Task processing duration determination method and device and electronic equipment
By obtaining and processing information from the service requester, service provider and pre-selected service executor, we can determine whether the task can be executed separately, and solve the problem of low accuracy of task processing time prediction in the prior art, improving user experience and task processing time accuracy.
Patent Information
- Application Number
- CN202411064101.5
- 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
The prior art has problems with low accuracy in task processing time prediction, especially due to negative migration problems caused by a single gating mechanism and time-segment prediction noise, which affects the accuracy of service fulfillment time.
By obtaining information from the service requester, the service provider and the pre-selected service executor, feature processing is performed to obtain time prediction features and execution parameters, and to determine whether the task can be executed separately, thereby determining the task processing time and improving prediction accuracy.
It improves the accuracy of task processing time, enhances the user experience, avoids the impact of low-probability events on the output results, and enhances the prediction accuracy of task processing time.
Smart Images

Figure CN118981380A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and more specifically, to a method, device and electronic device for determining task processing duration. Background Art
[0002] As a key component of various services, task processing time prediction is related to the task execution efficiency and user satisfaction of the task executor. Therefore, the accuracy of task processing time is crucial. Summary of the invention
[0003] In view of this, an embodiment of the present invention provides a method, device and electronic device for determining task processing duration, so as to refer to whether the task can be executed alone when predicting the task processing duration, thereby improving the accuracy of the task processing duration and further improving the user experience.
[0004] In a first aspect, an embodiment of the present invention provides a method for determining task processing duration, the method comprising:
[0005] Acquire task information, wherein the task information includes service requester information, service provider information, and pre-selected service executor information;
[0006] Performing feature processing on the information of the task to obtain duration prediction features and execution parameters, where the execution parameters are used to characterize the probability of the task being executed alone;
[0007] The task processing duration is determined according to the duration prediction feature and the execution parameter.
[0008] In a second aspect, an embodiment of the present invention provides a device for determining task processing duration, the device comprising:
[0009] An information acquisition unit is configured to acquire information of a task, wherein the information of the task includes information of a service requester, information of a service provider, and information of a pre-selected service executor;
[0010] A processing unit is configured to perform feature processing on the information of the task to obtain a duration prediction feature and an execution parameter, wherein the execution parameter is used to characterize a probability that the task is executed alone;
[0011] The duration determination unit is configured to determine the task processing duration according to the duration prediction feature and the execution parameter.
[0012] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in the first aspect of the embodiment of the present invention.
[0013] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect of the embodiment of the present invention is implemented.
[0014] In a fifth aspect, an embodiment of the present invention provides a computer program product. When the computer program product runs on a computer, the computer executes the method described in the first aspect of the embodiment of the present invention.
[0015] The embodiment of the present invention obtains task information including service requester information, service provider information, and pre-selected service executor information, performs feature processing on the task information, obtains duration prediction features and execution parameters corresponding to the task, and determines the task processing duration of the task according to the duration prediction features and execution parameters, wherein the execution parameters are used to characterize the probability of the task being executed alone. Therefore, this embodiment can refer to whether the task can be executed alone when predicting the task processing duration, thereby improving the accuracy of the task processing duration and further improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:
[0017] Figure 1 It is a structural diagram of a time estimation model for comparison ratio;
[0018] Figure 2 is a flow chart of a method for determining task duration according to an embodiment of the present invention;
[0019] Figure 3 is a flow chart of a method for determining a pre-selected service executor according to an embodiment of the present invention;
[0020] Figure 4 is a structural diagram of a duration determination model according to an embodiment of the present invention;
[0021] Figure 5 is a flow chart of another method for determining task duration according to an embodiment of the present invention;
[0022] Figure 6 is a structural diagram of another duration determination model according to an embodiment of the present invention;
[0023] Figure 7 is a structural diagram of another duration determination model according to an embodiment of the present invention;
[0024] Figure 8 is a schematic diagram of a device for determining task duration according to an embodiment of the present invention;
[0025] Fig. 9 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0026] The present application is described below based on embodiments, but the present application is not limited to these embodiments. In the detailed description of the present application below, some specific details are described in detail. It is possible for those skilled in the art to fully understand the present application without the description of these details. In order to avoid confusing the essence of the present application, known methods, processes, flows, components and circuits are not described in detail.
[0027] In addition, persons of ordinary skill in the art will appreciate that the drawings provided herein are for illustration purposes and are not necessarily drawn to scale.
[0028] Unless the context clearly requires otherwise, the words "include", "comprising" and similar words throughout the application should be interpreted as including rather than exclusive or exhaustive; that is, the meaning is "including but not limited to".
[0029] In the description of this application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" is two or more.
[0030] The solutions described in this specification and in the examples, if they involve the processing of personal information, will be processed on the premise of having a legal basis (such as obtaining the consent of the subject of personal information, or being necessary for the performance of a contract, etc.), and will only be processed within the scope of regulations or agreements. If a user refuses to process personal information other than the necessary information for basic functions, it will not affect the user's use of basic functions.
[0031] Before providing the corresponding service, estimating the service performance time is a key step. For example, in a food delivery application scenario or a same-city express delivery application scenario, the estimated delivery time is displayed when the user enters the bill of lading page of the order. For another example, in an online car-hailing application scenario, the travel time or estimated arrival time is displayed when the user enters the bill of lading page of the order. This embodiment is described in detail below mainly using the food delivery application scenario as an example, but it should be understood that this embodiment is not limited to this, and it can be applied to any application scenario where service performance time estimation is required. Furthermore, the "order" referred to in this embodiment corresponds to the "task", and both represent a corresponding service.
[0032] Figure 1It is a structural diagram of a comparative time estimation model. In a comparative example for implementing time estimation, the task fulfillment time corresponding to the service is divided into time segments according to the corresponding task processing nodes, and the duration prediction task of each time segment is used as a subtask of the multi-task model. The duration of each time segment is estimated through the multi-task model, and then the task fulfillment time is estimated. Taking the takeaway application field as an example, the task nodes of the takeaway task may include placing an order, receiving an order, arriving at the store, picking up the food, and delivering the food. The subtasks of the multi-task model may include the time prediction subtask of placing an order-receiving an order, the time prediction subtask of receiving an order-arriving at the store, and the time prediction subtasks of arriving at the store-picking up the food and picking up the food-delivering the food. Specifically, if Figure 1 As shown, taking the takeaway task as an example, category features (such as merchant category, product category, etc.), continuous features (such as historical average delivery time, meal delivery time, etc.), and discretized features after discretizing the continuous features are obtained.
[0033] Furthermore, the multi-task model of the comparative example includes a gated GATE, a DNN+BN+Leaky network, a DCN network and a repair module.
[0034] Among them, the gate GATE is used to dynamically determine which information should be retained or ignored based on contextual information. DNN+BN+Leaky is used to perform feature processing on categorical features and continuous features. Among them, DNN (Deep Neural Network) is a network composed of multiple layers of artificial neurons, which are connected together by weights and biases. BN (Batch Normalization) is a technology used to accelerate deep network training and improve model performance. Leaky ReLU (Leaky Rectified Linear Unit) is an activation function used to introduce nonlinearity to neural networks. DCN (Deep & Cross Network) is a network that combines deep neural networks (DNN) and cross networks (Cross Network) to solve the learning problem of large-scale sparse features. The correction module is used to make necessary adjustments and optimizations to the model, such as data cleaning and data screening.
[0035] In this comparative example, Figure 1As shown in the figure, the DNN+BN+Leaky network processes the category features and continuous features, and the DCN network processes the category features and discrete features to obtain the output results of each subtask. The gated GATE processes the category features and discrete features, and the repair module processes the category features and discrete features. Furthermore, the multi-task model obtains the product of the corresponding weights and the output results of the gated GATE and the output results of each subtask, and fuses the product and the output results of the repair module. Thus, the final fulfillment time can be obtained through the output results of each subtask, the above product and the fusion feature prediction.
[0036] Furthermore, the comparison ratio can also calculate the loss based on the above product and the output results of each subtask, the city dimension ME related features and the related compensation time and bottom-up logic to train the multi-task model.
[0037] Although this comparative example uses a multi-task time estimation model to improve the accuracy of service fulfillment time estimation, it uses a single gating mechanism to regulate multiple tasks, so that the underlying parameters are shared between each subtask, which is prone to negative transfer problems. In addition, this comparative example only divides the entire fulfillment time into multiple time segments as subtasks. It does not introduce additional effective information, but is prone to adding noise from the prediction of each time segment. Therefore, this comparative example still has the problem of low accuracy in fulfillment time estimation. Therefore, an embodiment of the present invention provides a method, device, and electronic device for determining task processing duration, so as to refer to whether the task can be executed alone when predicting the task processing duration, thereby improving the accuracy of the task processing duration, and further improving the user experience.
[0038] Figure 2 FIG. 1 is a flow chart of a method for determining task duration according to an embodiment of the present invention. Figure 2 As shown, the method for determining the task duration in an embodiment of the present invention includes the following steps:
[0039] Step S110, obtaining task information, wherein the task information includes service requester information, service provider information and pre-selected service executor information.
[0040] In an embodiment of the present invention, the service requester information may include the identification of the service requester and the location information input when requesting the service. Further optionally, the service requester information may also include the area information (such as business district information, POI information, AOI information, etc.) where the service requester is located, the service type, the target object type corresponding to the service, and / or historical evaluation information. Optionally, the area information corresponding to the service requester may include the area identification, the historical task volume information in the area, and / or the historical task execution time, etc. Further, the historical task volume information in the area may include the task volume and / or task execution time of a predetermined time segment (such as 10 minutes, 20 minutes, 30 minutes, 60 minutes, etc.) in the past day with the area as the task destination, and may also include the historical task volume and / or execution time in the same time segment as the current task in the historical predetermined time period (such as within a week, within half a month, within a month, etc.) with the area as the task destination, and may also include the historical daily average task volume and / or historical average execution time of the historical predetermined time period (such as within a week, within half a month, within a month, etc.) with the area as the task destination. It should be understood that this embodiment does not limit the specific information type in the service requester information. Taking the takeaway application scenario as an example, the service requester can be a user who orders food, and the service requester information can include a user ID and a delivery address. Optionally, the service type can be a group order type or a non-group order type, and the target object type corresponding to the service can be a food type.
[0041] The service provider information is used to provide the corresponding service or the target object corresponding to the service. Optionally, the service provider information includes a combination of one or more of the following: historical task volume information corresponding to the service provider, historical task time information, regional information of the service provider, historical evaluation information, etc. The historical task volume information of the service provider may include the task volume within a predetermined time segment in the past of the day, the task volume in the same time segment as the current task in the historical predetermined time period, or the daily average task volume in the historical predetermined time period, etc. The historical task time information of the service provider may include the historical task execution time and / or the estimated time of the historical task within a predetermined time segment in the past of the day, the historical task execution time and / or the estimated time of the historical task in the same time segment as the current task in the historical predetermined time period, or the daily average execution time of the tasks in the historical predetermined time period, etc.
[0042] The regional information of the service provider may include the regional identifier, historical task volume information in the region, and / or historical task execution time, etc. Further, the historical task volume information in the region may include the task volume and / or task execution time for a predetermined time segment in the past of the day with the region as the task starting point, and may also include the historical task volume and / or execution time in the same time segment as the current task in the historical predetermined time period with the region as the task starting point, and may also include the historical daily average task volume and / or historical average execution time for the historical predetermined time period with the region as the task starting point. In this embodiment, the information type of service providers of different services may be different, and the type of service provider information may be determined according to the specific service type and the actual situation of the application scenario.
[0043] Taking the takeaway application scenario as an example, the service provider is a merchant that provides food. The service provider information may include a combination of one or more of the following: the service provider's identification, the service provider's location, the area identification where the service provider's location is located, the service provider's order volume and order volume within the past scheduled time segments of the day, the service provider's historical meal delivery time, the corresponding average delivery time, the estimated delivery time of historical orders, and the historical order volume (for example, the order volume within the same time segment within the past scheduled time period, or the order volume per day within the past scheduled time period, etc.). In the takeaway application scenario, other information related to delivery can be applied to this embodiment, and will not be given one by one here.
[0044] Optionally, in this embodiment, each day can be divided into multiple identical or different time segments, a long time period can be divided into different time segments based on a day, week, or month, and can also be divided into types such as working days, weekends, and holidays based on different time types. The task information of this embodiment can also include a time segment identifier (used to identify which time segment in a day), a time identifier (used to identify which day in a month or the name of the week in a week), and / or a time type identifier, etc.
[0045] In an optional implementation, the task information of this embodiment also includes information obtained by preprocessing the original task information. Optionally, the preprocessing may include discrete processing. Thus, the task information of this embodiment also includes discrete features obtained by discrete processing the continuous features.
[0046] In an optional implementation, the task information of this embodiment may also include map data, environmental data such as weather, or road condition data, so as to further improve the prediction accuracy of the task processing time.
[0047] Figure 31 is a flow chart of a method for determining a pre-selected service executor according to an embodiment of the present invention. The pre-selected service executor is used to represent the executor that may be selected as the executor of the task. Figure 3 As shown, the method for determining the pre-selected service executor in this embodiment includes the following steps:
[0048] Step S210, recall multiple candidate service executors according to the location of the service provider. Optionally, recall multiple candidate service executors within a predetermined range around the service provider. It should be understood that this embodiment does not limit the specific value of the predetermined range, which can be determined according to the service type and different service time segments.
[0049] Step S220 , screening multiple candidate service executors according to the location and bound task information of each candidate service executor to obtain multiple pre-selected service executors.
[0050] Optionally, the bound task information of the candidate service executor may include the bound task amount, that is, the number of tasks being executed. This embodiment may screen according to the distance between the candidate service executor and the service provider and the bound task amount of each candidate service executor, and obtain the candidate service executors that meet the predetermined conditions as pre-selected service executors. Further optionally, the predetermined conditions may be: the distance between the candidate service executor and the service provider is less than a distance threshold, and / or the bound task amount is less than a task amount threshold. In other optional implementations, this embodiment may also score each candidate service executor based on the distance between the candidate service executor and the service provider and the bound task amount of each candidate service executor, and use the candidate service executor with a score greater than a predetermined value as the pre-selected service executor, or use a predetermined number of candidate service executors with the highest score as the pre-selected service executors. This embodiment can use any scoring method to calculate the score of each candidate service executor, for example, different distances have corresponding distance scores, different bound task amounts have corresponding task amount scores, and then obtain the final score of the candidate service executor based on the distance score and task amount score. It can also pre-set the functional relationship between the distance, bound task amount and score, and determine the final score of the candidate service executor based on the functional relationship. It should be understood that this embodiment does not limit the scoring method, which can be set according to actual conditions.
[0051] In another optional implementation, the bound task information of the candidate service executor of this embodiment may also include the location information of the bound task. Since the execution time of the current task is greatly affected by the execution of other bound tasks by the service executor, for example, if the location of the bound task of the service executor and the current task is located in two opposite directions of the service provider, then it is obvious that the service executor is not suitable for undertaking the current task. Therefore, this embodiment can also screen the recalled multiple candidate service executors based on information such as the distance between the candidate service executor and the service provider, the amount of bound tasks of each candidate service executor, the location information of the bound tasks, and the location of the service requester to obtain the pre-selected service executor.
[0052] In this embodiment, the pre-selected service executor information may include one or more of the following combinations: the pre-selected service executor's identification, location information, bound task information, historical average task volume information, historical average execution time information, historical evaluation information, historical execution timeout information, etc. Taking the takeaway scenario as an example, the service executor may be a rider who performs a delivery task.
[0053] The above examples only illustrate some of the service requester information, service provider information and service executor information. It should be understood that this embodiment does not limit various types of specific information. In a specific service type, any information that can affect the task execution time can be applied to this embodiment.
[0054] Step S120, feature processing is performed on the task information to obtain duration prediction features and execution parameters, wherein the execution parameters are used to characterize the probability of the task being executed alone.
[0055] Step S130, determining the task processing duration according to the duration prediction characteristics and the execution parameters.
[0056] In an optional implementation, regardless of the size of the execution parameter, it can participate in the calculation of the task processing duration to reduce the complexity of the data processing logic. In another optional implementation, since the impact on the execution time is greater only when the current task is executed alone, this embodiment allows the execution parameter to participate in the calculation of the task processing duration when it is determined that the current task is likely to be executed alone. Specifically, in response to the execution parameter being greater than a predetermined parameter threshold, this embodiment determines the task processing duration based on the duration prediction feature and the execution parameter, and in response to the execution parameter being not greater than the predetermined parameter threshold, determines the task processing duration based on the duration prediction feature. In this way, the influence of low-probability events on the output results can be avoided, further improving the prediction accuracy of the task processing duration.
[0057] In an optional implementation, the task processing duration of this embodiment is obtained by training a pre-trained duration determination model. Among them, the duration determination model of this embodiment includes a feature processing network, a duration prediction network and a parameter determination network. The feature processing network is used to perform feature processing on the information of the task and obtain feature information of the task. The parameter determination network is used to output the execution parameters corresponding to the task according to the feature information of the task. In an optional implementation, the duration prediction network is used to output the task processing duration according to the feature information of the task and the execution parameters of the task. In another optional implementation, the duration prediction network is used to output the task processing duration according to the feature information of the task and the execution parameters of the task when the execution parameter is greater than a predetermined parameter threshold, and output the task processing duration according to the feature information of the task when the execution parameter is not greater than the predetermined parameter threshold.
[0058] In an optional implementation, the duration prediction network and the parameter determination network in the duration determination model in this embodiment may share the same feature processing network or may use different feature processing networks, and this embodiment is not limited to this.
[0059] In an optional implementation, the duration determination model of this embodiment is a multi-task model. Among them, the duration prediction network and the parameter determination network are respectively used as corresponding subtask processing networks, that is, the duration prediction network is used to perform the task processing duration determination subtask, and the parameter determination network is used to perform the parameter determination subtask, that is, the parameter determination network focuses on learning the knowledge required to execute the parameters, and the duration prediction network focuses on learning the knowledge required to estimate the duration. Therefore, this embodiment can achieve relative isolation between different subtasks by setting independent specific expert networks for subtasks, avoid mutual influence on subtasks during feature information processing, and further improve the accuracy of duration prediction tasks.
[0060] It should be understood that this embodiment does not limit the specific network type used by the duration prediction network and the parameter determination network. It can be implemented using a fully connected network, a CNN network (Convolutional Neural Network), an RNN network (Recurrent Neural Network), a Transformer and other network types. This embodiment does not limit this and it can be selected based on the data characteristics of the specific application scenario.
[0061] Figure 4 Schematic diagram of a duration determination model according to an embodiment of the present invention. Figure 4As shown, the duration determination model 4 of this embodiment includes a feature processing network 41, a parameter determination network 42 and a duration prediction network 43. Among them, the feature processing network 41 performs feature processing on the acquired task information to obtain the feature information of the task, and the parameter determination network 42 processes the feature information of the task and outputs the execution parameter. The duration prediction network 43 processes the feature information of the task to obtain the duration prediction feature, and processes the duration prediction feature and the execution parameter to obtain the task processing duration T. In an optional implementation, when the duration prediction network 43 determines that the execution parameter is greater than the parameter threshold, the duration prediction feature is fused with the execution parameter to obtain the task processing duration T, and when the execution parameter is not greater than the parameter threshold, the duration prediction feature is processed to obtain the task processing duration T. In this way, the influence of low-probability events on the output results can be avoided, and the prediction accuracy of the task processing duration is further improved.
[0062] In a further optional implementation, the duration determination model of the present embodiment also includes a shared expert network. Since there may be a certain degree of correlation between different tasks, the duration determination model of the present embodiment can capture the common features between multiple tasks through a shared expert network, and avoid each task learning these features separately, thereby reducing the amount of calculation and the number of parameters. Thus, the duration determination model of the present embodiment can make the duration prediction network and the parameter determination network as a specific task expert network to focus on the learning of features unique to the duration prediction task and the execution parameter determination task, so that the shared expert network captures the common features between the duration prediction task and the execution parameter determination task, thereby further achieving relative isolation between tasks while reducing the amount of calculation and the number of parameters.
[0063] It should be understood that this embodiment does not limit the specific network type adopted by the shared expert network. It can be implemented by using a shared underlying network, a multi-gated hybrid expert network (MMoE), or an attention mechanism network. This embodiment does not limit this, and it can be selected based on the data characteristics of the specific application scenario.
[0064] Figure 5 FIG. 1 is a flow chart of another method for determining task duration according to an embodiment of the present invention. Figure 5 As shown, the method for determining the task duration in an embodiment of the present invention includes the following steps:
[0065] Step S310: performing feature processing on the task information based on the feature processing network to obtain feature information of the task.
[0066] Step S320: Generate general feature information based on the feature information of the task based on the shared expert network, wherein the general feature information is used to characterize the common features required by the duration prediction network and the parameter determination network.
[0067] Step S330 , obtaining duration prediction features based on the duration prediction network according to the feature information of the task and the general feature information.
[0068] Step S340: The parameter-based determination network acquires execution parameters according to the feature information and general feature information of the task.
[0069] Step S350, obtaining the task processing duration based on the duration prediction network according to the duration prediction features and execution parameters.
[0070] Figure 6 Schematic diagram of another duration determination model in an embodiment of the present invention. Figure 6 As shown, the duration determination model 6 of this embodiment includes a feature processing network 61, a parameter determination network 62, a duration prediction network 63 and a shared expert network 64. Among them, the feature processing network 61 performs feature processing on the acquired task information to obtain the feature information of the task, the shared expert network 64 processes the feature information of the task to generate general feature information, and the parameter determination network 62 processes the feature information and general feature information of the task and outputs the execution parameter. The duration prediction network 63 processes the feature information and general feature information of the task to obtain the duration prediction feature, and processes the duration prediction feature and the execution parameter to obtain the task processing duration T. In an optional implementation, when the duration prediction network 63 determines that the execution parameter is greater than the parameter threshold, the duration prediction feature and the execution parameter are fused to obtain the task processing duration T. When the execution parameter is not greater than the parameter threshold, the duration prediction feature is processed to obtain the task processing duration T. In this way, the influence of low-probability events on the output results can be avoided, and the prediction accuracy of the task processing duration is further improved.
[0071] In an optional implementation, the feature processing network of this embodiment can be any deep learning network, such as a DNN network, a CNN network, an RNN network, etc., or other network models based on a deep learning network as the infrastructure, and this embodiment is not limited to this.
[0072] In another optional implementation, the feature processing network of this embodiment may include a deep feature extraction module and a sparse data processing module. Furthermore, the feature processing network of this embodiment performs feature processing on the information of the task, and obtaining the feature information of the task may specifically be: inputting the information of the task into the deep feature extraction module for processing to obtain the first feature information of the task, and inputting at least part of the information of the task into the sparse data processing module for processing to obtain the second feature information of the task. Among them, all the information of the task can be input into the sparse data processing module for processing, or other information except the service executor information can be input into the sparse data processing module for processing, or only the sparse data in the task information can be input into the sparse data processing module for processing. This embodiment does not limit this, and it can be set according to actual conditions. Therefore, this embodiment can extract the feature information in the task information through the deep feature extraction module, and process the sparse data through the sparse data processing module, so as to capture the relationship between the sparse data features and further improve the prediction performance of the model.
[0073] In an optional implementation, the deep feature extraction module of this embodiment can use a DNN network. It should be understood that this embodiment is not limited to this, and it can also use a DNN network, a CNN network, an RNN network, etc., or other network models based on a deep learning network.
[0074] In an optional implementation, the sparse data processing module of this embodiment can be a FM model (Factorization Machines). The FM model is a machine learning algorithm for processing sparse prediction tasks, which can solve the prediction problem under large-scale sparse data, and can capture the nonlinear relationship between features by introducing interaction terms between features, thereby improving the prediction performance of the model. It should be understood that the sparse data processing module of this embodiment is not limited to the FM model, and other networks that can process sparse data, such as the DCN (Deep & Cross Network) model, can be applied to the embodiments of the present invention.
[0075] In an optional implementation, the duration prediction network of this embodiment includes at least one first task expert network module and a first output module. The parameter determination network includes at least one second task expert network module and a second output module. Further, the shared expert network includes at least one shared network module. Among them, the number of first task expert network modules, the number of second task expert network modules, and the number of shared network modules can be set according to the specific application scenario and the complexity of the data in the application scenario. Further optionally, the number of first task expert network modules, the number of second task expert network modules, and the number of shared network modules can be the same or different, and this embodiment does not limit the setting of the specific number.
[0076] In a further optional implementation, the first task expert network module includes a first task expert network layer and a first gating unit. The first task expert network layer is used to receive the output of the feature processing network or the output of the previous first task expert network module for processing. The first gating unit is used to receive the input and output of the first task expert network layer for processing. The second task expert network module includes a second task expert network layer and a second gating unit. The second task expert network layer is used to receive the output of the feature processing network or the output of the previous second task expert network module for processing. The second gating unit is used to receive the input and output of the second task expert network layer for processing.
[0077] Further optionally, the last layer of shared network modules in the shared expert network includes a shared network layer, and the shared network modules other than the last layer in the shared expert network include a shared network layer and a shared gating unit. The shared network layer is used to receive the output of the feature processing network or the output of the previous shared network module for processing, and the shared gating unit is used to receive the input and output of the corresponding shared network layer, the output of the corresponding first task expert network layer, and the output of the second task expert network layer in the corresponding second task expert network module for processing.
[0078] Further optionally, the first gating unit is used to receive the input and output of the first task expert network layer and the output of the corresponding shared network layer for processing.
[0079] In an optional implementation, the second output module of the parameter determination network is used to receive the output of the last second task expert network module for processing to obtain the execution parameters of the task. The first output module of the duration prediction network is used to receive the output of the last first task expert network module and the execution parameters output by the second output module for processing to obtain the task processing duration.
[0080] In another optional implementation, the second output module of the parameter determination network is used to receive the output of the last second task expert network module and the output of the sparse data processing module in the feature processing network for processing to obtain the execution parameters of the task. The first output module of the duration prediction network is used to receive the output of the last first task expert network module, the output of the sparse data processing module in the feature processing network, and the execution parameters output by the second output module for processing to obtain the task processing duration. Therefore, this embodiment can re-add shallow features to the output modules of the parameter determination network and the duration prediction network, avoiding the loss of shallow features during the processing process and further improving the accuracy of the task processing duration.
[0081] Figure 7 Schematic diagram of another duration determination model in an embodiment of the present invention. Figure 7 As shown, the embodiment of the present invention first obtains the information of the current task, wherein the information of the current task includes the service executor information dt1, the continuous feature information dt2 and the discrete feature information dt3. Among them, this embodiment obtains the service executor information dt1 by recalling and screening the service executor. It should be understood that the service executor information dt1 of this embodiment is obtained in the same way as Figure 3 The embodiments described are similar and will not be described again here.
[0082] Optionally, in this embodiment, the continuous feature information dt2 may include real-time statistical data and offline statistical data extracted from the service requester information, the service provider information and the service executor information, and may also include continuous feature information extracted from map data, environmental data, etc. The real-time statistical data may include the task volume and / or task execution time of a predetermined time segment in the past of the day with the area where the service requester is located as the task destination; the task volume, task execution time and / or task estimated time in a predetermined time segment in the past of the day corresponding to the service provider; the task volume and / or task execution time of a predetermined time segment in the past of the day with the area where the service provider is located as the task starting point. Offline statistical data may include the historical task volume and / or execution time in the same time segment as the current task in the historical scheduled time period, the historical daily average task volume and / or historical average execution time in the historical scheduled time period, etc., with the area where the service requester is located as the task destination; the task volume in the same time segment as the current task in the historical scheduled time period corresponding to the service provider, the historical task execution time, the estimated time of the historical task, or the daily average task volume and daily average execution time in the historical scheduled time period, etc.; with the area where the service provider is located as the task starting point, the historical task volume and / or execution time in the same time segment as the current task in the historical scheduled time period, the historical daily average task volume and / or historical average execution time in the historical scheduled time period, etc. It should be understood that the above-mentioned real-time statistical data and offline statistical data of continuous features are only exemplary, and any real-time statistical or offline statistical continuous features that can affect the task execution time in the relevant information of the task can be applied to this embodiment.
[0083] The discrete feature information dt3 may include discrete features after the continuous feature information dt2 is discretized, and also includes other discrete features in the task information, such as ID-type features, etc. Among them, the ID-type features may include the service requester ID, the service requester's region ID, the service requester's category ID, the service provider ID, the service provider's region ID, the time segment ID of the current task, the time type ID, the time ID (used to identify which day of the month or the name of the week in a week), etc. It should be understood that any discrete feature information that can affect the task execution time in the task-related information of this embodiment can be applied to this embodiment, and no examples will be given one by one here.
[0084] like Figure 7As shown, the duration determination model 7 of this embodiment includes a feature processing network 71, which is used to perform feature processing on the information of the task to obtain the feature information of the task. Among them, the feature processing network 71 includes a sparse data processing module 712. The deep feature extraction module 711 processes the service execution party information dt1, the continuous feature information dt2 and the discrete feature information dt3 to obtain the first feature information f1, and the sparse data processing module 712 processes the continuous feature information dt2 and the discrete feature information dt3 to obtain the second feature information f2.
[0085] The duration determination model 7 of this embodiment also includes a duration prediction network 72 and a parameter determination network 73. The parameter determination network 73 is used to output the execution parameters corresponding to the current task according to the characteristic information {f1, f2} of the task. The duration prediction network 72 is used to determine the task processing duration T according to the characteristic information {f1, f2} of the task and the execution parameters output by the parameter determination network 73. Furthermore, the duration determination model 7 also includes a shared expert network 74, which is used to determine the common characteristic information according to the characteristic information {f1, f2} of the task.
[0086] Specifically, Figure 7 As shown, the duration prediction network 72 includes at least one first task expert network module and a first output module. This embodiment is described in detail by taking the duration prediction network 72 including the first task expert network module 721, the first task expert network module 722 and the first output module 723 as an example. Among them, the first task expert network module 721 includes a first task expert network layer 721A and a first gating unit 721B. The first task expert network module 722 includes a first task expert network layer 722A and a first gating unit 722B. The first output module 723 includes a first tower network layer 723A and a first output layer 723B. Optionally, the duration prediction network of this embodiment can be implemented using a fully connected network, a CNN network (Convolutional Neural Network), an RNN network (Recurrent Neural Network), a Transformer and other network types. Therefore, the above-mentioned first task expert network layer 721A, first task expert network layer 722A, first tower network layer 723A and first output layer 723B can be implemented by a combination of one or more of the corresponding fully connected layers, convolutional neural network layers, recurrent neural network layers, encoders and decoder layers, and this embodiment is not limited to this.
[0087] The parameter determination network 73 includes at least one second task expert network module and a second output module. This embodiment is described in detail by taking the parameter determination network 73 including the second task expert network module 731, the second task expert network module 732 and the second output module 733 as an example. Among them, the second task expert network module 731 includes a second task expert network layer 731A and a second gating unit 731B. The second task expert network module 732 includes a second task expert network layer 732A and a second gating unit 732B. The second output module 733 includes a second tower network layer 733A and a second output layer 733B. Optionally, the parameter determination network 73 of this embodiment can be implemented using a fully connected network, a CNN network (Convolutional Neural Network), an RNN network (Recurrent Neural Network), a Transformer and other network types. Therefore, the above-mentioned second task expert network layer 731A, second task expert network layer 732A, second tower network layer 733A and second output layer 733B can be implemented by a combination of one or more of the corresponding fully connected layers, convolutional neural network layers, recurrent neural network layers, encoder and decoder layers, and this embodiment is not limited to this.
[0088] Further, the shared expert network 74 includes at least one shared network module. The last layer of shared network modules in the shared expert network 74 includes a shared network layer, and the shared network modules except the last layer in the shared expert network include a shared network layer and a shared gating unit. This embodiment is described in detail by taking the shared expert network 74 including a shared network module 741 and a shared network module 742 as an example. Among them, the shared network module 741 includes a shared network layer 741A and a shared gating unit 741B. The shared network module 742 includes a shared network layer 743A.
[0089] It should be understood that in this embodiment, the number of first task expert network modules, the number of second task expert network modules, and the number of shared network modules can be set according to the specific application scenario and the complexity of the data in the application scenario. Further, optionally, the number of first task expert network modules, the number of second task expert network modules, and the number of shared network modules can be the same or different, and this embodiment does not limit the setting of specific numbers.
[0090] In this embodiment, the first task expert network layer is used to receive the output of the feature processing network or the output of the previous first task expert network module for processing. The first gating unit is used to receive the input and output of the first task expert network layer, and the output of the corresponding shared network layer for processing. The second task expert network layer is used to receive the output of the feature processing network or the output of the previous second task expert network module for processing. The second gating unit is used to receive the input and output of the second task expert network layer for processing, and the output of the corresponding shared network layer for processing. The shared network layer is used to receive the output of the feature processing network or the output of the previous shared network module for processing, and the shared gating unit is used to receive the input and output of the corresponding shared network layer, the output of the corresponding first task expert network layer, and the output of the second task expert network layer in the corresponding second task expert network module for processing. In this embodiment, the corresponding network modules are located at the same layer of the multi-task model (that is, the duration determination model), for example, the first task expert network module 721, the second task expert network module 731 and the shared network module 741 are located at the same layer of the duration determination model, and the first task expert network module 722, the second task expert network module 732 and the shared network module 742 are located at the same layer of the duration determination model.
[0091] Specifically, Figure 7 As shown, the first task expert network layer 721A processes the feature information {f1, f2} of the task output by the feature processing network 71, and outputs the result to the first gating unit 721B and the shared gating unit 741B. The second task expert network layer 731A processes the feature information {f1, f2} of the task output by the feature processing network 71, and outputs the result to the second gating unit 731B and the shared gating unit 741B. The shared network layer 741A processes the feature information {f1, f2} of the task output by the feature processing network 71, and outputs the result to the shared gating unit 741B, the first gating unit 721B, and the second gating unit 731B.
[0092] The first gating unit 721B processes the received input (i.e., features {f1, f2}) and output of the first task expert network layer 721A, and the output of the shared network layer 741A, and outputs the processing results to the first task expert network module 722. The second gating unit 731B processes the received input (i.e., features {f1, f2}) and output of the second task expert network layer 731A, and the output of the shared network layer 741A, and outputs the processing results to the second task expert network module 732. The shared gating unit 741B processes the received input (i.e., features {f1, f2}) and output of the shared network layer 741A, the output of the first task expert network layer 721A, and the output of the second task expert network layer 731A, and outputs the processing results to the shared network module 742.
[0093] Further, the first task expert network layer 722A processes the processing result output by the previous first task expert network module 721 (i.e., the first gating unit 721B), and outputs the result to the first gating unit 722B. The second task expert network layer 732A processes the processing result output by the previous second task expert network module 731 (i.e., the second gating unit 731B), and outputs the result to the second gating unit 732B. The shared network layer 742A processes the processing result output by the previous shared network module 741 (i.e., the shared gating unit 741B), and outputs the processing result to the first gating unit 722B and the second gating unit 732B.
[0094] The first gating unit 722B processes the received input and output results of the first task expert network layer 722A and the output result of the shared network layer 742A, and outputs the processing results to the first output module 723. The second gating unit 732B processes the received input and output results of the second task expert network layer 732A and the output result of the shared network layer 742A, and outputs the processing results to the second output module 733.
[0095] Furthermore, the first tower network layer 723A in the first output module 723 processes the output result of the first gating unit 722B and the second feature information f2 output by the sparse data processing module 712, and outputs the processing result (i.e., the above-mentioned duration prediction feature) to the first output layer 723B. The second tower network layer 733A in the second output module 733 processes the output result of the second gating unit 732B and the second feature information f2 output by the sparse data processing module 712, and outputs the processing result to the second output layer 733B. Therefore, this embodiment can re-add shallow features to the output modules of the parameter determination network and the duration prediction network, avoiding the loss of shallow features during the processing process, and further improving the accuracy of the task processing duration.
[0096] The second output layer 733B decodes the processing result of the second tower network layer 733A to obtain the execution parameter. The execution parameter is used to characterize the probability of the current task being executed alone. Since the efficiency of a task being executed alone is much greater than the efficiency of multiple tasks being executed simultaneously by the task executor, the present embodiment can refer to the execution parameter of the current task to obtain the task processing time, thereby improving the prediction accuracy of the task processing time. Further, since the impact on the execution time is greater only when the current task is executed alone, the present embodiment allows the execution parameter to participate in the calculation of the task processing time when it is determined that the current task is likely to be executed alone. Specifically, when the execution parameter is greater than the predetermined parameter threshold, the first output layer 723B performs a fusion decoding process on the processing result (that is, the above-mentioned duration prediction feature) output by the first tower network layer 723A and the execution parameter to obtain the task processing time T. When the execution parameter is not greater than the predetermined parameter threshold, the first output layer 723B performs a decoding process on the processing result (that is, the above-mentioned duration prediction feature) output by the first tower network layer 723A to obtain the task processing time T. In this way, the influence of low-probability events on the output results can be avoided, further improving the prediction accuracy of the task processing time.
[0097] The duration determination model of this embodiment introduces the prediction of the service executor who will undertake the task in the future. Optionally, the duration determination model can be expressed by the following formula:
[0098]
[0099]
[0100]
[0101] Among them, h d Characterizing the parameters determining the last hidden layer of the network (e.g., the second output layer 733B described above), h TThe last hidden layer representing the duration prediction network (e.g., the first output layer 723B described above), The characterization parameters determine the output value of the network (that is, the execution parameters mentioned above), Represents the output value of the duration prediction network (i.e., the task processing duration T mentioned above).
[0102] Among them, this embodiment adds stop_grad to the execution parameter determination task, that is, the execution parameter that satisfies the corresponding constraint condition, so that when the execution parameter is greater than the parameter threshold VPT, the last hidden layer h of the duration prediction network T The fusion parameter determines the last hidden layer h of the network d , ensuring the relative isolation of the two tasks. At the same time, during the model training process, the embodiment of the present invention can control the gradient flow between the duration prediction network and the parameter determination network by setting stop_grad, thereby preventing the duration prediction network from affecting the parameter update of the parameter determination network, achieving the relative isolation of the two tasks, and further ensuring the accuracy of the trained parameter determination network.
[0103] Furthermore, the loss function of the duration determination model of this embodiment may be a fusion of the duration determination task and the parameter determination task, that is:
[0104]
[0105] Among them, loss is the loss of the model determined by the duration. is the loss of the duration prediction network, which can be based on the output of the duration prediction network corresponding to the training sample and the true value y of the sample T (that is, the actual processing time of the sample) is determined, The loss of the network is determined by the parameters, which can determine the output of the network based on the parameters corresponding to the training samples and the true value y of the sample d α is a loss coefficient, which can be preset according to actual conditions, or determined according to empirical values or tests. This embodiment does not limit the value of α and its determination method.
[0106] This embodiment is mainly described in detail by taking the duration determination model as a multi-task model as an example. It should be understood that in other optional implementations, the duration determination model can be composed of a mutually independent duration prediction network and parameter determination network, that is, the duration prediction network and the parameter determination network in the duration determination model can also be used as independent network modules, which can use the same, partially the same or different training sets for independent training and parameter adjustment. For example, the parameter determination network is first trained and adjusted based on the first training set, and then the parameters of the parameter determination network are maintained, and then the duration prediction model is trained and adjusted based on the second training set, etc. This embodiment is not limited to this.
[0107] The embodiment of the present invention obtains task information including service requester information, service provider information, and pre-selected service executor information, performs feature processing on the task information, obtains duration prediction features and execution parameters corresponding to the task, and determines the task processing duration of the task according to the duration prediction features and execution parameters, wherein the execution parameters are used to characterize the probability of the task being executed alone. Therefore, this embodiment can refer to whether the task can be executed alone when predicting the task processing duration, thereby improving the accuracy of the task processing duration and further improving the user experience.
[0108] Figure 8 Schematic diagram of a task duration determination device according to an embodiment of the present invention. Figure 8 As shown, the task processing duration determination device 8 according to the embodiment of the present invention includes an information acquisition unit 81 , a processing unit 82 and a duration determination unit 83 .
[0109] The information acquisition unit 81 is configured to acquire information of the task, the information of the task including information of the service requester, information of the service provider and information of the pre-selected service executor. The processing unit 82 is configured to perform feature processing on the information of the task, and acquire duration prediction features and execution parameters, the execution parameters being used to characterize the probability of the task being executed alone. The duration determination unit 83 is configured to determine the task processing duration according to the duration prediction features and the execution parameters.
[0110] In an optional implementation, the duration determination unit 83 is further configured to execute:
[0111] In response to the execution parameter being greater than a predetermined parameter threshold, determining the task processing duration according to the duration prediction feature and the execution parameter;
[0112] In response to the execution parameter being not greater than a predetermined parameter threshold, determining the task processing duration according to the duration prediction feature.
[0113] In an optional implementation, the task processing duration is determined by a pre-trained duration determination model, and the duration determination model includes a feature processing network, a duration prediction network and a parameter determination network. The feature processing network is used to perform feature processing on the information of the task and obtain feature information of the task. The parameter determination network is used to output the execution parameters according to the feature information of the task. The duration prediction network is used to output the task processing duration according to the feature information of the task and the execution parameters.
[0114] In an optional implementation, the duration determination model further includes a shared expert network, and the processing unit 82 is further configured to execute:
[0115] The information of the task is input into the feature processing network for processing to obtain task feature information, general feature information is generated according to the feature information of the task based on the shared expert network, the duration prediction feature is obtained according to the feature information of the task and the general feature information based on the duration prediction network, and the execution parameter is obtained according to the feature information of the task and the general feature information based on the parameter determination network.
[0116] In an optional implementation, the feature processing network includes a deep feature extraction module and a sparse data processing module, and the processing unit 82 is further configured to perform: inputting the information of the task into the deep feature extraction module for processing to obtain the first feature information of the task, inputting at least part of the information of the task into the sparse data processing module for processing to obtain the second feature information of the task.
[0117] In an optional implementation, the duration prediction network includes at least one first task expert network module and a first output module, and the parameter determination network includes at least one second task expert network module and a second output module.
[0118] In an optional implementation, the first task expert network module includes a first task expert network layer and a first gating unit. The first task expert network layer is used to receive the output of the feature processing network or the output of the previous first task expert network module for processing. The first gating unit is used to receive the input and output of the first task expert network layer for processing.
[0119] In an optional implementation, the shared expert network includes at least one shared network module, the last layer of shared network modules in the shared expert network includes a shared network layer, and the shared network modules other than the last layer in the shared expert network include a shared network layer and a shared gating unit. The shared network layer is used to receive the output of the feature processing network or the output of the previous shared network module for processing, and the shared gating unit is used to receive the input and output of the corresponding shared network layer, the output of the corresponding first task expert network layer, and the output of the second task expert network layer in the corresponding second task expert network module for processing.
[0120] In an optional implementation, the first gating unit is used to receive the input and output of the corresponding first task expert network layer and the output of the corresponding shared network layer for processing.
[0121] In an optional implementation, the first output module is used to receive the output of the last first task expert network module, the output of the sparse data processing module in the feature processing network, and the execution parameters output by the second output module for processing to obtain the task processing duration. The second output module is used to receive the output of the last second task expert network module, the output of the sparse data processing module in the feature processing network for processing to obtain the execution parameters.
[0122] In an optional implementation, the information acquisition unit 81 is further configured to perform: recalling multiple candidate service executors according to the location of the service provider, screening the multiple candidate service executors according to the location and binding task information of each candidate service executor, and obtaining multiple pre-selected service executors.
[0123] The embodiment of the present invention obtains task information including service requester information, service provider information, and pre-selected service executor information, performs feature processing on the task information, obtains duration prediction features and execution parameters corresponding to the task, and determines the task processing duration of the task according to the duration prediction features and execution parameters, wherein the execution parameters are used to characterize the probability of the task being executed alone. Therefore, this embodiment can refer to whether the task can be executed alone when predicting the task processing duration, thereby improving the accuracy of the task processing duration and further improving the user experience.
[0124] Fig. 9 Schematic diagram of an electronic device according to an embodiment of the present invention. In this embodiment, the electronic device 9 includes a server, a terminal, etc. Fig. 9As shown, the electronic device 9 includes: at least one processor 91; a memory 92 connected to the at least one processor 91 for communication; and a communication component 93 connected to the scanning device for communication, the communication component 93 receives and sends data under the control of the processor 91; wherein the memory 92 stores instructions executable by at least one processor 91, and the instructions are executed by at least one processor 91 to implement the above-mentioned task processing duration determination method.
[0125] Specifically, the electronic device includes: one or more processors 91 and a memory 92, Fig. 9 A processor 91 is taken as an example. The processor 91 and the memory 92 may be connected via a bus or other means. Fig. 9 In the example, the bus connection is used. The memory 92 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The processor 91 executes various functional applications and data processing of the device by running the non-volatile software programs, instructions and modules stored in the memory 92, that is, the above-mentioned task processing time determination method is realized.
[0126] The memory 92 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store a list of options, etc. In addition, the memory 92 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 92 may optionally include a memory remotely arranged relative to the processor 91, and these remote memories may be connected to an external device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0127] One or more modules are stored in the memory 92, and when executed by one or more processors 91, the task processing duration determination method in any of the above method embodiments is executed.
[0128] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.
[0129] The embodiment of the present invention obtains task information including service requester information, service provider information, and pre-selected service executor information, performs feature processing on the task information, obtains duration prediction features and execution parameters corresponding to the task, and determines the task processing duration of the task according to the duration prediction features and execution parameters, wherein the execution parameters are used to characterize the probability of the task being executed alone. Therefore, this embodiment can refer to whether the task can be executed alone when predicting the task processing duration, thereby improving the accuracy of the task processing duration and further improving the user experience.
[0130] Another embodiment of the present invention relates to a non-volatile storage medium for storing a computer-readable program, wherein the computer-readable program is used for a computer to execute part or all of the above method embodiments.
[0131] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0132] The embodiment of the present invention discloses A1, a method for determining task processing duration, the method comprising:
[0133] Acquire task information, wherein the task information includes service requester information, service provider information, and pre-selected service executor information;
[0134] Performing feature processing on the information of the task to obtain duration prediction features and execution parameters, where the execution parameters are used to characterize the probability of the task being executed alone;
[0135] The task processing duration is determined according to the duration prediction feature and the execution parameter.
[0136] A2. According to the method of A1, determining the task processing duration according to the duration prediction feature and the execution parameter comprises:
[0137] In response to the execution parameter being greater than a predetermined parameter threshold, determining the task processing duration according to the duration prediction feature and the execution parameter;
[0138] In response to the execution parameter being not greater than a predetermined parameter threshold, determining the task processing duration according to the duration prediction feature.
[0139] A3. According to the method described in A1, the task processing duration is determined by a pre-trained duration determination model, and the duration determination model includes a feature processing network, a duration prediction network and a parameter determination network. The feature processing network is used to perform feature processing on the information of the task and obtain the feature information of the task. The parameter determination network is used to output the execution parameters according to the feature information of the task. The duration prediction network is used to output the task processing duration according to the feature information of the task and the execution parameters.
[0140] A4. The method according to A3, wherein the duration determination model further includes a shared expert network;
[0141] The feature processing of the task information to obtain duration prediction features and execution parameters includes:
[0142] Inputting the task information into the feature processing network for processing to obtain task feature information;
[0143] Generate common feature information based on the feature information of the task based on the shared expert network;
[0144] Acquire the duration prediction feature based on the duration prediction network according to the feature information of the task and the general feature information;
[0145] Based on the parameter determination network obtains the execution parameter according to the feature information of the task and the general feature information.
[0146] A5. The method according to A4, wherein the feature processing network includes a deep feature extraction module and a sparse data processing module, and the feature processing of the task information to obtain the feature information of the task includes:
[0147] Inputting the task information into the deep feature extraction module for processing to obtain first feature information of the task;
[0148] At least part of the information of the task is input into the sparse data processing module for processing to obtain second characteristic information of the task.
[0149] A6. According to the method described in A4, the duration prediction network includes at least one first task expert network module and a first output module, and the parameter determination network includes at least one second task expert network module and a second output module.
[0150] A7. According to the method described in A6, the first task expert network module includes a first task expert network layer and a first gating unit;
[0151] The first task expert network layer is used to receive the output of the feature processing network or the output of the previous first task expert network module for processing;
[0152] The first gating unit is used to receive the input and output of the first task expert network layer for processing.
[0153] A8. The method according to A7, wherein the shared expert network includes at least one shared network module, the last layer of shared network modules in the shared expert network includes a shared network layer, and the shared network modules other than the last layer in the shared expert network include a shared network layer and a shared gating unit;
[0154] The shared network layer is used to receive the output of the feature processing network or the output of the previous shared network module for processing, and the shared gating unit is used to receive the corresponding input and output of the shared network layer, the corresponding output of the first task expert network layer, and the corresponding output of the second task expert network layer in the second task expert network module for processing.
[0155] A9. According to the method described in A8, the first gating unit is used to receive the input and output of the corresponding first task expert network layer and the output of the corresponding shared network layer for processing.
[0156] A10. According to the method described in A6, the first output module is used to receive the output of the last first task expert network module, the output of the sparse data processing module in the feature processing network, and the execution parameters output by the second output module for processing to obtain the task processing time;
[0157] The second output module is used to receive the output of the last second task expert network module and the output of the sparse data processing module in the feature processing network for processing to obtain the execution parameters.
[0158] A11. According to the method described in A1, the task information obtained includes:
[0159] Recalling multiple candidate service executors according to the location of the service provider;
[0160] The multiple candidate service executors are screened according to the location and bound task information of each of the candidate service executors to obtain multiple pre-selected service executors.
[0161] The embodiment of the present invention discloses B1, a device for determining task processing duration, the device comprising:
[0162] An information acquisition unit is configured to acquire information of a task, wherein the information of the task includes information of a service requester, information of a service provider, and information of a pre-selected service executor;
[0163] A processing unit is configured to perform feature processing on the information of the task to obtain a duration prediction feature and an execution parameter, wherein the execution parameter is used to characterize a probability that the task is executed alone;
[0164] The duration determination unit is configured to determine the task processing duration according to the duration prediction feature and the execution parameter.
[0165] An embodiment of the present invention discloses C1, an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement a method as described in any one of A1-A11.
[0166] The embodiment of the present invention discloses D1, a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of A1-A11 is implemented.
[0167] The embodiment of the present invention discloses E15, a computer program product, which, when executed on a computer, enables the computer to execute any one of the methods A1-A11.
[0168] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A method for determining task processing duration, characterized in that: The method comprises: Acquire task information, wherein the task information includes service requester information, service provider information, and pre-selected service executor information; Performing feature processing on the information of the task to obtain duration prediction features and execution parameters, where the execution parameters are used to characterize the probability of the task being executed alone; The task processing duration is determined according to the duration prediction feature and the execution parameter.
2. The method according to claim 1, characterized in that Determining the task processing duration according to the duration prediction feature and the execution parameter comprises: In response to the execution parameter being greater than a predetermined parameter threshold, determining the task processing duration according to the duration prediction feature and the execution parameter; In response to the execution parameter being not greater than a predetermined parameter threshold, determining the task processing duration according to the duration prediction feature.
3. The method according to claim 1, characterized in that The task processing duration is determined by a pre-trained duration determination model, which includes a feature processing network, a duration prediction network and a parameter determination network. The feature processing network is used to perform feature processing on the information of the task and obtain feature information of the task. The parameter determination network is used to output the execution parameters according to the feature information of the task. The duration prediction network is used to output the task processing duration according to the feature information of the task and the execution parameters.
4. The method according to claim 3, characterized in that The duration determination model also includes a shared expert network; The feature processing of the task information to obtain duration prediction features and execution parameters includes: Inputting the task information into the feature processing network for processing to obtain task feature information; Generate common feature information based on the feature information of the task based on the shared expert network; Acquire the duration prediction feature based on the duration prediction network according to the feature information of the task and the general feature information; Based on the parameter determination network obtains the execution parameter according to the feature information of the task and the general feature information.
5. The method according to claim 4, characterized in that The feature processing network includes a deep feature extraction module and a sparse data processing module, and the feature processing of the task information to obtain the feature information of the task includes: Inputting the task information into the deep feature extraction module for processing to obtain first feature information of the task; At least part of the information of the task is input into the sparse data processing module for processing to obtain second feature information of the task.
6. The method according to claim 4, characterized in that The duration prediction network includes at least one first task expert network module and a first output module, and the parameter determination network includes at least one second task expert network module and a second output module.
7. A device for determining task processing duration, characterized in that: The device comprises: An information acquisition unit is configured to acquire information of a task, wherein the information of the task includes information of a service requester, information of a service provider, and information of a pre-selected service executor; A processing unit is configured to perform feature processing on the information of the task to obtain a duration prediction feature and an execution parameter, wherein the execution parameter is used to characterize a probability that the task is executed alone; The duration determination unit is configured to determine the task processing duration according to the duration prediction feature and the execution parameter.
8. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
10. A computer program product, characterized in that When the computer program product is run on a computer, the computer is caused to execute the method according to any one of claims 1 to 6.
Citation Information
Cited By
Method and device for predicting order fulfillment duration
CN121119303A
Order fulfillment duration prediction method and prediction device
CN121119303B
