Method, device and equipment for predicting activation priority

Abstract

The invention provides an activation priority prediction method, device and equipment. The invention relates to the field of big data. The method for predicting the activation priority comprises the following steps: acquiring behavior data generated when a plurality of target objects perform business operation; on the basis of the attribute data and the behavior data of the multiple target objects, respective activation loss data of the multiple target objects are predicted; inputting the behavior data of the plurality of target objects into a dormancy risk prediction model to obtain respective dormancy probabilities of the plurality of target objects; and on the basis of the activation loss data and the dormancy probability, determining respective activation priority prediction results of the plurality of target objects.

Description

Technical Field

[0001] This application relates to the field of big data, and more specifically, to a method, apparatus, and device for predicting activation priorities. Background Technology

[0002] In high-frequency entry points such as retail, catering, travel, and social networking, "personal passes" are written into components such as NFC (Near Field Communication Chip) chips and eSIM (Embedded Subscriber Identity Module) in the form of digital keys or digital badges, becoming physical passes. When a user holds multiple physical passes issued by an institution, the backend needs to concurrently maintain multiple key chains, cache chains, and status bits. If some passes are not used for a long time, their key records, cache snapshots, and status logs continue to occupy non-volatile storage areas, causing the available space of the pass components to be exhausted prematurely. The writing of new passes triggers frequent garbage collection, increasing write latency.

[0003] Furthermore, under the conditions of limited computing power, storage, and bandwidth for physical passes, how to promptly identify which low-activity passes need to undergo processing operations such as compression, migration, or secure erasure of dormant data is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] In view of this, this application provides a method, apparatus and device for predicting activation priority.

[0005] One aspect of this application provides a method for predicting activation priority, comprising: acquiring behavioral data generated when multiple target objects perform business operations; predicting activation loss data for each of the multiple target objects based on attribute data and behavioral data; inputting the behavioral data of the multiple target objects into a dormancy risk prediction model to obtain the dormancy probability for each of the multiple target objects; and determining the activation priority prediction result for each of the multiple target objects based on the activation loss data and dormancy probability.

[0006] According to embodiments of this application, based on attribute data and behavior data of multiple target objects, the activation loss data of each target object is predicted, including: for each target object, extracting static features from attribute data to describe the target object; extracting dynamic features from behavior data to describe various behaviors of the target object; and based on the static and dynamic features, predicting the resources required for the target object to be activated in a dormant state to obtain activation loss data.

[0007] According to an embodiment of this application, extracting dynamic features from behavioral data to describe various behaviors of a target object includes: rearranging the behavioral data according to the time sequence information of the target object's business operations to obtain behavioral time sequence data; and extracting features from the behavioral time sequence data to obtain dynamic features describing the changes of various behaviors of the target object over time.

[0008] According to an embodiment of this application, the behavioral data of multiple target objects are input into a dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects, including: using an attention mechanism to filter out risky behavioral data from the behavioral data; inputting the risky behavioral data into the dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects.

[0009] According to an embodiment of this application, using an attention mechanism to filter out risky behavior data from behavioral data includes: classifying behavioral data based on multiple behavior types of a target object to obtain category data corresponding to each of the multiple behavior types; using an attention mechanism to assign attention weights to multiple behavior types through the category data; and determining the category data corresponding to the behavior type as risky behavior data when the attention weight is greater than the weight threshold.

[0010] According to an embodiment of this application, the activation priority prediction result of multiple target objects is determined based on activation loss data and dormancy probability, including: for each target object, fusing activation loss data and dormancy probability to obtain object activation data of the target object; and arranging multiple target objects based on the object activation data of the target objects to determine the activation priority prediction result of the target objects.

[0011] According to an embodiment of this application, for each target object, the dormancy probability and activation loss data are normalized to obtain probability data and loss data; the probability data and loss data are then fused to obtain object activation data.

[0012] According to an embodiment of this application, the method for predicting activation priority further includes: determining a target policy from a preset policy library based on the activation priority prediction result of the target object; and performing an activation operation on the target object based on the target policy.

[0013] Another aspect of this application provides an activation priority prediction device, comprising: an acquisition module for acquiring behavioral data generated when multiple target objects perform business operations; a loss prediction module for predicting activation loss data of each of the multiple target objects based on attribute data and behavioral data of the multiple target objects; a probability prediction module for inputting the behavioral data of the multiple target objects into a dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects; and a determination module for determining the activation priority prediction result of each of the multiple target objects based on the activation loss data and the dormancy probability.

[0014] Another aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method as described above.

[0015] Another aspect of this application provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the method described above.

[0016] Another aspect of this application provides a computer program product comprising computer-executable instructions which, when executed, are used to implement the method described above.

[0017] According to embodiments of this disclosure, the above steps involve parallel collection of behavioral and attribute data for each entity's pass, synchronously generating a two-dimensional vector including wake-up computational overhead and dormancy risk probability; and uniformly sorting all entity passes based on vector distance, enabling the subsequent execution of corresponding measures on the target object based on the sorting, such as hot key updates, cache compression, or data migration. Therefore, the system can reclaim storage resources before an entity pass enters deep dormancy, reducing memory usage, decreasing verification load before the entity pass completely enters dormancy, and eliminating confidentiality risks caused by long-term unupdated keys, thereby improving terminal storage efficiency, computational response speed, and data security. Attached Figure Description

[0018] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 The diagram illustrates an application scenario of the activation priority prediction method according to an embodiment of this application.

[0020] Figure 2 A flowchart illustrating a method for predicting activation priority according to an embodiment of this application is shown.

[0021] Figure 3 A data flow diagram illustrating a method for determining activation loss data according to an embodiment of this application is shown schematically.

[0022] Figure 4 The diagram illustrates a data flow diagram of a method for determining risk behavior data according to an embodiment of this application.

[0023] Figure 5 A schematic block diagram of an activation priority prediction device according to an embodiment of this application is shown.

[0024] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a prediction method for activation priority according to an embodiment of this application. Detailed Implementation

[0025] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0029] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all provide users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process enters the expert decision-making process.

[0030] Currently, with the rapid development of communication and computer technologies, physical passes have been widely penetrated into diverse scenarios such as retail, travel, social networking, and utility payments. Users frequently switch between different usage scenarios, resulting in highly volatile and short-cycle activity levels for physical passes. If the organization that creates the pass fails to detect signs of decline in time, the physical pass can easily become dormant within weeks.

[0031] Existing identification methods mostly adopt a "post-event statistics" model, which uses a fixed dormant period (e.g., no transactions for six consecutive months) as the sole criterion. This approach has a single data dimension, ignoring real-time behavioral information such as in-app clicks, location heatmaps, and external merchant interactions, which can lead to a delay in detection and miss the best intervention window.

[0032] In terms of implementing activation strategies, traditional solutions typically issue equivalent coupons or send SMS reminders to all silent physical accounts in bulk. This approach neither considers the differences in user value nor provides a quantitative assessment of activation costs, resulting in insufficient benefits for high-value customers and wasted resources for low-value customers, leading to a persistently low overall activation conversion rate.

[0033] Furthermore, existing methods make it difficult to allocate budgets and channel resources in advance, and resources can only be added after the physical pass has completely become dormant, creating a vicious cycle. If a uniform activation plan is blindly pushed out, the strategy will not match the characteristics of the customers, resulting in a waste of associated activation resources, increasing overall losses, and damaging the customer experience.

[0034] Furthermore, in related technologies, a full snapshot scan and integrity re-verification are only passively triggered after all entity passes have been inactive on the terminal side for a long time and the key records have filled the security element slots. At this time, the terminal resources are idle and the number of flash memory erases and writes increases sharply. The uniformly executed cache refresh or key rotation strategy does not take into account the difference in remaining storage blocks, nor does it take into account the device's power and computing power reserves, resulting in redundant calculations, invalid erases and writes, and additional power consumption. Instead, it accelerates the depletion of terminal resources and prolongs the response latency.

[0035] In view of this, embodiments of this application provide a method for predicting activation priority, comprising: acquiring behavioral data generated when multiple target objects perform business operations; predicting activation loss data of each of the multiple target objects based on attribute data and behavioral data of the multiple target objects; inputting the behavioral data of the multiple target objects into a dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects; and determining the activation priority prediction result of each of the multiple target objects based on the activation loss data and dormancy probability.

[0036] Figure 1 The diagram illustrates an application scenario of the activation priority prediction method according to an embodiment of this application. For example... Figure 1 As shown, application scenario 100 according to an embodiment of this application may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables. For example, a user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send information, etc.

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be electronic devices such as smartphones, wearable devices, personal computers, intelligent voice interaction devices, smart home appliances, intelligent vehicles, in-vehicle terminals, aircraft, unmanned vending terminals, and extended reality devices. Extended reality devices can include virtual reality devices, augmented reality devices, and mixed reality devices. A client application for the target application can be installed and run on the terminal devices. This target application can include, but is not limited to, financial transaction applications, payment applications, shopping applications, web browser applications, search applications, instant messaging tools, email clients, and social media platform software (these are just examples). Furthermore, this application embodiment does not limit the form of the target application, and it can include, but is not limited to, applications, mini-programs, etc., installed on the terminal devices, and can also be in the form of web pages.

[0038] Server 105 can be a server providing various services, such as a backend management server supporting websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services such as cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and basic cloud computing services such as big data. The server can be the backend server of the aforementioned target application, used to provide backend services to the clients of the target application.

[0039] It should be noted that the activation priority prediction method provided in this application embodiment can generally be executed by server 105 and / or terminal devices 101-103. Accordingly, the activation priority prediction device provided in this application embodiment can generally be set in server 105 and / or terminal devices 101-103.

[0040] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0041] Figure 2 A flowchart illustrating a method for predicting activation priority according to an embodiment of this application is shown.

[0042] like Figure 2 As shown, the method includes steps S210 to S240.

[0043] In step S210, behavioral data generated when multiple target objects perform business operations is obtained.

[0044] According to embodiments of this application, the target users can be active or inactive users who have created physical accounts. Business operations can include using physical accounts for payment, account login, password modification, balance inquiry, account transfer, credit limit adjustment, social sharing, etc. These business operations will generate corresponding behavioral data. For multiple target users, their individual behavioral data can be obtained separately.

[0045] In step S220, based on the attribute data and behavior data of multiple target objects, the activation loss data of each of the multiple target objects is predicted.

[0046] According to an embodiment of this application, the attribute data of the target object mentioned above can be profile data of the target object, such as the corresponding static information of the target object when establishing an entity pass.

[0047] According to the embodiments of this application, by inputting the above-mentioned behavioral data and attribute data into the activation loss prediction model, the quantitative resource amount required for each target object to change from a dormant state to an active state can be output. The quantitative resource amount is the activation loss data. This data is used to measure the total marketing cost, equity value and channel expenses that must be invested to re-engage the target object under the prediction confidence level, and serves as the basis for subsequent priority ranking, budget allocation or strategy matching decisions.

[0048] The activation loss prediction model can be trained using the following method: First, attribute data, behavioral data, and corresponding actual activation cost records from historical samples can be collected to construct a training set. Then, the resource input for successful activation can be used as a loss label to form a supervision signal. Next, regression training can be performed on the training set to minimize the loss between the predicted loss and the label. After the loss function converges, a shared-weight activation loss prediction model can be obtained, which can be used to output quantified resource amounts for new target objects.

[0049] In step S230, the behavioral data of multiple target objects are input into the hibernation risk prediction model to obtain the hibernation probability of each target object.

[0050] According to the embodiments of this application, the behavioral data of multiple target objects collected in the above steps within a preset time period can be input into the dormancy risk prediction model in batches, and the dormancy probability corresponding to each target object can be output. The dormancy probability can be used to quantify the possibility of the target object entering a dormant state in the future. The dormancy probability can be obtained by comprehensively extracting and mapping the temporal dependency features and global context features within the model. The numerical range of the dormancy probability can be between 0 and 1. The higher the value, the greater the risk of the target object remaining dormant without transactions or interactions in the subsequent observation period, which can be used as a ranking basis for early identification and intervention.

[0051] In step S240, based on activation loss data and dormancy probability, the activation priority prediction results for each of the multiple target objects are determined.

[0052] According to embodiments of this application, for the dormancy probability and activation loss data obtained above, for the same target object, the two types of data can be fused to form a fusion operator for the target object, or the two types of data can be concatenated to form a binary tuple for the target object. Both the fusion operator and the binary tuple can be used as the priority score for the target object. This priority score can be used to filter out high-priority objects with lower activation difficulty and fewer resources required; therefore, based on this score, all target objects can be sorted in descending order of score to obtain the activation priority prediction result.

[0053] According to embodiments of this disclosure, the above steps involve simultaneously collecting behavioral and attribute data of each entity's pass in parallel, and synchronously generating a two-dimensional vector including wake-up computational overhead and dormancy risk probability. All entity passes are then uniformly sorted based on vector distance, enabling the system to perform corresponding measures on the target objects according to the sorting, such as hot key updates, cache compression, or data migration. Therefore, the system can reclaim storage resources before an entity pass enters deep dormancy, reducing memory usage, decreasing the verification load before the entity pass completely enters dormancy, and eliminating the confidentiality risks caused by long-term unupdated keys. This improves terminal storage efficiency, computational response speed, and data security.

[0054] According to embodiments of this application, based on attribute data and behavior data of multiple target objects, the activation loss data of each target object is predicted, including: for each target object, extracting static features from attribute data to describe the target object; extracting dynamic features from behavior data to describe various behaviors of the target object; and based on the static and dynamic features, predicting the resources required for the target object to be activated in a dormant state to obtain activation loss data.

[0055] Figure 3 A data flow diagram illustrating a method for determining activation loss data according to an embodiment of this application is shown schematically.

[0056] According to the embodiments of this application, refer to Figure 3 For each of the aforementioned target objects, feature extraction can be performed based on the target object's attribute information to obtain static features. Attribute data represents data of the target object that does not change over time. Specifically, for each target object, fields that do not change over time are selected from its attribute data set and encoded to form a static feature vector representing the object's inherent attributes. The aforementioned static features may include one or more of the following: user profile, credit limit, card level, card opening duration, and risk rating. Static features can be used to complement dynamic features that change over time in subsequent models.

[0057] According to the embodiments of this application, refer to Figure 3 For each target object, data representing behaviors that change over time are selected from its behavioral data and encoded to form a dynamic feature vector characterizing the object's dynamic behavioral patterns in transactions, interactions, and scenarios. These dynamic features may include one or more of the following: transaction frequency decay rate, transaction amount standard deviation, recurring quota occupancy rate, function click entropy, message opening delay duration, page dwell time quantile, number of visits from partner merchants, geographical location heatmap blank periods, and number of third-party binding changes.

[0058] According to the embodiments of this application, refer to Figure 3After identifying the static and dynamic characteristics of the target object, predictions can be made based on these two characteristics. The prediction objective can be: if the target object is in a dormant state, the resources required to activate it, thus obtaining the activation loss data of the target object. Specifically, the static and dynamic characteristics of the same target object can be fused, then input into the activation loss prediction model. After model mapping, a quantified value is output. This value represents the amount of resources required to reactivate the target object from a dormant state, and is called the activation loss data.

[0059] According to embodiments of this application, the above steps simultaneously extract the static attributes and dynamic behaviors of an entity pass, fuse these two types of features, and feed them into a prediction model. This allows for the estimation of storage or communication resources required for future wake-up while the entity pass is still in a high-frequency calling phase. This shifts the resource assessment time point from traditional post-event backtracking to pre-event perception, avoiding unnecessary computational overhead. Static attributes preserve device differences, while dynamic behaviors reflect the key call frequency in the entity pass in real time. The complementarity of these two aspects makes the wake-up cost prediction closer to the actual system load, providing accurate input for subsequent priority ranking, cache quota allocation, and key lifecycle strategies. This reduces redundant computation and improves resource utilization and security response speed.

[0060] According to an embodiment of this application, extracting dynamic features from behavioral data to describe various behaviors of a target object includes: rearranging the behavioral data according to the time sequence information of the target object's business operations to obtain behavioral time sequence data; and extracting features from the behavioral time sequence data to obtain dynamic features describing the changes of various behaviors of the target object over time.

[0061] According to the embodiments of this application, the behavioral data generated by the target object during business operations will have a sequential order; therefore, in order to make the prediction results more accurate, multiple behavioral records generated by the target object during various business operations can be sorted in order from earliest to latest according to the timestamp of the operation to form a continuous behavioral time series, that is, behavioral time series data.

[0062] According to embodiments of this application, feature extraction is performed on the time-series behavioral data rearranged by timestamp, such as through statistics, aggregation, or model encoding; it is possible to generate quantitative indicators that can characterize the evolution trend of various behaviors such as the transaction frequency of the target object, changes in interactive objects, and changes in location over time, which are dynamic features.

[0063] According to embodiments of this application, the method of this application sorts cross-platform and cross-terminal events by using timestamps. The sorted behavioral time-series data can then be used for subsequent dynamic feature extraction and prediction of dormancy risk and activation loss. This can accurately capture the decay signals hidden in the target object's use of the physical communication certificate, making subsequent activation loss prediction based on real time-series dependencies and improving early identification accuracy. At the same time, time-series data can avoid feature noise caused by event misalignment, reduce model misjudgment, achieve early warning through activation priority prediction, and reduce the resources spent on further dormancy intervention.

[0064] According to an embodiment of this application, the behavioral data of multiple target objects are input into a dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects, including: using an attention mechanism to filter out risky behavioral data from the behavioral data; inputting the risky behavioral data into the dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects.

[0065] According to the embodiments of this application, by using an attention mechanism to weight and score different behaviors in the above-mentioned behavioral data, it is possible to highlight behavioral events that are highly related to dormancy risk, such as a decrease in transaction frequency, a decrease in login behavior, etc., and suppress irrelevant or redundant behavioral events, thereby filtering out risky behavioral data; then, only this part of high-weight risky behavioral data is fed into the dormancy risk prediction model, which enables the model to focus on key risk signals and quickly and accurately output the dormancy probability of each target object.

[0066] According to an embodiment of this application, using an attention mechanism to filter out risky behavior data from behavioral data includes: classifying behavioral data based on multiple behavior types of a target object to obtain category data corresponding to each of the multiple behavior types; using an attention mechanism to assign attention weights to multiple behavior types through the category data; and determining the category data corresponding to the behavior type as risky behavior data when the attention weight is greater than the weight threshold.

[0067] Figure 4 The diagram illustrates a data flow diagram of a method for determining risk behavior data according to an embodiment of this application.

[0068] According to the embodiments of this application, refer to Figure 4 Based on the various behaviors generated by the target object within a preset time period, the behavior can be pre-classified into multiple behavior types such as transaction, interaction, location, rights and interests, or third-party ecosystem according to business scenarios and technical attributes. Then, the behavior data can be classified according to the above-mentioned multiple behavior types to obtain the category data corresponding to each behavior type, such as the category data of transaction, the category data of interaction, etc.

[0069] According to the embodiments of this application, refer to Figure 4 The system uses an attention mechanism to assign attention weights to multiple behavior types based on categorical data. Specifically, after encoding behavior data into category vectors based on types such as transactions, interactions, locations, rights, and third-party ecosystems, the system calculates the contribution score related to dormancy risk for each category vector through self-attention or channel attention modules, and then normalizes it to obtain a weight coefficient. This weight coefficient is used to perform weighted fusion of category vectors, so that the behavior types most related to imminent dormancy receive greater model attention, while behavior types with weaker risk correlation are suppressed, thereby improving the accuracy and interpretability of subsequent dormancy probability predictions.

[0070] According to the embodiments of this application, refer to Figure 4 After the attention mechanism completes the weight allocation, a weight threshold can be preset to filter the behavior types that significantly contribute to the dormancy risk. When the attention weight of a certain behavior type is greater than the threshold, its corresponding category data can be determined as risk behavior data and included in the subsequent dormancy probability prediction, while behavior types with weights lower than the threshold are discarded or downweighted.

[0071] According to embodiments of this application, an attention mechanism is used to first weight behavior types based on the full set of behavioral data, then select category data of high-weight risk behavior types, and finally input the simplified category data into the dormancy risk prediction model. This reduces noise interference and computational load from irrelevant data, allowing the dormancy risk prediction model to focus on the most indicative key events for dormancy. On the one hand, this improves the accuracy and stability of dormancy probability estimation, and on the other hand, it reduces storage and model computation overhead, achieving fast and accurate dormancy identification. In addition, the above method uses an attention mechanism to filter risk behavior data, which can quantify the contribution of behavior categories such as transactions, interactions, and locations to dormancy risk, and complete the risk feature screening at the weight threshold. This allows high-weight categories to receive greater attention, improves the accuracy and interpretability of dormancy probability prediction, ensures the correct arrangement of activation priorities, and further enables early release of cache blocks and reduction of invalid key rotation, thereby reducing terminal resource occupation and flash memory erase / write cycles, thus simultaneously improving the resource utilization of the cloud and the terminal.

[0072] According to an embodiment of this application, the activation priority prediction result of multiple target objects is determined based on activation loss data and dormancy probability, including: for each target object, fusing activation loss data and dormancy probability to obtain object activation data of the target object; and arranging multiple target objects based on the object activation data of the target objects to determine the activation priority prediction result of the target objects.

[0073] According to embodiments of this application, for each target object, the corresponding activation loss data and dormancy probability can be fused. The activation loss data represents the quantitative resource amount required to reactivate from a dormant state, and the dormancy probability represents the likelihood of the target object entering dormancy in the future. Fusion can be performed in various ways, such as forming a binary tuple of the two types of data and concatenating them at the feature level, or using weighted summation and other methods to fuse the two one-dimensional data into two-dimensional object activation data. Object activation data can simultaneously carry dual-dimensional information of risk and cost.

[0074] According to the embodiments of this application, the object activation data of multiple target objects can be scored using the same evaluation function and sorted uniformly from high to low according to the scores, thereby determining the respective arrangement number of multiple target objects in the batch of objects, which can be used as the activation priority prediction result.

[0075] According to the embodiments of this application, by fusing the activation loss data and dormancy probability of each target object into unified object activation data, the risk level and cost can be measured simultaneously under the same dimension. This allows for the one-time sorting of multiple target objects in the batch, generating activation priority prediction results. The activation priority prediction results can then reflect which objects are easier to activate when dormant, thereby improving the activation conversion rate and reducing the resource investment for individual terminals on the cloud side of the organization.

[0076] According to an embodiment of this application, for each target object, the activation loss data and the dormancy probability are fused to obtain the target object's object activation data, including: for each target object, normalizing the dormancy probability and activation loss data respectively to obtain probability data and loss data; and fusing the probability data and loss data to obtain the object activation data.

[0077] According to an embodiment of this application, when fusing the above-mentioned dormancy probability and activation loss data, they can be normalized separately, so that the numerical values ​​and probabilities can be unified in terms of dimensions, and the object activation data that can be fused can be obtained.

[0078] According to an embodiment of this application, for example, the dormancy probability of target object A can be 0.82, and the normalized probability data can be 0.9; the original activation loss data of target object A can be 120, and the normalized loss data can be 0.3. Then, weighted linear fusion can be used to obtain the object activation data of target object A as 0.7×0.9+0.3×0.3=0.72. The weight of the dormancy probability can be set to 0.7, and the weight of the loss data can be set to 0.3. Furthermore, if the object activation data of target object B is 0.45, then the activation priority of target object A will be ranked ahead of target object B.

[0079] According to the embodiments of this application, the dormancy probability and activation loss data are first normalized to make them fall within the same numerical range, and then fused into object activation data. This fusion operation can eliminate differences in units and magnitudes, avoid large number features from masking small number features, improve the balance of fusion weights and the stability of sorting, thereby improving the accuracy and interpretability of activation priority prediction and reducing the sensitivity of subsequent models to abnormal values.

[0080] According to an embodiment of this application, the method for predicting activation priority further includes: determining a target policy from a preset policy library based on the activation priority prediction result of the target object; and performing an activation operation on the target object based on the target policy.

[0081] According to the embodiments of this application, after completing the activation priority prediction, the target strategy matching the priority level of each target object can be automatically retrieved from the preset strategy library; then, personalized activation operations are performed on the target object according to the reach channel, benefit amount, sending time and frequency limit specified by the target strategy; in addition, transaction receipts and cost data can be collected in real time.

[0082] According to an embodiment of this application, a corresponding target strategy can be assigned to multiple target objects in the order indicated in the activation priority prediction result. The target strategy can be selected from a preset strategy library, which may include providing personalized preferential strategies through communication with account managers, pushing promotional activities of partner merchants via SMS, etc.

[0083] According to embodiments of this application, the above method automatically maps terminal strategies based on an activation priority list. For example, high-priority passes immediately trigger hot key updates and cache compression, while low-priority passes are temporarily deferred. The entire process can allocate institutional resources to target objects with higher returns without manual intervention, thereby shortening the decision-making path and reducing idle computing power. Furthermore, it can collect transaction and spending data in real time and record it in a preset strategy library, updating and iterating on unreasonable strategies in the library to achieve online self-optimization of thresholds and equity amounts. This further reduces the invalid disturbance rate of objects, achieving a simultaneous improvement in computing resource utilization, customer experience, and operational efficiency.

[0084] According to embodiments of this application, furthermore, to protect the privacy of the target object, Laplace noise can be injected into the acquired behavioral data. Laplace noise injection can achieve differential privacy protection; specifically, random noise following a Laplace distribution is added to the original acquired behavioral data, so that the difference in output distribution between adjacent datasets (differences of only one record) is limited to a certain range. Within a factor of 1; thus, attackers cannot infer with high confidence whether a single record exists in the dataset, thereby protecting individual privacy while maintaining the overall availability of the data when publishing statistics or training models. The above... It can be equal to 0.5.

[0085] Based on the above-described activation priority prediction method, embodiments of this application also provide an activation priority prediction apparatus. The following will be combined with... Figure 5 The device is described in detail.

[0086] Figure 5 A schematic block diagram of a device for predicting activation priority according to an embodiment of this application is shown.

[0087] like Figure 5 As shown, the activation priority prediction device 500 of this embodiment includes an acquisition module 510, a loss prediction module 520, a probability prediction module 530, and a determination module 540.

[0088] The acquisition module 510 is used to acquire behavioral data generated when multiple target objects perform business operations. In one embodiment, the acquisition module 510 can be used to execute step S210 described above, which will not be repeated here.

[0089] The loss prediction module 520 is used to predict the activation loss data of each of the multiple target objects based on the attribute data and behavioral data of the multiple target objects. In one embodiment, the loss prediction module 520 can be used to perform the step S220 described above, which will not be repeated here.

[0090] The probability prediction module 530 is used to input the behavioral data of multiple target objects into the dormancy risk prediction model to obtain the dormancy probability of each target object. In one embodiment, the probability prediction module 530 can be used to perform step S230 described above, which will not be repeated here.

[0091] The determination module 540 is used to determine the activation priority prediction results for each of the multiple target objects based on activation loss data and dormancy probability. In one embodiment, the determination module 540 can be used to perform step S240 described above, which will not be repeated here.

[0092] According to an embodiment of this application, the loss prediction module 520 may include a first loss prediction submodule, a second loss prediction submodule, and a third loss prediction submodule.

[0093] The first loss prediction submodule is used to extract static features describing the target object from the attribute data for each target object.

[0094] The second loss prediction submodule is used to extract dynamic features from behavioral data to describe various behaviors of the target object.

[0095] The third loss prediction submodule is used to predict the resources required for a target object to be activated in a dormant state based on static and dynamic features, and to obtain activation loss data.

[0096] According to embodiments of this application, the second loss prediction submodule may include a first loss prediction unit and a second loss prediction unit.

[0097] The first loss prediction unit is used to rearrange the behavioral data according to the time-series information of the target object's business operations to obtain behavioral time-series data.

[0098] The second loss prediction unit is used to extract features from behavioral time-series data to obtain dynamic features that describe the changes in various behaviors of the target object over time.

[0099] According to an embodiment of this application, the probability prediction module 530 may include a first probability prediction submodule and a second probability prediction submodule.

[0100] The first probability prediction submodule is used to filter out risky behavior data from behavioral data using an attention mechanism.

[0101] The second probability prediction submodule is used to input risk behavior data into the dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects.

[0102] According to embodiments of this application, the first probability prediction submodule may include a first probability prediction unit, a second probability prediction unit, and a third probability prediction unit.

[0103] The first probability prediction unit is used to classify behavioral data based on multiple behavioral types of the target object, and obtain the category data corresponding to each of the multiple behavioral types.

[0104] The second probability prediction unit is used to assign attention weights to multiple behavior types using an attention mechanism based on category data.

[0105] The third probability prediction unit is used to determine the category data corresponding to the behavior type as risk behavior data when the attention weight is greater than the weight threshold.

[0106] According to an embodiment of this application, the determining module 540 may include a first determining submodule and a second determining submodule.

[0107] The first determination submodule is used to fuse activation loss data and dormancy probability for each target object to obtain the target object's object activation data.

[0108] The second determination submodule is used to arrange multiple target objects based on the object activation data of the target object and determine the activation priority prediction result of the target object.

[0109] According to embodiments of this application, the first determining submodule may include a first determining unit and a second determining unit.

[0110] The first determining unit is used to perform normalization operations on the dormancy probability and activation loss data for each target object to obtain probability data and loss data.

[0111] The second determining unit is used to fuse probability data and consumption data to obtain object activation data.

[0112] According to an embodiment of this application, the activation priority prediction device 500 may further include a target strategy determination module and an activation operation module.

[0113] The target strategy determination module is used to determine the target strategy from a preset strategy library based on the prediction results of the activation priority of the target object.

[0114] The activation operation module is used to perform activation operations on target objects based on the target strategy.

[0115] According to embodiments of this application, any multiple modules among the acquisition module 510, loss prediction module 520, probability prediction module 530, and determination module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the acquisition module 510, loss prediction module 520, probability prediction module 530, and determination module 540 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 510, loss prediction module 520, probability prediction module 530, and determination module 540 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0116] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a prediction method for activation priority according to an embodiment of this application.

[0117] like Figure 6As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory 602 or a program loaded from a storage portion 608 into a random access memory 603. The processor 601 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0118] Random access memory 603 stores various programs and data required for the operation of electronic device 600. Processor 601, read-only memory 602, and random access memory 603 are interconnected via bus 604. Processor 601 executes various steps of the method flow according to embodiments of this application by executing programs in read-only memory 602 and / or random access memory 603. It should be noted that the programs may also be stored in one or more memories other than read-only memory 602 and random access memory 603. Processor 601 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0119] According to embodiments of this application, the electronic device 600 may further include an input / output interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube, liquid crystal display, etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card, such as a local area network card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0120] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0121] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include the read-only memory 602 described above, and / or random access memory 603, and / or one or more memories other than read-only memory 602 and random access memory 603.

[0122] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0123] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0124] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by processor 601, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0125] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0126] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0127] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for predicting activation priority, characterized in that, include: Acquire behavioral data generated when multiple target objects perform business operations; Based on the attribute data and behavioral data of multiple target objects, predict the activation loss data of each of the multiple target objects; The behavioral data of multiple target objects are input into the hibernation risk prediction model to obtain the hibernation probability of each target object. Based on the activation loss data and the dormancy probability, the activation priority prediction results for each of the multiple target objects are determined.

2. The prediction method according to claim 1, characterized in that, The prediction of activation loss data for each of the multiple target objects based on attribute data and behavioral data includes: For each target object, static features describing the target object are extracted from the attribute data; Extract dynamic features from the behavioral data to describe various behaviors of the target object; Based on the static and dynamic features, the resources required to activate the target object in a dormant state are predicted, and the activation loss data is obtained.

3. The prediction method according to claim 2, characterized in that, The extraction of dynamic features from the behavioral data to describe various behaviors of the target object includes: The behavioral data is rearranged based on the timing information of the business operations performed on the target object to obtain behavioral timing data; Feature extraction is performed on the behavioral time-series data to obtain dynamic features that describe the changes of various behaviors of the target object over time.

4. The prediction method according to claim 1, characterized in that, The step of inputting the behavioral data of multiple target objects into the hibernation risk prediction model to obtain the hibernation probability of each target object includes: Use attention mechanisms to filter out risky behavior data from the behavioral data; The risk behavior data is input into the dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects.

5. The prediction method according to claim 4, characterized in that, The use of an attention mechanism to filter out risky behavior data from the behavioral data includes: Based on the multiple behavior types of the target object, the behavior data is classified to obtain category data corresponding to each behavior type. Using an attention mechanism, attention weights are assigned to multiple behavior types based on the category data; If the attention weight is greater than the weight threshold, the category data corresponding to the behavior type is determined as the risk behavior data.

6. The prediction method according to claim 1, characterized in that, The step of determining the activation priority prediction result for each of the multiple target objects based on the activation loss data and the dormancy probability includes: For each target object, the activation loss data and the dormancy probability are fused to obtain the target object's object activation data; Based on the object activation data of the target object, multiple target objects are arranged to determine the activation priority prediction result of the target object.

7. The prediction method according to claim 6, characterized in that, For each target object, the activation loss data and the dormancy probability are fused to obtain the target object's object activation data, including: For each target object, the dormancy probability and the activation loss data are normalized to obtain probability data and loss data, respectively. The probability data and the consumption data are fused together to obtain object activation data.

8. The prediction method according to claim 1, characterized in that, The method further includes: Based on the activation priority prediction results of the target object, the target strategy is determined from the preset strategy library; Based on the target strategy, an activation operation is performed on the target object.

9. An activation priority prediction device, comprising: The acquisition module is used to acquire behavioral data generated when multiple target objects perform business operations. The loss prediction module is used to predict the activation loss data of each of the multiple target objects based on the attribute data and the behavior data of the multiple target objects; The probability prediction module is used to input the behavioral data of multiple target objects into the dormancy risk prediction model to obtain the dormancy probability of each of the multiple target objects. as well as The determination module is used to determine the activation priority prediction result of each of the plurality of target objects based on the activation loss data and the dormancy probability.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 8.

Images