Method for predicting order cancellation probability, electronic device and readable storage medium

Abstract

The application provides a method for predicting order cancellation probability, an electronic device and a readable storage medium, and belongs to the technical field of data processing. In the method, by obtaining a first feature information set, a second feature information set and a third feature information set, multi-dimensional features related to a target source order, a first driver and their interaction relationship can be taken as inputs of a target prediction model, so that the target prediction model outputs a cancellation probability of the first driver canceling the target source order. In the application, not only the historical behavior of a target consignor to which the target source order belongs, the intrinsic attribute of the target source and the configuration requirement of the target source order are considered, but also the driving capacity feature, the historical behavior feature of the driver and the cross feature between the driver and the target consignor, the target source and the target source order are introduced, so that the target prediction model comprehensively covers diversified factors leading to cancellation of the target source order, and accurate prediction of the cancellation probability of the driver canceling the target source order is realized.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for predicting the probability of order cancellation, an electronic device, and a readable storage medium. Background Technology

[0002] In freight matching scenarios, matchmaking platforms collect and process information from both supply and demand sides, matching the freight demands posted by shippers with the transportation capacity resources provided by drivers to facilitate the completion of transportation orders. However, in actual operation, it is common for drivers to cancel orders after accepting them. Frequent order cancellations not only affect the user experience of shippers but also reduce the overall fulfillment efficiency of the matchmaking platform.

[0003] In related technologies, prediction models based on conversion rates are typically used to estimate the matching degree between drivers and cargo, thereby increasing the likelihood of a transaction during the recommendation process. However, these conversion rate-based prediction models primarily focus on the probability of a transaction and lack targeted prediction and control for subsequent fulfillment and cancellation processes. The main reasons drivers cancel orders after accepting them include: First, the driver's vehicle does not match the cargo's length, model, tonnage, etc., leading to cancellation after acceptance; second, there is a conflict between the cargo's loading / unloading time and the driver's transportation schedule, causing cancellation if the driver cannot arrive on time or wait; third, some drivers and cargo owners bypass the matching platform to conduct private transactions, resulting in online order cancellations; fourth, the cargo's remarks contain special requirements, such as the need for lifting equipment, tailgate vehicles, or truck-mounted assistance, which the accepting driver lacks, leading to cancellation. These diverse reasons can all contribute to a high order cancellation rate, thus affecting the fulfillment rate of the matching platform.

[0004] Therefore, in existing vehicle-cargo matching scenarios, the reasons for order cancellation are complex and diverse, and traditional prediction models based on completion rates are unable to accurately identify the risk of order cancellation, which has become an urgent problem to be solved. Summary of the Invention

[0005] This application provides a method, electronic device, and readable storage medium for predicting the probability of order cancellation, in order to solve the problem that in existing vehicle-cargo matching scenarios, the reasons for order cancellation are complex and diverse, and traditional prediction models based on completion rates are difficult to accurately identify the risk of order cancellation.

[0006] In a first aspect, this application provides a method for predicting the probability of order cancellation, the method comprising:

[0007] obtain a first feature information set related to the target cargo order, the first feature information set comprising first feature information related to historical behavior characteristics of a target consignor to which the target cargo order belongs, second feature information related to inherent attributes of a target cargo corresponding to the target cargo order, and third feature information related to a configuration of the target cargo order;

[0008] obtain a second feature information set related to the first driver, the second feature information set comprising fourth feature information related to a target vehicle corresponding to the first driver, and fifth feature information related to historical behavior characteristics of the first driver;

[0009] obtain a third feature information set related to the target cargo order and the first driver, the third feature information set comprising first cross feature information of the first driver and the target consignor, second cross feature information of the first driver and the target cargo, and third cross feature information of the first driver and the target cargo order;

[0010] process the first feature information set, the second feature information set, and the third feature information set by using a target prediction model to obtain a cancellation probability of the first driver canceling the target cargo order.

[0011] In a possible design, the first feature information comprises transaction information and cancellation information of the target consignor in a preset time period, and a credit score of the target consignor; the transaction information comprises transaction days and transaction orders of the target consignor; the cancellation information comprises cancellation days and cancellation orders of the target consignor; and the preset time period is before a publishing time of the target cargo order.

[0012] The second feature information comprises type information, shape information, volume information, weight information, and sensitive attribute information of the target cargo, and a quality score of the target cargo;

[0013] The third feature information comprises a transportation route and accessory conditions of the target cargo.

[0014] In a possible design, the fourth feature information comprises vehicle length information, vehicle type information, departure location information, destination information, and load information of the target vehicle.

[0015] The fifth feature information includes the number of exposures, the number of clicks, the number of transactions, the number of cancellations, and the number of fly orders of the first driver in the preset time period; the number of exposures is the number of times that the target platform displays a first cargo order to the first driver; the number of clicks is the number of times that the first driver clicks to view the first cargo order; the number of transactions is the number of times that the first driver accepts the first cargo order; the number of cancellations is the number of times that the first driver accepts the first cargo order but does not fulfill the first cargo order; the number of fly orders is the number of times that the first driver does not fulfill the first cargo order through the target platform after the transaction on the target platform; and the first cargo order is any cargo order of the target platform.

[0016] In a possible design, the first cross-feature information includes transaction information and cancellation information of the first driver and the target consignor in the preset time period.

[0017] The second cross-feature information includes the number of exposures, the number of clicks, the number of transactions, the number of cancellations, and the number of fly orders of the first driver to a second cargo order in the preset time period; the second cargo order corresponds to the target cargo source.

[0018] The third cross-feature information includes the number of cargo orders completed by the first driver along the transportation route in the preset time period.

[0019] In a possible design, before the first feature information related to the target cargo order is acquired, the method further includes:

[0020] A training sample set is constructed, the training sample set includes a plurality of training samples, each training sample includes a historical feature information set and label information corresponding to the historical feature information set; the historical feature information set includes a fourth feature information set related to a historical cargo order, a fifth feature information set related to a second driver corresponding to the historical cargo order, and a sixth feature information set related to the historical cargo order and the second driver; and the label information is used to indicate whether the second driver cancels the historical cargo order.

[0021] The initial prediction model is trained by using the training sample set to obtain the target prediction model.

[0022] In a possible design, the target prediction model is a tree model or a deep neural network model.

[0023] In a possible design, after the first feature information set, the second feature information set, and the third feature information set are processed by using the target prediction model to obtain the cancellation probability of the first driver canceling the target order, the method further includes:

[0024] determining whether the target order meets a tiered exposure condition, the tiered exposure condition being used to determine whether tiered exposure is effective in reducing the cancellation probability of the first driver canceling the target order;

[0025] in a case where the target order meets the tiered exposure condition, dividing the plurality of first drivers into high-cancellation-rate drivers, medium-cancellation-rate drivers, and low-cancellation-rate drivers based on the cancellation probability of each of the plurality of first drivers canceling the target order;

[0026] delaying the target order for a first time length for the high-cancellation-rate drivers, delaying the target order for a second time length for the medium-cancellation-rate drivers, and directly exposing the target order for the low-cancellation-rate drivers, where the first time length is greater than the second time length.

[0027] In a possible design, the dividing the plurality of first drivers into high-cancellation-rate drivers, medium-cancellation-rate drivers, and low-cancellation-rate drivers based on the cancellation probability of each of the plurality of first drivers canceling the target order includes:

[0028] determining a target quantity according to the quantity of the plurality of first drivers and a preset ratio, the target quantity being a total quantity of the high-cancellation-rate drivers and the medium-cancellation-rate drivers in the plurality of first drivers;

[0029] determining a first cancellation probability threshold according to a sum of a base cancellation probability and a first preset value, where the base cancellation probability is a median of the plurality of cancellation probabilities when the cancellation probabilities of the plurality of first drivers are arranged in descending order;

[0030] determining a second cancellation probability threshold according to a sum of the base cancellation probability and a second preset value, where the second preset value is less than the first preset value;

[0031] in a case where the cancellation probability of the first driver canceling the target order is greater than the first cancellation probability threshold and the quantity of the high-cancellation-rate drivers is less than the target quantity, determining the first driver as the high-cancellation-rate driver;

[0032] In a case where the cancel probability of the first driver canceling the target order is greater than the second cancel probability threshold, the cancel probability of the first driver canceling the target order is less than or equal to the first cancel probability, and the number of the high cancel rate drivers and the medium cancel rate drivers is less than the target number, the first driver is determined as the medium cancel rate driver.

[0033] The first driver of the plurality of first drivers other than the high cancel rate driver and the medium cancel rate driver is determined as the low cancel rate driver.

[0034] In a second aspect, the present application provides an order cancel probability prediction device, the device comprising:

[0035] A first acquisition module is configured to acquire a first feature information set related to a target order, the first feature information set comprising first feature information related to historical behavior characteristics of a target consigner to which the target order belongs, second feature information related to inherent attributes of a target consignment corresponding to the target order, and third feature information related to a configuration of the target order.

[0036] A second acquisition module is configured to acquire a second feature information set related to a first driver, the second feature information set comprising fourth feature information related to a target vehicle corresponding to the first driver, and fifth feature information related to historical behavior characteristics of the first driver.

[0037] A third acquisition module is configured to acquire a third feature information set related to the target order and the first driver, the third feature information set comprising first cross feature information of the first driver and the target consigner, second cross feature information of the first driver and the target consignment, and third cross feature information of the first driver and the target order.

[0038] A processing module is configured to process the first feature information set, the second feature information set, and the third feature information set by using a target prediction model to obtain a cancel probability of the first driver canceling the target order.

[0039] In a third aspect, the present application provides an electronic device, comprising a memory and at least one processor.

[0040] The memory stores computer execution instructions.

[0041] The at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the method as described in the above first aspect or various possible designs of the first aspect.

[0042] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, wherein computer execution instructions are stored in the computer readable storage medium, and when the computer execution instructions are executed, a method according to the first aspect or various possible designs of the first aspect is implemented.

[0043] In a fifth aspect, a computer program product is provided, wherein computer program codes are included in the computer program product, and when the computer program codes are run on a computer, the computer is caused to implement a method according to the first aspect or various possible designs of the first aspect.

[0044] In a sixth aspect, a chip is provided, comprising: an interface circuit and a logic circuit, the interface circuit is configured to receive a signal from another chip outside the chip and transmit the signal to the logic circuit, or send a signal from the logic circuit to another chip outside the chip, and the logic circuit is configured to implement a method according to the first aspect or various possible designs of the first aspect.

[0045] The embodiments of the present application provide an order cancellation probability prediction method, an electronic device and a readable storage medium. In the order cancellation probability prediction method, a first feature information set, a second feature information set and a third feature information set are obtained, multi-dimensional features related to a target cargo order and a first driver and an interaction relationship between the target cargo order and the first driver are obtained at the same time, the first feature information set, the second feature information set and the third feature information set are taken as inputs of a target prediction model for processing, and a cancellation probability of the first driver canceling the target cargo order is output. Compared with the prior art of predicting based on only a transaction rate, the method provided by the present application not only considers historical behaviors of a target consignor to which the target cargo order belongs, intrinsic attributes of the target cargo and configuration requirements of the target cargo order, but also introduces driving capacity features, historical behavior features of a driver and cross features between the driver and the target consignor, the target cargo and the target cargo order, so that the target prediction model can comprehensively cover diversified factors leading to cancellation of the target cargo order at a feature level. Therefore, the method can accurately predict the cancellation probability of the driver canceling the target cargo order, thereby solving the problem that a traditional prediction model based on a transaction rate cannot accurately identify order cancellation risks. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A flowchart of an order cancellation probability prediction method provided by an embodiment of the present application is shown in the figure;

[0047] Figure 2 A flowchart of another order cancellation probability prediction method provided by an embodiment of the present application is shown in the figure;

[0048] Figure 3A flowchart of another method for predicting order cancellation probability provided by an embodiment of the present application is shown in FIG. 1.

[0049] Figure 4 A structure diagram of a device for predicting order cancellation probability provided by an embodiment of the present application is shown in FIG. 2.

[0050] Figure 5 A structure diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 3. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the specification herein is for describing particular embodiments only and is not intended to be limiting of the application; the use of the terms "include," "includes" and "including" in the description and the claims and the use of terms such as "comprise", "comprises", "comprising" and the like are not intended to exclude or to be limited to the listed features only but rather are intended to encompass one or more features or a wide variety of features.

[0053] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase "an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive or alternative embodiments. It is expressly understood that the embodiments described herein are merely examples from a whole class of comparable embodiments which those skilled in the art will readily appreciate. It is further understood that the description and the drawings are not limited to the embodiments described and illustrated because such embodiments can encompass a wide variety of alternatives.

[0054] The term "and / or", merely describes an association relationship of associated objects, and means that there can be three relationships, for example, A and / or B, which means that there can be A, A and B at the same time, and B. In addition, the character " / " in the present application generally represents an "or" relationship between the front and rear associated objects.

[0055] In addition, the terms "first", "second", and the like in the specification and claims of the present application or the above drawings are used to distinguish different objects, and are not used to describe a specific order, and can explicitly or implicitly include one or more of the features.

[0056] In the description of this application, unless otherwise stated, "multiple" and "at least two" mean two or more (including two), and similarly, "multiple groups" and "at least two groups" mean two or more (including two groups).

[0057] In the description of this application, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, "connected" or "linked" can refer not only to a physical connection, but also to an electrical connection or a signal connection. For instance, it can be a direct connection, i.e., a physical connection, or an indirect connection through at least one intermediate component, as long as the circuit is connected. It can also refer to the internal connection between two components. A signal connection can refer not only to a signal connection through a circuit, but also to a signal connection through a medium, such as radio waves. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0058] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, different technical features in this application can be combined with each other.

[0059] In freight matching scenarios, matchmaking platforms collect and process information from both supply and demand sides, matching the freight demands posted by shippers with the transportation capacity resources provided by drivers to facilitate the completion of transportation orders. However, in actual operation, it is common for drivers to cancel orders after accepting them. Frequent order cancellations not only affect the user experience of shippers but also reduce the overall fulfillment efficiency of the matchmaking platform.

[0060] In related technologies, order cancellations between drivers and shippers are usually caused by a combination of factors, which can be summarized into the following categories:

[0061] First, there's the issue of mismatched vehicle parameters. In the recommendation logic of long-haul freight, the matching platform doesn't solely rely on identical vehicle length and model when recommending cargo. Instead, it allows for a certain degree of difference between vehicles and cargo; for example, a large vehicle might transport small-volume cargo, or a small vehicle might transport moderately oversized cargo. In such cases, some drivers may misjudge their vehicle's carrying capacity when accepting an order, only to discover later, after communicating with the shipper, that they cannot complete the transport, ultimately leading to order cancellation.

[0062] Second, the pickup time doesn't match. Orders posted by shippers on the matching platform include both immediate orders requiring drivers to arrive within a specified time and next-day orders with less urgent delivery times. If a driver's empty journey is too long to arrive within the stipulated time, they may cancel an immediate order; if a driver primarily engages in short-distance transportation and prefers quick transactions, they may cancel a next-day order due to the mismatch in timing requirements.

[0063] Third, private transactions. In some cases, drivers who find the prices offered by the matching platform too low after accepting an order may contact the shipper directly through means outside the platform. If the shipper also wants to avoid the service fees charged by the matching platform, they may complete the transaction with the driver, bypassing the platform, which could lead to the cancellation of the online order.

[0064] Fourth, special requirements were not met. When listing cargo, shippers often specify special requirements in the order notes, such as the need for lifting equipment, the need for truck closures, or the requirement for vehicles with tailgates. If the driver accepting the order does not meet these requirements, the order will be cancelled.

[0065] This demonstrates the diversity and complexity of reasons for order cancellations. Related technologies typically employ prediction models based on completion rates to estimate the matching degree between drivers and freight, thereby increasing the likelihood of a successful transaction during the recommendation process. However, these completion rate-based prediction models primarily focus on the probability of a successful transaction and struggle to accurately identify cancellation risks.

[0066] To address the problems existing in related technologies, this application provides a method, electronic device, and readable storage medium for predicting order cancellation probability. In the method for predicting order cancellation probability, by acquiring a first set of feature information, a second set of feature information, and a third set of feature information, multi-dimensional features related to the target cargo order and the first driver, as well as the interaction between the target cargo order and the first driver, can be simultaneously obtained. These three sets of feature information are then used as input to a target prediction model for processing, thereby outputting the probability of the first driver canceling the target cargo order. Compared to existing schemes that only predict based on the completion rate, the method provided in this application not only considers the historical behavior of the target cargo owner to which the target cargo order belongs, the inherent attributes of the target cargo, and the configuration requirements of the target cargo order, but also introduces the driver's capacity characteristics, historical behavior characteristics, and cross-features between the driver and the target cargo owner, the target cargo, and the target cargo order. This allows the target prediction model to comprehensively cover the diverse factors leading to the cancellation of target cargo orders at the feature level. Therefore, this method can accurately predict the probability of drivers canceling orders for target cargo, thus solving the problem that traditional prediction models based on completion rates cannot accurately identify order cancellation risks.

[0067] Next, through some specific embodiments and accompanying drawings, we will describe in detail how this application solves the problem that in the above-mentioned vehicle-cargo matching scenario, the reasons for order cancellation are complex and diverse, and traditional prediction models based on the completion rate are difficult to accurately identify the risk of order cancellation.

[0068] Figure 1 This is a flowchart illustrating a method for predicting order cancellation probability provided in an embodiment of this application. Figure 1 As shown, the method for predicting the order cancellation probability provided in this application embodiment specifically includes S101 to S104, and S101 to S104 will be described in detail below.

[0069] It should be noted that the execution subject of the order cancellation probability prediction method provided in this application embodiment can be an electronic device. The electronic device can be a mobile phone, a laptop computer, etc., and this embodiment does not specifically limit it.

[0070] S101. Obtain the first set of feature information related to the target source order.

[0071] The first set of feature information includes first feature information related to the historical behavior characteristics of the target cargo owner to which the target cargo order belongs, second feature information related to the inherent attributes of the target cargo corresponding to the target cargo order, and third feature information related to the configuration of the target cargo order.

[0072] It should be noted that the first set of feature information is used to describe the target source order from multiple dimensions. The first set of feature information includes first feature information, second feature information, and third feature information.

[0073] The first feature information is generated by the target platform based on historical transaction data. It is used to reflect the behavioral characteristics of the target cargo owner within a given time range, thereby reflecting the target cargo owner's transaction habits and stability during modeling.

[0074] The second characteristic information consists of the description, quantity, specifications, transportation requirements, etc. of the target cargo source filled in by the target cargo owner when publishing the target cargo source, which can reflect the basic situation of the target cargo source in terms of physical characteristics, usage conditions or business needs.

[0075] The third characteristic information mainly involves the operational constraints set for the target freight order on the target platform, which reflects the specific requirements for drivers and vehicles during the execution of the target freight order. The target platform is a matching platform.

[0076] The first feature information set does not rely on single-dimensional feature information, and can provide a more comprehensive characterization of the target cargo order, providing input basis for the subsequent target prediction model to calculate the probability prediction of the first driver canceling the target cargo order.

[0077] In a possible implementation, the first characteristic information includes transaction information and cancellation information of the target consignor in a preset time period, and a credit score of the target consignor.

[0078] The preset time period is before a publishing time of the target cargo order.

[0079] It should be noted that the preset time period refers to a time range for statistics of behavior data, and is before a statistics time (the publishing time of the target cargo order).

[0080] The preset time period can be set by a manager of the target platform, and the embodiment does not make a specific limitation.

[0081] For example, the preset time period is the last 7 days, the last 14 days, or the last 30 days.

[0082] It should be noted that the credit score of the target consignor is used to represent a numerical index obtained by the target platform after comprehensively quantitatively evaluating the credit status of the target consignor based on historical behavior data, transaction performance, and default record of the target consignor.

[0083] The transaction information includes transaction days and transaction numbers of the target consignor.

[0084] It should be noted that the transaction information is used to represent a situation that cargo orders published by the target consignor in the preset time period are taken by drivers.

[0085] The transaction days refer to a quantity of natural days in which there is at least one cargo order published by the target consignor and taken by a driver in the preset time period.

[0086] The transaction numbers refer to a total quantity of cargo orders published by the target consignor and taken by drivers in the preset time period.

[0087] The cancellation information includes cancellation days and cancellation numbers of the target consignor.

[0088] It should be noted that the cancellation information is used to represent a situation that cargo orders published by the target consignor in the preset time period are cancelled after being taken by drivers.

[0089] The cancellation days refer to a quantity of natural days in which there is at least one cargo order published by the target consignor and cancelled after being taken by a driver in the preset time period.

[0090] The cancellation numbers refer to a total quantity of cargo orders published by the target consignor and cancelled after being taken by drivers in the preset time period.

[0091] The second characteristic information includes type information, form information, volume information, weight information, and sensitive attribute information of the target cargo, and a quality score of the target cargo.

[0092] It should be noted that the type information of the target cargo source is used to represent the category to which the target cargo source belongs, such as steel, building materials, electronic products, food, etc.

[0093] The form information of the target cargo source is used to represent the physical state of the target cargo source in the transportation process, such as bulk, whole vehicle, pallet or liquid, etc.

[0094] The volume information of the target cargo source is used to represent the size of the space occupied by the target cargo source, which can be described by the length, width and height of the target cargo source.

[0095] The weight information of the target cargo source is used to represent the weight or tonnage of the target cargo source.

[0096] The sensitive attribute information of the target cargo source is used to represent whether the target cargo source has special attributes, such as fragile, flammable, refrigeration or dangerous goods.

[0097] The quality score of the target cargo source is used to represent a numerical index obtained by the target platform after comprehensively quantitatively evaluating the completeness, standardization and historical performance of the target cargo source.

[0098] The third feature information includes the transportation route and accessory conditions of the target cargo source.

[0099] It should be noted that the transportation route of the target cargo source is used to represent the geographical path between the starting point and the destination of the target cargo source, which can include the starting city, the ending city and the information of the stopover.

[0100] The accessory conditions of the target cargo source are used to represent the additional requirements set by the target consignor in the target cargo order, such as whether hoisting equipment is needed, whether car pressing is needed or whether tailboard vehicle is needed, etc.

[0101] In the embodiments of the present application, by introducing the transaction information, cancellation information and credit score of the target consignor in the first feature information, and combining the attribute characteristics, quality score, transportation route and note information of the target cargo order in the second feature information and the third feature information, the input data of the target prediction model can be more comprehensive, which not only considers the historical transaction behavior of the target consignor, but also includes the inherent attributes of the target cargo source and the configuration requirements of the target cargo order, thereby effectively improving the recognition ability and prediction accuracy of the target prediction model on the cancellation risk of the target consignor canceling the target cargo order.

[0102] S102, acquiring a second feature information set related to the first driver.

[0103] The second feature information set includes fourth feature information related to the target vehicle corresponding to the first driver, and fifth feature information related to the historical behavior characteristics of the first driver.

[0104] It should be noted that the second feature information set is used to comprehensively describe the capacity conditions and historical behavior characteristics of the first driver. The second feature information set includes fourth feature information and fifth feature information.

[0105] The fourth feature information is obtained by the first driver when registering a vehicle or updating vehicle information on the target platform, and is used to represent the basic situation of the target vehicle in terms of physical parameters and operating conditions, thereby reflecting the adaptation ability of the first driver to different freight orders.

[0106] The fifth feature information is derived from the historical operation data records of the first driver within a specified time range recorded by the target platform, and is used to represent the behavior patterns of the first driver in the order receiving, transportation and performance stages, so as to reflect the transaction habits and performance stability of the first driver when modeling.

[0107] In a possible embodiment, the fourth feature information includes vehicle length information, vehicle type information, departure location information, destination information and load information of the target vehicle.

[0108] It should be noted that the vehicle length information of the target vehicle is used to represent the numerical value or interval information of the length of the target vehicle. The length information is recorded in meters, for example, 4.2m, 6.8m, 7.6m, 9.6m, 13m, 17.5m, etc.; or can be discretely recorded according to the preset gear of the target platform.

[0109] The vehicle type information of the target vehicle is used to represent the category information of the body type of the target vehicle, such as van, flatbed, dump, refrigerated, dangerous chemical, etc. The vehicle type information of the target vehicle can be stored in the form of enumeration or one-hot encoding.

[0110] The departure location information of the target vehicle is used to represent the preference of the starting location or the frequently starting location of the target vehicle, which can be represented in the form of administrative division (province / city / district), geographic code or preset region code.

[0111] The destination information of the target vehicle is used to represent the preference of the arrival location or the frequently arrival location of the target vehicle. The representation method of the destination information of the target vehicle is consistent with the representation method of the target vehicle of the departure location information of the target vehicle.

[0112] The load information of the target vehicle is used to represent the carrying capacity of the target vehicle, which can be the rated load, and the unit can be ton or kilogram.

[0113] Specifically, the departure location information and the destination information of the target vehicle can be configured by the first driver in the vehicle information of the target platform, or can be obtained through the statistical results of the historical completed orders of the first driver or the route preference of the first driver recorded by the target platform.

[0114] For example, the fourth feature information includes that the length of the target vehicle is 9.6 m, the vehicle type is a van, the frequently-visited place is A province A1 city, the frequently-visited place is B province B2 city, and the load information is no more than 10 tons.

[0115] The fifth feature information includes the exposure times, the click times, the transaction times, the cancellation times and the fly order times of the first driver in the preset time period.

[0116] The exposure times are the number of times that the target platform displays the first freight order to the first driver in the preset time period.

[0117] The first freight order is any freight order of the target platform.

[0118] It should be noted that the target platform displays the first freight order to the first driver once, which is counted as 1 time; the same first freight order is displayed to the first driver again at different times or different entrances, which is counted separately.

[0119] The click times are the number of times that the first driver clicks to view the first freight order in the preset time period.

[0120] It should be noted that the first driver performs a click event of “entering the order detail page” on the first freight order once, which is counted as 1 time; the first driver clicks on the same first freight order multiple times, which is counted separately.

[0121] The transaction times are the number of times that the first driver accepts the first freight order in the preset time period.

[0122] It should be noted that the “order acceptance success” event recorded by the target platform is used as the standard, and the same first freight order is counted separately or only the first valid order acceptance is counted if the order is canceled and then accepted again according to the rules of the target platform.

[0123] The cancellation times are the number of times that the first driver accepts but does not fulfill the first freight order in the preset time period.

[0124] The fly order times are the number of times that the first driver does not fulfill the first freight order through the target platform after the transaction on the target platform in the preset time period.

[0125] For example, in the preset time period of the last 30 days, the exposure times of the first driver are 120 times, the click times are 35 times, the transaction times are 12 times, the cancellation times are 3 times, and the fly order times are 1 time. The counting of the above exposure times, click times, transaction times, cancellation times and cancellation times is based on the event log of the target platform, and the same first freight order is counted separately.

[0126] In the embodiment of the present application, by introducing the length, model, departure place, destination and load of the target vehicle in the second feature information set, and combining the behavior statistical data of the first driver in the preset time period, such as the exposure times, click times, transaction times, cancellation times and fly single times, the input features of the target prediction model on the driver side can be more comprehensive, which can not only reflect the adaptability of the corresponding vehicle of the driver in the aspects of carrying capacity and route selection, but also reflect the historical performance of the driver in the platform interaction and performance process, thereby improving the discrimination and prediction accuracy of the target prediction model on the cancellation risk of the target order cancellation by different drivers.

[0127] S103, acquire a third feature information set related to the target order and the first driver.

[0128] The third feature information set includes first cross feature information of the first driver and the target consignor, second cross feature information of the first driver and the target order, and third cross feature information of the first driver and the target order.

[0129] It should be noted that the third feature information set is used to represent the characteristics of the first driver and the target order in the multi-dimensional interaction relationship. The third feature information set includes the first cross feature information, the second cross feature information and the third cross feature information.

[0130] The first cross feature information is generated by the target platform based on the cooperation record of the first driver and the target consignor in the historical order, and is used to reflect the transaction frequency, cooperation stability and historical performance between the first driver and the target consignor.

[0131] The second cross feature information is obtained by the target platform according to the interaction of the first driver with the target order in the preset time period, and is used to reflect the familiarity and historical performance of the first driver to the target order.

[0132] The third cross feature information involves the matching of the first driver and the target order in the transportation route, time arrangement or execution condition, and is used to represent the adaptability of the first driver in the specific order scene.

[0133] By acquiring the third feature information set, the interaction characteristics between the first driver and the target order can be described from different angles such as cooperation relationship, order preference and order execution condition, thereby providing more comprehensive feature basis for the input of the cancellation probability prediction model.

[0134] In one possible embodiment, the first cross feature information includes transaction information and cancellation information of the first driver and the target consignor in the preset time period.

[0135] It should be noted that the first cross feature information is used to represent the historical cooperation between the first driver and the target consignor within the preset time period, and at least includes transaction information and cancellation information.

[0136] The transaction information of the first driver and the target consignor within the preset time period is used to represent the situation that the order of the target consignor is taken by the first driver within the preset time period.

[0137] The transaction day number of the first driver and the target consignor within the preset time period is used to indicate the number of natural days in which at least one order is successfully taken between the target consignor and the first driver within the preset time period.

[0138] The transaction order number of the first driver and the target consignor within the preset time period is used to indicate the total number of orders successfully taken between the target consignor and the first driver within the preset time period.

[0139] The cancellation information of the first driver and the target consignor within the preset time period is used to represent the situation that the order of the target consignor is taken by the first driver and then canceled within the preset time period.

[0140] The cancellation day number of the first driver and the target consignor within the preset time period is used to indicate the number of natural days in which at least one order is canceled after being taken between the target consignor and the first driver within the preset time period.

[0141] The cancellation order number of the first driver and the target consignor within the preset time period is used to indicate the total number of orders canceled after being taken between the target consignor and the first driver within the preset time period.

[0142] For example, within the preset time period of the last 30 days, the transaction day number between the first driver and the target consignor is 3 days, the transaction order number is 5, the cancellation day number is 1 day, and the cancellation order number is 1.

[0143] The second cross feature information includes the exposure number, the click number, the transaction number, the cancellation number and the fly order number of the second consignment order of the first driver within the preset time period.

[0144] The second consignment order corresponds to the target consignment.

[0145] It should be noted that the second cross feature information is used to represent the interaction behavior of the first driver to the order corresponding to the target consignment within the preset time period.

[0146] The exposure number of the second consignment order of the first driver within the preset time period is the number of times that the target platform displays the second consignment order to the first driver within the preset time period.

[0147] The number of clicks of the second cargo order by the first driver within the preset time period is the number of times that the first driver clicks to view the details of the second cargo order within the preset time period.

[0148] The number of transactions of the second cargo order by the first driver within the preset time period is the number of times that the first driver performs order confirmation on the second cargo order within the preset time period.

[0149] The number of cancellations of the second cargo order by the first driver within the preset time period is the number of times that the first driver cancels the second cargo order that has been accepted but not fulfilled and is recorded by the platform within the preset time period.

[0150] The number of fly orders of the second cargo order by the first driver within the preset time period is the number of times that the first driver accepts the second cargo order on the target platform within the preset time period, but does not fulfill the second cargo order through the target platform.

[0151] The third cross-feature information includes the number of cargo orders completed by the first driver along the transportation route within the preset time period.

[0152] The number of cargo orders completed by the first driver along the transportation route within the preset time period refers to the total number of orders that the first driver has fulfilled within the preset time period, and the starting point and destination of the orders match the transportation route.

[0153] The matching rule can be city-level accurate matching, or in implementation, equivalent matching is determined by area code, radius aggregation, etc., and is not limited to a single implementation.

[0154] Specifically, when there is no record of some interaction in the first cross-feature information, the second cross-feature information, and the third cross-feature information, the count can be recorded as 0.

[0155] In the embodiments of the present application, by introducing the historical transaction and cancellation data of the first driver and the target consignor, the exposure, click, transaction, cancellation, and fly order interaction data of the first driver to the target cargo, and the number of orders completed by the first driver along the transportation route within the preset time period into the third feature information set, the interaction features of the driver and the target cargo order can be comprehensively described from the three aspects of cooperation relationship, cargo preference, and route matching, the input data of the target prediction model is more comprehensive, and thus the identification ability of the target prediction model for the cancellation risk between the first driver and the target cargo order is improved.

[0156] S104, processing the first feature information set, the second feature information set, and the third feature information set by using the target prediction model to obtain the cancellation probability of the first driver canceling the target cargo order.

[0157] The input of the target prediction model is the first feature information set, the second feature information set and the third feature information set, and the output of the target prediction model is a probability value for representing the possibility of the first driver canceling the target order.

[0158] The cancellation probability of the first driver canceling the target order is generally in the interval [0, 1], wherein the closer the cancellation probability of the first driver canceling the target order is to 1, the greater the possibility of the first driver canceling the target order is; the closer the cancellation probability of the first driver canceling the target order is to 0, the smaller the possibility of the first driver canceling the target order is.

[0159] In the embodiments of the present application, by obtaining the first feature information set, the second feature information set and the third feature information set, multi-dimensional features related to the target order, the first driver, and the interaction relationship between the target order and the first driver can be obtained at the same time, and the first feature information set, the second feature information set and the third feature information set are processed as the input of the target prediction model, so as to output the cancellation probability of the first driver canceling the target order. Compared with the prior art of predicting based on only the transaction rate, the method provided in the present application not only considers the historical behavior of the target consignor to which the target order belongs, the inherent attributes of the target consignment, and the configuration requirements of the target order, but also introduces the driving capacity features, historical behavior features of the driver, and the cross features between the driver and the target consignor, the target consignment and the target order, so that the target prediction model can comprehensively cover the diversified factors leading to the cancellation of the target order at the feature level. Therefore, the present method can realize accurate prediction of the cancellation probability of the driver canceling the target order, thereby solving the problem that the traditional prediction model based on the transaction rate cannot accurately identify the order cancellation risk.

[0160] In the above embodiments, the electronic device needs to process the first feature information set, the second feature information set and the third feature information set by using the target prediction model to obtain the cancellation probability of the first driver canceling the target order. Next, the specific process of the electronic device obtaining the target prediction model will be described in detail.

[0161] Figure 2 Another flowchart of the prediction method of the order cancellation probability provided in the embodiments of the present application is shown in FIG. 2. Figure 2 As shown in FIG. 2, in a possible embodiment, before the method steps shown in S101, the prediction method further includes S201 and S202, which will be described in detail below.

[0162] S201, constructing a training sample set.

[0163] The training sample set includes a plurality of training samples, and each training sample includes a historical feature information set and label information corresponding to the historical feature information set.

[0164] It should be noted that the training sample set is used to support offline / on-line training of the initial prediction model, and the training sample set includes a plurality of training samples, and each training sample is composed of a historical feature information set and label information corresponding to the historical feature information set.

[0165] The historical feature information set includes a fourth feature information set related to a historical cargo order, a fifth feature information set related to a second driver corresponding to the historical cargo order, and a sixth feature information set related to the historical cargo order and the second driver.

[0166] It should be noted that the fourth feature information set includes feature information similar to the feature information included in the first feature information set, and the embodiment will not be repeated.

[0167] The fourth feature information set includes feature information related to historical behavior characteristics of a consignor to which the historical cargo order belongs, feature information related to inherent attributes of a cargo source corresponding to the historical cargo order, and feature information related to a configuration of the historical cargo order.

[0168] The fifth feature information set includes feature information similar to the feature information included in the second feature information set, and the embodiment will not be repeated.

[0169] The fifth feature information set includes feature information related to a vehicle of a third driver corresponding to the historical cargo order, and feature information related to historical behavior characteristics of the third driver.

[0170] The sixth feature information set includes feature information similar to the feature information included in the third feature information set, and the embodiment will not be repeated.

[0171] The third feature information set includes cross feature information of the third driver and a consignor to which the historical cargo order belongs, cross feature information of the third driver and a cargo source corresponding to the historical cargo order, and cross feature information of the third driver and the historical cargo order.

[0172] The label information is used to indicate whether the second driver cancels the historical cargo order.

[0173] It should be noted that the label information is used to indicate whether the second driver cancels the historical cargo order, denoted as a binary variable y∈{0,1}.

[0174] y=1 indicates that the historical cargo order is canceled after being ordered by the second driver.

[0175] y = 0 means that the historical cargo order is not canceled (e.g., in a fulfillment completion or other non-cancellation state) after being accepted by the second driver.

[0176] S202, training the initial prediction model using the training sample set to obtain a target prediction model.

[0177] In a possible embodiment, the target prediction model is a tree model or a deep neural network model.

[0178] Wherein, the initial prediction model can be a tree-based binary classification model (such as a gradient boosting class model) or a deep neural network model, and the specific network structure, feature interaction mode and hyperparameter setting of the initial prediction model are not limited to specific implementations.

[0179] It should be noted that the process of training the initial prediction model using the training sample set to obtain the target prediction model is similar to the existing model training process, and this embodiment will not be repeated.

[0180] In the embodiments of the present application, by constructing the training sample set, and simultaneously containing the feature information set of the historical cargo order, the driver feature information set corresponding to the historical cargo order and the feature information set of the interaction between the two in the training sample, and taking whether the driver cancels the historical cargo order as label information, the target prediction model can learn the relationship between the driver cancellation behavior and the multi-dimensional features in the training stage. Further, by training the initial prediction model using the training sample set and generating the target prediction model, the target prediction model can output the corresponding cancellation probability based on the features of the target cargo order and the first driver in the inference stage, so as to realize the effective identification and prediction of the cancellation risk of the target prediction model in actual application.

[0181] In the above embodiments, the electronic device needs to obtain the cancellation probability of the first driver canceling the target cargo order. Next, the method executed by the electronic device after obtaining the cancellation probability of the first driver canceling the target cargo order will be described in detail.

[0182] Figure 3 Another flowchart of the order cancellation probability prediction method provided by the embodiments of the present application is shown in FIG. 10. As shown in FIG. 10, in a possible embodiment, after the method steps shown in S104, the prediction method further includes S105 to S107, which will be described in detail below. Figure 3

[0183] S105, determine whether the target cargo satisfies the layered exposure condition.

[0184] Wherein, the layered exposure condition is used to determine whether the layered exposure is effective in reducing the cancellation probability of the first driver canceling the target cargo order.

[0185] ​It should be noted that the layered exposure condition is used to determine whether the layered exposure strategy for the first driver is an effective strategy in the current target cargo order scenario.

[0186] In a possible implementation, whether the target cargo source meets the layered exposure condition can be determined based on the control group and the intervention group.

[0187] In the offline training phase, the situation in which the layered exposure is performed on the cargo source is taken as the intervention group, the situation in which the layered exposure is not performed is taken as the control group, and the cargo order fulfillment is taken as the target result, so as to construct an estimation model of the fulfillment rate improvement.

[0188] Specifically, the difference between the intervention group and the control group can be estimated by a gain (Uplift) model, and the Uplift value is defined as: .

[0189] wherein represents performing the layered exposure, represents not performing the layered exposure, represents the cargo order fulfillment, and the Uplift value is used to represent the net influence of the layered exposure on the fulfillment rate.

[0190] In the modeling process of the estimation model, the training set and the test set can be divided, and the proportion of the intervention group and the control group in the two groups is kept consistent, so as to ensure the stability of the estimation result.

[0191] Further, X-Learner can be selected as the learning method, the Uplift value is taken as the optimization target for training, and the Uplift tree is combined to split the features, so as to extract the target feature mode meeting the layered exposure condition.

[0192] The target feature mode refers to a condition combination capable of representing the relationship between the cargo source feature and the layered exposure effect, which is extracted from the split result of the Uplift model through modeling analysis on historical cargo order data.

[0193] In the offline training process, the target platform takes the multi-dimensional features of the historical cargo order as input, trains the Uplift model by using the X-Learner method, and combines the Uplift tree for splitting. At the leaf node of the Uplift tree, a combination composed of feature conditions can be formed, for example, the cargo quality score is greater than a certain value and the remark information contains the requirement of a tail plate, or the transportation distance exceeds a certain mileage and the historical cancellation rate of the cargo owner is lower than a certain proportion. Each feature condition combination corresponds to a Uplift value, which is used to represent the net promotion effect of performing the layered exposure on the fulfillment rate under the feature condition combination.

[0194] When the Uplift value of a certain feature condition combination exceeds the first preset threshold value set by the target platform, and the sample size and confidence index meet the preset requirements, the feature condition combination is confirmed as the target feature mode. The target feature mode is essentially a rule set, which describes that performing hierarchical exposure under this type of cargo feature condition is effective.

[0195] In the online determination stage, for a target cargo, it can be determined whether the feature vector corresponding to the target cargo meets any target feature mode generated by the offline training stage according to the feature vector corresponding to the target cargo, and then it is determined whether the target cargo meets the hierarchical exposure condition.

[0196] Specifically, in the case that the feature vector corresponding to the target cargo matches the target feature mode, and the Uplift value of the target feature mode matched by the feature vector corresponding to the target cargo is greater than the first preset threshold value, it is determined that the target cargo meets the hierarchical exposure condition; otherwise, it is determined that the target cargo does not meet the hierarchical exposure condition.

[0197] In another possible implementation, when the target cargo is in a special category, a special route or a special note set configured by the target platform, it is directly determined that the target cargo meets the hierarchical exposure condition.

[0198] S106, in the case that the target cargo meets the hierarchical exposure condition, the plurality of first drivers are divided into high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers based on the cancellation probability of each first driver canceling the target cargo order.

[0199] In the case that the target cargo meets the hierarchical exposure condition, the plurality of first drivers are divided into high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers based on the cancellation probability of each first driver.

[0200] In a possible implementation, the plurality of first drivers can be classified by using a fixed threshold method based on the cancellation probability of each first driver.

[0201] The threshold value is set as θh>θm≥0. When the cancellation probability of the first driver is pi≥θh, the first driver is divided into a high cancellation rate driver; when the cancellation probability of the first driver is θh>pi≥θm, the first driver is divided into a medium cancellation rate driver; and when the cancellation probability of the first driver is pi<θm, the first driver is divided into a low cancellation rate driver.

[0202] In another possible embodiment, a proportion constraint can be additionally added on the basis of the fixed threshold method to control the size of each type of driver, such as limiting the number of high cancellation rate drivers to no more than α% of the total number of the plurality of first drivers, and α is a configurable parameter.

[0203] S107, exposing the target freight order to the high cancellation rate driver after a first time delay, exposing the target freight order to the medium cancellation rate driver after a second time delay, and directly exposing the target freight order to the low cancellation rate driver.

[0204] The first time is longer than the second time.

[0205] It should be noted that different exposure timing control parameters are configured for drivers of different levels, and the display of the target freight order is scheduled.

[0206] Specifically, the exposure timing control parameter configured for the high cancellation rate driver is the first time T1, the exposure timing control parameter configured for the medium cancellation rate driver is the second time T2, and the exposure timing control parameter configured for the low cancellation rate driver is the third time T0; T1>T2, and T0=0.

[0207] For the high cancellation rate driver, the target freight order is added to the visible queue of the high cancellation rate driver after a first time delay T1; for the medium cancellation rate driver, the target freight order is added to the visible queue of the medium cancellation rate driver after a second time delay T2; and for the low cancellation rate driver, the target freight order is directly exposed.

[0208] The first time T1 and the second time T2 can be set by the target platform manager, and the present embodiment does not make specific limitations thereto.

[0209] For example, the first time T1 is 10 minutes, and the second time T2 is 5 minutes.

[0210] It should be noted that the target platform maintains an exposure scheduling queue for each first driver, and when the exposure delay time corresponding to the first driver is reached, the target freight order is pushed to the front end of the queue or inserted into a suitable position according to a predetermined sorting factor (such as distance, time efficiency, platform sorting points, etc.). If the state of the target freight order changes (such as being picked up by another driver, being cancelled, or being completed) during the exposure delay period, the target freight order will not be exposed to the first driver.

[0211] In the embodiments of the present application, by introducing a hierarchical exposure condition determination step, and when the target freight meets the hierarchical exposure condition, the candidate drivers are divided into high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers based on the cancellation probability of each candidate driver, and different delay exposure times are configured for drivers of different cancellation rate levels, which can further realize the differentiated management and control of the driver group based on the output of the target prediction model, so that the exposure scheduling link can be processed in stages according to the cancellation risk level, thereby forming inhibition of high-risk drivers and preferential support for low-risk drivers in the order allocation process of the target platform.

[0212] In the above embodiment, the electronic device needs to divide the plurality of first drivers into high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers based on the cancellation probability of each first driver in the plurality of first drivers canceling the target cargo order. Next, the specific process of the electronic device dividing the plurality of first drivers into high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers will be described in detail.

[0213] In a possible embodiment, the method step shown in S106 can be implemented by S1061 to S1066, which will be described in detail below.

[0214] S1061, determine the target quantity according to the number of the plurality of first drivers and the preset ratio.

[0215] Among them, the target quantity is the total number of high cancellation rate drivers and medium cancellation rate drivers in the plurality of first drivers.

[0216] It should be noted that the preset ratio can be configured by the manager of the target platform, and the present embodiment does not make specific limitation thereto.

[0217] For example, the preset ratio is 20%.

[0218] If the number of the plurality of first drivers is N, and the preset ratio is 20%, then the target quantity M=Nx20%.

[0219] For example, the total number of first drivers is 50, and the preset ratio is set to 20%, then the target quantity is 10, that is, at most 10 first drivers can be divided into high cancellation rate drivers or medium cancellation rate drivers.

[0220] S1062, determine the first cancellation probability threshold according to the sum of the base cancellation probability and the first preset value.

[0221] Among them, the base cancellation probability is the median of the plurality of cancellation probabilities when the plurality of cancellation probabilities of the plurality of first drivers are arranged in descending order.

[0222] It should be noted that if N is even, then the base cancellation probability is the average of the cancellation probability at N / 2 and the cancellation probability at (N / 2+1) when the plurality of cancellation probabilities of the plurality of first drivers are arranged in descending order.

[0223] The first preset value is a positive offset value preset by the manager of the target platform, which is used to move the base cancellation probability by the first preset value.

[0224] For example, the base cancellation probability is 0.7, and the first preset value is 0.15, then the first cancellation probability threshold is 0.85.

[0225] S1063, determining a second cancellation probability threshold according to a sum of the base cancellation probability and a second preset value.

[0226] The second preset value is less than the first preset value.

[0227] The second preset value is a positive offset value preset by a manager of the target platform, and is used to move the second preset value upward on the basis of the base cancellation probability.

[0228] For example, the base cancellation probability is 0.7, and the second preset value is 0.05. Then the second cancellation probability threshold is 0.75.

[0229] S1064, in a case where the cancellation probability of the first driver canceling the target order is greater than the first cancellation probability threshold, and the number of high cancellation rate drivers is less than the target number, determining the first driver as a high cancellation rate driver.

[0230] In a case where the cancellation probability of the first driver canceling the target order is greater than the first cancellation probability threshold, and the number of high cancellation rate drivers has not reached the target number, determining the first driver as a high cancellation rate driver.

[0231] S1065, in a case where the cancellation probability of the first driver canceling the target order is greater than the second cancellation probability threshold, the cancellation probability of the first driver canceling the target order is less than or equal to the first cancellation probability, and the number of high cancellation rate drivers and medium cancellation rate drivers is less than the target number, determining the first driver as a medium cancellation rate driver.

[0232] In a case where the cancellation probability of the first driver canceling the target order is greater than the second cancellation probability threshold, and is less than or equal to the first cancellation probability threshold, and the number of high cancellation rate drivers and medium cancellation rate drivers has not reached the target number, determining the first driver as a medium cancellation rate driver.

[0233] S1066, determining the first drivers other than the high cancellation rate drivers and the medium cancellation rate drivers among the plurality of first drivers as low cancellation rate drivers.

[0234] Determining the remaining first drivers that are not divided into high cancellation rate drivers and medium cancellation rate drivers as low cancellation rate drivers.

[0235] In the embodiments of the present application, the number of first drivers, the preset proportion, the base cancellation probability, the first preset value and the second preset value can be set to form clear hierarchical standards among the plurality of first drivers, so as to reasonably divide the plurality of first drivers into three categories of high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers, and the stability and controllability of the division result is strong, and the subsequent hierarchical exposure strategy can be implemented on the basis of clear risk distribution.

[0236] In another possible embodiment, when the electronic device divides the plurality of first drivers into high cancellation rate drivers, medium cancellation rate drivers and low cancellation rate drivers based on the cancellation probability of each first driver in the plurality of first drivers cancelling the target cargo order, the following method can also be used.

[0237] Specifically, first, the first drivers are filtered in order of the cancellation probability of the plurality of first drivers from large to small, and after each filtering operation, the performance rate value corresponding to the remaining first driver set is calculated. By recording the change of the performance rate as the number of filtered first drivers gradually increases, a relationship curve of the number of filtered drivers and the performance rate can be drawn.

[0238] On the relationship curve of the number of filtered drivers and the performance rate, the position of the inflection point of the curve is determined, and the inflection point of the curve indicates that the decline amplitude of the performance rate appears steep when the number of filtered first drivers continues to increase. Therefore, the number of filtered drivers x can be set as the number of first drivers corresponding to the inflection point of the curve. In addition, the number of filtered drivers x can also be set as 20% of the total number of the plurality of first drivers.

[0239] Further, to avoid abnormal fluctuations in the performance rate statistical results caused by too many filtered drivers, a constraint condition can also be set for the performance rate change amplitude. For example, the relative change amplitude of the performance rate before and after filtering can be limited to not more than a second preset threshold, and when the relative change amplitude of the performance rate before and after filtering exceeds the second preset threshold, the number of drivers classified as high cancellation rate drivers or medium cancellation rate drivers is no longer increased.

[0240] In this case, the first drivers with the top x cancellation probabilities in the plurality of first drivers are classified as high cancellation rate drivers or medium cancellation rate drivers, and the remaining first drivers that are not classified are classified as low cancellation rate drivers.

[0241] Figure 4 A structural schematic diagram of an order cancellation probability prediction device provided by an embodiment of the present application is provided. As shown in Figure 4 The order cancellation probability prediction device 400 provided by the embodiment includes a first acquisition module 401, a second acquisition module 402, a third acquisition module 403 and a processing module 404.

[0242] The first acquisition module 401 is configured to acquire a first feature information set related to the target cargo order, the first feature information set including first feature information related to historical behavior characteristics of a target consignor to which the target cargo order belongs, second feature information related to inherent attributes of a target cargo corresponding to the target cargo order, and third feature information related to a configuration of the target cargo order.

[0243] The second acquisition module 402 is configured to acquire a second feature information set related to the first driver, the second feature information set including fourth feature information related to a target vehicle corresponding to the first driver and fifth feature information related to a historical behavior feature of the first driver.

[0244] The third acquisition module 403 is configured to acquire a third feature information set related to the target cargo order and the first driver, the third feature information set including first cross feature information of the first driver and a target consignor, second cross feature information of the first driver and a target cargo, and third cross feature information of the first driver and the target cargo order.

[0245] The processing module 404 is configured to process the first feature information set, the second feature information set and the third feature information set by using a target prediction model to obtain a cancellation probability of the first driver canceling the target cargo order.

[0246] It should be understood that the respective modules perform the above-mentioned corresponding processes, which have been described in detail in the above method embodiments. For brevity, the details are not described herein.

[0247] Figure 5 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 5. As shown in FIG. 5, the electronic device 500 provided by the embodiment includes a memory 501 and a processor 502. Figure 5

[0248] The memory 501 can be an independent physical unit and can be connected to the processor 502 through a bus 503. The memory 501 and the processor 502 can also be integrated together and implemented by hardware. The memory 501 is configured to store program instructions, and the processor 502 invokes the program instructions to perform the operations of the electronic device in any of the above method embodiments.

[0249] ​Optionally, when part or all of the method in the above embodiments is implemented by software, the electronic device 500 can only include the processor 502. The memory 501 for storing programs is located outside the electronic device 500, and the processor 502 is connected with the memory through circuitry / wires for reading and executing the programs stored in the memory. The processor 502 can be a central processing unit (CPU), a network processor (NP), or a combination of the CPU and the NP. The processor 502 can further include a hardware chip. The hardware chip can be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0250] The memory 501 can include volatile memory, such as random-access memory (RAM); the memory can also include non-volatile memory, such as flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); and the memory can also include a combination of the above-mentioned kinds of memory.

[0251] Exemplarily, the present application provides a chip, comprising: an interface circuit and a logic circuit, the interface circuit is used for receiving signals from other chips outside the chip and transmitting to the logic circuit, or sending signals from the logic circuit to other chips outside the chip, and the logic circuit is used for executing the operations performed by the electronic device in the above method embodiments.

[0252] Exemplarily, the present application provides a computer-readable storage medium, which stores computer program instructions, and the computer program instructions are run by the processor of the electronic device to make the electronic device execute the operations performed by the electronic device in the above method embodiments.

[0253] Exemplarily, the present application provides a computer program product, which, when running on an electronic device, causes the electronic device to perform the operations performed by the electronic device in the above method embodiments.

[0254] The foregoing is merely illustrative of the principles of the application and various modifications can be made by persons skilled in the art without departing from the scope and nature of the application disclosed herein. The above embodiments are presented by way of example only and are not intended to limit the principles of the application. The intent is to conform the claims to the practice of the application, and appropriate modifications to the embodiments chosen by the inventors to embody their application will be recognized by those skilled in the art. Therefore, the true scope of the application should be determined only by reference to the appended claims and their equivalents, instead of by reference to the description and accompanying drawings.

Claims

1. A method for predicting order cancellation probability, characterized in that, The method comprises: obtaining a first feature information set related to a target cargo order, the first feature information set comprising first feature information related to historical behavior characteristics of a target consignor to which the target cargo order belongs, second feature information related to inherent attributes of a target cargo corresponding to the target cargo order, and third feature information related to a configuration of the target cargo order; obtaining a second feature information set related to a first driver, the second feature information set comprising fourth feature information related to a target vehicle corresponding to the first driver, and fifth feature information related to historical behavior characteristics of the first driver; obtaining a third feature information set related to the target cargo order and the first driver, the third feature information set comprising first cross feature information of the first driver and the target consignor, second cross feature information of the first driver and the target cargo, and third cross feature information of the first driver and the target cargo order; processing the first feature information set, the second feature information set, and the third feature information set using a target prediction model to obtain a cancellation probability of the first driver canceling the target cargo order; determining whether the target cargo satisfies a tiered exposure condition, the tiered exposure condition being used to determine whether tiered exposure is effective in reducing the cancellation probability of the first driver canceling the target cargo order; in a case where the target cargo satisfies the tiered exposure condition, dividing a plurality of first drivers into high cancellation rate drivers, medium cancellation rate drivers, and low cancellation rate drivers based on a cancellation probability of each of the first drivers canceling the target cargo order; delaying the target cargo order for a first time length for the high cancellation rate drivers, delaying the target cargo order for a second time length for the medium cancellation rate drivers, and directly exposing the target cargo order for the low cancellation rate drivers; wherein the first time length is greater than the second time length.

2. The method of claim 1, wherein, The first feature information comprises transaction information and cancellation information of the target consignor within a preset time period, and a credit score of the target consignor; wherein the transaction information comprises transaction days and transaction numbers of the target consignor; the cancellation information comprises cancellation days and cancellation numbers of the target consignor; and the preset time period is before a publishing time of the target cargo order; The second feature information comprises type information, form information, volume information, weight information, and sensitive attribute information of the target cargo, and a quality score of the target cargo; The third feature information comprises a transportation route and accessory conditions of the target cargo.

3. The method of claim 2, wherein, The fourth feature information comprises vehicle length information, vehicle type information, departure location information, destination information, and load information of the target vehicle. The fifth feature information includes the number of exposures, the number of clicks, the number of transactions, the number of cancellations, and the number of fly orders of the first driver within the preset time period; the number of exposures is the number of times that the target platform displays a first cargo order to the first driver; the number of clicks is the number of times that the first driver clicks to view the first cargo order; the number of transactions is the number of times that the first driver accepts the first cargo order; the number of cancellations is the number of times that the first driver accepts but does not fulfill the first cargo order; the number of fly orders is the number of times that the first driver does not fulfill the first cargo order through the target platform after the transaction; and the first cargo order is any cargo order of the target platform.

4. The method of claim 3, wherein, The first cross-feature information includes transaction information and cancellation information of the first driver and the target consignor within the preset time period; The second cross-feature information includes the number of exposures, the number of clicks, the number of transactions, the number of cancellations, and the number of fly orders of the first driver to a second cargo order within the preset time period; the second cargo order corresponds to the target cargo source; The third cross-feature information includes the number of cargo orders completed by the first driver along the transportation route within the preset time period.

5. The method of claim 1, wherein, Before obtaining the first feature information related to the target cargo order, the method further includes: constructing a training sample set, the training sample set including a plurality of training samples, each training sample including a historical feature information set and label information corresponding to the historical feature information set; wherein the historical feature information set includes a fourth feature information set related to a historical cargo order, a fifth feature information set related to a second driver corresponding to the historical cargo order, and a sixth feature information set related to the historical cargo order and the second driver; the label information is used to indicate whether the second driver cancels the historical cargo order; training an initial prediction model using the training sample set to obtain the target prediction model.

6. The method of claim 1, wherein, The target prediction model is a tree model or a deep neural network model.

7. The method of claim 1, wherein, The target prediction model is a tree model or a deep neural network model. The target prediction model is a tree model or a deep neural network model. According to the number of the plurality of first drivers and a preset ratio, a target number is determined, the target number being the total number of the high cancellation rate drivers and the medium cancellation rate drivers among the plurality of first drivers; determining a first cancellation probability threshold according to the sum of a base cancellation probability and a first preset value; wherein the base cancellation probability is the median of the plurality of cancellation probabilities when the cancellation probabilities of the plurality of first drivers are arranged in descending order; determining a second cancellation probability threshold according to the sum of the base cancellation probability and a second preset value; wherein the second preset value is less than the first preset value; In a case where the cancel probability of the first driver canceling the target order is greater than the first cancel probability threshold, and the number of the high cancel rate drivers is less than the target number, the first driver is determined as the high cancel rate driver; In a case where the cancel probability of the first driver canceling the target order is greater than the second cancel probability threshold, the cancel probability of the first driver canceling the target order is less than or equal to the first cancel probability, and the number of the high cancel rate drivers and the medium cancel rate drivers is less than the target number, the first driver is determined as the medium cancel rate driver; The first drivers of the plurality of first drivers, except the high cancel rate drivers and the medium cancel rate drivers, are determined as the low cancel rate drivers.

8. An electronic device, comprising: Comprise: a memory and at least one processor; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method as claimed in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed, the method as claimed in any one of claims 1 to 7 is implemented.

Images