Method, device, equipment, medium and product for generating personalized recommendation scheme

Abstract

This application relates to the field of data compliance governance technology, providing a method, apparatus, device, medium, and product for generating personalized recommendation schemes. The scheme includes receiving user authorization change information, updating the authorization status of each node in the user status tree, and constructing a corresponding mask vector accordingly; then, combining this with the user's initial feature vector for constraint processing to obtain a target feature vector that conforms to the current authorization range; and generating a personalized recommendation scheme based on this target feature vector. This scheme can quickly transmit authorization changes to the feature usage stage, achieving refined compliance control over feature dimensions, preventing invalid authorization information from continuing to participate in recommendations, and improving recommendation consistency, real-time performance, and compliance governance effectiveness while retaining legally usable features.

Description

Technical Field

[0001] This application relates to the field of data compliance governance technology, and in particular to a method, apparatus, device, medium and product for generating personalized recommendation schemes. Background Technology

[0002] In financial apps, user profiles are typically composed of multi-dimensional features such as geolocation, transaction history, asset information, and behavioral preferences. These are then analyzed using deep learning models to achieve personalized recommendations, such as financial products, promotional activities, and nearby merchant services. However, as users increasingly demand greater autonomy in granting privacy permissions—for example, users may revoke their geolocation authorization at any time or refuse specific categories of marketing recommendations—the situation is evolving.

[0003] Traditional solutions rely on offline computing architectures, generating user profile features through batch processing of historical user data and synchronizing them to an online cache for real-time recommendation calls. Alternatively, a unified master switch for personalized recommendations or a coarse-grained category switch is set in the front-end business system. When a user's instruction to turn off the switch is received, the application logic layer directly intercepts and blocks the recommendation interface calls of the corresponding business module for that user, returning a globally default fallback content.

[0004] However, current solutions exhibit a time lag in responding to changes in user authorization status, and online caching struggles to promptly synchronize the latest authorization results, making it easy for revoked features to continue to be used for a period of time. Furthermore, existing control methods largely remain at the application-layer switch level, making precise management by feature dimension difficult. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and product for generating personalized recommendation schemes. Addressing scenarios with dynamically changing user authorization status, it links the user authorization result with the personalized recommendation scheme generation process. When user authorization changes, it can dynamically constrain and adjust the user features used for recommendations based on the latest authorization status, ensuring that the generated recommendation scheme remains consistent with the current authorization scope. This improves the real-time compliance of recommendation feature usage, the finer control, and the consistency of online and offline data.

[0006] In a first aspect, embodiments of this application provide a method for generating personalized recommendation schemes, the method comprising:

[0007] Upon receiving an information authorization change message from the target user, update the authorization status of the node in the status tree corresponding to the target user based on the information authorization change message;

[0008] Generate a target mask vector based on the authorization status of each node in the target user's current state tree;

[0009] Obtain the initial feature vector of the target user;

[0010] A target feature vector is generated based on the initial feature vector and the target mask vector, and then a personalized recommendation scheme for the target user is generated using the target feature vector.

[0011] In one possible embodiment, the state tree includes nodes corresponding to various feature information of the target user, and each node includes parameters identifying the authorization status; updating the authorization status of the nodes in the state tree corresponding to the target user based on the information authorization change message includes:

[0012] When the information authorization change message includes a message indicating that the target user changes the authorization status of a certain feature information from allowed to prohibited, the node in the status tree corresponding to the feature information of the target user is updated from the first parameter to the second parameter.

[0013] When the information authorization change message includes a message indicating that the target user has changed the authorization status of a certain feature information from prohibited to permitted, the node in the status tree corresponding to the feature information for the target user is updated from the second parameter to the first parameter.

[0014] In one possible embodiment, the target mask vector is of the same length as the initial feature vector. The target mask vector is generated based on the authorization status of each node in the target user's current state tree, including:

[0015] Traverse the nodes of the target user's state tree in a specified order, which is the same as the order of the features in the initial feature vector.

[0016] Obtain the authorization status parameters corresponding to each node;

[0017] Arrange the parameters corresponding to each node in a specified order to generate the target mask vector.

[0018] In one possible embodiment, generating a target feature vector based on an initial feature vector and a target mask vector includes:

[0019] Perform a bitwise multiplication operation on the initial feature vector and the target mask vector to invalidate the feature values ​​corresponding to unauthorized user information in the target feature vector.

[0020] In one possible embodiment, the initial feature vector is stored in a preset cache, and before obtaining the initial feature vector of the target user, the method further includes:

[0021] Periodically update the initial feature vectors corresponding to each user in the preset cache;

[0022] Periodically update the initial feature vectors for each user in the preset cache, including:

[0023] Periodically acquire each user's historical user behavior and current target mask vector;

[0024] Generate user profile feature vectors for each user based on historical user behavior;

[0025] Perform bitwise multiplication operations between the user profile feature vector corresponding to each user and the current target mask vector to generate the initial feature vector corresponding to each user, and store it in the preset cache area.

[0026] In one possible embodiment, it also includes:

[0027] Construct a state tree for each user. The state tree is a multi-branch structure with the user identifier as the root node. The multi-branch structure includes at least one first-level feature domain branch, and each first-level feature domain branch includes multiple corresponding nodes.

[0028] In one possible embodiment, receiving an information authorization change message from the target user includes:

[0029] The system receives information authorization change messages from target users through a pre-defined distributed message queue. These messages are generated based on information authorization change requests triggered by the target users.

[0030] In one possible embodiment, a personalized recommendation scheme for the target user is generated using the target feature vector, including:

[0031] Input the target feature vector into the preset depth recommendation model;

[0032] A pre-defined deep recommendation model is used to perform flexible degradation inference based on the target feature vector, and a personalized recommendation scheme for the target user is output. Flexible degradation inference means that the recommendation calculation is completed based on the effective features when some features are invalid.

[0033] Secondly, embodiments of this application provide an apparatus for generating personalized recommendation schemes, comprising:

[0034] The update module is used to respond to the received information authorization change message of the target user and update the authorization status of the node in the status tree corresponding to the target user based on the information authorization change message;

[0035] The generation module is used to generate a target mask vector based on the authorization status of each node in the target user's current state tree;

[0036] The acquisition module is used to obtain the initial feature vector of the target user;

[0037] The generation module is also used to generate a target feature vector based on the initial feature vector and the target mask vector, so as to generate a personalized recommendation scheme for the target user using the target feature vector.

[0038] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0039] The memory stores instructions that the computer executes;

[0040] The processor executes computer execution instructions stored in memory, causing the processor to perform the methods described above.

[0041] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided above.

[0042] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method provided above.

[0043] The personalized recommendation scheme generation method, apparatus, equipment, medium, and product provided in this application, responds to the received information authorization change message from the target user, updates the authorization status of nodes in the target user's corresponding state tree based on the information authorization change message, and achieves centralized and structured management of user authorization status, supporting fine-grained feature-level authorization changes. It generates a target mask vector based on the authorization status of each node in the target user's current state tree, transforming the tree-like authorization status into a computable mask, adapted to the vector input structure used for recommendations. The mask is dynamically generated and updated in real time, significantly improving the response rate. By combining the target user's initial feature vector to generate a target feature vector, the feature content involved in the personalized recommendation scheme generation corresponds to the target user's current authorization status in real time, achieving a second-level circuit breaker architecture and single-feature-level authorization control, avoiding the risk of illegal marketing. It improves the compliance and real-time performance of recommendation feature usage and the consistency of online and offline data in the case of dynamic changes in user authorization. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0045] Figure 1 An application scenario diagram of the personalized recommendation scheme generation method provided in this application;

[0046] Figure 2 A flowchart illustrating a method for generating a personalized recommendation scheme according to an embodiment of this application;

[0047] Figure 3 A flowchart illustrating a method for generating a personalized recommendation scheme according to another embodiment of this application;

[0048] Figure 4 A schematic diagram of the structure of a personalized recommendation scheme generation device provided in an embodiment of this application;

[0049] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0050] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0051] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0052] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0053] It should be noted that the personalized recommendation scheme generation method, apparatus, equipment, medium and product provided in this application can also be used in the field of fintech, or in any field other than fintech. The application field of the personalized recommendation scheme generation method, apparatus, equipment, storage medium and program product in this application is not limited.

[0054] To clearly understand the technical solution of this application, the solutions of the prior art will be described in detail first.

[0055] Personalized recommendations in the fintech field are widely used in financial apps such as banking, securities, and insurance companies for precise push notifications of wealth management products, payment discounts, insurance services, and nearby merchants. This typically relies on the collaborative operation of mobile devices, business gateways, user profiling services, offline computing platforms, online caching, and recommendation models. User profiles are built based on multi-source data such as geolocation, transaction records, asset levels, and behavioral preferences. Feature vectors are generated through offline batch processing and synchronized to the online cache for real-time use by the recommendation model. Under stringent regulatory requirements and high privacy sensitivity, users can change the authorization status of individual information items at any time. This necessitates the system accurately matching authorization status and feature usage scope under high concurrency and low latency conditions. Existing solutions rely on offline computing and cache loading modes, resulting in time lags in authorization changes and cache updates. Revoked features can still be used illegally, making it difficult to respond in real-time to users' right to be forgotten. Application-layer switches only support coarse-grained control at the business or interface level, failing to achieve fine-grained control at the feature dimension. This can easily lead to blanket blocking, resulting in the overuse of legitimate features, weakening recommendation effectiveness, and impacting marketing conversion, user experience, and data compliance.

[0056] Therefore, to address the technical challenges of existing technologies and resolve issues such as the inability to synchronize user authorization changes in real time and the chaotic state management, a state tree structure is used to uniformly maintain the authorization status of user information across various dimensions. Upon receiving an authorization change message, the authorization status of the corresponding node in the state tree is precisely updated, enabling real-time effectiveness of authorization changes, fine-grained control, and clear, scalable hierarchical structure. Since authorization status cannot be efficiently combined with recommendation features and it is difficult to accurately mask unauthorized features, a corresponding target mask vector is generated based on the authorization enabled / disabled status of each node in the state tree. This mask vector represents which feature dimensions are available and which must be masked, achieving standardized transmission of authorization rules to the computation layer. Then, by calculating the initial feature vector and the target mask vector, feature information from unauthorized dimensions is filtered out, resulting in a target feature vector that is only within the authorized scope. Thus, personalized recommendation schemes are generated using the target feature vector, ensuring that the entire calculation is performed within the user's authorized scope, guaranteeing recommendation effectiveness while meeting permission compliance and privacy security requirements.

[0057] Figure 1 This is an application scenario diagram illustrating the method for generating personalized recommendation schemes provided in this application, such as... Figure 1 As shown, the scenario diagram corresponding to the personalized recommendation scheme generation method provided in this application includes: a terminal device 101, a server 102, and a state tree storage repository 103. The server 102 integrates a personalized recommendation scheme generation device and a recommendation model.

[0058] Specifically, the user performs an authorization change operation on the interface of terminal device 101 to initiate an information authorization change request, which is sent to server 102. Server 102 generates an information authorization change message for the target user based on the information authorization change request, and then retrieves the state tree corresponding to the target user from the state tree repository 103 based on the information authorization change message. Server 102 updates the authorization status of nodes in the state tree corresponding to the target user based on the information authorization change message, and generates a target mask vector according to the authorization status of each node in the target user's current state tree. Then, server 102 obtains the initial feature vector of the target user and generates a target feature vector based on the initial feature vector and the target mask vector. The target feature vector is input into the recommendation model so that the recommendation model uses the target feature vector to generate a personalized recommendation scheme for the target user. Server 102 sends the personalized recommendation scheme to terminal device 101 for the user to view.

[0059] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0060] Figure 2 A flowchart illustrating a method for generating a personalized recommendation scheme according to an embodiment of this application is shown below. Figure 2 As shown, the execution entity in this embodiment is a personalized recommendation scheme generation device. This device can be implemented through a computer program, or through a medium storing the relevant computer program, such as a USB flash drive and / or optical disc, or through a physical device integrating or installing the relevant computer program, such as a chip or electronic device. The electronic device can be a server, server cluster, etc. The personalized recommendation scheme generation method provided in this embodiment includes the following steps:

[0061] S201. In response to receiving the information authorization change message of the target user, update the authorization status of the node in the state tree corresponding to the target user based on the information authorization change message.

[0062] Among them, the information authorization change message refers to the event information used to carry the change of the authorization status of the target user's information.

[0063] Optionally, the information authorization change message can be generated by the mobile device when the user enters the privacy settings page, recommended authorization settings page, or specific data category authorization control page, and when the user performs the operation of enabling, disabling, adjusting the scope, withdrawing consent, or restoring authorization. It can also be sent to the backend service through the business gateway, or asynchronously delivered to the user profile service or authorization processing service through a preset distributed message queue.

[0064] Among them, the state tree refers to a tree-like data structure used to represent the authorization status of various characteristic information of a target user. Its specific implementation can take the user identifier as the root node, map different information domains such as location information, transaction behavior, asset level, device information, page access behavior, risk preference, portfolio structure, and marketing reach preference to first-level nodes, and map the subdivided dimensions under each information domain to child nodes or leaf nodes.

[0065] Among them, the node authorization status refers to whether the feature information corresponding to a certain node in the status tree is allowed to be used. For example, a boolean value, an enumeration value, or a binary value can be used to record the allowed and prohibited status.

[0066] In practice, upon receiving an authorization change message from a target user, the system parses the user identifier, change time, change source, authorized object identifier, change action type, and latest authorization result from the message, and performs a validity check on the message. Based on the user identifier, the system locates the user's corresponding dedicated state tree, and then, according to the feature information in the message, locates the matching node in the state tree.

[0067] Optionally, the authorized object identifier in the message can directly correspond to a leaf node in the state tree, or it can be mapped to the feature field encoding first, and then located to the target node in the state tree according to the preset node mapping table.

[0068] Furthermore, after location is completed, the change action carried in the information authorization change message is written into the state tree corresponding to the target user. If the user revokes the permission to use a certain type of information, the authorization status of the node corresponding to that type of information is switched from allowed to prohibited. If the user re-grants the permission, the authorization status of the corresponding node is switched from prohibited to allowed. If the change targets an upper-level node, the status can also be propagated downwards according to the tree hierarchy, so that the child nodes under that node are updated synchronously, or only the current node is updated according to the preset inheritance rules while retaining the original status of the child nodes.

[0069] Optionally, an update method combining independent node-level updates and parent-child node consistency checks can be used. That is, the target node is updated first, and then the predefined authorization constraints between its parent and child nodes are checked. When the parent node is in a prohibited state, even if the child node retains its original permitted flag, it is considered unusable in subsequent processing, thus ensuring the semantic consistency of the tree structure.

[0070] Optionally, after updating the node authorization status, the updated state tree can be written to an in-memory database, a distributed key-value store, or a graph database, and the authorization status version number and effective time can be recorded so that the current valid status can be directly read in the subsequent mask generation process.

[0071] S202. Generate a target mask vector based on the authorization status of each node in the target user's current state tree.

[0072] The target mask vector is an ordered sequence of labels transformed from the authorization states of each feature node in the state tree. It is used to quickly identify which features can be used and which features must be prohibited during recommendation calculation.

[0073] In practice, after reading the target user's current state tree, the authorization nodes corresponding to each feature dimension are traversed according to a preset feature order. If the authorization status of a node corresponding to a certain feature dimension is allowed, the corresponding element in the target mask vector is set to an available flag. If the authorization status of the corresponding node is prohibited, the corresponding element is set to an unavailable flag.

[0074] For example, the available flag can be set to 1, and the unavailable flag can be set to 0.

[0075] Optionally, the node order in the state tree can be directly aligned with the dimensional order of the initial feature vector. In this case, when generating the target mask vector, the authorization state parameters can be written sequentially according to the node order. Alternatively, the node order in the state tree may not be completely consistent with the feature dimension order. In this case, a rearrangement mapping can be performed based on the feature mapping table. First, determine the nodes that each feature dimension depends on, and then fill the node authorization states into the corresponding positions of the target mask vector.

[0076] Optionally, after generating the target mask vector, it can be bound to the user identifier and status version number, and written to the online feature cache or temporary context storage for subsequent generation of target feature vectors.

[0077] S203. Obtain the initial feature vector of the target user.

[0078] The initial feature vector refers to a complete vector set containing all personalized feature data of the target user. For example, it can be generated by an offline computing platform based on multi-source data such as user historical transaction records, account asset changes, financial product browsing behavior, payment behavior, device attributes, geographical activity patterns, rights and benefits redemption status, insurance consultation behavior, and page dwell preferences. This data is then synchronized to an online feature cache for the recommendation model service to access in real time when a user initiates a recommendation request.

[0079] In practice, after the authorization status is updated and the target mask vector is generated, the initial feature vector can be obtained by accessing the preset cache based on the target user identifier. The initial feature vector can be read directly from the online feature cache. If the complete feature vector corresponding to the target user does not exist in the cache or the feature has expired, it can be further retrieved from the profiling service, offline feature warehouse, or real-time feature calculation service to reassemble the initial feature vector, and the retrieved result can be written back to the cache.

[0080] Optionally, the preset cache can be a distributed memory cache, a feature service database, a key-value storage system, or a real-time feature repository maintained by the feature gateway backend.

[0081] Understandably, to ensure a clear correspondence between the initial feature vector and the state tree authorization control, each feature dimension in the feature vector is pre-bound with a feature domain label and a node mapping identifier. For example, a feature dimension related to geographic location can be bound to a location node, and a feature dimension related to transaction behavior can be bound to a transaction node.

[0082] S204. Generate a target feature vector based on the initial feature vector and the target mask vector, and use the target feature vector to generate a personalized recommendation scheme for the target user.

[0083] The target feature vector refers to the set of compliant feature data obtained after filtering the initial feature vector through a permission mask, retaining only the feature information that the user has authorized to use.

[0084] Personalized recommendation schemes refer to recommendation results for target users based on target feature vectors. Specifically, they can include lists of financial products, credit card benefits and activities, payment discounts, and insurance products.

[0085] In practice, the initial feature vector and the target mask vector are subjected to element-wise operations in the same dimensional order. If the target feature vector is generated using bitwise multiplication, then for the i-th feature, if the i-th element in the target mask vector is 1, the i-th dimension of the target feature vector retains the original value from the initial feature vector. If the i-th element in the target mask vector is 0, the i-th dimension of the target feature vector is set to a predetermined invalid value.

[0086] For example, the target feature vector Zi = initial feature vector Xi × target mask vector Mi, where Zi represents the i-th dimension of the target feature vector, Xi represents the i-th dimension of the initial feature vector, and Mi represents the i-th dimension of the target mask vector. When Mi is 1, Zi = Xi; when Mi is 0, Zi = 0.

[0087] Furthermore, after generating the target feature vector, this target feature vector is passed to the recommendation model service for inference calculation. The recommendation model service can be an online inference service built on deep neural networks, gradient boosting trees, dual-tower recall models, ranking models, or multi-stage recommendation frameworks. After receiving the target feature vector, the model service calculates the matching score, click probability, conversion probability, or comprehensive ranking score of the candidate recommendation objects, and outputs the recommendation results that meet the current business objectives.

[0088] The personalized recommendation scheme generation method provided in this application, in response to receiving an information authorization change message from a target user, updates the authorization status of nodes in the target user's corresponding state tree based on the information authorization change message, achieving centralized and structured management of user authorization status and supporting fine-grained feature-level authorization changes. A target mask vector is generated based on the authorization status of each node in the target user's current state tree, transforming the tree-like authorization status into a computable mask adapted to the vector input structure used for recommendations. The mask is dynamically generated and updated in real time, significantly improving the response rate. By combining the target user's initial feature vector to generate a target feature vector, the feature content involved in generating the personalized recommendation scheme can correspond in real time with the target user's current authorization status, achieving a second-level circuit breaker architecture and single-feature-level authorization control, avoiding the risk of illegal marketing. This improves the compliance and real-time performance of recommendation feature usage and the consistency of online and offline data in the event of dynamic changes in user authorization.

[0089] As an optional implementation, based on the above embodiments, the state tree includes nodes corresponding to various feature information of the target user, and each node includes parameters identifying the authorization status; updating the authorization status of the nodes in the state tree corresponding to the target user based on the information authorization change message includes:

[0090] When the information authorization change message includes a message indicating that the target user changes the authorization status of a certain feature information from allowed to prohibited, the node in the status tree corresponding to the feature information of the target user is updated from the first parameter to the second parameter.

[0091] When the information authorization change message includes a message indicating that the target user has changed the authorization status of a certain feature information from prohibited to permitted, the node in the status tree corresponding to the feature information for the target user is updated from the second parameter to the first parameter.

[0092] Among them, the parameter storing the authorization status in the node is used to characterize whether the feature information can participate in the subsequent recommendation calculation. The first parameter represents the allowed status, and the second parameter represents the prohibited status. Both can also be implemented using enumerated values, Boolean values, or preset status codes, as long as they can correspond one-to-one with the authorization semantics.

[0093] In actual processing, upon receiving an authorization change message from the target user, the system first parses the feature information identifier and its status change content within the message. Then, it locates the node corresponding to that feature information in the state tree and writes the parameter values ​​of that node in place. When the change message indicates that the user has changed a feature from allowed to prohibited, the authorization parameters stored in that node are uniformly modified from the first parameter indicating allowed to the second parameter indicating prohibited, completing the status update of prohibited authorization. When the change message indicates that the user has changed a feature from prohibited to allowed, the authorization parameters stored in that node are uniformly modified from the second parameter indicating prohibited to the first parameter indicating allowed, completing the status update of restored authorization.

[0094] The personalized recommendation method provided in this application, by parsing change messages and locating users and nodes, can accurately identify the authorized change object, ensuring that status updates point to the correct user and correct features, avoiding erroneous or missed modifications, and improving the accuracy of authorization management. Changing parameters can promptly update the corresponding feature nodes when user authorization changes, preventing prohibited features from continuing to participate in recommendation calculations, while retaining information on other features still in a permitted state. This improves the accuracy and real-time performance of authorization control and reduces the risk of inconsistencies between recommendation results and the current authorization status.

[0095] As an optional implementation, based on the above embodiments, the target mask vector has the same length as the initial feature vector. The target mask vector is generated according to the authorization status of each node in the target user's current state tree, including:

[0096] Traverse the nodes of the target user's state tree in a specified order, which is the same as the order of the features in the initial feature vector.

[0097] Obtain the authorization status parameters corresponding to each node;

[0098] Arrange the parameters corresponding to each node in a specified order to generate the target mask vector.

[0099] The target mask vector and the initial feature vector maintain the same dimension length so that they can be correlated bitwise.

[0100] The specified order is a pre-determined node access order, which corresponds one-to-one with the arrangement order of each feature in the initial feature vector, thereby avoiding misalignment between mask elements and feature elements.

[0101] In actual implementation, upon receiving an update to the target user's authorization status, the system first reads the nodes corresponding to each feature in the state tree and traverses the nodes in the same order as the initial feature vector. During the traversal, authorization status parameters are extracted from each node. These parameters can correspond to allowed or prohibited states, or to different numerical codes, to express the meaning of reservation or masking at the vector level.

[0102] Furthermore, the extracted authorized state parameters are sequentially written into the vector storage area in a predetermined order to form a target mask vector of the same length as the initial feature vector. To ensure the stability of the vector dimension, the number of nodes remains consistent with the initial feature dimension during modeling. When adding or adjusting the feature domain, the state tree nodes and feature vector dimensions can be expanded simultaneously to maintain a one-to-one mapping relationship between them.

[0103] Optionally, the encoding method of the authorized status parameter can be set according to the input format of the recommended model, such as using 1 to represent available and 0 to represent unavailable. In practical applications, other numerical forms can also be selected for this encoding method, and this application embodiment does not limit this.

[0104] The personalized recommendation generation method provided in this application traverses nodes in the same specified order as the feature vectors, ensuring strict alignment between the authorization status and the initial features in terms of position, avoiding misalignment and mismatch, and structurally guaranteeing the accuracy of subsequent compliance processing. It generates equal-length mask vectors by arranging parameters in a specified order, converting the tree-like authorization structure into a vector form that the model can directly process, adapting to the input format of deep learning recommendation systems without modifying the model structure. This reduces mask mismatch problems caused by disordered node order or inconsistent dimensions, and improves the consistency between authorization control and feature usage, thereby enhancing the compliance, stability, and interpretability of personalized recommendation results.

[0105] As an optional implementation, based on the above embodiments, generating a target feature vector based on the initial feature vector and the target mask vector includes:

[0106] Perform a bitwise multiplication operation on the initial feature vector and the target mask vector to invalidate the feature values ​​corresponding to unauthorized user information in the target feature vector.

[0107] Among them, positional multiplication refers to the operation of multiplying the elements at the same position of two vectors of equal length one by one. The result can set the feature values ​​of unauthorized dimensions to an invalid state, while retaining the original values ​​of authorized dimensions.

[0108] Feature invalidation refers to the forced invalidation of the value corresponding to an unauthorized feature, so that it cannot have an impact in subsequent model calculations, thus achieving compliant blocking of privacy features.

[0109] Optionally, the initial feature vector read from the online feature cache can be represented as a real number vector, an integer vector, or a normalized floating-point vector.

[0110] Specifically, after receiving two vectors, a multiplication calculation is performed on each dimension. When the mask value corresponding to a certain dimension is zero, the feature value of that dimension is directly suppressed to zero or an equivalent invalid value. When the mask value is one, the feature value of that dimension remains unchanged. The resulting target feature vector can be directly input into the recommendation model service for subsequent similarity calculation, recall matching, or ranking prediction.

[0111] Optionally, if zero is used as the failure value, it is easier for the model to identify this feature as an invalid input. In practical applications, the failure value can also be other values ​​that can represent an unusable state. This application does not limit this.

[0112] The personalized recommendation method provided in this application ensures precise matching between authorization markers and original features through strict vector alignment, avoiding feature misalignment, omissions, and miscontrols, thus improving the accuracy of compliance processing. Real-time feature-level circuit breaking is implemented at the recommendation calculation entry point, immediately invalidating unauthorized features and preventing them from participating in model calculations, meeting privacy compliance requirements from the source. While restricting illegal features, the method fully retains the feature information allowed by the user, avoiding a blanket shutdown of all recommendations and ensuring recommendation effectiveness and business conversion rates. Privacy isolation is achieved in a non-intrusive and low-computational-cost manner, without modifying the cache or altering the model structure, resulting in minimal impact on the original system performance. The method provides the recommendation model with input data that fully conforms to the user's authorization intentions, ensuring that the recommended content and authorization status are consistent in real time, eliminating compliance delays and the risk of unauthorized use.

[0113] As an optional implementation, based on the above embodiments, the initial feature vector is stored in a preset cache area, and before obtaining the initial feature vector of the target user, the following steps are also included:

[0114] Periodically update the initial feature vectors corresponding to each user in the preset cache;

[0115] Periodically update the initial feature vectors for each user in the preset cache, including:

[0116] Periodically acquire each user's historical user behavior and current target mask vector;

[0117] Generate user profile feature vectors for each user based on historical user behavior;

[0118] Perform bitwise multiplication operations between the user profile feature vector corresponding to each user and the current target mask vector to generate the initial feature vector corresponding to each user, and store it in the preset cache area.

[0119] The preset cache area refers to the online cache area, which is used to store the initial feature vectors currently available to each user so that they can be quickly read when a recommendation request arrives.

[0120] Historical user behavior refers to historical data generated by users during the use of financial services, such as transaction records, browsing behavior, operating habits, device information, and geographical location, which can be used to build user profiles.

[0121] Among them, the user profile feature vector refers to the full feature data sequence that fully reflects the user's attributes and preferences, formed after offline statistics, calculation and processing based on the user's historical behavior data. Its dimensions correspond to the target mask vector.

[0122] In practical implementation, historical user behavior can be retrieved from behavior logs, transaction records, or user profile services at preset time intervals, such as updating daily at midnight. Simultaneously, the target mask vector corresponding to each user's current authorization status is read. After cleaning, normalizing, and aggregating features of historical user behavior through offline computation, user profile feature vectors corresponding one-to-one with each user dimension are generated. Before storing these vectors in the preset cache, the user profile feature vector for each user is multiplied at corresponding positions with its latest target mask vector, invalidating unauthorized features and filtering out feature data that users are not allowed to use. After processing, the compliant and filtered data is used as the initial feature vector and stored in the preset cache one by one according to the user dimension, overwriting the original old data, completing this round of periodic cache updates.

[0123] The personalized recommendation method provided in this application preprocesses data compliance by performing periodic cache updates, reducing online real-time computation pressure, ensuring low-latency response to recommendation requests, and improving the system's high-concurrency support capabilities. Before data is stored, unauthorized features are invalidated, achieving compliance isolation from the storage source and preventing non-compliant features from entering the online cache, thus reducing real-time operational risks. Compliant data is overwritten into a preset cache area, achieving periodic refresh of cached data and a compliance closed loop, preventing revoked features from repeatedly entering the online system and improving the overall architecture's compliance robustness.

[0124] As an optional implementation, based on the above embodiments, it further includes:

[0125] Construct a state tree for each user. The state tree is a multi-branch structure with the user identifier as the root node. The multi-branch structure includes at least one first-level feature domain branch, and each first-level feature domain branch includes multiple corresponding nodes.

[0126] Among them, the user identifier refers to the unique identity mark that distinguishes different users, and serves as the starting root node of the corresponding state tree so as to independently save the authorization status and feature mapping relationship of each user.

[0127] The first-level feature domain branch refers to the first-level classification branch extending directly from the root node. It is used to classify user features into broad categories and is the top-level structure for feature classification management. Each first-level feature domain branch corresponds to a type of feature information, such as geographical location, transaction behavior, asset level, device attributes, or preference tags. Multiple nodes are set under each first-level feature domain branch. Each node corresponds to a specific feature item in the feature domain and can record authorized or prohibited authorization status parameters.

[0128] In practical implementation, upon receiving user account opening information, profile initialization data, or authorization initialization results, an independent tree-like data structure can be generated for the target user, and the user identifier can be written to the root node. Based on the feature domain partitioning rules, multiple first-level feature domain branches are established under the root node, and nodes corresponding one-to-one with specific feature items are then established under each branch. Each node can store Boolean flags, state enumeration values, or encoded parameters to indicate whether the current feature is allowed to participate in the recommendation calculation.

[0129] Optionally, the state tree can be stored in the memory structure, database table structure, or distributed cache structure corresponding to the user profile service, so as to quickly locate the target node and complete the update when the user's authorization status changes; in practical applications, the state tree can also be implemented using other tree encoding methods, which are not limited in this embodiment.

[0130] The personalized recommendation method provided in this application generates a state tree by constructing a state tree with the user identifier as the root node. This binds each user to the authorization structure, ensuring clear ownership of authorization data, accurate querying, and avoiding confusion of authorization information between different users. A multi-branch tree structure is used to organize authorization information, managing features and permissions hierarchically. This structure is clear, highly scalable, and facilitates the addition, deletion, or adjustment of feature items. A first-level feature domain branch is set up to categorize features according to business dimensions, facilitating batch management, batch authorization, and batch querying, thus improving system processing and maintenance efficiency. Multiple corresponding nodes are set under the first-level feature domain, refining the granularity of authorization control to the single-feature level and supporting independent switching on / off for each feature, achieving precise compliance control.

[0131] As an optional implementation, based on the above embodiments, receiving the target user's information authorization change message includes:

[0132] The system receives information authorization change messages from target users through a pre-defined distributed message queue. These messages are generated based on information authorization change requests triggered by the target users.

[0133] Among them, the preset distributed message queue refers to the asynchronous message channel used to carry authorization change events, which can reliably transmit authorization modification information triggered by the front end to the back end processing unit, thereby reducing the coupling between authorization changes and subsequent profile updates.

[0134] Among them, the information authorization change request refers to the authorization status adjustment operation initiated by the target user on the client interface. It is usually generated by the interface control capturing the user's modification of the access permission for a certain type of information. The information authorization change message is the message body after the request is encapsulated as an event, which carries the target user identifier, the change object identifier, and the content of the authorization status change, so that the backend can accurately identify the corresponding authorization change item.

[0135] In the specific implementation, after the target user initiates an authorization adjustment via a mobile terminal, the front-end application transforms the user's operation into an information authorization change request. It then generates a message containing the user identifier, feature information identifier, pre-change state, and post-change state according to a preset message format, and sends this message to a preset distributed message queue. The application receives the corresponding message from the message queue via a subscription method and performs message routing and association matching based on the user identifier in the message to ensure a one-to-one correspondence between the authorization change message and the target user.

[0136] Optionally, the distributed message queue can be implemented using message middleware with peak-shaving and valley-filling capabilities, which facilitates maintaining the continuity of message transmission and the timeliness of reception in high-concurrency authorization change scenarios; in practical applications, other models of the message queue can also be selected, and this application embodiment does not limit this.

[0137] The personalized recommendation method provided in this application receives messages through a distributed message queue, enabling asynchronous, decoupled, and high-throughput transmission of authorization change information. This avoids the blocking and latency caused by synchronous calls and adapts to high-concurrency requests in financial scenarios. This improves the consistency between the personalized recommendation process and the user's current authorization status, reducing the risk of revoked information continuing to participate in recommendation calculations.

[0138] As an optional implementation, based on the above embodiments, a personalized recommendation scheme for target users is generated using target feature vectors, including:

[0139] Input the target feature vector into the preset depth recommendation model;

[0140] A pre-defined deep recommendation model is used to perform flexible degradation inference based on the target feature vector, and a personalized recommendation scheme for the target user is output. Flexible degradation inference means that the recommendation calculation is completed based on the effective features when some features are invalid.

[0141] Among them, the preset deep recommendation model refers to a deep learning recommendation model that receives target feature vectors and outputs personalized recommendation results. It can perform non-linear association modeling on input features and continue to use the remaining effective features to complete inference when some input features become invalid due to changes in authorization.

[0142] Among them, flexible degradation inference refers to the model still participating in scoring, ranking or recall calculations based on effective features when the feature dimensions are incomplete, so as to avoid the complete interruption of recommendations due to the failure of local features.

[0143] In practical implementation, the target feature vector is directly fed into the input layer of a pre-defined deep recommendation model after generation. The model can use embedding layers, fully connected layers, attention layers, or cross-feature layers to learn the representation of the vector and output the relevance score of the candidate recommendation object. For invalid feature dimensions, the model treats them as zero or default values ​​in the forward computation and does not participate in the aggregation of effective features after weight updates, thus maintaining the continuity of the inference process. When there are few invalid features, the model can still rely on the remaining features to complete normal recommendations. When there are many invalid features, the model can adjust the activation intensity of intermediate layers or the output ranking distribution based on the confidence of the remaining features, but still output recommendation schemes with business value. In practical applications, this deep recommendation model can also choose other network structures, which are not limited in this embodiment.

[0144] The personalized recommendation scheme generation method provided in this application embodiment ensures that the recommendation calculation remains consistent with the user's current authorization status. Expired information will not continue to participate in model inference, while the retained valid features can still support the generation of recommendation results. This improves the compliance, continuity, and business availability of recommendation services in financial scenarios and reduces the risk of recommendation interruption caused by partial feature failure.

[0145] Figure 3 This is a flowchart illustrating a method for generating a personalized recommendation scheme according to another embodiment of this application. The method in this embodiment involves a user changing the authorization status of a certain feature information from allowed to prohibited, such as... Figure 3 As shown, the method for generating a personalized recommendation scheme provided in this embodiment includes the following steps:

[0146] S301. Construct the state tree corresponding to each user. The state tree is a multi-branch structure with the user identifier as the root node. The multi-branch structure includes at least one first-level feature domain branch, and the first-level feature domain branch includes multiple corresponding nodes.

[0147] S302. Receive the information authorization change message of the target user through a preset distributed message queue. The information authorization change message is generated based on the information authorization change request triggered by the target user.

[0148] S303. Update the node in the state tree corresponding to the feature information of the target user from the first parameter to the second parameter.

[0149] S304. Traverse the nodes of the target user's state tree in the specified order, which is the same as the order of the features in the initial feature vector.

[0150] S305. Obtain the authorization status parameters corresponding to each node.

[0151] S306. Arrange the parameters corresponding to each node in a specified order to generate the target mask vector.

[0152] S307. Obtain the initial feature vector of the target user.

[0153] S308. Perform bitwise multiplication on the initial feature vector and the target mask vector to generate the target feature vector.

[0154] S309. Input the target feature vector into the preset deep recommendation model, so that the preset deep recommendation model can perform flexible degradation inference based on the target feature vector and output a personalized recommendation scheme for the target user.

[0155] In this embodiment, the implementation method and technical effect of S301-S309 are similar to the implementation method of the corresponding solution in the above embodiments, and will not be repeated here.

[0156] Figure 4 A schematic diagram of the structure of the device for generating personalized recommendation schemes provided in this application is shown below. Figure 4 As shown, the personalized recommendation scheme generation device 40 provided in this embodiment includes: an update module 41, a generation module 42, and an acquisition module 43.

[0157] The update module 41 is used to update the authorization status of the node in the state tree corresponding to the target user in response to receiving the information authorization change message of the target user; the generation module 42 is used to generate a target mask vector according to the authorization status of each node in the current state tree of the target user; the acquisition module 43 is used to acquire the initial feature vector of the target user; the generation module 42 is also used to generate a target feature vector based on the initial feature vector and the target mask vector, so as to generate a personalized recommendation scheme for the target user using the target feature vector.

[0158] The personalized recommendation generation device provided in this embodiment can execute... Figure 2 The implementation principles and technical effects of the methods shown are similar, and will not be repeated here.

[0159] Optionally, the state tree includes nodes corresponding to each feature information of the target user, and each node includes a parameter identifying the authorization status; the update module 41, when updating the authorization status of nodes in the state tree corresponding to the target user based on the information authorization change message, is specifically used to: when the information authorization change message includes the target user changing the authorization status of a certain feature information from allowed to prohibited, update the node in the state tree corresponding to the feature information in the target user's state tree from the first parameter to the second parameter; when the information authorization change message includes the target user changing the authorization status of a certain feature information from prohibited to allowed, update the node in the state tree corresponding to the feature information in the target user's state tree from the second parameter to the first parameter.

[0160] Optionally, the target mask vector is of the same length as the initial feature vector. When generating the target mask vector based on the authorization status of each node in the target user's current state tree, the generation module 42 is specifically used to: traverse each node of the target user's state tree in a specified order, the specified order being the same as the arrangement order of each feature in the initial feature vector; obtain the authorization status parameters corresponding to each node; and arrange the parameters corresponding to each node in a specified order to generate the target mask vector.

[0161] Optionally, when generating the target feature vector based on the initial feature vector and the target mask vector, the generation module 42 is specifically used to: perform a bitwise multiplication operation on the initial feature vector and the target mask vector so that the feature value corresponding to the unauthorized user information in the target feature vector becomes invalid.

[0162] Optionally, the update module 41 is also used to periodically update the initial feature vectors corresponding to each user in the preset cache, including: periodically acquiring the historical user behavior and the current target mask vector of each user; generating the user profile feature vector corresponding to each user based on the historical user behavior; performing bitwise multiplication operations on the user profile feature vector corresponding to each user with the current target mask vector to generate the initial feature vector corresponding to each user, and storing it in the preset cache.

[0163] Optionally, the personalized recommendation scheme generation device provided in this embodiment further includes a construction module.

[0164] Correspondingly, the construction module is used to construct the state tree corresponding to each user. The state tree is a multi-branch structure with the user identifier as the root node. The multi-branch structure includes at least one first-level feature domain branch, and the first-level feature domain branch includes multiple corresponding nodes.

[0165] Optionally, the personalized recommendation scheme generation device provided in this embodiment further includes a receiving module.

[0166] Correspondingly, the receiving module is used to receive the information authorization change message of the target user through a preset distributed message queue. The information authorization change message is generated based on the information authorization change request triggered by the target user.

[0167] Optionally, the personalized recommendation scheme generation device provided in this embodiment further includes an input module.

[0168] Correspondingly, the input module is used to input the target feature vector into the preset deep recommendation model; the preset deep recommendation model performs flexible degradation inference based on the target feature vector and outputs a personalized recommendation scheme for the target user. Flexible degradation inference means that the recommendation calculation is completed based on the effective features when some features fail.

[0169] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes a processor 51 and a memory 52. ​​The processor 51 and the memory 52 are connected via a bus and communicate with each other.

[0170] In the specific implementation process, the processor 51 executes the computer execution instructions stored in the memory 52, causing the processor 51 to perform the above-described method.

[0171] The specific implementation process of processor 51 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0172] In the above embodiments, it should be understood that the processor 51 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0173] The memory 52 may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0174] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0175] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0176] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0177] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0178] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0179] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0180] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0182] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0183] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0184] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for generating a personalized recommendation scheme, characterized in that, The method includes: Upon receiving an information authorization change message from the target user, update the authorization status of the node in the status tree corresponding to the target user based on the information authorization change message; Generate a target mask vector based on the authorization status of each node in the target user's current state tree; Obtain the initial feature vector of the target user; Based on the initial feature vector and the target mask vector, a target feature vector is generated, and a personalized recommendation scheme for the target user is generated using the target feature vector.

2. The method according to claim 1, characterized in that, The state tree includes nodes corresponding to various feature information of the target user, and each node includes parameters identifying the authorization status; updating the authorization status of the nodes in the state tree corresponding to the target user based on the information authorization change message includes: When the information authorization change message includes a message indicating that the target user changes the authorization status of a certain feature information from allowed to prohibited, the node in the status tree corresponding to the feature information of the target user is updated from the first parameter to the second parameter. When the information authorization change message includes a message indicating that the target user has changed the authorization status of a certain feature information from prohibited to permitted, the node in the status tree corresponding to the feature information for the target user is updated from the second parameter to the first parameter.

3. The method according to claim 2, characterized in that, The target mask vector has the same length as the initial feature vector. Generating the target mask vector based on the authorization status of each node in the target user's current state tree includes: The nodes of the target user's state tree are traversed in a specified order, the specified order being the same as the order of the features in the initial feature vector; Obtain the authorization status parameters corresponding to each node; Arrange the parameters corresponding to each node in a specified order to generate the target mask vector.

4. The method according to claim 1, characterized in that, The step of generating the target feature vector based on the initial feature vector and the target mask vector includes: Perform a bitwise multiplication operation on the initial feature vector and the target mask vector to invalidate the feature values ​​corresponding to unauthorized user information in the target feature vector.

5. The method according to claim 1, characterized in that, The initial feature vector is stored in a preset cache area. Before obtaining the initial feature vector of the target user, the process further includes: Periodically update the initial feature vectors corresponding to each user in the preset cache; The periodic updating of the initial feature vectors corresponding to each user in the preset cache includes: Periodically acquire each user's historical user behavior and current target mask vector; Based on the historical user behavior, generate user profile feature vectors for each user; Perform bitwise multiplication operations between the user profile feature vector corresponding to each user and the current target mask vector to generate the initial feature vector corresponding to each user, and store it in the preset cache area.

6. The method according to claim 1, characterized in that, Also includes: Construct a state tree corresponding to each user. The state tree is a multi-branch structure with the user identifier as the root node. The multi-branch structure includes at least one first-level feature domain branch, and the first-level feature domain branch includes multiple corresponding nodes.

7. The method according to claim 1, characterized in that, The received information authorization change message from the target user includes: The system receives information authorization change messages from target users through a pre-set distributed message queue. These information authorization change messages are generated based on information authorization change requests triggered by the target users.

8. The method according to claim 1, characterized in that, The personalized recommendation scheme for generating target users using the target feature vector includes: Input the target feature vector into the preset depth recommendation model; A pre-defined deep recommendation model is used to perform flexible degradation inference based on the target feature vector, and a personalized recommendation scheme for the target user is output. The flexible degradation inference is to complete the recommendation calculation based on the effective features when some features fail.

9. A device for generating personalized recommendation schemes, characterized in that, include: The update module is used to respond to the received information authorization change message of the target user and update the authorization status of the node in the status tree corresponding to the target user based on the information authorization change message; The generation module is used to generate a target mask vector based on the authorization status of each node in the target user's current state tree; The acquisition module is used to obtain the initial feature vector of the target user; The generation module is further configured to generate a target feature vector based on the initial feature vector and the target mask vector, so as to generate a personalized recommendation scheme for the target user using the target feature vector.

10. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-8.

Images