry historical dispute rate data (at least 3 years) obtained from the fraud event database (including case amount, type, disposal result).

[0090] Define risk levels based on dispute rate size and divide the corresponding coefficient interval, for example, high risk, medium risk and low risk, corresponding coefficient intervals are[ b 1min , b 1max ], [ b 2min , b 2max ] and[ b 3min , b 3max ].

[0091] Dynamic coefficient calculation of risk preference coefficient: I fx = b imin +Δ b i x (1-dispute rate / 10%), b imin is the lower limit of the coefficient interval, and Δ b i is the upper and lower limit difference value of the coefficient interval. Taking the dispute rate of 7% as an example, first determine the risk level (high risk) according to the dispute rate, and then calculate the risk preference coefficient I fx according to the corresponding coefficient interval.

[0092] (3) Settlement cycle coefficient Izg : Settlement cycle is divided into T+0 (immediate service) and monthly settlement (long-term project) two categories, according to the actual situation directly determine the corresponding type of benchmark coefficient.

[0093] Dynamic coefficient calculation of settlement cycle coefficient: I zg = benchmark coefficient x (1 + account period deviation), account period deviation = (standard account period - actual account period) / standard account period. To prevent data errors, set the constraint condition: account period shortening (actual < standard): deviation > 0, coefficient up (maximum + 20%); account period extension (actual > standard): deviation < 0, coefficient down (minimum - 30%).

[0094] (4) Geoeconomic attribute coefficient R dl : Through the benchmark interval calculation algorithm to obtain static interval[ a min , a max ]. Based on the economic data of the National Bureau of Statistics, the city is divided into multiple levels, and a number of corresponding coefficient intervals[ b imin , b imax ] are divided, i representing the i coefficient interval.

[0095] Dynamic coefficient calculation of geoeconomic attribute coefficient: R dl = b imin x (1 + economic vitality factor), economic vitality factor = w1 x (GDP growth rate / national average growth rate) + w2 x (consumption growth rate / 10%) + w3 x (newly registered enterprises / industry average). w1, w2 and w3 are the weights corresponding to the relevant factors.

[0096] Constraint condition for dynamic coefficient calculation: economic vitality factor range: -0.15 to +0.25, final R not more than ±20% of static interval.

[0097] (5) Special area difference coefficient R sp : Set rules table, related content includes policy type, benchmark addition, superposition rules, effective range, determine the last three according to policy type. Through the monitoring of policy effective period, select the maximum number of benchmark addition (3) policy addition, addition based on superposition rules, and control the number of policies to prevent excessive superposition. Through the program, the policy addition is automatically completed, the input is the policy list enjoyed by the enterprise (including type, value, effective period), and the output is the comprehensive policy addition coefficient. The specific procedure is as follows: Policy effective period monitoring: def policy_factor(policies): active_policies = [p for p in policies if p['expire_date']>today] # Take the maximum 3 policy bonuses (prevent over stacking) sorted_policies = sorted(active_policies, key=lambda x:x['value'],reverse=True)[:3] return 1.0 + sum([p['value'] for p in sorted_policies]) / 100.

[0098] (6) Holiday nature coefficient H n : The holiday nature is classified, and the static interval is obtained by the benchmark interval calculation algorithm, but the corresponding data source is selected according to the holiday nature, and the dynamic coefficient calculation uses the dynamic adjustment formula which is different according to the holiday nature, and the specific is shown in Table 1.

[0099] Table 1: Dynamic coefficient calculation method table based on holiday nature type

[0100] (7) Holiday period coefficient H c : According to the holiday name in a certain time range, the holiday period is determined and classified into long holiday model and short holiday model, and according to the date, it is determined which stage of the holiday period the current time belongs to, and the stage is divided into four: 15-8 days before the holiday, 7-1 day before the holiday, during the holiday, and 1-7 days after the holiday. The transaction volatility and risk occurrence rate generated thereby are determined by analyzing the historical behavior of the corresponding time period. The data source used in the above analysis is the UnionPay daily transaction report and the e-commerce real-time monitoring system.

[0101] The static interval of different holiday stages for different holiday nature types is calculated based on historical data and benchmark interval calculation algorithm a imin , a imax ], i represents the i th stage, in the growth state, the dynamic coefficient calculation of the geographical economic attribute coefficient: holiday period coefficient H c = a imax ×period effective coefficient, in the reduction state, the dynamic coefficient calculation of the geographical economic attribute coefficient: holiday period coefficient H c = a iminX period effective coefficient. Period effective coefficient = T*+λ×Δ t , T* is a static reference value, λ is a correction coefficient, Δ t is the difference between the time interval of the current time in the corresponding stage and the earliest time threshold value, for example, if the current time is 10, it belongs to the stage of 15-8 days before the holiday, then Δ t =15-10=5. λ takes 0.02, T* takes 1.0. Here, an adjustment range constraint is also introduced, the upper limit of the adjustment range of the upper limit of the static interval is 1.2, and the lower limit of the adjustment range of the lower limit of the static interval is 0.8, that is, the calculation result of the period effective coefficient is greater than 1.2, taking 1.2, and the calculation result of the period effective coefficient is less than 0.8, taking 0.8.

[0102] (8) Influence level coefficient H i : According to the influence factors of holidays on the influence level, a comprehensive threshold value is determined, for example, the influence factors of national major holidays include the number of national holidays, the amount of cross-provincial population flow, and the amount of national media special reports; The influence factors of regional characteristic holidays include whether the provincial government organizes it officially, the proportion of regional participating population, and the local media heat index; The influence factors of industry vertical holidays include the TOP3 industry transaction volume growth rate, industry exclusive promotion budget, and vertical field search volume proportion. After determining the influence level of the holiday, the corresponding static interval is calculated based on the type by the reference interval calculation algorithm[ a imin , a imax ], i , which represents the i th influence level, and further based on the lower limit of the static interval, the corresponding static basic coefficient is calculated, and the algorithm is as follows according to the type: ① National major holidays: static basic coefficient = a imin +0.3×(cross-provincial flow / 100 million); ② Regional characteristic holidays: static basic coefficient = a imin +0.3×(local culture budget / provincial average); ② Industry vertical holidays: static basic coefficient = a imin +0.3×(industry transaction proportion / 50%).

[0103] The above basic coefficients are required to be less than or equal to a imax , that is, the corresponding upper limit constraint is carried out. Finally, the dynamic coefficient calculation of the influence level coefficient is H i =static basic coefficient×dynamic factor. The dynamic factor is obtained according to historical data analysis.

[0104] Step 3: Final threshold calculation: Substitute the calculated industry adjustment factor I, regional adjustment factor R, holiday adjustment factor H, and the pre-set base threshold T0 and risk correction factor V into the base formula of the risk control threshold, to calculate the final risk control threshold T as the standard for human resource risk control during the current holiday period.

[0105] The management process after threshold calculation includes the following: (I) Monitoring and dynamic adjustment: including real-time data monitoring, regular dynamic adjustment, and predictive adjustment Real-time data monitoring: Continuously collect enterprise business data and compare it with the calculated threshold.

[0106] Regular dynamic adjustment: When the deviation between actual business data and threshold exceeds the set range (such as ±10%), automatically trigger the threshold adjustment process. This process recalculates the adjustment factors of each dimension and updates the risk control threshold. For example, if the actual payment amount exceeds the threshold by 12%, the system recalculates the industry, regional, and holiday adjustment factors to obtain a new threshold to adapt to business changes.

[0107] Predictive adjustment: Use machine learning algorithms (such as LSTM, ARIMA) to deeply analyze historical holiday data and predict future business trends and risk changes. Before the holiday arrives, adjust the threshold in advance based on the prediction results.

[0108] (II) Special situation response and threshold correction: including emergency response and policy and regulation change handling.

[0109] Emergency response: When monitoring natural disasters, public health events, and other emergencies, the system automatically starts the emergency threshold adjustment process. According to the impact range and degree of the event, adjust the risk correction factor V and then recalculate the risk control threshold. For example, during the epidemic, for enterprises in severely affected areas, set the risk correction factor V to 0.7, significantly lower the payment threshold, and strictly control the risk of capital outflow.

[0110] Policy and regulation change handling: Through the policy and regulation monitoring module, real-time track changes in national and local policies and regulations. When policies related to labor regulations, tax policies, and other factors affecting enterprise human cost and capital flow are introduced, the system automatically analyzes the policy impact and adjusts the relevant threshold. For example, after the introduction of tax preferential policies, according to the cost saved by enterprises, appropriately increase the payment threshold to support enterprises to expand their operations.

[0111] (III) Effect evaluation and optimization iteration: Effect evaluation: Establish an evaluation system including risk warning accuracy, capital risk control effect, enterprise operation efficiency and other indicators to evaluate the implementation effect of the scheme regularly.

[0112] Optimization iteration: according to the evaluation results and enterprise feedback, the coefficient value, calculation formula and data acquisition dimension are optimized, and the threshold calculation model is improved.

[0113] The application is described above according to the drawings, and it is obvious that the specific implementation of the application is not limited by the above manner, as long as various non-essential improvements are made by adopting the inventive concept and technical solution of the application, or the inventive concept and technical solution of the application is directly applied to other occasions without improvement, which is within the protection scope of the application.

Claims

1. A system for payment security of crowdsourced labor platforms based on multi-dimensional risk control, characterized by: include: The payment verification module for the receiving party is used for identity verification and dynamic transaction verification. The tiered transaction restriction engine includes an evaluation submodule and a risk control threshold module. The evaluation submodule uses a multi-dimensional credit scoring model to evaluate and obtain a credit score. The risk control threshold submodule calculates dynamic thresholds for risk control from multiple aspects such as industry, region, and holiday type. It classifies credit levels and controls limits based on credit levels by comparing credit scores with dynamic thresholds. A transfer monitoring system monitors transfer activities. It also includes its geolocation and IP verification module, and a system for verifying the consistency between the recipient and the payee.

2. The system for payment security of crowdsourced human resource platforms based on multi-dimensional risk control according to claim 1, characterized in that: Evaluation Submodule: Employs a multi-dimensional credit scoring model. Based on the standardized feature values, a weighted summation algorithm is used to calculate the credit score. The calculation formula is as follows: ,in, Let i be the weight of the i-th dimension. These are standardized feature values; the data dimensions specifically include basic data dimensions and external data dimensions. The basic data dimensions include task completion rate, dispute rate, and evaluation score, while the external data dimensions include central bank credit information and business registration information.

3. The system for payment security of a crowdsourced human resource platform based on multi-dimensional risk control as described in claim 2, characterized in that: The risk control threshold submodule performs detailed dynamic threshold calculations based on various dimensions such as industry, region, and holiday type. The risk control threshold submodule includes: The dimension setting unit is used to set the various dimensions; Dynamic threshold calculation engine, used to calculate dynamic thresholds as risk control thresholds; Security module: Security is ensured through delayed payment processing combined with manual review.

4. The system for payment security of a crowdsourced human resource platform based on multi-dimensional risk control as described in claim 2 or 3, characterized in that: The various dimensions include industry classification, regional economic attributes, holiday nature classification, holiday cycle characteristics, and holiday influence levels. The basic formula of the dynamic threshold calculation engine is: Risk control threshold T = Basic threshold T0 × Industry adjustment coefficient I × Regional adjustment coefficient R × Holiday adjustment coefficient H × Risk correction coefficient V. Among them, the basic threshold T0 refers to the industry general standard or the company's past historical data to set the initial basic threshold for different business scenarios. The risk correction coefficient V is used to deal with special situations such as sudden risks and policy changes. The initial value is set to 1 and is dynamically adjusted according to the actual situation.

5. The system for payment security of a crowdsourced human resource platform based on multi-dimensional risk control according to claim 4, characterized in that: Industry classification dimensions include cash flow characteristics, risk appetite, and settlement cycle; regional economic attributes include first-tier cities and urban clusters, new first-tier cities and provincial capitals, and small and medium-sized cities and counties; festival nature classification includes traditional consumer festivals, statutory commemorative festivals, religious and folk festivals, and e-commerce-manufactured festivals; festival cycle characteristics include long holidays and short holidays; and festival influence levels include major national festivals and regional characteristic festivals. The formula for calculating the industry adjustment coefficient I is: Industry Adjustment Coefficient I = Cash Flow Characteristic Coefficient I xj × Risk Preference Coefficient I fx ×Settlement Cycle Coefficient I zq The formula for calculating the regional adjustment coefficient R is: Regional adjustment coefficient R = Geographical economic attribute coefficient R dl ×Special regional difference coefficient R sp The formula for calculating the holiday adjustment factor H is: Holiday adjustment factor H = Holiday nature factor H n ×Holiday Cycle Coefficient H c ×Festival Influence Coefficient H i .

6. The system for payment security of a crowdsourced human resource platform based on multi-dimensional risk control according to claim 1, characterized in that: The transfer monitoring system includes the following functions: Time-Amount Matrix: The monitoring amount threshold is adjusted according to the time period to form a time-amount matrix; Anomaly detection includes anomaly detection in two aspects: transaction frequency and fund flow. Transaction frequency alert: Calculate the transaction frequency index by summing the number and amount of daily transactions for each user; set a threshold of 3 times the historical baseline of the transaction frequency index to trigger a high-frequency transaction alert; Funds flow tracking: Correlation analysis is performed on the recipients of large transfers. If the recipient is a newly registered account or has no historical transaction records, the transfer will be blocked.

7. A method for payment security in crowdsourced labor platforms based on multi-dimensional risk control, characterized by: The system for payment security of a crowdsourced human resource platform based on multi-dimensional risk control, as described in any of claims 1-6, includes a preparation process before threshold calculation, a threshold calculation process, and a management process after threshold calculation. The preparation process before threshold calculation includes: Step 1: Data collection and preprocessing; Step 2: Parameter setting and model initialization. The threshold calculation process includes: Step 1: Preliminary calculation of the calculation model; Step 2: Calculation of various adjustment coefficients; Step 3: Final threshold calculation. The management process after threshold calculation includes: monitoring and dynamic adjustment, handling of special situations and threshold correction, and effect evaluation and optimization iteration.

8. The method for payment security of crowdsourced human resource platforms based on multi-dimensional risk control according to claim 7, characterized in that: The threshold calculation process, in its second step, includes two parts: static interval calculation and dynamic coefficient calculation. These include: Holiday nature coefficient H n The nature of holidays is classified, and static intervals are obtained through a baseline interval calculation algorithm. However, the corresponding data source and dynamic adjustment formula are selected according to the nature of the holidays. Holiday cycle coefficient H c The system determines the holiday cycle based on the holiday names within a certain time range at the current time, classifying them into long holiday models and short holiday models. It also determines which stage of the holiday cycle the current time belongs to based on the date, and calculates the static intervals for different stages of holidays of different natures based on historical data and a baseline interval calculation algorithm. a imin , a imax ], i Indicates the first i Each stage, the festival cycle coefficient H c的 The dynamic coefficient calculation is divided into two cases: in the growth state, the holiday cycle coefficient H... c = a imax ×Effective coefficient of the cycle, when in a reduced state, the festival cycle coefficient H c = a imin ×Effective coefficient of period; Effective coefficient of period = T* + λ×Δ t T* is the static baseline value, λ is the correction coefficient, and Δ t It is the difference between the current time and the earliest time threshold in the corresponding time interval of the current stage. At the same time, an adjustment range constraint is introduced, setting the upper limit and lower limit of the adjustment range of the static interval.

9. The method for payment security of a crowdsourced human resource platform based on multi-dimensional risk control according to claim 7, characterized in that: The second step in the threshold calculation process also includes: the influence level coefficient H. i Based on a comprehensive threshold judgment of the influencing factors of holidays at different levels of influence, the influence level of holidays is determined. Then, a corresponding static interval is calculated using a baseline interval calculation algorithm based on the type. Further, a corresponding static basic coefficient is calculated based on the lower limit of the static interval. All basic coefficients are subject to an upper limit constraint, meaning they cannot exceed the upper limit constraint of the static interval. Finally, the dynamic coefficient of the influence level coefficient is calculated: H i = Static base coefficient × Dynamic factor; The dynamic factor is obtained based on historical data analysis.

Images