Performance prediction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a performance prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining the order related characteristics of a target delivery order in response to a performance evaluation request for the target delivery order, obtaining the user real-time behavior characteristics of a user corresponding to the target delivery order after the order is placed, and obtaining the performance of the target delivery order according to the user real-time behavior characteristics; and based on the order related features and the user real-time behavior features, generating the timeliness cancellation probability of the target delivery order. By adopting the mode, the real-time behavior characteristics of the user are obtained as main reference for reflecting whether the user is satisfied with the timeliness of the current order or not, and the cancellation probability of the current performance of the target delivery order user is accurately predicted in combination with the related characteristics of the order.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, electronic device, and storage medium for predicting contract performance. Background Art

[0002] Instant delivery services utilize modern technology to deliver goods from merchants to consumers as quickly as possible. Their primary goal is to improve delivery efficiency and service quality, enhance the user experience, and meet consumer demands for timeliness and convenience. However, in the actual fulfillment process, various factors can lead to problems, the most common of which is delayed delivery. This not only degrades the user experience but can also lead to user dissatisfaction and complaints. Therefore, real-time monitoring of the delivery process and proactive identification of potential timeliness issues has become a key challenge. Summary of the Invention

[0003] The main purpose of this application is to provide a fulfillment prediction method, device, electronic device, and storage medium, which aims to determine the probability of users canceling orders due to timeliness issues, thereby better providing users with personalized services and recommendations, improving user satisfaction and conversion rate. The technical solution is as follows:

[0004] In a first aspect, an embodiment of the present application provides a performance prediction method, comprising:

[0005] In response to a fulfillment evaluation request for a target delivery order, obtaining order-related features of the target delivery order;

[0006] Obtaining real-time behavior characteristics of the user corresponding to the target delivery order after placing the order;

[0007] Based on the order-related features and the user's real-time behavior features, a time-limited cancellation probability of the target delivery order is generated.

[0008] In a second aspect, an embodiment of the present application provides a performance prediction method, comprising:

[0009] Obtain training feature data and training label data for sample delivery orders; the training feature data includes real-time behavioral features of sample users after placing an order and features related to the sample order; the training label data includes whether the sample delivery order has a timeliness issue;

[0010] Inputting the sample order-related features and the sample user real-time behavior features into the feature embedding layer of the fulfillment prediction model to extract sample feature embedding data;

[0011] Inputting the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time cancellation probability of the sample delivery order;

[0012] The predicted loss value of the performance prediction model is determined based on the sample time cancellation probability and the training label data, and the performance prediction model is iteratively trained based on the predicted loss value to obtain a trained performance prediction model.

[0013] In a third aspect, an embodiment of the present application provides a contract performance prediction device, comprising:

[0014] an acquiring unit, configured to acquire order-related features of a target delivery order in response to a fulfillment evaluation request for the target delivery order;

[0015] A real-time data collection unit is used to obtain real-time behavior characteristics of the user corresponding to the target delivery order after the user places the order;

[0016] A probability prediction unit is used to generate a time-limited cancellation probability of the target delivery order based on the order-related characteristics and the user's real-time behavior characteristics.

[0017] In a fourth aspect, an embodiment of the present application provides a contract performance prediction device, comprising:

[0018] A sample acquisition unit is configured to acquire training feature data and training label data for sample delivery orders; the training feature data includes real-time behavioral features of sample users after placing an order and features related to the sample order; the training label data includes whether the sample delivery order has a timeliness issue;

[0019] An embedding unit, configured to input the sample order-related features and the sample user real-time behavior features into a feature embedding layer of a fulfillment prediction model to extract sample feature embedding data;

[0020] A sample prediction unit, configured to input the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time-limit cancellation probability for the sample delivery order;

[0021] A training unit is used to determine the predicted loss value of the performance prediction model based on the sample time cancellation probability and the training label data, and iteratively train the performance prediction model based on the predicted loss value to obtain a trained performance prediction model.

[0022] In a fifth aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the above method when executed by the processor.

[0023] In a sixth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, and the computer program implements the steps of the above method when executed by a processor.

[0024] In an embodiment of the present application, by responding to a fulfillment evaluation request for a target delivery order, obtaining the order-related features of the target delivery order, obtaining the real-time behavioral features of the user corresponding to the target delivery order after placing the order, and generating the target delivery order's cancellation probability based on the order-related features and the user's real-time behavioral features. In this way, the user's real-time behavioral features (such as the number of views on the order details page, the number of clicks on the reminder button, etc.) are obtained in real time as the main reference for reflecting whether the user is satisfied with the current order timeliness and the degree of emotional anxiety, thereby accurately predicting the user's cancellation probability for each delivery order in real time online, which can better provide users with personalized services and recommendations, and improve user satisfaction and conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 This is a scenario diagram of a performance prediction method provided by an embodiment of the present application;

[0027] Figure 2 This is a flow chart of a performance prediction method provided in an embodiment of the present application;

[0028] Figure 3 This is a flow chart of a performance prediction method provided in an embodiment of the present application;

[0029] Figure 4 1 is a flow chart of a method for training a performance prediction model provided in an embodiment of the present application;

[0030] Figure 5 This is a schematic diagram of the structure of a contract performance prediction device provided in an embodiment of the present application;

[0031] Figure 6 This is a schematic diagram of the structure of a contract performance prediction device provided in an embodiment of the present application;

[0032] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this specification.

[0034] In the description of this specification, it should be understood that the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance. In the description of this specification, it should be noted that, unless otherwise expressly specified and limited, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices. For those of ordinary skill in the art, the specific meanings of the above terms in this specification can be understood according to the specific circumstances. In addition, in the description of this specification, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0035] The fulfillment prediction device in the embodiment of the present application can be a terminal device such as a mobile phone, computer, tablet computer or vehicle-mounted device, or it can be a module in the terminal device for implementing the fulfillment prediction method. The fulfillment prediction device can obtain order-related features of the target delivery order, obtain the real-time behavior features of the user corresponding to the target delivery order after placing the order, and generate the time-limit cancellation probability of the target delivery order based on the order-related features and the real-time behavior features of the user.

[0036] See Figure 1 , a scenario diagram of a fulfillment prediction method is provided for an embodiment of the present application. The fulfillment prediction device can be a server, and the server can be an online service platform that provides delivery service applications. After obtaining the order request of the target delivery order, the server can request the user's real-time behavior characteristics after placing the order from the terminal device. The server can also obtain order-related characteristics of the target delivery order, and then generate the time-limit cancellation probability of the target delivery order based on the order-related characteristics and the user's real-time behavior characteristics.

[0037] The terminal device may be a device with data acquisition, storage and / or transmission functions, and may include a terminal used by any user, such as a terminal used by an ordering user or a terminal used by an order receiving user.

[0038] The real-time behavioral characteristics of users can be collected by the terminal device through its own sensors, or data obtained by external sensors can be collected, or data stored in its own memory can be read. In some embodiments, the sensor may include a positioning device, a sound sensor, an image sensor, a temperature and humidity sensor, a position sensor, a pressure sensor, a distance sensor, a speed sensor, an acceleration sensor, a gravity sensor, a displacement sensor, a torque sensor, a gyroscope, or any combination thereof. The various types of data collected by the terminal device can be used to determine the user's anxiety level during the service execution process, that is, to determine whether the user is dissatisfied with the timeliness of the delivery, and then determine the probability that the user may cancel the order.

[0039] It should be noted that the fulfillment prediction method of this application can be applied to instant delivery scenarios, but it should be understood that this is only an exemplary embodiment. The fulfillment prediction method may be applicable to any other on-demand service, such as taxi services, express delivery services, etc.

[0040] The performance prediction method provided in this specification is described in detail below with reference to specific embodiments.

[0041] See Figure 2 , provides a flow chart of a performance prediction method for an embodiment of the present application. Figure 2 As shown, the method of the embodiment of the present application may include the following steps S101 to S103.

[0042] S101, in response to a fulfillment evaluation request for a target delivery order, obtaining order-related features of the target delivery order;

[0043] In one embodiment, the target delivery order can be generated after the user selects the delivery address on the product settlement page of the client and places an order. A cancellation probability prediction can be automatically initiated for each target delivery order, or the delivery orders can be manually screened and a fulfillment evaluation request for a specific order (target delivery order) can be initiated. The fulfillment evaluation request is used to request a prediction of the cancellation probability in terms of user fulfillment. It can be understood that the main parties involved in whether the order is successfully fulfilled are the transaction party, the delivery party and the user. Only when the fulfillment is successful at the end user can the profit of the order be obtained. Therefore, this application focuses on the cancellation probability of the user's fulfillment this time, and responds to the fulfillment evaluation request for the target delivery order, and starts to execute the cancellation probability prediction for the target delivery order.

[0044] Order-related features are various features related to the target delivery order, such as order time, starting and ending points, product information, promotional offers, order channels, transaction party information, delivery party information, user information, etc.

[0045] S102, obtaining real-time behavior characteristics of the user corresponding to the target delivery order after placing the order;

[0046] In one embodiment, since the cancellation of the target delivery order occurs after the order is placed, real-time monitoring of user behavior after the order is placed helps to immediately identify user dissatisfaction.

[0047] Among them, the user's real-time behavior characteristics may include the user's operational behavior data in the service platform (ordering software) after placing an order (such as browsing the order details page, pressing the reminder button, contacting the delivery party / business party), facial behavior information, language behavior information, etc. Understandably, when a user opens the order details page multiple times to check the order status, it indicates that the user may be more concerned about the order status and hopes for early or quick delivery. Similarly, if a user repeatedly clicks on the reminder button, contacts the business party to urge the food to be served quickly, or contacts the delivery party for quick delivery, etc., it can directly reflect the user's emotional anxiety level.

[0048] S103: Generate a time-limited cancellation probability of the target delivery order based on the order-related features and the user's real-time behavior features.

[0049] In one embodiment, order-related features can be used to determine whether the order itself has time-limit anomalies, which may lead to user dissatisfaction. For example, during the fulfillment process, the transaction party exceeds the expected product preparation time, and the delivery party exceeds the expected pickup time. These situations may cause the order to exceed the expected delivery time, resulting in time-limit issues. User real-time behavior features can intuitively reflect whether the current user is dissatisfied from the user's perspective. Combining these two aspects can accurately predict the probability of users canceling orders due to order time-limit issues. In a feasible implementation, the probability of time-limit cancellation can be predicted based on order-related features and user real-time behavior features by training prediction models, such as Wide&Deep, DeepFM, xDeepFM, DIN, DIEN, etc.

[0050] It's understandable that the "time-limit cancellation probability" refers to the probability of an order being canceled due to time-limit issues. Alternatively, it can also refer to the probability of complaints, order returns, and other adverse after-sales issues caused by time-limit issues.

[0051] In an embodiment of the present application, by responding to a fulfillment evaluation request for a target delivery order, obtaining order-related features of the target delivery order, obtaining the real-time behavioral features of the user corresponding to the target delivery order after placing the order, and generating a cancellation probability for the target delivery order based on the order-related features and the real-time behavioral features of the user. This approach uses the real-time behavioral features of the user as the primary reference for reflecting whether the user is satisfied with the timeliness of the current order. Combined with the order-related features, the cancellation probability of the user's current fulfillment of the target delivery order can be accurately predicted. This provides the operations team with a precise basis for decision-making, allowing them to develop effective pre-emptive intervention plans, ensure the smooth performance of services, and improve user satisfaction.

[0052] See Figure 3 , is a flow chart of a fulfillment prediction method provided by an embodiment of the present application. In one embodiment, order-related features include at least environmental features, order features, transaction party features, and user basic features. Figure 3 As shown, the method of the embodiment of the present application may include the following steps S201 to S208.

[0053] S201, in response to a fulfillment evaluation request for a target delivery order, obtaining environmental features based on the order request of the target delivery order;

[0054] In one embodiment, in response to a fulfillment evaluation request for a target delivery order, environmental characteristics are obtained based on the order request for the target delivery order. The order request may include information such as the order time and order number of the target delivery order. Based on the order time, current weather information, traffic congestion information, and order time period can be determined. Based on the order number, detailed order data such as the transaction party's location and the user's delivery address can be obtained.

[0055] Environmental characteristics include geographical location characteristics, timeliness characteristics, weather characteristics, etc. Geographic location characteristics include cities, business districts, order distances, etc.; timeliness characteristics include order time periods, whether it is a holiday, etc.; weather characteristics include whether it is bad weather, etc.

[0056] S202, determining order features of the target delivery order based on the order data of the target delivery order;

[0057] In one embodiment, the order data may include order code, order details, and other content. Thus, the required order features can be determined based on the order data, such as the transaction type (catering / retail), the number of diners, the order distance, whether it is a group order, whether it is online payment, whether it is an anonymous order, whether it is an environmentally friendly order, whether it is a member order, whether it is a new customer on the platform, whether there is invoice information, whether it is a new retail order, whether it is a corporate order, whether it is a large shopping cart, price, whether there are any promotional offers, product category, order channel, etc. Furthermore, the order status can be continuously tracked based on the order number to obtain order fulfillment features, such as delivery type, fulfillment stage, order time span (payment time - estimated delivery time), whether it exceeds the meal time (estimated meal delivery time - actual meal delivery time), whether the transport capacity is slow to accept the order (transport capacity acceptance time > order creation time), whether the delivery party is slow to accept the order (delivery party acceptance time > order acceptance time), whether the meal is slow to pick up (meal pickup time > rider arrival time), whether it exceeds the predicted delivery time (actual delivery time - predicted delivery time), whether a second call is made, etc.

[0058] S203, collecting historical order data, and generating offline feature data based on the historical order data;

[0059] In one embodiment, historical order data refers to the records and related information of all delivery orders over a period of time. It includes detailed information about completed or processed orders and can be obtained by regularly collecting transaction logs, shipping order logs, and user fulfillment logs.

[0060] For example, historical order data may include order number, order time, user information, product / delivery content, delivery address, delivery status, delivery time, cost information, delivery party information, special requirements, etc.

[0061] Feature engineering is performed based on historical order data to extract offline feature data. Offline feature data can be summarized from historical order data to generate characteristics such as average order delivery time, user tags, user preferences, user personality, and user timeliness sensitivity. Offline feature data can be derived from short-term or long-term historical orders, depending on the specific requirements of feature construction. Furthermore, during feature construction, data cleaning and other operations can be performed on historical order data to improve the accuracy of feature data.

[0062] S204, acquiring transaction party characteristics of the transaction party corresponding to the target delivery order and user basic characteristics of the user corresponding to the target delivery order from the offline feature data;

[0063] In one embodiment, offline feature data can be filtered to identify the transaction parties and corresponding user feature data corresponding to the target delivery order. Transaction party features refer to characteristics of the transaction party derived from historical order summaries, including transaction party processing speed, transaction party ID, and transaction party performance quality. Basic user features include user time sensitivity, ID, overall spending power, discount sensitivity, occupation, life stage, education level, asset level, identity tags, city residency, membership tier, and address scenario. User time sensitivity refers to the user's sensitivity to the delivery time of a product, service, or delivery. In other words, it reflects the importance a user places on whether a service or product can be delivered within the specified or expected timeframe. For example, on an e-commerce platform, some users may be very sensitive to delivery times and demand a short delivery timeframe, while others may be less demanding and willing to wait longer. Therefore, varying time sensitivity can also affect the probability of time cancellation.

[0064] S205, obtaining real-time behavior characteristics of the user corresponding to the target delivery order after placing the order;

[0065] In one embodiment, the real-time behavior characteristics of users include at least one of order details page browsing data, order reminder behavior data, delivery contact behavior data, and transaction party contact behavior data. Exemplarily, a button for users to urge orders may be provided on the service platform, and the order reminder behavior data may include the number of clicks on the order reminder button. The order details page may display the status of the order, the location of the goods, etc., and the order details page browsing data may include the number of views (such as the number of times), the content of browsing (such as browsing logistics information), etc. The delivery contact behavior data may include the number of times the delivery party is contacted, the content of the contact (whether it contains keywords, such as "hurry up", "urgent", the number of times the keywords appear), etc. Similarly, the transaction party contact behavior data may include the number of times the transaction party is contacted, the content of the contact, etc.

[0066] S206, inputting the order-related features and the user's real-time behavior features into the feature embedding layer of the trained fulfillment prediction model to extract feature embedding data;

[0067] In one embodiment, a fulfillment prediction model is pre-trained. After acquiring a target delivery order, the fulfillment prediction model is used to generate a time-limited cancellation probability for the order user. The fulfillment prediction model can be constructed using deep learning, or alternatively, based on logistic regression models, decision trees, or other methods, depending on actual needs. The fulfillment prediction model includes a feature embedding layer, which extracts vector representations of order-related features and real-time user behavior features, thereby generating feature embedding data.

[0068] S207, inputting the feature embedding data into the prediction layer of the fulfillment prediction model to generate a time-limited cancellation probability of the target delivery order;

[0069] In one embodiment, the fulfillment prediction model also includes a prediction layer. These feature data are processed through each layer of the model. The model learns from historical order fulfillment patterns and patterns to predict the probability of a target delivery order being cancelled due to time constraints. At the prediction layer, the model calculates a probability value based on the feature data, indicating the likelihood that the order will not be fulfilled on time and will be cancelled under given conditions. This probability value provides a basis for delivery decisions, helping to identify high-risk orders in advance and take appropriate measures.

[0070] Optionally, the cancellation probability levels can be divided according to the time-limited cancellation probability, such as extremely unlikely to cancel, unlikely to cancel, somewhat likely to cancel, very likely to cancel and extremely likely to cancel, so as to facilitate decision makers in user classification.

[0071] It is understandable that the fulfillment prediction model can adopt an incremental training method. Based on the existing pre-trained model, a small amount of update is performed every once in a while through incremental data (newly added order data) to continuously improve the model's prediction performance and adaptability.

[0072] Optionally, in one embodiment, the contract performance prediction method provided by the embodiment of the present application further includes the following steps S2071-S2073:

[0073] S2071, inputting the feature embedding data into the factorization machine module of the performance prediction model to generate a first prediction result;

[0074] S2072, inputting the feature embedding data into the deep neural network module of the performance prediction model to generate a second prediction result;

[0075] S2073: Determine the timeliness cancellation probability of the target delivery order based on the first prediction result and the second prediction result.

[0076] It is understandable that, generally speaking, shallow models such as Logistic Regression (LR) and Factorization Machine (FM) have strong memory capabilities; while deep models such as Deep Neural Network (DNN) can discover the hidden information in the data through multiple combinations of features. This is the generalization ability that shallow models lack. Therefore, these two models can be combined to build a performance prediction model to improve the model's prediction accuracy.

[0077] In one feasible implementation, the performance prediction model is trained based on DeepFM, with a shallow linear part (factor decomposition machine module) capturing feature interactions and a deep neural network part (deep neural network module) enhancing nonlinear expression. Finally, the two are connected in parallel, wherein the factor decomposition machine module and the deep neural network module share a set of embedding vectors. The feature embedding data output by the feature embedding layer is input into the factor decomposition machine module and the deep neural network module respectively. These two parts work in parallel in the model to obtain the corresponding first prediction result and second prediction result. The outputs of the two parts are then connected through a fully connected layer to make a final prediction to obtain the probability of time cancellation.

[0078] Optionally, determining the time-limited cancellation probability of the target delivery order based on the first prediction result and the second prediction result includes: adding the first preset result and the second prediction result to obtain the sum of the prediction results; and normalizing the sum of the prediction results to obtain the time-limited cancellation probability of the target delivery order. In a feasible embodiment, the first prediction result and the second prediction result can be directly added together. The first prediction result and the second prediction result can be the scores output by the model. The sum of the prediction results is then normalized to obtain an output result between 0 and 1 as the time-limited cancellation probability. By summing up rather than other complex fusion methods, the architecture of the model will be simpler. In some tasks, a simple summation method may be easier to understand and implement than complex fusion methods (such as weighted average, weighted summation, etc.).

[0079] S208: Determine a decision solution for the timeliness issue of the user based on the timeliness cancellation probability.

[0080] In one embodiment, after determining the cancellation probability, a time-sensitive solution can be generated for the target delivery order user based on the cancellation probability. Different time-sensitive solution decisions can be implemented for users with different cancellation probabilities. For example, a solution with a higher cost can be implemented for users with a high cancellation probability, a solution with a medium cost can be implemented for users with a medium cancellation probability, and no solution can be implemented for users with a low cancellation probability.

[0081] Optionally, timeliness issues include at least one of user reminders, user returns, user cancellations, user negative reviews, and user complaints.

[0082] Optionally, the decision-making solution for timeliness issues can include customer service reminders, sending subsidies / red envelopes, providing premium services, etc. Among them, customer service reminders refer to the service platform customer service helping to supervise the service progress of the transaction party or delivery party.

[0083] It's understandable that the decision-making process for timeliness issues can also be supplemented by factors such as the user's other identity tags and the delivery scenario. For example, in a late-night snack scenario where the user's identity tag is high-value, if a certain probability of cancellation is detected for the current order, a premium service option can be displayed in advance. The user can purchase the premium service to speed up delivery, thereby improving user satisfaction and conversion rate.

[0084] In an embodiment of the present application, order-related features include at least environmental features, order features, transaction party features, and user basic features. By responding to a fulfillment evaluation request for a target delivery order, environmental features are obtained based on the order request of the target delivery order, order features of the target delivery order are determined based on the order data of the target delivery order, historical order data is collected, and offline feature data is generated based on the historical order data. The transaction party features of the transaction party corresponding to the target delivery order and the user basic features of the user corresponding to the target delivery order are obtained from the offline feature data, and the real-time behavior features of the user corresponding to the target delivery order after placing the order are obtained. The order-related features and the real-time behavior features of the user are input into the feature embedding layer of the trained fulfillment prediction model, feature embedding data is extracted, and the feature embedding data is input into the prediction layer of the fulfillment prediction model to generate the time-limited cancellation probability of the target delivery order. Using this approach, we can obtain rich real-time and offline features related to delivery order fulfillment. Real-time features can reflect the changes occurring in the delivery process, while offline features are based on historical data and provide a long-term and stable perspective, which can help the model understand past performance under similar conditions. These features are input into the trained fulfillment prediction model, feature embedding data is extracted, and the timeliness cancellation probability of the target delivery order is output through the prediction layer. The model can comprehensively consider current dynamic changes and long-term historical patterns when making predictions, improve the accuracy and robustness of the predictions, and thus improve overall operational efficiency and customer satisfaction.

[0085] See Figure 4 , is a flow chart of a training method for a performance prediction model provided in an embodiment of the present application. Figure 4 As shown, the method of the embodiment of the present application may include the following steps S301-S304.

[0086] S301, obtaining training feature data and training label data of sample delivery orders;

[0087] In one embodiment, a sample delivery order may be selected from historical delivery orders, and training feature data of the sample delivery order may be obtained and training label data of the sample delivery order may be confirmed. It is understood that the sample delivery order may be a set of sample delivery orders, that is, including multiple sample delivery orders.

[0088] The training feature data includes real-time behavioral features of sample users after they place an order and features related to sample orders. The real-time behavioral features of sample users may include at least one of order details page browsing data, order reminder behavior data, delivery contact behavior data, and transaction contact behavior data. Sample order-related features include at least environmental features, order features, transaction party features, and basic user features. The training label data includes whether a sample delivery order has a timeliness issue, and this training label serves as the ground truth for model training.

[0089] Optionally, after obtaining training feature data and training label data, you can also define initial parameters for the performance prediction model, such as the learning rate, embedding size, deep network parameters, activation function, and so on. Hyperparameter optimization can be performed using methods such as grid search and random search. Furthermore, more advanced optimization techniques such as Bayesian optimization can more efficiently identify hyperparameter combinations within the parameter space. Model design options for the performance prediction model include deep neural networks (DNNs), decision trees, random forests, or XGBoost.

[0090] It is understood that a sample dataset can be generated based on historical delivery order data. Furthermore, the sample dataset can be divided into a training set, a validation set, and a test set. Each training data point in the training set contains training feature data and training label data for a sample delivery order. The fulfillment prediction model is iteratively trained using the training data, and the validation set is used to tune hyperparameters.

[0091] S302: Input the sample order-related features and the sample user real-time behavior features into the feature embedding layer of the fulfillment prediction model to extract and obtain sample feature embedding data;

[0092] In one embodiment, the fulfillment prediction model uses an embedding layer to map sample order features and sample user real-time behavior features into a low-dimensional space. This improves the model's expressiveness, especially when processing high-dimensional, sparse data. The embedded data generated in this process serves as input to the fulfillment prediction model.

[0093] S303: Input the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time-limit cancellation probability of the sample delivery order;

[0094] In one embodiment, the prediction layer is the part that predicts the probability of time-limit cancellation based on the input features. The specific implementation of the prediction layer can be a deep neural network, decision tree, random forest, XGBoost, etc. The specific model to be selected depends on the actual needs.

[0095] It is understood that the identification or determination result of time-limit cancellation may include whether the time limit is cancelled and a quantitative representation of the probability of cancellation. As an example only, the determination result may be no cancellation. Alternatively, the determination result may be that there is a possibility of cancellation and the probability of cancellation.

[0096] Optionally, in one embodiment, DeepFM is selected as the model architecture of the fulfillment prediction model. The prediction layer may include a factorization machine (FM) module and a deep neural network (DNN) module. The sample feature embedding data is input into the prediction layer of the fulfillment prediction model. The generation of the sample time cancellation probability of the sample delivery order may include the following steps S3031-S3033:

[0097] S3031, inputting the sample feature embedding data into the factorization machine module of the performance prediction model to generate a first sample prediction result;

[0098] In one embodiment, DeepFM, selected as the performance prediction model, automatically performs feature crosstalk, eliminating the need for manual feature selection and generalization. Furthermore, DeepFM is an end-to-end model, allowing users to train directly from input data to prediction targets, simplifying the model training process. The factorization machine module in DeepFM is primarily responsible for linear feature interaction, generating the first sample prediction result.

[0099] S3032: Input the sample feature embedding data into the deep neural network module of the performance prediction model to generate a second sample prediction result;

[0100] In one embodiment, the deep neural network module is responsible for automatically learning complex feature representations from the original data and generating a second sample prediction result based on the sample feature embedding data.

[0101] S3033: Determine the sample timeliness cancellation probability of the sample delivery order based on the first sample prediction result and the second sample prediction result.

[0102] In one embodiment, the prediction results of the two modules can be combined into a comprehensive prediction value. Common methods include simple averaging and weighted averaging. The combined prediction value is the time-limited cancellation probability of the sample delivery order. The time-limited cancellation probability of the sample delivery order can be a value between 0 and 1.

[0103] In one feasible implementation, the output of the factorization machine can be summed with the output of the deep neural network and normalized. This allows for a good combination of low- and high-order features, allowing the model to account for both simple and complex feature interactions during learning and prediction. This approach can also improve the model's stability, expressiveness, and interpretability, while also accelerating convergence, avoiding overfitting, and enhancing its ability to handle sparse data.

[0104] S304: Determine a predicted loss value of the performance prediction model based on the sample time-limited cancellation probability and the training label data, and iteratively train the performance prediction model based on the predicted loss value to obtain a trained performance prediction model.

[0105] In one embodiment, the sample cancellation probability is compared with the training label to determine the difference between the sample cancellation probability and the actual result, thereby obtaining a predicted loss value. The predicted loss value is then used to adjust the parameters of the fulfillment prediction model and iteratively train it to obtain a trained fulfillment prediction model. A loss function, such as cross-entropy loss, can be selected to measure the difference between the predicted probability and the actual label. The smaller the loss value, the better the model's prediction performance. Generally, the training process includes the following steps: Backpropagation: By calculating the gradient of the loss function, the backpropagation algorithm adjusts the model parameters (such as the weights in the neural network). Gradient Descent: Using a gradient descent optimization algorithm (such as Adam or SGD) to minimize the loss function. During each training iteration, the model updates its parameters based on feedback from the loss function to reduce prediction error. Stopping criteria: Training typically stops when a maximum number of iterations is reached, the loss value no longer decreases significantly, or the performance on the validation set meets the requirements. After multiple rounds of iterative training, the model will gradually learn how to more accurately predict the cancellation probability of sample delivery orders.

[0106] In an embodiment of the present application, by obtaining training feature data and training label data of a sample delivery order, wherein the training feature data includes the real-time behavioral features of the sample user after the sample user places an order and the sample order-related features, and the training label data includes whether the sample delivery order has a time-limit problem, the sample order-related features and the sample user's real-time behavioral features are input into the feature embedding layer of the fulfillment prediction model, the sample feature embedding data is extracted, the sample feature embedding data is input into the prediction layer of the fulfillment prediction model, the sample time-limit cancellation probability of the sample delivery order is generated, the predicted loss value of the fulfillment prediction model is determined based on the sample time-limit cancellation probability and the training label data, and the fulfillment prediction model is iteratively trained based on the predicted loss value to obtain a trained fulfillment prediction model. The end-to-end training method can automatically learn the most useful features and patterns in the data and realize the prediction of the time-limit cancellation probability.

[0107] The following will be combined with the Figure 5-6 , the performance prediction device provided by the embodiment of the present application is introduced in detail. Figure 5-6 The performance prediction device in the Figure 1-Figure 4 For the convenience of explanation, only the part related to the embodiment of the present application is shown. For the specific technical details not disclosed, please refer to this specification. Figure 1-Figure 4 The embodiment shown.

[0108] See Figure 5 , which shows a schematic diagram of the structure of a performance prediction device provided by an exemplary embodiment of the present application. The performance prediction device can be implemented as all or part of a device through software, hardware, or a combination of both. The device 1 includes an acquisition unit 11, a real-time data acquisition unit 12, and a probability prediction unit 13.

[0109] An acquiring unit 11 is configured to acquire order-related features of a target delivery order in response to a fulfillment evaluation request for the target delivery order;

[0110] A real-time data collection unit 12 is used to obtain real-time behavior characteristics of the user corresponding to the target delivery order after the user places the order;

[0111] The probability prediction unit 13 is used to generate the time-limited cancellation probability of the target delivery order based on the order-related characteristics and the user's real-time behavior characteristics.

[0112] Optionally, the order-related features include at least environmental features, order features, transaction party features, and user basic features. The acquisition unit 11 is specifically configured to respond to a fulfillment evaluation request for a target delivery order and acquire the environmental features based on the order request for the target delivery order.

[0113] Determining order features of the target delivery order based on the order data of the target delivery order;

[0114] Collecting historical order data, and generating offline feature data based on the historical order data;

[0115] The transaction party characteristics of the transaction party corresponding to the target delivery order and the user basic characteristics of the user corresponding to the target delivery order are obtained from the offline feature data.

[0116] Optionally, the user's real-time behavior characteristics include at least one of order details page browsing data, order reminder behavior data, delivery party contact behavior data, and transaction party contact behavior data.

[0117] Optionally, the probability prediction unit 13 is specifically configured to input the order-related features and the user's real-time behavior features into a feature embedding layer of a trained fulfillment prediction model to extract feature embedding data;

[0118] The feature embedding data is input into the prediction layer of the fulfillment prediction model to generate the time-limited cancellation probability of the target delivery order.

[0119] Optionally, the probability prediction unit 13 is specifically configured to input the feature embedding data into a factorization machine module of the performance prediction model to generate a first prediction result;

[0120] Inputting the feature embedding data into the deep neural network module of the performance prediction model to generate a second prediction result;

[0121] The time-limit cancellation probability of the target delivery order is determined based on the first prediction result and the second prediction result.

[0122] Optionally, the probability prediction unit 13 is specifically configured to add the first preset result and the second prediction result to obtain a total prediction result;

[0123] The sum of the prediction results is normalized to obtain the time-limited cancellation probability of the target delivery order.

[0124] Optionally, the probability prediction unit 13 is further configured to determine a timeliness issue decision plan for the user based on the timeliness cancellation probability.

[0125] Optionally, the timeliness issue includes at least one of user reminders, user returns, user cancellations, user negative reviews, and user complaints.

[0126] See Figure 6, which shows a schematic diagram of the structure of a performance prediction device provided by an exemplary embodiment of the present application. The performance prediction device can be implemented as all or part of the device through software, hardware, or a combination of both. The device 2 includes a sample acquisition unit 21, an embedding unit 22, a sample prediction unit 23, and a training unit 24.

[0127] The sample acquisition unit 21 is used to acquire training feature data and training label data of sample delivery orders; the training feature data includes real-time behavioral features of sample users after placing an order and features related to the sample order; the training label data includes whether the sample delivery order has a timeliness issue;

[0128] An embedding unit 22 is configured to input the sample order-related features and the sample user real-time behavior features into a feature embedding layer of a fulfillment prediction model to extract sample feature embedding data;

[0129] A sample prediction unit 23 is configured to input the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time-limit cancellation probability of the sample delivery order;

[0130] The training unit 24 is used to determine the predicted loss value of the performance prediction model based on the sample time cancellation probability and the training label data, and iteratively train the performance prediction model based on the predicted loss value to obtain a trained performance prediction model.

[0131] Optionally, the sample prediction unit 23 is specifically configured to input the sample feature embedding data into a factorization machine module of the performance prediction model to generate a first sample prediction result;

[0132] Inputting the sample feature embedding data into the deep neural network module of the performance prediction model to generate a second sample prediction result;

[0133] The sample time-limit cancellation probability of the sample delivery order is determined based on the first sample prediction result and the second sample prediction result.

[0134] It should be noted that the performance prediction device provided in the above embodiment only uses the division of the above functional modules as an example when executing the performance prediction method or the training method of the performance prediction model. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the performance prediction device provided in the above embodiment and the performance prediction method embodiment belong to the same concept. The implementation process is detailed in the method embodiment and will not be repeated here.

[0135] The serial numbers of the embodiments of the present application are for descriptive purposes only and do not represent the merits of the embodiments. In some cases, the actions or steps described in the claims may be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0136] The embodiment of the present application further provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned Figure 1-Figure 4 The method of the embodiment shown, the specific execution process can be found in Figure 1-Figure 4 The detailed description of the illustrated embodiment will not be repeated here.

[0137] Please refer to Figure 7 , which shows a schematic diagram of the structure of an electronic device provided by an exemplary embodiment of this specification. The electronic device described in this specification may include one or more of the following components: a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, the memory 120, the input device 130, and the output device 140 may be connected via the bus 150.

[0138] The processor 110 may include one or more processing cores. The processor 110 utilizes various interfaces and circuits to connect various components within the electronic device. It executes instructions, programs, code sets, or instruction sets stored in the memory 120, as well as accesses data stored in the memory 120, to perform various functions and process data for the terminal 100. Optionally, the processor 110 may be implemented using at least one of the following hardware forms: a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 110 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interfaces, and applications; the GPU is responsible for rendering and drawing display content; and the modem handles wireless communications. It is understood that the modem may not be integrated into the processor 110 and may be implemented separately via a communication chip.

[0139] The memory 120 may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory 120 includes a non-transitory computer-readable medium (Non-Transitory Computer-Readable Storage Medium). The memory 120 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc. The operating system may be an Android system, including a system deeply developed based on the Android system, an iOS system developed by Apple, including a system deeply developed based on the iOS system or other systems.

[0140] The memory 120 can be divided into an operating system space and a user space. The operating system runs in the operating system space, and native and third-party applications run in the user space. In order to ensure that different third-party applications can achieve better operating results, the operating system allocates corresponding system resources to different third-party applications. However, the requirements for system resources in different application scenarios in the same third-party application are also different. For example, in the local resource loading scenario, the third-party application has higher requirements for disk reading speed; in the animation rendering scenario, the third-party application has higher requirements for GPU performance. The operating system and the third-party application are independent of each other, and the operating system often cannot perceive the current application scenario of the third-party application in a timely manner, resulting in the operating system being unable to perform targeted system resource adaptation according to the specific application scenario of the third-party application.

[0141] In order for the operating system to distinguish the specific application scenarios of third-party applications, it is necessary to open up data communication between third-party applications and the operating system so that the operating system can obtain the current scenario information of third-party applications at any time, and then perform targeted system resource adaptation based on the current scenario.

[0142] Among them, the input device 130 is used to receive input instructions or data, and the input device 130 includes but is not limited to a keyboard, a mouse, a camera, a microphone or a touch device. The output device 140 is used to output instructions or data, and the output device 140 includes but is not limited to a display device and a speaker. In one example, the input device 130 and the output device 140 can be combined, and the input device 130 and the output device 140 are touch screen displays. The touch screen display can be designed as a full screen, a curved screen or a special-shaped screen. The touch screen display can also be designed as a combination of a full screen and a curved screen, or a combination of a special-shaped screen and a curved screen, which is not limited in the embodiments of the present application.

[0143] In addition, those skilled in the art will understand that the structures of the electronic devices shown in the above figures do not limit the electronic devices. The electronic devices may include more or fewer components than shown, or may combine certain components or arrange the components differently. For example, the electronic devices may also include radio frequency circuits, input units, sensors, audio circuits, WiFi modules, power supplies, Bluetooth modules, and other components, which will not be described in detail here.

[0144] exist Figure 7 In the electronic device shown, the processor 110 may be configured to call a computer application stored in the memory 120 and specifically perform the following operations:

[0145] In response to a fulfillment evaluation request for a target delivery order, obtaining order-related features of the target delivery order;

[0146] Obtaining real-time behavior characteristics of the user corresponding to the target delivery order after placing the order;

[0147] Based on the order-related features and the user's real-time behavior features, a time-limited cancellation probability of the target delivery order is generated.

[0148] In one embodiment, the order-related characteristics include at least environmental characteristics, order characteristics, transaction party characteristics, and user basic characteristics. When the processor 110 obtains the order-related characteristics of the target delivery order in response to the fulfillment evaluation request for the target delivery order, the processor 110 specifically performs the following operations:

[0149] In response to a fulfillment evaluation request for a target delivery order, obtaining environmental features based on an order request for the target delivery order;

[0150] Determining order features of the target delivery order based on the order data of the target delivery order;

[0151] Collecting historical order data, and generating offline feature data based on the historical order data;

[0152] The transaction party characteristics of the transaction party corresponding to the target delivery order and the user basic characteristics of the user corresponding to the target delivery order are obtained from the offline feature data.

[0153] In one embodiment, the user real-time behavior characteristics include at least one of order detail page browsing data, order reminder behavior data, delivery party contact behavior data, and transaction party contact behavior data.

[0154] In one embodiment, when the processor 110 generates the time-limit cancellation probability of the target delivery order based on the order-related features and the user's real-time behavior features, the processor 110 specifically performs the following operations:

[0155] Inputting the order-related features and the user's real-time behavior features into the feature embedding layer of the trained fulfillment prediction model to extract feature embedding data;

[0156] The feature embedding data is input into the prediction layer of the fulfillment prediction model to generate the time-limited cancellation probability of the target delivery order.

[0157] In one embodiment, when the processor 110 inputs the feature embedding data into the prediction layer of the fulfillment prediction model to generate the time-limited cancellation probability of the target delivery order, the processor 110 specifically performs the following operations:

[0158] Inputting the feature embedding data into a factorization machine module of the performance prediction model to generate a first prediction result;

[0159] Inputting the feature embedding data into the deep neural network module of the performance prediction model to generate a second prediction result;

[0160] The time-limit cancellation probability of the target delivery order is determined based on the first prediction result and the second prediction result.

[0161] In one embodiment, when determining the time-limit cancellation probability of the target delivery order based on the first prediction result and the second prediction result, the processor 110 specifically performs the following operations:

[0162] Adding the first preset result and the second predicted result to obtain a total predicted result;

[0163] The sum of the prediction results is normalized to obtain the time-limited cancellation probability of the target delivery order.

[0164] In one embodiment, the processor 110 further performs the following operations:

[0165] A decision solution for the timeliness problem of the user is determined based on the timeliness cancellation probability.

[0166] In one embodiment, the timeliness issue includes at least one of user reminders, user returns, user cancellations, user negative reviews, and user complaints.

[0167] In one embodiment, the processor 110 further performs the following operations:

[0168] Obtain training feature data and training label data for sample delivery orders; the training feature data includes real-time behavioral features of sample users after placing an order and features related to the sample order; the training label data includes whether the sample delivery order has a timeliness issue;

[0169] Inputting the sample order-related features and the sample user real-time behavior features into the feature embedding layer of the fulfillment prediction model to extract sample feature embedding data;

[0170] Inputting the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time cancellation probability of the sample delivery order;

[0171] The predicted loss value of the performance prediction model is determined based on the sample time cancellation probability and the training label data, and the performance prediction model is iteratively trained based on the predicted loss value to obtain a trained performance prediction model.

[0172] In one embodiment, when the processor 110 inputs the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate the sample time-limit cancellation probability of the sample delivery order, the processor 110 specifically performs the following operations:

[0173] Inputting the sample feature embedding data into the factorization machine module of the performance prediction model to generate a first sample prediction result;

[0174] Inputting the sample feature embedding data into the deep neural network module of the performance prediction model to generate a second sample prediction result;

[0175] The sample time-limit cancellation probability of the sample delivery order is determined based on the first sample prediction result and the second sample prediction result.

[0176] In an embodiment of the present application, by responding to a fulfillment evaluation request for a target delivery order, obtaining order-related features of the target delivery order, obtaining the real-time behavioral features of the user corresponding to the target delivery order after placing the order, and generating a cancellation probability for the target delivery order based on the order-related features and the real-time behavioral features of the user. This approach uses the real-time behavioral features of the user as the primary reference for reflecting whether the user is satisfied with the timeliness of the current order. Combined with the order-related features, the cancellation probability of the user's current fulfillment of the target delivery order can be accurately predicted. This provides the operations team with a precise basis for decision-making, allowing them to develop effective pre-emptive intervention plans, ensure the smooth performance of services, and improve user satisfaction.

[0177] Furthermore, order-related features include at least environmental features, order features, transaction party features and user basic features. By responding to a fulfillment evaluation request for a target delivery order, environmental features are obtained based on the order request of the target delivery order, order features of the target delivery order are determined based on the order data of the target delivery order, historical order data is collected, and offline feature data is generated based on the historical order data. The transaction party features of the transaction party corresponding to the target delivery order and the user basic features of the user corresponding to the target delivery order are obtained from the offline feature data, and the real-time behavior features of the user corresponding to the target delivery order after placing the order are obtained. The order-related features and the real-time behavior features of the user are input into the feature embedding layer of the trained fulfillment prediction model, feature embedding data is extracted, and the feature embedding data is input into the prediction layer of the fulfillment prediction model to generate the time cancellation probability of the target delivery order. Using this approach, we can obtain rich real-time and offline features related to delivery order fulfillment. Real-time features can reflect the changes occurring in the delivery process, while offline features are based on historical data and provide a long-term and stable perspective, which can help the model understand past performance under similar conditions. These features are input into the trained fulfillment prediction model, feature embedding data is extracted, and the timeliness cancellation probability of the target delivery order is output through the prediction layer. The model can comprehensively consider current dynamic changes and long-term historical patterns when making predictions, improve the accuracy and robustness of the predictions, and thus improve overall operational efficiency and customer satisfaction.

[0178] Furthermore, by obtaining training feature data and training label data for sample delivery orders, wherein the training feature data includes the real-time behavioral features of sample users after placing an order and sample order-related features, and the training label data includes whether the sample delivery order has a time-limit issue, the sample order-related features and the sample user's real-time behavioral features are input into the feature embedding layer of the fulfillment prediction model, and the sample feature embedding data is extracted. The sample feature embedding data is input into the prediction layer of the fulfillment prediction model to generate the sample time-limit cancellation probability of the sample delivery order. Based on the sample time-limit cancellation probability and the training label data, the predicted loss value of the fulfillment prediction model is determined. Based on the predicted loss value, the fulfillment prediction model is iteratively trained to obtain a trained fulfillment prediction model. Using an end-to-end training method, the most useful features and patterns in the data can be automatically learned to predict the time-limit cancellation probability.

[0179] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0180] The above disclosure is only a preferred embodiment of this specification, and certainly cannot be used to limit the scope of rights of this specification. Therefore, equivalent changes made according to the claims of this specification are still within the scope covered by this specification.

Claims

1. A performance prediction method, characterized in that: include: In response to a fulfillment evaluation request for a target delivery order, obtaining order-related features of the target delivery order; Obtaining real-time behavior characteristics of the user corresponding to the target delivery order after placing the order; Based on the order-related features and the user's real-time behavior features, a time-limited cancellation probability of the target delivery order is generated.

2. The method according to claim 1, wherein The order-related characteristics include at least environmental characteristics, order characteristics, transaction party characteristics and user basic characteristics; The step of obtaining, in response to the fulfillment evaluation request for the target delivery order, order-related features of the target delivery order includes: In response to a fulfillment evaluation request for a target delivery order, obtaining environmental features based on an order request for the target delivery order; Determining order features of the target delivery order based on the order data of the target delivery order; Collecting historical order data, and generating offline feature data based on the historical order data; The transaction party characteristics of the transaction party corresponding to the target delivery order and the user basic characteristics of the user corresponding to the target delivery order are obtained from the offline feature data.

3. The method according to claim 1, wherein The user's real-time behavior characteristics include at least one of order details page browsing data, order reminder behavior data, delivery party contact behavior data, and transaction party contact behavior data.

4. The method according to claim 1, wherein Generating the time-limit cancellation probability of the target delivery order based on the order-related features and the user's real-time behavior features includes: Inputting the order-related features and the user's real-time behavior features into the feature embedding layer of the trained fulfillment prediction model to extract feature embedding data; The feature embedding data is input into the prediction layer of the fulfillment prediction model to generate the time-limited cancellation probability of the target delivery order.

5. The method according to claim 4, wherein Inputting the feature embedding data into the prediction layer of the fulfillment prediction model to generate the time-limited cancellation probability of the target delivery order includes: Inputting the feature embedding data into a factorization machine module of the performance prediction model to generate a first prediction result; Inputting the feature embedding data into the deep neural network module of the performance prediction model to generate a second prediction result; The time-limit cancellation probability of the target delivery order is determined based on the first prediction result and the second prediction result.

6. A training method for a performance prediction model, characterized in that: include: Obtaining training feature data and training label data for sample delivery orders; the training feature data includes real-time behavioral features of sample users after they place an order and features related to the sample order; The training label data includes whether the sample delivery order has a timeliness issue; Inputting the sample order-related features and the sample user real-time behavior features into the feature embedding layer of the fulfillment prediction model to extract sample feature embedding data; Inputting the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time cancellation probability of the sample delivery order; The predicted loss value of the performance prediction model is determined based on the sample time cancellation probability and the training label data, and the performance prediction model is iteratively trained based on the predicted loss value to obtain a trained performance prediction model.

7. A contract performance prediction device, characterized in that: The device comprises: an acquiring unit, configured to acquire order-related features of a target delivery order in response to a fulfillment evaluation request for the target delivery order; A real-time data collection unit is used to obtain real-time behavior characteristics of the user corresponding to the target delivery order after the user places the order; A probability prediction unit is used to generate a time-limited cancellation probability of the target delivery order based on the order-related characteristics and the user's real-time behavior characteristics.

8. A contract performance prediction device, characterized in that: The device comprises: A sample acquisition unit is configured to acquire training feature data and training label data for sample delivery orders; the training feature data includes real-time behavioral features of sample users after placing an order and features related to the sample order; the training label data includes whether the sample delivery order has a timeliness issue; An embedding unit, configured to input the sample order-related features and the sample user real-time behavior features into a feature embedding layer of a fulfillment prediction model to extract sample feature embedding data; A sample prediction unit, configured to input the sample feature embedding data into the prediction layer of the fulfillment prediction model to generate a sample time-limit cancellation probability for the sample delivery order; A training unit is used to determine the predicted loss value of the performance prediction model based on the sample time cancellation probability and the training label data, and iteratively train the performance prediction model based on the predicted loss value to obtain a trained performance prediction model.

9. An electronic device, characterized in that: include: processor and memory; The memory stores a computer program, which is suitable for being loaded by the processor and executing the steps of the method according to any one of claims 1 to 6.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images