Behavior prediction method, behavior prediction model construction method and related device
By constructing a behavior prediction model, using the historical behavior sequence and candidate behavior nodes of the distribution capacity, the model is trained to predict the performance behavior type, which solves the problem of insufficient prediction accuracy of performance behavior in the intelligent scheduling system and improves the prediction accuracy.
Patent Information
- Application Number
- CN202411064665.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-19
AI Technical Summary
When upgrading strategies and algorithms, the existing intelligent scheduling system relies on online experiments, resulting in loss of business benefits and user experience, and lacks accuracy in predicting the actual performance behavior of distribution capacity.
Construct a behavior prediction model, and by obtaining the historical behavior sequence and candidate behavior nodes of the distribution capacity, train the model to predict the performance behavior type of the distribution capacity, and improve the prediction accuracy.
It can predict performance behaviors in the scheduling simulation system that meet the true intention of distribution capacity in the scheduling simulation system, and improve the accuracy of prediction of actual performance behaviors for distribution capacity.
Smart Images

Figure CN118982116A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of order delivery, and in particular to a behavior prediction method, a method for constructing a behavior prediction model, a computer program product, an electronic device, and a computer-readable storage medium. Background Art
[0002] Delivery services are widely used in online shopping, food delivery, errands and other scenarios. The delivery service scenario includes business platforms, merchants, delivery capacity and users. The business platform has a server and provides users with a client, through which users can use the services provided by the business platform; different from the client for users, the business platform also provides merchants with a client for merchants, through which merchants can use the services provided by the business platform. Delivery capacity refers to a party with delivery capabilities, including but not limited to delivery personnel, such as so-called riders. Delivery capacity can communicate with the server through the client used for delivery capacity. Among them, each user can trade with the merchant through the client and can initiate delivery orders.
[0003] In related technologies, in order to plan more reasonable routes for delivery capacity to improve fulfillment efficiency, improve user and delivery capacity experience, and adapt to more and more business developments, an intelligent scheduling system is used to plan the delivery capacity's fulfillment behavior for orders in advance, so as to more reasonably assign orders to the delivery capacity or guide the next action of the delivery capacity. However, the online experiments that the intelligent scheduling system relies on when upgrading and iterating strategies and algorithms may result in a loss of more business revenue and user experience. Therefore, a scheduling simulation system is used to replace online experiments to simulate the actual fulfillment behavior of the delivery capacity at a low cost and realistically. However, the current prediction accuracy of the actual fulfillment behavior of the delivery capacity needs to be improved. Summary of the invention
[0004] In view of the above technical problems, the present disclosure provides a behavior prediction method, a method for constructing a behavior prediction model, a computer program product, an electronic device, and a computer-readable storage medium. The technical solutions are as follows:
[0005] According to a first aspect of an embodiment of this specification, a behavior prediction method is provided, the method comprising:
[0006] Obtaining a current behavior sequence of the delivery transport capacity, wherein the current behavior sequence includes behavior nodes of the delivery transport capacity from the most recent order acceptance time to the current time, and the behavior nodes represent different types of fulfillment behaviors of the delivery transport capacity;
[0007] Determine candidate behavior nodes corresponding to each behavior node in the current behavior sequence, wherein the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the delivery transport capacity can perform after the behavior node;
[0008] Input each behavior node in the current behavior sequence and the candidate behavior node corresponding to each behavior node into a preset behavior prediction model, and the behavior prediction model predicts the expected performance behavior type of the delivery capacity after the current moment based on the input behavior nodes and the candidate behavior nodes corresponding to each behavior node;
[0009] The behavior prediction model is trained by utilizing the historical behavior sequence of the distribution capacity to learn the actual fulfillment behavior of the distribution capacity after each behavior node in the historical behavior sequence.
[0010] According to a second aspect of an embodiment of this specification, a method for constructing a behavior prediction model is provided, the method comprising:
[0011] Acquire a historical behavior sequence of the delivery transport capacity, wherein the historical behavior sequence includes behavior nodes of the delivery transport capacity between two adjacent order acceptance moments, and the behavior nodes represent different types of fulfillment behaviors of the delivery transport capacity;
[0012] Determine candidate behavior nodes corresponding to each behavior node in the historical behavior sequence, wherein the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the delivery capacity can perform after the behavior node;
[0013] Each behavior node in the historical behavior sequence and the candidate behavior node corresponding to each behavior node are respectively input into a preset behavior prediction model. The behavior prediction model predicts the expected predicted behavior of the distribution capacity after each behavior node based on the input of each behavior node and the candidate behavior node corresponding to each behavior node, and is trained according to the error between the predicted behavior of the behavior node and the actual fulfillment behavior of the distribution capacity in the candidate behavior node corresponding to the behavior node.
[0014] According to a third aspect of the embodiments of this specification, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, the method described in the first aspect or the second aspect is implemented.
[0015] According to a fourth aspect of the embodiments of this specification, an electronic device is provided, the electronic device comprising:
[0016] processor;
[0017] a memory for storing processor-executable instructions;
[0018] The processor is configured to implement the method as described in the first aspect or the second aspect.
[0019] According to a fifth aspect of the embodiments of this specification, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the method described in the first aspect or the second aspect are implemented.
[0020] The technical solutions provided by the embodiments of this specification may have the following beneficial effects:
[0021] In the embodiments of the present specification, based on each behavior node in the input historical behavior sequence and the candidate behavior node corresponding to each behavior node, the expected predicted behavior of the distribution capacity after each behavior node is predicted, and the behavior prediction model is trained according to the error between the predicted behavior and the actual performance behavior of the distribution capacity in the candidate behavior node corresponding to the behavior node. The behavior prediction model is trained with a large amount of historical data of the distribution capacity, and only learns the actual performance behavior of the distribution capacity without considering other purposes. It can predict the performance behavior that meets the real intention of the distribution capacity in the scheduling simulation system better than the operations optimization algorithm, thereby improving the prediction accuracy of the actual performance behavior of the distribution capacity.
[0022] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the technical solutions in the embodiments of this specification or related technologies, the drawings required for use in the embodiments or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0024] Figure 1 It is a schematic diagram of a delivery scenario in an embodiment of this specification;
[0025] Figure 2 It is a flowchart of a method for constructing a behavior prediction model according to an embodiment of the present specification;
[0026] Figure 3A It is a schematic diagram of the behavior sequence structure of an embodiment of this specification;
[0027] Figure 3B is a schematic diagram of a behavior sequence structure of another embodiment of the present specification;
[0028] Figure 3C is a schematic diagram of a behavior sequence structure of another embodiment of the present specification;
[0029] Figure 3D is a schematic diagram of a behavior sequence structure of another embodiment of the present specification;
[0030] Figure 4 is a schematic diagram of the structure of a behavior prediction model of an embodiment of this specification;
[0031] Figure 5 is a flow chart of a behavior prediction method according to an embodiment of the present specification;
[0032] Figure 6 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present specification. DETAILED DESCRIPTION
[0033] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this specification. Instead, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims.
[0034] The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this specification. The singular forms "a", "the" and "the" used in this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0035] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this specification, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0036] In the field of e-commerce, an order can refer to a purchase request submitted by a user, which can include information such as the product, quantity, price, etc. selected by the user, and the transaction is completed through online payment. The e-commerce platform will pass this information to the corresponding seller, and the seller will then ship the goods according to the order information.
[0037] See also Figure 1, taking the order in the delivery service scenario as an example, the delivery scenario of the embodiment of this specification is first introduced. The delivery service is widely used in scenarios such as online shopping, food delivery, and errands. The delivery service scenario includes a business platform, merchants, delivery capacity, and users. Among them, the business platform has a server, and provides users with a client, and users can use the services provided by the business platform through the client; different from the client for users, the business platform also provides merchants with a client for merchants, and merchants can use the services provided by the business platform through the client. Delivery capacity refers to a party with delivery capabilities, including but not limited to delivery personnel, such as the so-called riders. The delivery capacity can communicate with the server through the client used for the delivery capacity. In other examples, the delivery capacity can also include unmanned delivery equipment, such as drones, unmanned vehicles, etc. Among them, each user can trade with the merchant through the client, and can initiate a delivery order; the business party can allocate delivery capacity for the instant delivery order.
[0038] For example, after a user places an order, the service platform server (such as an e-commerce platform or other delivery platform that cooperates with it) needs to generate a corresponding delivery waybill for the delivery items and assign the delivery waybill to the delivery transport. Therefore, after the delivery transport accepts the delivery waybill, it will go to the pick-up location to pick up the goods, and deliver them to the user's designated location after successful pick-up.
[0039] For example, in a food delivery scenario, a user places an order through a client to a physical store on the food delivery platform. The food delivery platform generates a corresponding food delivery order and then allocates a delivery waybill to the delivery capacity (such as the client device used by the rider). The rider then goes to the physical store (i.e., the pick-up location for the delivery items) to pick up the food and deliver it to the location specified by the user.
[0040] It can be seen that the execution process of the delivery waybill includes the stage of picking up the goods at the pick-up location and the stage of leaving the pick-up location after picking up the goods and delivering the goods to the user's designated location.
[0041] In actual business, it is often necessary to locate the location of the delivery capacity. For example, to determine whether the delivery capacity has actually arrived at the merchant's location, or to accurately determine the time when the delivery capacity arrives at the merchant.
[0042] In the related technology, in order to plan a more reasonable delivery route for the delivery capacity to improve the fulfillment efficiency, improve the user and delivery capacity experience, and adapt to the development of more and more businesses, an intelligent scheduling system is used to plan the fulfillment behavior of the delivery capacity for the delivery order in advance, so as to more reasonably assign the delivery capacity or guide the next action of the delivery capacity. The intelligent scheduling system needs to continuously upgrade and iterate the strategy and algorithm to adapt to the rapid development and changes of the delivery business. When upgrading and iterating the strategy and algorithm, the intelligent scheduling system may need to rely on online experiments, but online experiments may lose a lot of business revenue and user experience, and the time period of online experiments is long, and there are problems such as difficulty in controlling the consistency of supply and demand between experimental groups. Therefore, a scheduling simulation system can be used to replace the online experiment. The scheduling simulation system uses historical data and the scheduling strategy and algorithm of the intelligent scheduling system. By simulating the fulfillment behavior of the delivery capacity, it realizes the simulation of the intelligent scheduling system and the delivery capacity in the real world, and simulates the actual fulfillment behavior of the delivery capacity at a low cost and realistically.
[0043] However, the current prediction accuracy of the actual performance of distribution capacity needs to be improved. Specifically, in the related technology, the distribution path is often designed through the operation optimization algorithm to guide the performance of the distribution capacity. The purpose of the operation optimization algorithm is to design the best path to guide the performance of the distribution capacity. This identity first needs to design an optimization goal (for example: the shortest distance and the shortest time goals), and then obtain the best path through a large number of calculations. However, the optimization goal and the actual distribution goal of the distribution capacity are often not completely consistent, that is, the best path is often not equal to the actual distribution path of the distribution capacity. The scheduling simulation system hopes to imitate the performance of the distribution capacity as realistically as possible. Therefore, the operation optimization algorithm in the related technology is not suitable for the prediction of the performance of the distribution capacity in the scheduling simulation system. The current prediction accuracy of the actual performance of the distribution capacity based on the scheduling simulation system needs to be improved.
[0044] In response to the above problems, the embodiments of this specification provide a method for constructing a behavior prediction model, which can predict the performance behavior that meets the real intention of the distribution capacity in the scheduling simulation system better than the operations optimization algorithm, thereby improving the prediction accuracy of the actual performance behavior of the distribution capacity. Figure 2 As shown, the method comprises the following steps:
[0045] S201. Obtain the historical behavior sequence of distribution capacity.
[0046] The historical behavior sequence includes behavior nodes of the delivery capacity between two adjacent order acceptance moments, and the behavior nodes represent different types of fulfillment behaviors of the delivery capacity.
[0047] S202: Determine candidate behavior nodes corresponding to each behavior node in the historical behavior sequence.
[0048] Among them, the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the distribution capacity can perform after the behavior node;
[0049] S203: Input each behavior node in the historical behavior sequence and the candidate behavior node corresponding to each behavior node into a preset behavior prediction model respectively.
[0050] The behavior prediction model is based on inputting each behavior node and the candidate behavior node corresponding to each behavior node, predicting the expected predicted behavior of the distribution capacity after each behavior node, and is trained according to the error between the predicted behavior of the behavior node and the actual fulfillment behavior of the distribution capacity in the candidate behavior node corresponding to the behavior node.
[0051] In the embodiments of the present specification, based on each behavior node in the input historical behavior sequence and the candidate behavior node corresponding to each behavior node, the expected predicted behavior of the distribution capacity after each behavior node is predicted, and the behavior prediction model is trained according to the error between the predicted behavior and the actual performance behavior of the distribution capacity in the candidate behavior node corresponding to the behavior node. The behavior prediction model is trained with a large amount of historical data of the distribution capacity, and only learns the actual performance behavior of the distribution capacity without considering other purposes. It can predict the performance behavior that meets the real intention of the distribution capacity in the scheduling simulation system better than the operations optimization algorithm, thereby improving the prediction accuracy of the actual performance behavior of the distribution capacity.
[0052] Before building a behavior prediction model, you first need to obtain sample data for training the model. Sample data includes multiple types of data and can be obtained in multiple ways. As an example, sample data can be a historical behavior sequence of delivery capacity, which can contain behavior nodes of delivery capacity between two adjacent order-taking moments. The behavior nodes can characterize different types of fulfillment behaviors of delivery capacity. By obtaining the historical behavior sequence of delivery capacity, you can obtain the behavior nodes of delivery capacity between two adjacent order-taking moments as sample data for the behavior prediction model. Among them, the order-taking moment refers to the moment when the delivery capacity receives the delivery order.
[0053] It can be understood that the above two adjacent order-receiving times are both historical order-receiving times of the delivery capacity; the two adjacent order-receiving times can be understood as the order-receiving time from one order-receiving time to the next order-receiving time in the historical order-receiving time of the delivery capacity.
[0054] As an example, the number of behavior nodes between two adjacent order receiving moments may be one or more, and there is no specific limitation on the number of behavior nodes between two adjacent order receiving moments.
[0055] As an example, the number of delivery orders that can be undertaken by the same delivery capacity at the same time of order acceptance can be one or more, and there is no specific limitation on this.
[0056] As an example, during the process of fulfilling delivery orders, the delivery capacity can continue to accept delivery orders until the number of delivery orders currently accepted by the delivery capacity reaches the upper limit of the number of delivery orders that the delivery capacity can accept.
[0057] It is understandable that the behavior nodes between two adjacent order-taking moments have a sequence according to the time sequence.
[0058] Considering that if the amount of sample data used to build the model is too large, it may increase the computational burden of the model and reduce training efficiency.
[0059] In response to the above problems, the historical behavior sequence of delivery capacity can be obtained in a variety of ways. As an example, the behavior sequence of delivery capacity divided by historical delivery waves can be first obtained, and the behavior sequence belonging to the same delivery wave can be divided according to the order acceptance time in the behavior sequence divided by historical delivery waves to obtain the above historical behavior sequence.
[0060] It is worth noting that the above introduction to the method of obtaining sample data is only an exemplary display. In practical applications, other methods of obtaining sample data are not excluded and are not specifically limited to this.
[0061] As an example, a delivery wave of delivery capacity can be the process of delivery capacity accepting, picking up and delivering orders for the same wave of orders, that is, the process from the time when the delivery capacity is assigned to the order to the time when the delivery capacity does not accept the order. For example, from the time when the delivery capacity does not accept the delivery order to the time when the delivery capacity delivers all the delivery orders it has accepted (that is, the delivery capacity does not accept the order again), this period of time can be defined as a delivery wave of delivery capacity. As another example, a historical delivery wave contains a complete delivery path of the delivery capacity within a historical period of time; as another example, the behavior sequence of the delivery capacity can be divided into multiple groups of behavior sequences by dividing it according to the historical delivery waves, and each group of behavior sequences belongs to the same historical delivery wave.
[0062] It is understandable that a group of behavior sequences of delivery capacity belonging to the same historical delivery wave may include multiple order-taking moments, and a group of behavior sequences belonging to the same historical delivery wave may be segmented according to two adjacent order-taking moments, thereby obtaining segmented behavior sequences, i.e., the above-mentioned historical behavior sequences. Dividing the behavior sequences by historical delivery waves and further segmenting them by order-taking moments can shorten the length of a single behavior sequence, turning a long sequence into a short sequence, thereby reducing the amount of data for a single behavior sequence and reducing the difficulty of model calculation.
[0063] Taking the takeaway delivery scenario as an example, considering that in the relevant technology, only the delivery capacity's meal pickup and meal delivery behaviors are considered when planning the delivery route of the delivery capacity, but in the actual delivery process, there are many cases where the delivery capacity arrives at the store and fails to pick up the meal of the current order due to the slow delivery of the merchant, and performs the fulfillment behavior of the next order. In this case, the delivery capacity's real intention is to pick up the meal, but the other behaviors that the delivery capacity performs because it did not pick up the meal in the end are caused by interference from the merchant, and these behaviors are not the behaviors expected by the delivery capacity. If the sample data used to train the model only selects the delivery capacity's meal pickup and meal delivery behaviors for learning, the learning data may be mixed with the uncertainty of the merchant's meal delivery, so the results learned by the model cannot represent the delivery capacity's true delivery mentality.
[0064] In response to the above problems, as an example, the different types of fulfillment behaviors of delivery capacity represented by behavior nodes may include: store arrival behavior, meal pickup behavior and / or meal delivery behavior; as another example, the different types of fulfillment behaviors of delivery capacity represented by behavior nodes may include one or more of store arrival behavior, meal pickup behavior and meal delivery behavior, and at least store arrival behavior. Predicting the delivery path of delivery capacity is actually predicting the order in which the delivery capacity goes to the pickup and delivery points of different orders. In order to more accurately restore the real behavior of the delivery capacity, the arrival of the rider at the store is also regarded as an action of the rider. A complete delivery path of the delivery capacity is abstracted into a behavioral sequence of fulfillment behaviors including store arrival behavior, meal pickup behavior and meal delivery behavior. This modeling method eliminates the influence of the merchant's meal delivery information and only focuses on the delivery habits of the delivery capacity, which can make the results predicted by the model more in line with the actual delivery intention of the delivery capacity.
[0065] See also Figure 3A , below, taking the delivery capacity as a rider as an example, a delivery wave of the delivery capacity, that is, a complete delivery path of the delivery capacity and the behavior nodes contained in the delivery path are exemplified:
[0066] like Figure 3A As shown in the figure, A represents the behavior of going to the store, P represents the behavior of picking up the food, D represents the behavior of delivering the food, Oi represents the delivery order i, and the order acceptance time is recorded as I0. Among them, when the delivery capacity continuously undertakes multiple delivery orders in the same time period, the time when the delivery capacity undertakes the last delivery order can be used as the order acceptance time of the multiple delivery orders. For example, Figure 3A In the beginning, the rider takes three delivery orders in succession, recorded as O1, O2, and O3, that is, Figure 3AIn the figure, "Order = {1, 2, 3}", the time when the rider accepts the delivery order O3 is recorded as the order acceptance time l0. After delivering the delivery order O3, the rider accepts the delivery order O4, that is, "Order = {4}" in Figure 3. The time when the rider accepts the delivery order O4 is recorded as the order acceptance time l0. ’ 0, such as Figure 3A , until the rider completes all the delivery orders (O1, O2, O3, O4) he / she has undertaken, the behavior nodes of the rider's actual delivery path, including the arrival at the store, picking up the food, and delivering the food, are displayed in order, and a long sequence of the actual delivery path can be obtained, that is, a long behavior sequence. It can be understood that the long sequence is a behavior sequence belonging to the same delivery wave.
[0067] Considering that at each order-taking moment, the delivery orders currently undertaken by the delivery capacity may change, and the delivery capacity may re-plan the order of going to the store, picking up the food, and delivering the food for the current delivery orders, the behavior sequences belonging to the same delivery wave can be segmented according to the order-taking moment, for example Figure 3B As shown, it can be Figure 3A The long sequence shown is split into multiple short sequences, such as Figure 3B The short sequence Sequence1 and the short sequence Sequence2 in
[0068] In order to further reduce the amount of data in the behavior sequence, thereby reducing the computational burden of the model, as an example, when the behavior sequence belonging to the same delivery wave is segmented according to the order receiving time, if the store arrival behavior and the meal pickup behavior of the same delivery order are continuous, the store arrival behavior and the meal pickup behavior of the delivery order can be combined into one meal pickup behavior during segmentation. For example: Figure 3A Middle A O1 , P O1 If it is a continuous in-store behavior and meal pickup behavior, A can be split into O1 , P O1 Merged Figure 3C P shown in Sequence 2 O1 .
[0069] As another example, when the behavior sequences belonging to the same delivery wave are segmented according to the order receiving time, if the store arrival behavior and the meal pickup behavior of the same delivery order are not continuous, the meal pickup behavior of the delivery order can be deleted during segmentation, and only the store arrival behavior of the delivery order is retained. For example, Figure 3A Middle A O4 , P O4 For discontinuous in-store and food pickup behaviors, P O1 Delete and keep A O4 , then divided into Figure 3B as well as Figure 3C A contained in the short sequence Sequence1 O4 When the store arrival behavior and meal pickup behavior of the same delivery order are not continuous, it can be considered that the delivery capacity is originally inclined to perform the fulfillment behavior of the delivery order first, but because the merchant’s meal delivery time is too late after arriving at the store, the delivery capacity arrives at the store but fails to pick up the meal. Therefore, the store arrival behavior of the delivery order can be regarded as the real fulfillment behavior of the delivery capacity and the meal pickup behavior of the delivery order can be ignored. In this way, a short sequence is generated by segmentation, which not only reduces the data volume of the behavior sequence but also takes into account the real delivery intention reflected by the store arrival behavior of the delivery capacity.
[0070] It is understandable that after Figure 3A The long sequence in is split according to the order receiving time, and the behavior nodes are selected according to the continuity of the store arrival behavior and the meal pickup behavior of the same delivery order. The short sequence (such as Figure 3C Sequence1 and Sequence2) can be used as the above-mentioned historical behavior sequence, that is, sample data for building a behavior prediction model.
[0071] Before building a behavior prediction model, the required sample data may include, in addition to the historical behavior sequence of the delivery capacity, candidate behavior nodes corresponding to each behavior node in the historical behavior sequence. Therefore, after obtaining the historical behavior sequence of the delivery capacity, the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence of the delivery capacity may be determined, wherein the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the delivery capacity can perform after the behavior node. For example, when the behavior node is the pick-up behavior of delivery order A, the fulfillment behavior that the delivery capacity can perform after the behavior node may be the delivery behavior of delivery order A or the fulfillment behavior of other delivery orders (such as the in-store behavior or pick-up behavior of delivery order B).
[0072] It can be understood that the candidate behavior nodes corresponding to each behavior node may include one or more. For example, the candidate behavior nodes corresponding to each behavior node may be any combination of store visit behavior, meal pickup behavior, and meal delivery behavior.
[0073] The candidate behavior nodes corresponding to each behavior node in the historical behavior sequence of the delivery capacity can be determined in a variety of ways. As an example, one of the determination methods may include: if there are store-going behaviors and meal pickup behaviors for the same order in the fulfillment behaviors that the delivery capacity can perform after the behavior node, the meal pickup behavior of the order is deleted, and the store-going behavior of the order is retained as the candidate behavior node of the behavior node. This can further reduce the amount of data in the historical behavior sequence, thereby reducing the model calculation burden.
[0074] It is worth noting that the above introduction to the method of determining the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence of distribution capacity is only an exemplary display. In practical applications, other determination methods are not excluded and no specific limitation is made to this.
[0075] It is understandable that the above-mentioned in-store and food pickup behaviors of the same order are not continuous, for example Figure 3A A shown O4 With P O4 Both are discontinuous performance behaviors.
[0076] As another example, another method of determining the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence of the delivery capacity may include: if there are pick-up behaviors and delivery behaviors for the same order in the fulfillment behaviors that the delivery capacity can perform after the behavior node, the delivery behavior of the order is deleted, and the pick-up behavior of the order is retained as a candidate behavior node for the behavior node.
[0077] See also Figure 3A-3D , the following is an exemplary introduction to the candidate behavior phase corresponding to each behavior node determined by the historical behavior sequence:
[0078] like Figure 3D As shown, based on Figure 3A By splitting the long sequence in , we can get an exemplary historical behavior sequence: ’ 0→A O4 →D O4 →D O2 Based on the historical behavior sequence, the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence can be determined, such as the order acceptance time I ’ Candidate behavior node corresponding to 0: D O1 , D O2 , A O4 ; A O4 Corresponding candidate behavior node: D O1 , D O2 , D O4 ;D O1 Corresponding candidate behavior node: D O2 , D O4 ;D O2 Corresponding candidate behavior node: D O4 .
[0079] As an example, the sample data used to build the behavior prediction model needs to carry positive labels to conduct supervised training on the behavior prediction model; as another example, the actual fulfillment behavior of the delivery capacity in the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence can carry the above positive labels. Figure 3D As shown, for the historical behavior sequence: I ’ 0→A O4 →D O4 →D O2 Among the candidate behavior nodes corresponding to each behavior node, the one marked in red is the actual fulfillment behavior of the delivery capacity carrying positive labels.
[0080] After the sample data, i.e., each behavior node in the historical behavior sequence, and the candidate behavior nodes corresponding to each behavior node are respectively input into the preset behavior prediction model, the behavior prediction model can predict the expected predicted behavior of the distribution capacity after each behavior node based on the input behavior nodes and the candidate behavior nodes corresponding to each behavior node. There can be multiple prediction methods. As an example, one prediction method may include: determining the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node, and predicting the expected predicted behavior of the distribution capacity after each behavior node based on the similarity between each behavior node and each candidate behavior node corresponding to the behavior node.
[0081] There are many ways to determine the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node. As an example, one of the determination methods may include: extracting the first feature of each behavior node in the historical behavior sequence and the second feature of the candidate behavior node corresponding to each behavior node, and based on the first feature of each behavior node and the second feature of the candidate behavior node corresponding to each behavior node, determining the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node.
[0082] As another example, the first feature of each behavior node in the historical behavior sequence or the second feature of the candidate behavior node corresponding to each behavior node is obtained by extracting the behavior environment characteristics of the behavior node, the distribution capacity characteristics of the behavior node, and the fulfillment behavior type characteristics of the behavior node, and splicing the behavior environment characteristics of the behavior node, the distribution capacity characteristics of the behavior node, and the fulfillment behavior type characteristics of the behavior node.
[0083] As another example, the behavior environment feature of the behavior node may be the weather feature of the behavior node, such as sunny or rainy day, etc., or the time of the behavior node, such as Monday, Tuesday, etc.
[0084] As another example, the delivery capacity feature of the behavior node may represent the delivery level of the delivery capacity, or the average delivery speed of the delivery capacity.
[0085] As another example, the fulfillment behavior type feature of the behavior node can characterize what type of fulfillment behavior the behavior node belongs to, such as the in-store behavior, meal pickup behavior, and meal delivery behavior mentioned above.
[0086] It is worth noting that the above introduction to the specific implementation of the first feature or the second feature is only an exemplary display. In practical applications, other specific implementations are not excluded and are not specifically limited to this.
[0087] As another example, another method of determining the similarity between each behavior node in a historical behavior sequence and each candidate behavior node corresponding to the behavior node may include: obtaining the contextual correlation features that characterize each behavior node in the historical behavior sequence and other behavior nodes in the historical behavior sequence through a recurrent neural network, and fusing the corresponding contextual correlation features with the first features of each behavior node in the historical behavior sequence to obtain the fused features of each behavior node, and then determining the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node based on the fused features of each behavior node and the second features of the candidate behavior node corresponding to each behavior node.
[0088] It can be understood that the above-mentioned correlation can be understood as the order relationship between each behavior node in the historical behavior sequence and other behavior nodes in the historical behavior sequence. For example, Figure 3D As shown, for the historical behavior sequence: I ’ 0→A O4 →D O4 →D O2 In the example, node A is O4 For example, A O4 The correlation with other behavior nodes can be referred to as A O4 with I ’ 0.D O4 , D O2 The order of I ’ 0 in A O4 The previous one, D O4 In A O4 The last one, D O2 In A O4 The last two digits.
[0089] As an example, after determining the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node, the expected predicted behavior of the delivery capacity after the behavior node can be predicted based on the fulfillment behavior type represented by the candidate behavior node with the highest similarity among the candidate behavior nodes corresponding to the behavior node.
[0090] It can be understood that the type of fulfillment behavior represented by the candidate behavior node with the highest similarity among the candidate behavior nodes corresponding to the behavior node is the type of predicted behavior expected by the predicted delivery capacity after the behavior node.
[0091] The structure of the above-mentioned behavior prediction model can have multiple specific implementations. As an example, the above-mentioned behavior prediction model can include a first feature extraction layer, a second feature extraction layer, a recurrent neural network layer, and a network layer based on cross attention (CrossAttention); the first feature extraction layer can be used to extract the first feature of each behavior node in the above-mentioned historical behavior sequence; the second feature extraction layer is used to extract the second feature of the candidate behavior node corresponding to each behavior node; the recurrent neural network layer is used to obtain the before-after correlation feature that characterizes each behavior node in the above-mentioned historical behavior sequence and other behavior nodes in the historical behavior sequence, and fuses the corresponding before-after correlation features for the first feature of each behavior node in the historical behavior sequence to obtain the fused feature of each behavior node; the cross-attention-based network layer is used to calculate the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node, and then determine the expected predicted behavior after each behavior node.
[0092] Below Figure 4 Taking the example of FIG. 1 as an example, the structure of the behavior prediction model of the embodiment of this specification is exemplarily introduced:
[0093] As an example, the first feature extraction layer can be used as follows Figure 4 As another example, the second feature extraction layer may also be implemented as follows: Figure 4 The two DNNs shown in Figure 4 The two blue rectangular boxes DNN*2 are shown in the figure.
[0094] As another example, the first feature of each behavior node in the historical behavior sequence or the second feature of the candidate behavior node corresponding to each behavior node is extracted by DNN to obtain the behavior environment feature of the behavior node (such as Figure 4 The “environmental characteristics” shown in the figure), the distribution capacity characteristics of the behavior nodes (such as Figure 4 The "rider characteristics" shown in the figure) and the fulfillment behavior type characteristics of the behavior node (such as Figure 4 The “current node characteristics” shown in the figure are obtained by splicing the characteristics of the behavior environment of the behavior node, the distribution capacity characteristics of the behavior node, and the fulfillment behavior type characteristics of the behavior node.
[0095] As another example, the above recurrent neural network layer can be used as follows Figure 4The Long Short-Term Memory (LSTM) network shown in FIG. Since the historical behavior sequence input to the model includes multiple behavior nodes, a fixed number of nodes can be set. Therefore, the LSTM network used by the recurrent neural network layer may include multiple neurons (cells), wherein the number of neurons may correspond to the number of behavior nodes in the behavior sequence, and each neuron may extract the fusion features of the corresponding behavior node.
[0096] As another example, a criss-cross attention-based network layer may include Figure 4 The Attention network shown. Similarly, the number of neurons (cells) included in the Attention network also corresponds to the number of behavior nodes in the behavior sequence. The input of each neuron included in the Attention network is the fused features of the neuron output of the corresponding LSTM network, and the second feature of the candidate behavior node corresponding to each behavior node output by the left DNN*2. The output is the similarity between each behavior node in the historical behavior sequence and each candidate behavior node corresponding to the behavior node, and the expected predicted behavior after each behavior node determined based on the similarity between each behavior node and each candidate behavior node corresponding to the behavior node.
[0097] As another example, for the similarity between each behavior node in the historical behavior sequence output by the Attention network and each candidate behavior node corresponding to the behavior node, the probability value corresponding to each candidate behavior node can be determined based on the similarity (e.g. Figure 4 The “probabilities” shown in the figure can represent the possibility of each candidate behavior node being predicted as the above-mentioned predicted behavior.
[0098] like Figure 4 As shown in , the output dimension of each layer of the behavior prediction model can be identified, where B is the batch_size, which can be understood as the number of samples that can be input at one time when training the behavior prediction model, and T is the length of the short sequence (i.e., the historical behavior sequence) (e.g. Figure 4 The length of the example is 4, that is, 4 behavior nodes), F is the number of features of the behavior node, and C is the number of candidate behavior nodes corresponding to each behavior node (for example Figure 4 The "candidate node features" on the left are the features of the candidate behavior nodes corresponding to each behavior node, where C1 to C8 can represent that the behavior nodes of the historical behavior sequence correspond to a total of 8 candidate behavior nodes, that is, the number of candidate behavior nodes C = 8), and D is the length of each feature after embedding (can be set to 16 in the experiment).
[0099] Consider the length of the historical behavior sequence of each input behavior prediction model (such as Figure 4 The length of medium and short sequences T) may be inconsistent, and the number of candidate behavior nodes C corresponding to each behavior node may also be inconsistent, but the data input required by the feature extraction layer of the behavior prediction model may be of fixed length, that is, the number of neurons (cells) contained in the LSTM network and Attention network set in the model is fixed.
[0100] In response to the above problem, as an example, if the above-mentioned first feature extraction layer is a feature extraction layer that receives a first preset input number of behavior nodes, then the historical behavior sequence can be firstly subjected to a padding operation so that the number of behavior nodes in the historical behavior sequence after the padding operation reaches the above-mentioned first preset input number, and then the historical behavior sequence subjected to the padding operation is input into the above-mentioned first feature extraction layer.
[0101] As another example, if the second feature extraction layer is a feature extraction layer that receives a second preset input number of behavior nodes, a padding operation can be first performed on the candidate behavior nodes corresponding to the behavior nodes in the historical behavior sequence, so that the number of candidate behavior nodes corresponding to the behavior nodes in the historical behavior sequence after the padding operation reaches the second preset input number, and then the candidate behavior nodes after the padding operation are input into the second feature extraction layer.
[0102] As an example, assuming that the historical behavior sequence input into the behavior prediction model includes 8 behavior nodes, the behavior prediction model will predict the expected predicted behavior of the delivery capacity after each of the 8 behavior nodes. The 8 behavior nodes correspond to 8 predicted behaviors in total, and errors will occur between the 8 predicted behaviors and the positive labels, that is, each behavior node will have its corresponding positive label (label), and each behavior node will calculate the error with its corresponding label; Figure 4 As shown in , each behavior node independently corresponds to each cell of LSTM and each cell of Attention. Therefore, during the training process, the LSTM and Attention networks can adjust parameters based on the loss function results of the behavior nodes corresponding to each cell. Figure 4 The DNN layer shown does not correspond to each behavior node independently. During the training process, this part of the network can adjust parameters based on the results of the loss function of multiple behavior nodes.
[0103] In the above DNN layer, there are multiple ways to adjust parameters based on the results of the loss functions of multiple behavior nodes. As an example, the average value of the results of the loss functions of multiple behavior nodes can be calculated, and the parameters can be adjusted based on the average value. There is no specific limitation on the way to adjust parameters based on the results of the loss functions of multiple behavior nodes.
[0104] As an example, the behavior prediction model can predict the behavior nodes expected by the delivery capacity, that is, predict the behavior expected by the delivery capacity, such as predicting the delivery capacity's pick-up behavior, store behavior or delivery behavior for the delivery waybill; then, based on the behavior predicted by the behavior prediction model, the location information to which the delivery capacity needs to go can be determined in combination with the information of the corresponding delivery order; for example, if it is predicted that the delivery capacity is about to pick up food for a certain order, then the merchant location of the order is the location to which the delivery capacity is about to go; if it is predicted that the delivery capacity is about to deliver food to a certain order, then the user's delivery address of the order is the location to which the delivery capacity is about to go; further, the delivery route of the delivery capacity can be determined based on the determined location information, and so on.
[0105] Based on the behavior prediction model constructed by the method for constructing the behavior prediction model described in any of the above embodiments, that is, the preset behavior prediction model, the embodiment of this specification also provides a behavior prediction method, such as Figure 5 As shown, the method comprises the following steps:
[0106] S501. Obtain the current behavior sequence of the distribution capacity.
[0107] The current behavior sequence includes behavior nodes of the delivery capacity from the most recent order acceptance moment to the current moment, and the behavior nodes represent different types of fulfillment behaviors of the delivery capacity.
[0108] S502: Determine candidate behavior nodes corresponding to each behavior node in the current behavior sequence.
[0109] Among them, the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the distribution capacity can perform after the behavior node.
[0110] S503. Input each behavior node in the current behavior sequence and the candidate behavior node corresponding to each behavior node into a preset behavior prediction model respectively, and the behavior prediction model predicts the expected fulfillment behavior type of the delivery capacity after the current moment based on the input behavior nodes and the candidate behavior nodes corresponding to each behavior node.
[0111] The behavior prediction model is trained by utilizing the historical behavior sequence of the distribution capacity to learn the actual fulfillment behavior of the distribution capacity after each behavior node in the historical behavior sequence.
[0112] The construction method of the above-mentioned behavior prediction model can refer to the construction method of the behavior prediction model described in any of the embodiments above, and the details will not be repeated here.
[0113] It should be noted that, in the reasoning and application stage of the behavior prediction model, the above-mentioned current moment can be any moment, for example, it can be an order-accepting moment or a non-order-accepting moment, that is, at the current moment, the delivery capacity may or may not be able to accept delivery orders, and there is no specific limitation on this.
[0114] It should be noted that there are differences in the output results of the behavior prediction model during the training phase and the inference application phase. Figure 4 As shown in the figure, assuming that the behavior prediction model has the ability to output 8 probabilities, then in the training stage of the behavior prediction model, the 8 cells of the Attention network will output 8 prediction result probabilities respectively, and the 8 prediction result probabilities are used as the basis for calculating the error. In the reasoning and application stage of the behavior prediction model, only the behavior node corresponding to the current moment in the 8 cells of the Attention network will be taken, and the prediction result probabilities output by the corresponding cell will be used as the output result of the behavior prediction model.
[0115] Taking into account that the prediction output result of the behavior prediction model is the result obtained by only considering the delivery willingness of the delivery capacity, in actual applications, the merchant’s meal delivery time may directly affect whether the delivery capacity can pick up the meal.
[0116] To address this issue, as an example, if the predicted result of the type of fulfillment behavior expected by the delivery capacity after the current moment is the in-store behavior for the target order, then the estimated meal delivery time of the merchant corresponding to the in-store behavior of the target order and the estimated store arrival time of the delivery capacity corresponding to the store behavior of the target order are obtained, and then the predicted result of the type of fulfillment behavior expected after the current moment is adjusted based on the time difference between the merchant's estimated meal delivery time and the estimated store arrival time of the delivery capacity. By adjusting the predicted result of the type of fulfillment behavior expected after the current moment based on the time difference between the merchant's estimated meal delivery time and the estimated store arrival time of the delivery capacity, the predicted result can be calibrated to more realistically restore the fulfillment process of the delivery capacity.
[0117] There are many ways to adjust the prediction results based on the time difference between the merchant's estimated meal delivery time and the estimated store arrival time of the delivery capacity. As an example, if the time difference between the merchant's estimated meal delivery time and the estimated store arrival time of the delivery capacity is less than or equal to the preset threshold, the prediction result of the type of fulfillment behavior expected after the current moment can be changed to the pick-up behavior for the above-mentioned target order; as another example, if the time difference between the merchant's estimated meal delivery time and the estimated store arrival time of the delivery capacity is greater than the preset threshold, the prediction result of the type of fulfillment behavior expected after the current moment can be changed to the fulfillment behavior for orders other than the above-mentioned target order. For example: for order A, if the above time difference is less than or equal to the preset threshold, the prediction result can be changed to the pick-up behavior for order A. If the above time difference is greater than the preset threshold, the prediction result can be changed to the pick-up behavior for order A after the preset time, or the prediction result can be changed to the fulfillment behavior for order B (such as store arrival, pick-up, delivery, etc.) after the preset time. It is worth noting that the above introduction of the method of adjusting the prediction results based on the time difference between the merchant's estimated meal delivery time and the delivery capacity's estimated arrival time at the store is only an exemplary display. In actual application, other adjustment methods are not excluded and are not specifically limited to this.
[0118] As an example, after the behavior prediction model predicts the type of fulfillment behavior that the delivery capacity is expected to take after the current moment, it can determine the location information where the delivery capacity needs to go based on the information of the delivery order corresponding to the behavior node of the predicted fulfillment behavior type, and predict the delivery path of the delivery capacity based on the determined location information.
[0119] It can be seen from the above embodiments that in order to solve the problem of generating rider delivery paths by imitating the fulfillment process of takeaway riders in scheduling simulation, the embodiments of this specification abstract the fulfillment process of riders in the takeaway scenario into a sequence of "arriving at the store, picking up food, and delivering food". Through LSTM and Attention model training, the riders' willingness and behavioral habits to go to different destinations throughout the process are learned, and the predicted rider behavior is output. This method is trained with a large amount of historical data, and only learns the real rider behavior without considering other purposes. It can generate a path sequence that is more in line with the rider's wishes in the simulation system than the operations optimization algorithm, and the model structure is simple, and has a better computing speed than the graph neural network algorithm. At the same time, it can use the merchant's meal delivery time calibration to more flexibly reproduce the fulfillment behavior of real-world riders.
[0120] As an example, the behavior prediction method described in any of the above embodiments can be applied to a scheduling simulation system, which can be used to generate a predicted delivery path that meets the delivery capacity expectations based on the prediction results of the expected fulfillment behavior type of the delivery capacity after the above-mentioned current moment.
[0121] The embodiments of this specification also provide a computer program product, including a computer program, which, when executed by a processor, implements the behavior prediction method or the method for constructing a behavior prediction model described in any of the embodiments above.
[0122] The present specification also provides an electronic device, such as Figure 6 As shown, the electronic device includes:
[0123] Processor 601;
[0124] Memory 602 for storing processor executable instructions;
[0125] The processor 601 is configured to implement the behavior prediction method or the method for constructing a behavior prediction model described in any of the above embodiments.
[0126] The embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the behavior prediction method or the method for constructing a behavior prediction model described in any of the embodiments above is implemented.
[0127] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this specification. A person of ordinary skill in the art can understand and implement it without paying creative labor.
[0128] The above embodiments can be applied to one or more computer devices, where the computer device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions. The hardware of the computer device includes but is not limited to a microprocessor, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), an embedded device, etc.
[0129] The computer device may be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), a game console, an interactive network television (IPTV), a smart wearable device, etc.
[0130] The computer device may also include a network device and / or a user device, wherein the network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud consisting of a large number of hosts or network servers based on cloud computing.
[0131] The network where the computer device is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.
[0132] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.
[0133] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this patent; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the protection scope of this application.
[0134] The description of "specific examples" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of this specification. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0135] Those skilled in the art will readily appreciate other embodiments of the specification after considering the specification and practicing the invention claimed herein. The specification is intended to cover any variations, uses or adaptations of the specification that follow the general principles of the specification and include common knowledge or customary techniques in the art that are not claimed in the specification. The specification and examples are to be considered exemplary only, and the true scope and spirit of the specification are indicated by the following claims.
[0136] It should be understood that the present description is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present description is limited only by the appended claims.
[0137] The above description is only a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this specification should be included in the scope of protection of this specification.
Claims
1. A behavior prediction method, comprising: Obtaining a current behavior sequence of the delivery transport capacity, wherein the current behavior sequence includes behavior nodes of the delivery transport capacity from the most recent order acceptance time to the current time, and the behavior nodes represent different types of fulfillment behaviors of the delivery transport capacity; Determine candidate behavior nodes corresponding to each behavior node in the current behavior sequence, wherein the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the delivery transport capacity can perform after the behavior node; Input each behavior node in the current behavior sequence and the candidate behavior node corresponding to each behavior node into a preset behavior prediction model, and the behavior prediction model predicts the expected performance behavior type of the delivery capacity after the current moment based on the input behavior nodes and the candidate behavior nodes corresponding to each behavior node; The behavior prediction model is trained by utilizing the historical behavior sequence of the distribution capacity to learn the actual fulfillment behavior of the distribution capacity after each behavior node in the historical behavior sequence.
2. According to the method of claim 1, the behavior prediction model is constructed by: Acquire a historical behavior sequence of the delivery transport capacity, wherein the historical behavior sequence includes behavior nodes of the delivery transport capacity between two adjacent order acceptance moments, and the behavior nodes represent different types of fulfillment behaviors of the delivery transport capacity; Determine the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence, wherein: The candidate behavior nodes corresponding to each behavior node represent each type of fulfillment behavior that the delivery capacity can perform after the behavior node; Each behavior node in the historical behavior sequence and the candidate behavior node corresponding to each behavior node are respectively input into a preset behavior prediction model. The behavior prediction model predicts the expected predicted behavior of the distribution capacity after each behavior node based on the input of each behavior node and the candidate behavior node corresponding to each behavior node, and is trained according to the error between the predicted behavior of the behavior node and the actual fulfillment behavior of the distribution capacity in the candidate behavior node corresponding to the behavior node.
3. According to the method of claim 2, the step of obtaining the historical behavior sequence of the distribution capacity comprises: The behavior sequence of the delivery capacity divided by historical delivery waves is obtained, and the behavior sequence belonging to the same delivery wave is segmented according to the order receiving time in the behavior sequence divided by historical delivery waves to obtain the historical behavior sequence.
4. According to the method of claim 2, the different types of fulfillment behaviors of the delivery capacity include: The delivery capacity’s in-store visit, pick-up and / or delivery behavior.
5. According to the method of claim 4, the step of determining the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence comprises: If there are store-arrival behaviors and meal pickup behaviors for the same order among the fulfillment behaviors that the delivery capacity can perform after the behavior node, the meal pickup behavior for the same order will be deleted, and the store-arrival behavior for the same order will be retained as a candidate behavior node for the behavior node.
6. According to the method of claim 4, the step of determining the candidate behavior nodes corresponding to each behavior node in the historical behavior sequence comprises: If there are pick-up behaviors and delivery behaviors for the same order in the fulfillment behaviors that the delivery capacity can perform after the behavior node, the delivery behavior of the same order will be deleted, and the pick-up behavior of the same order will be retained as a candidate behavior node for the behavior node.
7. A method for constructing a behavior prediction model, the method comprising: Acquire a historical behavior sequence of the delivery transport capacity, wherein the historical behavior sequence includes behavior nodes of the delivery transport capacity between two adjacent order acceptance moments, and the behavior nodes represent different types of fulfillment behaviors of the delivery transport capacity; Determine candidate behavior nodes corresponding to each behavior node in the historical behavior sequence, wherein the candidate behavior node corresponding to each behavior node represents each type of fulfillment behavior that the delivery capacity can perform after the behavior node; Each behavior node in the historical behavior sequence and the candidate behavior node corresponding to each behavior node are respectively input into a preset behavior prediction model. The behavior prediction model predicts the expected predicted behavior of the distribution capacity after each behavior node based on the input of each behavior node and the candidate behavior node corresponding to each behavior node, and is trained according to the error between the predicted behavior of the behavior node and the actual fulfillment behavior of the distribution capacity in the candidate behavior node corresponding to the behavior node.
8. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
9. An electronic device, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method according to any one of claims 1 to 7 when executed by a processor.
