Online car-hailing order advance decision-making method and system

Abstract

The invention relates to an online car-hailing order advance decision-making method and system, and the method comprises the steps: firstly, calling order data of which payment is not completed in a current period from an order center, extracting loss rate influence factors in combination with corresponding driver side service data, and dividing the loss rate influence factors into a first type and a second type according to the influence level; inputting the factors into a loss rate model, marking the order as a non-advance order when any first type of factor is triggered, otherwise, calculating a loss rate based on a second type of factor, and dividing the order into a non-advance order, a to-be-manually-audited order or an advance order according to a relationship with a preset first loss rate threshold and a preset second loss rate threshold; the order to be manually checked is manually rechecked and the mark is updated, the advanced payment process is executed on the advanced payment order, and the result is fed back to the driver side, so that the intelligent and refined risk assessment of the single order is realized, and the advanced payment loss is reduced.

Description

Technical Field

[0001] This application relates to the field of ride-hailing technology, and in particular to a ride-hailing order advance payment decision method and system. Background Technology

[0002] In the daily operations of ride-hailing platforms, the processing of unpaid orders directly impacts driver income and platform financial security. To ensure drivers receive their service fees promptly, platforms typically implement an advance payment mechanism for unpaid orders. However, while this advance payment protects drivers' rights, it inevitably exposes the platform to bad debt risks. Current industry practices, such as fixed advance payment limits or manual analysis, have significant shortcomings. Fixed limit strategies cannot adapt to the varying risks of different orders. For example, in high-risk scenarios like abnormally high order amounts, excessively short billing times, or driver-passenger identity linkage, fixed limits are insufficient to effectively prevent potential losses. Furthermore, for regular orders, fixed limits may impose unnecessary payment restrictions. Manual review methods are inefficient, costly, prone to delays when processing massive volumes of orders, and their subjective review standards lack standardized quantitative criteria, leading to unstable risk assessment results. Summary of the Invention

[0003] To at least partially overcome the problems of low efficiency, high cost, and potential delays when processing massive orders, as well as the highly subjective nature of the review standards and the lack of unified quantitative basis, which leads to unstable risk assessment results in ride-hailing platforms, this application provides a ride-hailing order advance payment decision-making method and system.

[0004] The proposed solution is as follows: According to a first aspect of the embodiments of this application, a method for making decisions on advance payments for ride-hailing orders is provided, including: Retrieve order data from the order center that has not been paid within the current period as pending orders; Retrieve the driver-side service data corresponding to the pending orders, and extract the factors affecting the loss rate from the pending orders and driver-side service data; The factors affecting the loss rate are classified into two categories based on their level of influence: Category I factors and Category II factors. The loss rate influencing factors are input into the loss rate model, and the loss rate of the pending order is output. When the pending order triggers any of the first type of loss rate influencing factors, the pending order is marked as an order that cannot be prepaid. When the pending order does not trigger the first type of loss rate influencing factors, the loss rate of the pending order is calculated based on the second type of loss rate influencing factors. If the loss rate of an order to be processed is higher than the first loss rate threshold, the order to be processed will be marked as an order that cannot be prepaid; if the loss rate of an order to be processed is between the first loss rate threshold and the second loss rate threshold, the order to be processed will be marked as an order pending manual review; if the loss rate of an order to be processed is lower than the second loss rate threshold, the order to be processed will be marked as an order that can be prepaid. Orders awaiting manual review are sent to the manual review department. Based on the feedback from the manual review department, the order status is updated to either "orders that cannot be prepaid" or "orders that can be prepaid". The system sends non-payable orders and their loss rates to the driver's end, executes the payment process for payable orders, and sends the payment results for payable orders to the driver's end.

[0005] Preferably, the method further includes: Receive appeal requests and supporting materials submitted by drivers for orders that cannot be prepaid; The appeal request and supporting materials are sent to the manual review department. Based on the feedback from the manual review department, the order mark that cannot be prepaid is updated to an order that can be prepaid or an order that has been reviewed as unprepaid.

[0006] Preferably, the first type of loss rate influencing factors include: order amount, actual billing duration, actual number of billing trajectory points, whether the driver and passenger have the same identity, whether the audio recording file during the trip is missing, and whether the order was placed by scanning a QR code; The second category of loss rate influencing factors includes: toll fees, parking fees, order amount, difference between actual and estimated mileage, driver complaint rate, and the ratio of orders where drivers take detours.

[0007] Preferably, the conditions that trigger the first type of loss rate influencing factors include: The order amount is greater than the preset order amount, the actual billing time is less than the preset billing time, the actual number of billing trajectory points is less than the preset billing trajectory points, the driver and passenger have the same identity, the audio recording file during the trip is missing, and the current order is a QR code order. The actual number of billing trajectory points is calculated based on the number of billing trajectory points in the order statistics. Whether the driver and passenger have the same identity is determined by the mobile phone numbers of the driver and the passenger in the order; Whether the audio recordings during the trip are missing can be determined by whether the audio recording file link exists in the order; Whether an order was placed via QR code depends on the order channel.

[0008] Preferably, the loss rate of the orders to be processed is calculated based on the second type of loss rate influencing factors, including: The factors affecting the second type of loss rate were normalized. The initial weights of each factor influencing the second type of loss rate were determined by historical regression analysis. The final weights of each factor influencing the second type of loss rate were determined by a weighted average method, taking into account the expert scoring results and the initial weights. The loss rate of the orders to be processed is calculated based on the normalized values ​​of each of the second-category loss rate influencing factors and their corresponding final weights.

[0009] Preferably, the method further includes: After the current period ends, the percentage of orders that cannot be prepaid during the current period is calculated. If the proportion of non-payable orders in the current period is greater than the preset ratio, then the first loss rate threshold and / or the second loss rate threshold for the next period will be increased. If the proportion of non-payable orders in the current period is not greater than the preset ratio, then the first loss rate threshold and / or the second loss rate threshold for the next period will be reduced.

[0010] Preferably, the method further includes: After the current period ends, the average actual loss rate of all pending orders within the current period is calculated. The average actual loss rate is compared with the preset target loss rate; If the average actual loss rate is greater than the preset target loss rate, then the first loss rate threshold and / or the second loss rate threshold for the next period are increased. If the average actual loss rate is not greater than the preset target loss rate, then the first loss rate threshold and / or the second loss rate threshold for the next period are reduced.

[0011] Preferably, the method further includes: Based on the business city, order type, and time period of the order to be processed, determine the corresponding city configuration parameters, order type configuration parameters, and time period configuration parameters; When adjusting the first loss rate threshold and / or the second loss rate threshold, the city configuration parameter, order type configuration parameter, and time period configuration parameter are introduced to correct the first loss rate threshold and / or the second loss rate threshold.

[0012] Preferably, the final weights of each factor influencing the second type of loss rate are determined, including: Based on the second type of loss rate influencing factors and the corresponding actual loss results recorded in the historical order samples, a regression analysis sample set is constructed, and the influence coefficient of each second type of loss rate influencing factor on the loss rate is obtained through multivariate regression analysis. The influence coefficient is used as the initial weight of each second type of loss rate influencing factor. Obtain expert scores for each factor influencing the second type of loss rate; The initial weights and expert scoring results are normalized to obtain normalized initial weights and normalized expert weights. The weighting ratio of data-driven results and expert experience results is pre-configured, and the normalized initial weights and normalized expert weights are weighted and summed according to the weighting ratio to obtain the final weights of each second-type loss rate influencing factor.

[0013] According to a second aspect of the embodiments of this application, a ride-hailing order advance payment decision system is provided, comprising: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute a ride-hailing order prepayment decision method as described in any of the above.

[0014] The technical solution provided in this application may include the following beneficial effects: This technical solution constructs a layered risk assessment mechanism, combining real-time order data with driver service behavior characteristics, to achieve refined risk quantification and decision-making for unpaid ride-hailing orders. Specifically, the rigid rules designed for the first category of loss rate influencing factors quickly intercept high-risk orders, reducing wasted computational resources. The quantitative calculation of the second category of loss rate influencing factors, combined with threshold range-based routing, automatically rejects high-risk orders, manually reviews medium-risk orders, and automatically advances payments for low-risk orders. This prevents the platform from incurring excessive losses while protecting the legitimate rights and interests of drivers. Thus, the system accurately controls the platform's bad debt risk while ensuring driver income, effectively solving the problem of inaccurate risk assessment caused by extensive management methods such as fixed advance payment amounts and manual analysis, and improving the targeting and decision-making efficiency of risk assessment.

[0015] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] Figure 1 This is a flowchart illustrating a ride-hailing order advance payment decision-making method according to one embodiment of this application; Figure 2 This is a schematic diagram of the structure of a ride-hailing order advance payment decision system provided in one embodiment of this application.

[0018] Reference numerals: Processor-21; Memory-22. Detailed Implementation

[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0020] In traditional ride-hailing order payment management, the use of fixed payment amounts and manual analysis results in a lack of granular quantification in risk assessment, hindering the systematic identification and tiered processing of order risk characteristics. Specifically, the fragmented use of basic order information and driver service behavior data makes it difficult to comprehensively capture potential risks. For example, abnormal features such as driver detours or missing recordings are not included in the assessment system, leading to insufficient basis for risk decisions. Furthermore, this problem directly results in the platform's inability to differentiate order risk levels during payment decisions. High-risk orders are incorrectly paid for while low-risk orders face delays in review, leading to resource misallocation and decreased decision-making efficiency, ultimately affecting the timeliness and effectiveness of the payment mechanism.

[0021] For example, in scenarios where ride-hailing platforms handle a large number of incomplete payment orders during peak hours, the order center continuously receives newly generated pending orders, while driver-side service data includes real-time information such as trip recordings and route deviation records. When an order shows a high degree of overlap between driver and passenger identity information and involves abnormally detoured routes, the manual review department, due to limited data integration capabilities, cannot promptly link order characteristics with service behavior data, leading to the misclassification of high-risk orders as eligible for advance payment. Furthermore, the fixed advance payment limit in such scenarios results in the platform adopting a uniform processing strategy for all orders, failing to dynamically adjust decision-making paths based on the actual risk of each order. This leads to an increased bad debt rate after advance payments for high-risk orders, while the backlog of manual reviews causes delays in advance payments for low-risk orders, significantly impacting the effectiveness of the driver income protection mechanism.

[0022] If the aforementioned issues are not addressed, the platform's bad debt risk will continue to accumulate and be difficult to effectively contain. The crude nature of its risk assessment mechanism will lead to a gradual increase in the proportion of high-risk orders, thereby damaging the platform's financial stability. Furthermore, the lack of transparency in the decision-making process for advance payments will create a crisis of trust among drivers, diminishing the credibility of the driver rights protection mechanism and prompting drivers to question the platform's operating rules. In addition, the continuously increasing workload of manual review will reduce overall decision-making efficiency, and key system performance indicators such as risk interception accuracy and advance payment response speed will show a deteriorating trend, ultimately putting the platform in a dilemma between protecting driver rights and controlling risks.

[0023] In response, this application proposes a decision-making method for advance payment of ride-hailing orders, referring to... Figure 1 ,include: S11: Retrieve order data that has not been paid within the current period from the order center as pending orders; S12: Retrieve the driver-side service data corresponding to the pending orders, and extract the factors affecting the loss rate from the pending orders and driver-side service data; S13: Classify the factors affecting the loss rate into the first category of factors affecting the loss rate and the second category of factors affecting the loss rate according to their level of influence. S14: Input the loss rate influencing factors into the loss rate model and output the loss rate of the pending order; when the pending order triggers any of the first type of loss rate influencing factors, the loss rate model marks the pending order as an order that cannot be prepaid; when the pending order does not trigger the first type of loss rate influencing factors, the loss rate of the pending order is calculated based on the second type of loss rate influencing factors. S15: If the loss rate of the pending order is higher than the first loss rate threshold, mark the pending order as an order that cannot be prepaid; if the loss rate of the pending order is between the first loss rate threshold and the second loss rate threshold, mark the pending order as an order awaiting manual review; if the loss rate of the pending order is lower than the second loss rate threshold, mark the pending order as an order that can be prepaid. S16: Send the orders pending manual review to the manual review department, and update the marking of the orders pending manual review to either non-payable or payable orders based on the feedback from the manual review department; S17: Send non-payable orders and their loss rates to the driver's end, execute the payment process for payable orders, and send the payment results of payable orders to the driver's end.

[0024] This application relates to a decision-making method for advance payment of ride-hailing orders, which aims to solve the problem that ride-hailing platforms use extensive management methods such as fixed advance payment amounts and manual analysis when making advance payments for passengers' unpaid orders, resulting in inaccurate risk assessment and thus increasing the platform's bad debt risk while protecting the rights and interests of drivers.

[0025] In practical applications, loss rate influencing factors refer to key data features that can quantify the potential loss risk of an order. These features can be extracted from order data, such as trip time period and regional risk level, and from driver service data, such as driver historical service rating and vehicle type. For example, relevant fields can be obtained through database joint queries or external data interfaces. The main purpose is to achieve multi-dimensional identification of the sources of order risk.

[0026] Classifying loss rate influencing factors into Category I and Category II loss rate influencing factors according to their impact level refers to hierarchical classification of influencing factors based on the severity of risk. This is mainly to achieve rapid filtering of high-risk orders and refined evaluation of low-risk orders.

[0027] In practice, the first type of loss rate is affected by factors such as: order amount, actual billing duration, number of actual billing trajectory points, whether the driver and passenger have the same identity, whether the audio recording file during the trip is missing, and whether the order was placed by scanning a QR code. The second category of factors affecting the loss rate includes: toll fees, parking fees, order amount, difference between actual and estimated mileage, driver complaint rate, and the ratio of orders where drivers take detours.

[0028] Based on this, the conditions that trigger the first type of loss rate influencing factors include: The order amount is greater than the preset order amount, the actual billing time is less than the preset billing time, the actual number of billing trajectory points is less than the preset billing trajectory points, the driver and passenger have the same identity, the audio recording file during the trip is missing, and the current order is a QR code order. The actual number of billing trajectory points is calculated based on the number of billing trajectory points in the order statistics. Whether the driver and passenger have the same identity is determined by the mobile phone numbers of the driver and the passenger in the order; Whether the audio recordings during the trip are missing can be determined by whether the audio recording file link exists in the order; Whether an order was placed via QR code depends on the order channel.

[0029] The order amount refers to the total cost of the order, which can be retrieved directly from the order system database. Its purpose is to quickly identify abnormally high-priced orders and prevent the platform from incurring bad debts due to covering fraudulent orders. The actual billing duration refers to the actual billing time during the trip, which can be calculated based on GPS track point timestamps. Its purpose is to effectively distinguish between normal trips and short-term fraudulent orders, ensuring trip credibility. The actual number of billing track points refers to the number of GPS track points recorded during the trip, which can be calculated by counting the total number of track points stored in the order system. Its purpose is to check the integrity of track data and ensure trip traceability. Driver and passenger are... "Identity discrepancy" refers to verifying the consistency of the driver and passenger identities. This can be achieved by comparing the driver's account-linked mobile phone number with the passenger's mobile phone number in the order. The purpose is to accurately identify the risk of fraudulent orders and prevent internal cheating. "Whether the trip recording file is missing" refers to verifying the integrity of the trip recording evidence. This can be achieved by checking whether the recording file link associated with the order exists or whether its storage status is valid. The purpose is to ensure the availability of key evidence and avoid misjudgments due to missing evidence. "Whether it was ordered by scanning a QR code" refers to the ordering channel attribute. This can be achieved by judging based on the channel identifier recorded when the order was created. The purpose is to implement preventative rejection measures for high-risk channels. Toll fees refer to the tolls incurred during the trip, which can be calculated using data extracted from the order's additional fee details. Their purpose is to more comprehensively reflect the actual expenditure of the order. Parking fees refer to the parking fees incurred during the trip, which can also be calculated using data extracted from the order's additional fee details. Their purpose is to supplement the order's true cost information. Actual vs. Estimated Distance Difference refers to the difference between the actual distance traveled and the estimated distance. This can be calculated by comparing the cumulative distance on the GPS track with the estimated distance. Their purpose is to quantify the additional losses caused by detours. Driver Complaint Rate refers to the frequency of driver complaints. This can be calculated using the monthly complaint ratio based on historical complaint records. Their purpose is to predict driver risk tendencies. Driver Detour Order Ratio refers to the proportion of detour orders out of total orders. This can be calculated using the ratio based on historical order trajectory analysis. Their purpose is to assess the risk level of driver behavior patterns.

[0030] The loss rate model refers to the calculation logic used to output the loss rate of orders to be processed, which is mainly for the purpose of realizing automatic graded processing of orders.

[0031] Sending orders awaiting manual review to the manual review department means transferring medium-risk orders to the manual review stage. Specifically, a task queue system can be used to automatically allocate orders to the reviewers' terminals, such as pushing tasks to be processed through an internal workflow engine. This is mainly to combine human experience to handle boundary cases that are difficult for the model to determine.

[0032] This application constructs a tiered risk assessment mechanism, setting high-risk factors as hard rules for rapid interception, calculating loss rates for low-risk orders based on a quantitative model, and forming a closed-loop decision-making process by combining threshold comparison and manual review. Thus, based on real-time order data and driver service behavior characteristics, it achieves refined quantification of the risk of unpaid ride-hailing orders, thereby effectively controlling the platform's bad debt risk while ensuring the stability of driver income.

[0033] As a preferred implementation, the comparison process between the output of the loss rate model and the threshold is executed using a parallel computing architecture. For example, multiple pending orders are processed simultaneously through distributed processing units. This is mainly to improve decision-making efficiency and ensure the timeliness of system response. Specifically, the operation of sending non-payable orders and their loss rates to the driver's end is completed through a message push service, such as using instant messaging protocols to transmit structured data. This ensures that drivers are informed of the processing status and basis in a timely manner, maintaining drivers' right to know and the transparency of platform services.

[0034] During the ride-hailing order processing, the system first retrieves the order data of orders that have not been paid within the current period from the order center as pending orders. This ensures that the processing targets are focused on real-time unpaid risk orders, avoiding interference from historical data or paid orders, and making the risk assessment timely and targeted.

[0035] Furthermore, the driver-side service data corresponding to the pending orders is retrieved, and the factors affecting the loss rate are extracted from the pending orders and driver-side service data. This process uses driver behavior data to supplement the basic information of the orders, which can more comprehensively capture potential risk points, such as abnormal behavior characteristics of drivers during the service process, thus reflecting the real risks more accurately than a single order data source.

[0036] The factors influencing the loss rate are categorized into two levels based on their impact: Category 1 and Category 2. This tiered approach sets high-risk factors as hard rules. When any Category 1 factor is triggered, the system immediately marks the order as non-payable, quickly filtering out obviously high-risk orders. If no Category 1 factor is triggered, the system proceeds to the loss rate calculation stage. Specifically, the loss rate influencing factors are input into the loss rate model, which outputs the loss rate for the order. This model dynamically selects the processing path based on whether a Category 1 factor is triggered, avoiding redundant calculations and ensuring that high-risk orders are quickly intercepted while medium- and low-risk orders receive accurate assessments.

[0037] Therefore, the system classifies orders based on the comparison between the loss rate and preset thresholds: if the loss rate is higher than the first loss rate threshold, it is marked as an order that cannot be prepaid; if it is between the first and second loss rate thresholds, it is marked as an order awaiting manual review; if it is lower than the first loss rate threshold, it is marked as an order that can be prepaid. For example, when an order pending processing is detected to have triggered a first-category loss rate influencing factor, the system immediately marks it as an order that cannot be prepaid, without needing to perform loss rate calculation; while for orders that have not triggered the first-category factor, the system calculates the loss rate based on the second-category loss rate influencing factors and then automatically classifies them according to the threshold range.

[0038] Furthermore, once the order awaiting manual review is sent to the manual review department, the marking is updated based on the feedback results. This step introduces boundary cases that are difficult for the human experience model to judge, and combines expert judgment to improve the reliability of decision-making and make up for the limitations of pure algorithm models.

[0039] Finally, orders that cannot be prepaid and their loss rates are sent to the driver's end, while orders that can be prepaid undergo the prepayment process. The prepayment results are simultaneously sent to the driver's end, ensuring that the driver is informed of the processing status and basis in a timely manner, maintaining the driver's right to know and trust. At the same time, the automated execution of the prepayment process ensures the stability of the driver's income.

[0040] This technical solution constructs a layered risk assessment mechanism, combining real-time order data with driver service behavior characteristics, to achieve refined risk quantification and decision-making for unpaid ride-hailing orders. Specifically, the rigid rules designed for the first category of loss rate influencing factors quickly intercept high-risk orders, reducing wasted computational resources. The quantitative calculation of the second category of loss rate influencing factors, combined with threshold range-based routing, automatically rejects high-risk orders, manually reviews medium-risk orders, and automatically advances payments for low-risk orders. This prevents the platform from incurring excessive losses while protecting the legitimate rights and interests of drivers. Thus, the system accurately controls the platform's bad debt risk while ensuring driver income, effectively solving the problem of inaccurate risk assessment caused by extensive management methods such as fixed advance payment amounts and manual analysis, and improving the targeting and decision-making efficiency of risk assessment.

[0041] Example 2 In practical applications, in the automatic decision-making mechanism for ride-hailing order advance payment based on the loss rate model, after the system marks an order as non-payable, drivers lack a channel to appeal against misjudged orders. This leads to the inability to correct potential erroneous decisions in a timely manner, which may result in reasonable orders being wrongly rejected for advance payment, affecting drivers' income protection and the fairness of platform decision-making.

[0042] In this regard, this application further proposes: Receive appeal requests and supporting materials submitted by drivers for orders that cannot be prepaid; Send the appeal request and supporting materials to the manual review department. Based on the feedback from the manual review department, the order mark will be updated to "order that cannot be prepaid" or "order that has been reviewed and is not prepaid".

[0043] Specifically, the appeal request and supporting materials refer to the appeal information and related evidence submitted by the driver to challenge the determination that the order cannot be prepaid. This can be achieved through trip records, audio files, or written descriptions uploaded by the driver via mobile terminal. The purpose is to provide supplementary information to identify special scenarios that the automated model may overlook. The human review department can be understood as a review team that specializes in handling order appeals. This can be achieved by an internal review group composed of experienced customer service personnel. The purpose is to use human judgment to conduct in-depth analysis of boundary cases of automated decision-making. In practical applications, updating the order label refers to modifying the order status based on the review results. This can be achieved by the system automatically changing the order status from "non-prepayable" to "prepayable" or "reviewed but not prepayable". The purpose is to dynamically adjust the decision to protect the driver's rights or maintain the authority of the decision.

[0044] Specifically, the proposed solution involves receiving appeals and supporting documents from drivers after an order is marked as non-payable. These documents serve as supplementary evidence for the initial decision and are sent to a human review department for in-depth analysis. The order marking is updated based on the feedback from the human review department, thus forming a closed-loop processing mechanism. This mechanism ensures that drivers proactively provide supplementary information such as trip details. The human review department then makes professional judgments on boundary cases based on this information, ultimately dynamically correcting misjudged orders through status updates. This effectively compensates for the shortcomings of automated decision-making systems that cannot adjust misjudged orders in a timely manner.

[0045] As a specific embodiment, the solution of this application is implemented as follows: The driver discovers that the order has been marked as non-payable through the mobile application of the ride-hailing platform, and then submits an appeal request and uploads the audio recording of the trip as supporting material; after receiving the material, the manual review department confirms that the passenger was temporarily out of contact during the trip, which is not the driver's responsibility, and therefore updates the order to be marked as a payable order; the system automatically executes the payment process based on the feedback.

[0046] Through the above solution, this application resolves the issue of drivers lacking appeal channels, ensures that misjudged orders can be corrected in a timely manner, effectively protects drivers' reasonable income, and maintains the fairness and authority of the platform's decision-making.

[0047] Example 3 In some of the embodiments described above in this application, the loss rate of pending orders is calculated based on the second type of loss rate influencing factors. However, in its implementation, the calculation of the loss rate may rely only on simple rules or subjective experience, lacking scientific quantitative basis. This makes it impossible to accurately distinguish the degree of influence of different factors, and the evaluation results are easily affected by differences in data scale and human bias, which in turn leads to errors in advance payment decisions and fails to effectively balance the protection of drivers' rights and the control of bad debt risks on the platform.

[0048] In this regard, this application further proposes to calculate the loss rate of pending orders based on the second type of loss rate influencing factors, including: The factors affecting the second type of loss rate were normalized. The initial weights of each factor influencing the second type of loss rate were determined by historical regression analysis. The final weights of each factor influencing the second type of loss rate were determined by combining the expert scoring results and the initial weights using a weighted average method. The loss rate of the orders to be processed is calculated based on the normalized values ​​of each of the second-category loss rate influencing factors and their corresponding final weights.

[0049] Normalization refers to converting loss rate influencing factors with different dimensions and value ranges to a uniform and comparable scale, with the aim of eliminating evaluation biases caused by differences in units or magnitudes in data such as toll fees and parking fees.

[0050] In this embodiment, the Log(x+1) + Min-Max normalization method is used. The implementation principle of Log(x+1) + Min-Max is as follows: For example: Calculated based on toll fees Assume toll fees are 0 yuan, 5 yuan, 50 yuan, and 500 yuan: First, process it according to Log(x+1) to get: 0 yuan -> log(1) = 0 5 yuan -> log(6) ≈ 1.79 50 yuan -> log(51) ≈ 3.93 500 yuan -> log(501) ≈ 6.22 Continuing the Min-Max calculation: data with a value of 0 is removed. min=1.79, max=6.22 5 yuan->log(6) ≈1.79 ->(1.79-min) / (max-min)=0.0 50 yuan->log(51)≈3.93->(3.93-min) / (max-min) = 0.48 500 yuan->log(501)≈6.22->(6.22-min) / (max-min) =1.0.

[0051] Historical regression analysis can be understood as building a statistical model based on historical order data to quantify the influence of factors. It can be achieved through methods such as multiple linear regression or ridge regression. The purpose is to objectively extract the statistical correlation between various factors and loss rates from historical samples. The expert rating results specifically refer to the subjective evaluation values ​​of domain experts on the importance of each loss rate influencing factor. These can be obtained through structured questionnaires or expert meetings. The purpose is to introduce in-depth experience from business scenarios to make up for the limitations of pure data-driven approaches. Weighted average refers to a mathematical processing method that integrates multi-source weighted information. It can be implemented using methods such as linear weighting or geometric average. Its purpose is to balance the contribution of data-driven results and expert experience results through a preset ratio. Calculating the loss rate is a quantitative process that integrates normalized factor values ​​with the final weights. It can be achieved through weighted summation or exponential weighting, with the aim of generating a unified and comparable loss risk indicator.

[0052] Substitute the normalized data into the loss rate model formula: f(x)=x1*0.10+x2*0.10+x3*0.20+x4*0.25+x5*0.15+x5*0.2.

[0053] Where f(x) represents the loss rate; x1 represents the toll fee; x2 represents the parking fee; x3 represents the order amount; x4 represents the difference between the actual and estimated mileage; x5 represents the driver complaint rate; and x6 represents the ratio of orders where the driver takes a detour.

[0054] Weights: w1 = 0.10; w2 = 0.10; w3 = 0.20; w4 = 0.25; w5 = 0.15; w6 = 0.20.

[0055] In one feasible approach, unpaid orders are categorized into three types based on their loss rate: Loss rate > 70%: Orders that cannot be prepaid; Orders with a loss rate between 40% and 70% are pending manual review. Loss rate < 40%: Orders can be advanced.

[0056] Through the above technical solutions, the objectivity and accuracy of loss rate assessment are significantly improved, effectively distinguishing the impact of different factors such as toll fees and driver complaint rates, avoiding assessment bias caused by differences in data scale or subjective experience, thereby supporting the platform's accurate identification of orders that can be prepaid, and effectively suppressing bad debt risks while protecting drivers' income.

[0057] In some of the embodiments described above in this application, the final weights of each second-class loss rate influencing factor are determined to calculate the order loss rate and support advance payment decisions. However, in its implementation, the determination of the initial weights relies only on simple historical regression analysis. The combination of expert scoring results and expert experience results lacks a standardized process, resulting in fixed weight ratios that cannot adapt to dynamic changes in business, causing weight deviations and inaccurate loss rate assessments, which in turn affect the reliability of advance payment decisions and the accuracy of risk control.

[0058] In this regard, this application further proposes to determine the final weights of each factor influencing the second type of loss rate, including: Based on the second type of loss rate influencing factors and the corresponding actual loss results recorded in historical order samples, a regression analysis sample set is constructed, and the influence coefficient of each second type of loss rate influencing factor on the loss rate is obtained through multivariate regression analysis. The influence coefficient is used as the initial weight of each second type of loss rate influencing factor. Obtain expert scores for each factor influencing the second type of loss rate; The initial weights and expert scores are normalized to obtain normalized initial weights and normalized expert weights. The weighting ratios of data-driven results and expert experience results are pre-configured. The normalized initial weights and normalized expert weights are then weighted and summed according to the weighting ratios to obtain the final weights of each factor affecting the second type of loss rate.

[0059] Among them, the regression analysis sample set refers to the data set used for statistical analysis built based on historical order data. It can be implemented using a structured database table that contains the values ​​of the second type of loss rate influencing factors in historical orders and the corresponding actual loss results. Its purpose is to ensure that the analysis sample reflects the loss pattern of real business scenarios. Multivariate regression analysis is a statistical analysis method that considers the influence of multiple independent variables on the dependent variable at the same time. It can be implemented using algorithms such as least squares, ridge regression, or LASSO regression. Its purpose is to capture the interaction effects and nonlinear relationships between factors and to fully quantify the independent contribution of each factor. The influence coefficient is a numerical parameter obtained through regression analysis that reflects the degree of influence of each factor on the loss rate. It can be expressed as the coefficient value in the regression equation. Its purpose is to transform the objective laws of historical data into the quantitative basis of the initial weights. Expert scoring results refer to the subjective assessment of the importance of each type II loss rate influencing factor by domain experts. These results can be obtained using expert scoring sheets, the Delphi method, or the analytic hierarchy process. The purpose is to introduce expert judgments on emerging risks or special scenarios to compensate for insufficient coverage of historical data. Normalization refers to the mathematical processing of converting data of different dimensions or scales to a unified range. It can be achieved by methods such as min-max normalization, Z-score standardization, or decimal scaling normalization. Its purpose is to eliminate the differences in the dimensions of weights from different sources and make data-driven results comparable to expert experience on a unified benchmark. Weighting ratio refers to the relative importance ratio of data-driven results and expert experience results in the final weighting calculation. It can be configured as a fixed ratio or dynamically adjusted according to the business stage. Its purpose is to allow for a flexible balance between objective data and subjective experience based on data quality and business needs. Weighted summation is a method of combining normalized weights according to a preset ratio. It can be implemented using algorithms such as linear weighting, exponential weighting, or adaptive weighting. Its purpose is to generate more robust final weights by scientifically balancing weights from different sources.

[0060] Specifically, the proposed solution addresses the subjectivity and adaptability issues in weight determination by systematically integrating data-driven approaches with expert experience. First, a regression analysis sample set is constructed based on historical order data to ensure the samples reflect the loss patterns of real-world business scenarios. Second, multivariate regression analysis is used to obtain influence coefficients as initial weights, objectively quantifying the independent contribution of each factor. Simultaneously, expert scoring results are obtained to supplement insufficient historical data coverage. Based on this, the initial weights and expert scoring results are normalized to eliminate differences in dimensions and scale effects. Subsequently, the weight ratios of data-driven results and expert experience results are pre-configured, dynamically adjusting their proportions according to business stage or data quality. Finally, the normalized initial weights and normalized expert weights are weighted and summed according to their weight ratios to generate the final weights for each of the second-category loss rate influencing factors. This process ensures that weight determination is based on the objective patterns of historical data while adapting to dynamic business changes, thereby improving the accuracy of loss rate assessment.

[0061] As a specific implementation method, the solution of this application is implemented as follows: In the order processing system of the ride-hailing platform, the system first extracts order data from the historical order database for the past year, and constructs a regression analysis sample set containing the second type of loss rate influencing factors and their actual loss results, such as toll fees, parking fees, order amount, difference between actual and estimated mileage, driver complaint rate, and driver detour order ratio; then, multivariate regression analysis is performed using the least squares method to obtain the influence coefficient of each factor as the initial weight; at the same time, a risk control expert team is organized to score the importance of each factor and obtain the expert score results; next, the initial weights and expert score results are respectively subjected to minimum-maximum normalization processing to map the values ​​to a unified interval; the system pre-configures the weight ratio of data-driven results to expert experience results as 7:3, and this ratio can be dynamically adjusted according to the richness of data; finally, the normalized initial weights and normalized expert weights are weighted and summed according to the configured weight ratio to obtain the final weight of each second type of loss rate influencing factor, which is used to calculate the loss rate of the order to be processed.

[0062] Through the above-mentioned solution, this application can effectively solve the problems of insufficient subjectivity and adaptability in the weight determination process. In the scenario of advance payment decision for ride-hailing orders, this solution ensures that the weights not only reflect the objective laws of historical data, but also adapt to dynamic changes in business, thereby improving the accuracy of loss rate assessment, reducing the problem of inaccurate loss rate assessment caused by weight deviation, and ultimately improving the reliability of advance payment decision and the accuracy of risk control.

[0063] Example 4 In some of the embodiments described above in this application, a prepayment decision mechanism based on a fixed threshold is proposed to determine whether an order is eligible for prepayment. However, in its implementation, the threshold cannot be dynamically adjusted according to actual business data, resulting in inaccurate decision-making when order risk characteristics fluctuate, which may lead to excessive bad debt risk on the platform or insufficient protection of drivers' rights.

[0064] In one feasible approach, the method also includes: After the current period ends, calculate the percentage of orders that cannot be prepaid during the current period; If the proportion of non-payable orders in the current period is greater than the preset ratio, then the first loss rate threshold and / or the second loss rate threshold for the next period will be increased. If the proportion of non-payable orders in the current period is not greater than the preset ratio, then the first loss rate threshold and / or the second loss rate threshold for the next period will be reduced.

[0065] Specifically, the current period refers to the business evaluation time period set by the system. It can use a fixed duration such as 24 hours, 7 days, or 30 days as the statistical unit, aiming to provide a consistent time benchmark for performance evaluation. The proportion of non-payable orders refers to the ratio of the number of orders judged as non-payable by the system to the total number of pending orders within the current period. This ratio can be automatically calculated by the order management module, objectively reflecting the strictness of the current threshold setting. In practical applications, the preset ratio refers to a pre-configured threshold ratio reference value, which can be set based on historical business data or risk management strategies, serving as a basis for dynamically adjusting the threshold. Specifically, raising or lowering the first loss rate threshold and / or the second loss rate threshold refers to adaptively adjusting key parameters in the decision-making model. This can be achieved using fixed-step adjustments, proportional adjustments, or machine learning-based optimization algorithms, aiming to enable the system to automatically optimize risk control accuracy based on actual business performance.

[0066] Specifically, the proposed solution automatically calculates the percentage of orders that cannot be prepaid at the end of each business cycle and compares this percentage with a preset ratio. Based on the comparison result, the loss rate threshold for the next cycle is dynamically adjusted. When the percentage exceeds the preset ratio, the system identifies that the current threshold setting is too strict, resulting in insufficient protection of drivers' rights, and therefore raises the threshold to relax the prepayment conditions. When the percentage does not exceed the preset ratio, the system identifies that the current threshold setting is too lenient, posing a risk of bad debts, and therefore lowers the threshold to tighten the prepayment conditions. This mechanism forms a closed-loop feedback system, enabling decision parameters to continuously adapt to changes in the business environment, thereby improving the accuracy and adaptability of risk assessment.

[0067] As a specific implementation method, the solution of this application is implemented as follows: The system sets the current period as a daily period, and automatically calculates the percentage of orders that cannot be prepaid at the end of each day. If the percentage is higher than a preset ratio, the system appropriately increases the loss rate threshold for the next period; if the percentage is not higher than the preset ratio, the system appropriately decreases the loss rate threshold for the next period. For example, when the percentage is significantly higher than the preset ratio, the threshold is increased by a large margin; when the percentage is slightly higher than the preset ratio, the threshold is increased by a small margin.

[0068] Through the above technical solution, this application can dynamically adjust the advance payment decision threshold based on historical business data, effectively avoiding the problem of rigid decision-making when fixed thresholds fluctuate during business periods. This ensures the income rights of drivers while reasonably controlling the platform's bad debt risk, achieving refined and intelligent risk control.

[0069] In another possible implementation, the method also includes: At the end of the current period, calculate the average actual loss rate of all pending orders during the current period; Compare the average actual loss rate with the preset target loss rate; If the average actual loss rate is greater than the preset target loss rate, then the first loss rate threshold and / or the second loss rate threshold for the next period will be increased. If the average actual loss rate is not greater than the preset target loss rate, then the first loss rate threshold and / or the second loss rate threshold for the next period will be reduced.

[0070] Specifically, the average actual loss rate refers to the arithmetic mean of the actual loss rates of all pending orders in the current period. It can be achieved by using a simple average method or an average method based on order amount weighting. Its purpose is to objectively reflect the overall risk distribution status within the period. The preset target loss rate can be understood as the risk control benchmark value set in advance by the platform. It can be set based on historical bad debt rate data or business strategic goals. Its purpose is to provide a quantifiable optimization direction for threshold adjustment. Raising or lowering the first and second loss rate thresholds refers to adjusting the threshold parameters used to determine the eligibility for order advance payment. This can be achieved by adjusting the threshold parameters using a fixed step size, a proportional coefficient, or an adaptive step size based on risk volatility. The purpose is to dynamically calibrate the advance payment decision criteria to match the actual business risk level.

[0071] Specifically, the proposed solution ensures that threshold adjustments are based on actual order execution results by statistically analyzing the average actual loss rate at the end of the current cycle, thus avoiding decision-making biases caused by data lag. It establishes a business-goal-oriented adjustment benchmark by comparing the average actual loss rate with a preset target loss rate, providing clear business logic support for threshold changes. When the average actual loss rate exceeds the preset target loss rate, the threshold for the next cycle is raised, tightening payment conditions to suppress payment for high-risk orders and effectively reducing bad debt risk accumulation. When the average actual loss rate is not greater than the preset target loss rate, the threshold is lowered, relaxing payment conditions to expand the range of orders eligible for payment and ensuring timely driver income. These steps are executed sequentially and form a closed-loop feedback loop, ensuring that threshold settings are dynamically synchronized with changes in periodic business risks.

[0072] As a specific implementation method, at the end of each weekly cycle, the system automatically calculates the average actual loss rate of all pending orders for that week; compares this average with the preset target loss rate; if the average actual loss rate is higher than the preset target loss rate, the system can appropriately increase the first loss rate threshold and the second loss rate threshold for the next cycle; if the average actual loss rate is not greater than the preset target loss rate, the system can appropriately decrease the first loss rate threshold and the second loss rate threshold for the next cycle.

[0073] The above technical solution realizes a dynamic adjustment mechanism for the loss rate threshold, enabling the threshold setting to adapt to changes in order risk distribution. This controls the platform's bad debt risk while optimizing driver income protection, and improves the accuracy of advance payment decisions and business adaptability.

[0074] It should be noted that in practice, only one of the two dynamic adjustment mechanisms for the loss rate thresholds mentioned above needs to be chosen.

[0075] In practical applications, some of the embodiments described above in this application propose a method for dynamically adjusting the loss rate threshold based on global statistics to optimize advance payment decisions. However, in its implementation, since the threshold adjustment only relies on the overall data within the period and does not distinguish the differences in business cities, order types, and time periods, the adjustment strategy is too general and cannot adapt to the risk characteristics distribution of different scenarios. For example, in high-risk cities or specific time periods, the unified threshold may result in insufficient or excessively strict risk control, affecting the accuracy of advance payment decisions and the platform's risk balance.

[0076] Based on this, the method also includes: Based on the business city, order type, and time period of the order to be processed, determine the corresponding city configuration parameters, order type configuration parameters, and time period configuration parameters; When adjusting the first loss rate threshold and / or the second loss rate threshold, city configuration parameters, order type configuration parameters, and time period configuration parameters are introduced to correct the first loss rate threshold and / or the second loss rate threshold.

[0077] Among them, the business city refers to the geographical area covered by the ride-hailing service, which can be achieved by city level classification or risk characteristic clustering. The purpose is to distinguish the economic level and fraud risk characteristics of different regions. Order type refers to different categories of ride-hailing services, which can be classified by service mode or price range. The purpose is to identify the differences in risk characteristics of different service modes. Time period refers to the specific time interval in which an order occurs. It can be implemented using fixed or dynamic time period divisions, with the aim of capturing the patterns of risk fluctuations over time. In practical applications, city configuration parameters refer to quantitative indicators reflecting the risk characteristics of a specific city. They can be implemented using historical bad debt rate statistics or expert ratings, with the aim of providing regional risk references for threshold adjustments. Order type configuration parameters are quantitative indicators that reflect the risk characteristics of a specific order type. They can be implemented using the historical average loss rate of that type of order or a risk coefficient table. Their purpose is to distinguish the risk levels of different service models. Time period configuration parameters refer to quantitative indicators that reflect the risk characteristics of a specific time period. They can be implemented using historical risk data or dynamic risk indices for that time period, with the aim of adapting to risk changes over time.

[0078] Specifically, the proposed solution first determines the corresponding configuration parameters based on the business city, order type, and time period of the order to be processed. These parameters capture the differences in risk characteristics across geographical, service mode, and time dimensions, respectively. Then, when adjusting the loss rate threshold, these configuration parameters are introduced as correction factors into the threshold calculation process, allowing the threshold adjustment to dynamically adapt to the risk distribution characteristics of different business scenarios. When the system detects that an order belongs to a high-risk city, the city configuration parameter triggers an upward adjustment of the threshold to prevent excessive advance payments in that region. When the order type is a low-risk ride-hailing service, the order type configuration parameter triggers a downward adjustment of the threshold to promote reasonable advance payments. When the order occurs during a high-risk nighttime period, the time period configuration parameter automatically tightens the threshold to address potential risks. This multi-dimensional parameter-driven threshold correction mechanism ensures that the loss rate threshold closely matches the risk characteristics of actual business scenarios, thereby achieving refined control of advance payment decisions.

[0079] As a specific implementation method, the system can classify business cities into three risk levels: first-tier cities are configured as high-risk, tourist cities as medium-risk, and other cities as low-risk. Order types are divided into three categories: express, premium, and carpooling, with premium services configured as low-risk. Time periods are divided into four intervals: morning peak, off-peak, evening peak, and nighttime, with nighttime configured as a high-risk period. During threshold adjustment, when processing premium orders in tourist cities, the system calls the city configuration parameters, order type configuration parameters, and time period configuration parameters, and corrects the basic threshold through multiplication to obtain an adjusted threshold suitable for the scenario.

[0080] Through the above solution, this application can implement differentiated threshold adjustment strategies for different business scenarios based on their risk characteristics, avoiding decision-making biases caused by a globally uniform threshold, significantly improving the accuracy of advance payment decisions and risk control capabilities, and effectively reducing the platform's bad debt risk while protecting the rights and interests of drivers.

[0081] Example 5 A ride-hailing order advance payment decision system, referring to Figure 2 ,include: Processor 21 and memory 22; Processor 21 and memory 22 are connected via a communication bus: The processor 21 is used to call and execute the program stored in the memory 22; The memory 22 is used to store a program, which is at least used to execute a ride-hailing order advance payment decision method as described in any of the above embodiments.

[0082] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0083] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0084] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0085] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0086] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0087] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0088] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0089] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0090] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for making decisions on advance payments for ride-hailing orders, characterized in that, include: Retrieve order data from the order center that has not been paid within the current period as pending orders; Retrieve the driver-side service data corresponding to the pending orders, and extract the factors affecting the loss rate from the pending orders and driver-side service data; The factors affecting the loss rate are classified into two categories based on their level of influence: Category I factors and Category II factors. The loss rate influencing factors are input into the loss rate model, and the loss rate of the pending order is output. When the pending order triggers any of the first type of loss rate influencing factors, the pending order is marked as an order that cannot be prepaid. When the pending order does not trigger the first type of loss rate influencing factors, the loss rate of the pending order is calculated based on the second type of loss rate influencing factors. If the loss rate of pending orders exceeds the first loss rate threshold, the pending orders will be marked as non-payable orders. If the loss rate of an order to be processed is between the first loss rate threshold and the second loss rate threshold, the order to be processed will be marked as an order to be manually reviewed. If the loss rate of pending orders is lower than the second loss rate threshold, the pending orders will be marked as orders that can be prepaid. Orders awaiting manual review are sent to the manual review department. Based on the feedback from the manual review department, the order status is updated to either "orders that cannot be prepaid" or "orders that can be prepaid". The system sends non-payable orders and their loss rates to the driver's end, executes the payment process for payable orders, and sends the payment results for payable orders to the driver's end.

2. The method according to claim 1, characterized in that, The method further includes: Receive appeal requests and supporting materials submitted by drivers for orders that cannot be prepaid; The appeal request and supporting materials are sent to the manual review department. Based on the feedback from the manual review department, the order mark that cannot be prepaid is updated to an order that can be prepaid or an order that has been reviewed as unprepaid.

3. The method according to claim 1, characterized in that, The first category of loss rate influencing factors include: order amount, actual billing duration, number of actual billing trajectory points, whether the driver and passenger have the same identity, whether the audio recording file during the trip is missing, and whether the order was placed by scanning a QR code; The second category of loss rate influencing factors includes: toll fees, parking fees, order amount, difference between actual and estimated mileage, driver complaint rate, and the ratio of orders where drivers take detours.

4. The method according to claim 3, characterized in that, The conditions that trigger factors affecting the first type of loss rate include: The order amount is greater than the preset order amount, the actual billing time is less than the preset billing time, the actual number of billing trajectory points is less than the preset billing trajectory points, the driver and passenger have the same identity, the audio recording file during the trip is missing, and the current order is a QR code order. The actual number of billing trajectory points is calculated based on the number of billing trajectory points in the order statistics. Whether the driver and passenger have the same identity is determined by the mobile phone numbers of the driver and the passenger in the order; Whether the audio recordings during the trip are missing can be determined by whether the audio recording file link exists in the order; Whether an order was placed via QR code depends on the order channel.

5. The method according to claim 1, characterized in that, The loss rate of pending orders is calculated based on the second category of loss rate influencing factors, including: The factors affecting the second type of loss rate were normalized. The initial weights of each factor influencing the second type of loss rate were determined by historical regression analysis. The final weights of each factor influencing the second type of loss rate were determined by a weighted average method, taking into account the expert scoring results and the initial weights. The loss rate of the orders to be processed is calculated based on the normalized values ​​of each of the second-category loss rate influencing factors and their corresponding final weights.

6. The method according to claim 1, characterized in that, The method further includes: After the current period ends, the percentage of orders that cannot be prepaid during the current period is calculated. If the proportion of non-payable orders in the current period is greater than the preset ratio, then the first loss rate threshold and / or the second loss rate threshold for the next period will be increased. If the proportion of non-payable orders in the current period is not greater than the preset ratio, then the first loss rate threshold and / or the second loss rate threshold for the next period will be reduced.

7. The method according to claim 1, characterized in that, The method further includes: After the current period ends, the average actual loss rate of all pending orders within the current period is calculated. The average actual loss rate is compared with the preset target loss rate; If the average actual loss rate is greater than the preset target loss rate, then the first loss rate threshold and / or the second loss rate threshold for the next period are increased. If the average actual loss rate is not greater than the preset target loss rate, then the first loss rate threshold and / or the second loss rate threshold for the next period are reduced.

8. The method according to any one of claims 6 or 7, characterized in that, The method further includes: Based on the business city, order type, and time period of the order to be processed, determine the corresponding city configuration parameters, order type configuration parameters, and time period configuration parameters; When adjusting the first loss rate threshold and / or the second loss rate threshold, the city configuration parameter, order type configuration parameter, and time period configuration parameter are introduced to correct the first loss rate threshold and / or the second loss rate threshold.

9. The method according to claim 5, characterized in that, Determine the final weights of each factor influencing the second type of loss rate, including: Based on the second type of loss rate influencing factors and the corresponding actual loss results recorded in the historical order samples, a regression analysis sample set is constructed, and the influence coefficient of each second type of loss rate influencing factor on the loss rate is obtained through multivariate regression analysis. The influence coefficient is used as the initial weight of each second type of loss rate influencing factor. Obtain expert scores for each factor influencing the second type of loss rate; The initial weights and expert scoring results are normalized to obtain normalized initial weights and normalized expert weights. The weighting ratio of data-driven results and expert experience results is pre-configured, and the normalized initial weights and normalized expert weights are weighted and summed according to the weighting ratio to obtain the final weights of each second-type loss rate influencing factor.

10. A ride-hailing order prepayment decision-making system, characterized in that, include: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute the ride-hailing order advance payment decision method according to any one of claims 1-9.

Images