Target service determination method and device based on heterogeneous graph neural network
By constructing a heterogeneous graph neural network and using social information to calculate the attention weight of the target to be evaluated, the problem of inaccurate trust prediction when information is insufficient is solved, and accurate service strategy formulation and data security are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI QIYUE INFORMATION TECH CO LTD
- Filing Date
- 2022-07-04
- Publication Date
- 2026-06-02
AI Technical Summary
When information is limited, existing technologies struggle to accurately predict the trust level of a target being evaluated, especially for newly registered service users, as they cannot provide accurate service strategies.
By acquiring the social information of the target to be evaluated, a heterogeneous graph neural network is constructed, attention weights are calculated, and the data are input into a security assessment model to determine the level of trust. Service strategies are then formulated based on the trust trend strategy.
With limited user data, it improved the accuracy of trust prediction, ensured the matching of service strategies, avoided losses caused by service mismatch, and protected data security.
Smart Images
Figure CN115310762B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer information processing technology, and more specifically relates to a method, apparatus, electronic device and computer-readable medium for determining target services based on heterogeneous graph neural networks. Background Technology
[0002] For service providers, a comprehensive and in-depth analysis of service users can help them offer higher-quality services. However, in many cases, the information submitted by service users is insufficient, especially for newly registered users, where service providers can only obtain basic user information. Therefore, integrating service user data scattered across different locations and institutions has become an important trend.
[0003] In the field of data services, data insecurity is a concern. Therefore, organizations often train scoring models using a large amount of historical sample data to determine the trustworthiness of the target being evaluated. However, when organizations collect limited information, it becomes very difficult to determine the services to be provided, leading to inaccurate predictions. Inaccurate predictions, in turn, make it difficult to develop a matching service strategy. Summary of the Invention
[0004] (a) Technical problems to be solved
[0005] The present invention aims to solve the technical problem of how to accurately predict the level of trust in a target to be evaluated when there is limited information.
[0006] (II) Technical Solution
[0007] To address the aforementioned technical problems, one aspect of the present invention proposes a method for determining target services based on heterogeneous graph neural networks, comprising:
[0008] Obtain social information about the target to be evaluated;
[0009] Based on the social information, construct a heterogeneous graph of social relationships with different types of social relationships, and determine the attention weights corresponding to the target to be evaluated;
[0010] The attention weights are input into a preset security assessment model to calculate the trust level of the target to be assessed;
[0011] Based on the comparison results between the trust level and the trust level trend strategy, the service strategy for the target to be evaluated is determined and pushed.
[0012] According to a preferred embodiment of the present invention, the step of constructing a heterogeneous social relationship graph with different types of social relationships based on the social information and determining the attention weight corresponding to the target to be evaluated further includes:
[0013] Targets that have social relationships with the target to be evaluated are designated as neighbor nodes, and the social information between targets is designated as edges to construct a heterogeneous graph of social relationships.
[0014] Calculate the node attention weights between different neighbor nodes and the target to be evaluated based on the heterogeneous social relationship graph;
[0015] The edge attention weights between the target to be evaluated and each of its neighboring nodes are calculated based on the node attention weights.
[0016] According to a preferred embodiment of the present invention, inputting the attention weights into a preset security assessment model to calculate the trust level of the target to be assessed further includes:
[0017] The node attention weights are weighted and summed to obtain the node attention weight vector;
[0018] The edge attention weights are weighted and summed to obtain the edge attention weight vector;
[0019] The node attention weight vector and edge attention weight vector of the target to be evaluated are input into the security assessment model to obtain the trust level of the target to be evaluated.
[0020] According to a preferred embodiment of the present invention, determining the service strategy of the target to be evaluated and pushing it based on the comparison results of the trust trend strategy further includes:
[0021] Set different trust threshold ranges;
[0022] Set corresponding service policies for different trust threshold ranges;
[0023] Provide corresponding service strategies based on the trust threshold range in which the trust level of the target to be evaluated falls.
[0024] According to a preferred embodiment of the present invention, obtaining the social information of the target to be evaluated further includes:
[0025] Obtain social relationship information of the target to be evaluated, including call records, contact information, and geographical location information.
[0026] According to a preferred embodiment of the present invention, before constructing a heterogeneous graph of social relationships with different types of social relationships based on the social information and determining the attention weight corresponding to the target to be evaluated, the method further includes:
[0027] Use historical users' social information as training samples;
[0028] A heterogeneous social relationship graph is constructed based on the training samples, and the attention weight vector of the historical user is calculated.
[0029] The attention weight vector is input into the convolutional neural network model for training to obtain the security assessment model.
[0030] According to a preferred embodiment of the present invention, the step of inputting the attention weight vector into a convolutional neural network model for training to obtain the security assessment model further includes:
[0031] The attention weight vector of historical users is input into the neural network model, and the output security score is compared with the actual security score of the historical users.
[0032] By adjusting the model parameters using a loss function, the error between the security score output by the model and the actual security score of the historical user is minimized, thus obtaining the security assessment model.
[0033] According to a preferred embodiment of the present invention, the method further includes:
[0034] The training samples are updated periodically, and the security assessment model is updated based on the updated training samples.
[0035] A second aspect of the present invention provides a target service determination device based on heterogeneous graph neural networks, comprising:
[0036] The information acquisition module is used to acquire the social information of the target to be evaluated;
[0037] The attention weight calculation module is used to construct a social relationship heterogeneous graph with different types of social relationships based on the social information, and determine the attention weight corresponding to the target to be evaluated.
[0038] The evaluation module is used to input the attention weights into a preset security evaluation model to calculate the trust level of the target to be evaluated;
[0039] The service determination module is used to determine the service strategy of the target to be evaluated and push it based on the comparison result between the trust level and the trust level trend strategy.
[0040] According to a preferred embodiment of the present invention, the attention weight calculation module further includes:
[0041] The social relationship heterogeneous graph creation unit is used to construct a social relationship heterogeneous graph by taking targets that have social relationships with the target to be evaluated as neighbor nodes and the social information between targets as edges.
[0042] The node attention weight calculation unit is used to calculate the node attention weights between different neighbor nodes and the target to be evaluated based on the heterogeneous social relationship graph.
[0043] The edge attention weight calculation unit is used to calculate the edge attention weight between the target to be evaluated and each neighboring node based on the node attention weight.
[0044] According to a preferred embodiment of the present invention, the evaluation module further includes:
[0045] The node attention weight vector calculation unit is used to calculate the node attention weights by weighted summation to obtain the node attention weight vector.
[0046] The edge attention weight vector calculation unit is used to calculate the edge attention weights by weighted summation to obtain the edge attention weight vector.
[0047] The evaluation unit is used to input the node attention weight vector and edge attention weight vector of the target to be evaluated into the security evaluation model to obtain the trust level of the target to be evaluated.
[0048] According to a preferred embodiment of the present invention, the service determination module further includes:
[0049] The trust threshold range setting unit is used to set different trust threshold ranges;
[0050] The service strategy formulation unit is used to set corresponding service strategies for different trust threshold ranges.
[0051] The service determination unit is used to provide a corresponding service strategy based on the trust threshold range in which the trust level of the target to be evaluated falls.
[0052] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory being used to store a computer-executable program, wherein when the computer program is executed by the processor, the processor performs the method described thereon.
[0053] A fourth aspect of the present invention also provides a computer-readable medium storing a computer-executable program, which, when executed, implements the method described.
[0054] (III) Beneficial Effects
[0055] This invention acquires the social information of the target to be evaluated, constructs a heterogeneous social relationship graph with different types of social relationships based on the social information, determines the attention weight corresponding to the target to be evaluated, and inputs the attention weight vector into a preset security assessment model to calculate the trust level of the target to be evaluated. In this solution, the trust level obtained from the social relationships of the target to be evaluated enables the service provider to make timely response strategies before providing services, avoids losses caused by mismatch in the services provided, and ensures the data security of the services. Attached Figure Description
[0056] Figure 1 This is a system block diagram of a target service determination method and apparatus based on a heterogeneous graph neural network according to an embodiment of the present invention;
[0057] Figure 2 This is a schematic flowchart of a target service determination method based on a heterogeneous graph neural network according to an embodiment of the present invention;
[0058] Figure 3 This is a schematic diagram of a target service determination device based on a heterogeneous graph neural network according to an embodiment of the present invention. Detailed Implementation
[0059] In the description of specific embodiments, detailed descriptions of structures, performance, effects, or other features are provided to enable those skilled in the art to fully understand the embodiments. However, this does not preclude those skilled in the art from implementing the present invention with technical solutions that do not contain the aforementioned structures, performance, effects, or other features under specific circumstances.
[0060] The same reference numerals in the accompanying drawings denote the same or similar elements, components, or parts, and therefore, repeated descriptions of the same or similar elements, components, or parts may be omitted below. It should also be understood that although terms such as first, second, third, etc., indicating numbers may be used herein to describe various devices, elements, components, or parts, these devices, elements, components, or parts should not be limited by these terms. That is, these terms are only used to distinguish one from another. For example, a first device may also be referred to as a second device, without departing from the essential technical solution of the invention. Furthermore, the terms "and / or" and "and / or" refer to all combinations including any one or more of the listed items.
[0061] This invention proposes a target service determination method based on heterogeneous graph neural networks. By constructing a heterogeneous social relationship graph from the social information of the target to be identified, the corresponding attention weights are calculated from the node dimension and the view dimension. Finally, the attention weight vector is input into a trained classification model to calculate the user's trust level.
[0062] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.
[0063] Figure 1 This is a system block diagram illustrating a target service determination method and apparatus based on a heterogeneous graph neural network according to an exemplary embodiment.
[0064] like Figure 1 As shown, system architecture 100 may include one or more of user terminals 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between user terminals 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.
[0065] It should be understood that Figure 1 The number of user terminals, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of user terminals, networks, and servers. For example, server 105 could be a server cluster composed of multiple servers.
[0066] Users can use user terminals 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. User terminals 101, 102, and 103 can be various electronic devices with displays, including but not limited to smartphones, tablets, laptops, and desktop computers, etc.
[0067] In some embodiments, the device security level identification method provided in this invention is generally executed by server 105, and correspondingly, the device security level identification device is generally disposed in server 105. In other embodiments, some terminals may have functions similar to those of the server to execute this method. Therefore, the device security level identification method provided in this invention is not limited to execution on the server side.
[0068] Figure 2 This is a schematic flowchart of a target service determination method based on a heterogeneous graph neural network, which is one embodiment of the present invention.
[0069] like Figure 2 As shown, this method includes:
[0070] S201. Obtain the social information of the target to be evaluated.
[0071] In this embodiment, the target to be evaluated can be an individual user or an enterprise user. When the target to be evaluated is an individual user, the social information may include basic information authorized by the user, such as call information, telephone address book information, social software address book, user's geographical location information, transaction information, interaction information, etc.; social information may also include behavioral information, such as transaction information between users, communication information between users, etc.; when the target to be evaluated is an enterprise user, the social information may be equity penetration information between enterprises, news information between enterprises, social information of managers within the enterprise, etc.; the specific content of the social information can be determined according to the actual application scenario and is not limited here.
[0072] In a specific application, the node to be identified can be a user, and social information refers to the interaction information between that user and other users. This interaction information can include: interactions on the internet, simultaneous appearances and travel together at the same geographical location, sharing the same hardware device, etc. Whether there is interaction is used to determine if there is a connection between users. Then, the method in this application is used to generate associated users for the application. Furthermore, based on the behavior of the associated users (e.g., whether there are any violations, whether there are any negative records, etc.), it is determined whether the associated users are suspicious individuals who could affect user information security.
[0073] In this embodiment of the application, the node to be identified may also be a terminal device or a server, wherein the social information may be data interaction information, data transmission information, device status information, etc. in the terminal device or server.
[0074] In a specific application, the node to be identified can be a terminal device, and social information can be other terminals that have transmitted data with that terminal device, other terminals on the same network segment as that terminal device, or other terminals under the same company name as that terminal device. The associated terminals of a given terminal device can be determined by whether data has been transmitted between terminal devices or whether there is a physical connection between them.
[0075] In this embodiment, social information may be follow information or communication information between user-used terminal devices or social accounts; or transaction information between different targets; or communication records and contact information stored in the device with user authorization; or communication information between users collected from video data; or it may be job information of employees within an enterprise or information about the department to which employees belong, etc.
[0076] In this solution, taking the user as an example, the social information of the user to be evaluated is first obtained with the user's authorization. Social information may include transaction information, call information, online interactions, simultaneous appearances and travels at the same geographical location, sharing the same hardware device, etc. The aforementioned social information can be publicly available data from users using communication devices on internet service platforms. In this embodiment, social associations can be temporary, such as users who had a phone call with the user on a particular day, or those geographically close; or they can be long-term, such as family associations, users in the phone's address book, workplace associations, etc., without specific limitations. User privacy can be protected by deleting or anonymizing user information. Anonymization can be achieved through data encryption.
[0077] In this approach, since users living in close proximity may have social connections, historical users who are geographically close to the target being evaluated can be obtained, such as those who are less than one kilometer away from the user's location or those who live in the same community.
[0078] S202. Construct a heterogeneous social relationship graph with different types of social relationships based on the social information, and determine the attention weight corresponding to the target to be evaluated.
[0079] In this embodiment of the invention, the security assessment model is trained using a large amount of historical sample data. Before constructing a heterogeneous social relationship graph with different types of social relationships based on the social information and determining the attention weights corresponding to the target to be assessed, a large amount of user information from historical users is first obtained, and the social information of each historical user is used as a training sample. A heterogeneous social relationship graph is constructed based on the training samples, and the attention weight vectors of the historical users are calculated. Since the security status of historical users is known, the known security status is used as a label. The attention weight vectors obtained from these training samples are input into the model for training. The output security score is compared with the actual security score of the historical user. The model parameters are adjusted using a loss function to minimize the error between the model's output security score and the actual security score of the historical user, ultimately obtaining the security assessment model. The models used in this embodiment of the invention include, but are not limited to, convolutional neural network models.
[0080] In some embodiments, after obtaining the social information of the target to be evaluated, a heterogeneous social relationship graph is constructed based on the social information. The mobile phone number of the target to be evaluated is used as the user to be evaluated, each user who has a social relationship with the target to be evaluated is used as a neighbor node, and the social relationships between users are used as edges to construct the heterogeneous social relationship graph.
[0081] Social relationships can be categorized into different types, such as phone call relationships, contact list relationships, and geographical location relationships.
[0082] This invention calculates the attention weight of the target to be evaluated from two aspects: node dimension and view dimension.
[0083] 1) Node dimension: The attention weights for the node dimension are calculated using the following formula:
[0084] k contains
[0085] Where u represents the user to be evaluated, and i represents the neighboring node. This represents the feature vector of neighbor node i in the current heterogeneous social relationship graph. This represents the attention vector of neighbor node i that needs to be learned in the current heterogeneous social relationship graph. This represents the set of neighbor nodes of the user u to be evaluated in the current heterogeneous social relationship graph. This represents the feature vector of neighbor node k in the current heterogeneous social relationship graph. This represents the attention vector of the neighbor node k that needs to be learned in the current heterogeneous social relationship graph.
[0086] The attention vector is represented by the vector formed by the social relationship between the user to be evaluated and the neighbor node i. Furthermore, the vector can also be formed according to the scores corresponding to the social relationship. For example, there may be a variety of different social relationships between the user to be evaluated and the neighbor node. In this scheme, a corresponding score can be given to different social relationships, or a scoring function can be set to score each different social relationship, and the scores corresponding to the different social relationships can be combined to form the attention vector.
[0087] The calculated attention weights α ui Normalization is performed.
[0088] After obtaining the attention coefficient of the user to be evaluated, it is weighted and summed with the features of the surrounding neighboring nodes to obtain the node attention weight vector of the user to be evaluated, which is expressed by the following formula:
[0089] k contains
[0090] Where, α uk e represents the attention weight between the user to be evaluated and neighbor node k. uk This represents the feature vector of the neighbor node k of the user to be evaluated.
[0091] Node-level comparisons of user influence between neighboring nodes and the user to be evaluated within the same user's social relationships include comparisons of neighboring nodes in a user's call history, neighboring nodes in a user's address book, and neighboring nodes near a user's geographical location.
[0092] 2) View Dimensions:
[0093] Calculate the edge attention weights that differ from the social relationships of the user being evaluated based on the node attention weights. Input the node attention weight vectors into the fully connected layer of the neural network.
[0094]
[0095] Where ReLU is a linear rectified function. Let be the node attention weight vector, where wi is the feature of neighbor node i; b is the bias term, with an initial value of a preset constant; the node attention weight vector is obtained by rectifying the function of the preceding connection layer through a linear rectifier function.
[0096] Then calculate the edge attention weight coefficients: The calculation method is the same as for calculating the node attention coefficients. Finally, a weighted sum is performed, and the edges are concatenated for each view to obtain the final edge attention weight vector. k contains
[0097] in, Let i be the edge attention weights of the edges between the target node and its neighbor node i. Let i be the i-th element in the node attention weight vector. Let be the feature vector of the edge between the target to be evaluated and its neighbor node i. The k-th element in the node attention weight vector Let be the feature vector of the edge between the target to be evaluated and its neighbor node k. This represents the edge attention weight of the edge between the user to be evaluated and its neighbor node k. Let be the k-th element in the node attention weight vector, and concat represents the merging function.
[0098] The view dimension compares the influence of nodes in different users' social relationships. For example, it compares the influence of nodes in a user's call information and nodes in a user's address book on the target to be evaluated, or it compares the influence of nodes in a user's address book and nodes near a user's geographical location on the target to be evaluated.
[0099] In this embodiment, after calculating the side attention weights between each neighboring node and the user to be evaluated, the side attention weights between neighboring nodes with the same type of social relationship and the user to be evaluated can be aggregated to obtain the average side attention weight of nodes with that type of social relationship. This average side attention weight is used as the side attention weight between each neighboring node with that type of social relationship and the user to be evaluated, reducing the impact of excessive attention to a single target on the calculation. Through this step, the average attention of the user to targets of different types of social relationships can be obtained, allowing for a macroscopic observation of the user's attention to targets of different social relationships, reducing the impact of excessive attention to a single target on the overall picture, and improving prediction accuracy. The aggregation method can include mean calculation, harmonic mean, weighted average, etc. Furthermore, the variance or volatility of the side attention weights between nodes with the same type of social relationship and the user to be evaluated can be determined. Based on the variance or volatility, the attention of the user to nodes with that type of social relationship can be effectively determined, thus determining the impact of nodes with different types of social relationships on the user to be evaluated.
[0100] S203. Input the attention weight into a preset security assessment model to calculate the trust level of the target to be assessed.
[0101] The node attention weight vector and edge attention weight vector of the target to be evaluated, obtained by weighted summation of attention weight coefficients, are respectively input into the security evaluation model trained in the above embodiment to obtain the trust level of the target to be evaluated between the same social relationship and between different social relationships. Finally, the user's trust level is obtained by comprehensive evaluation through different strategies.
[0102] Preferably, in order to ensure the accuracy of the results output by the text classification model, the training samples need to be updated regularly. The training samples with more recent time are selected, the trust level of the samples is calculated, and the security assessment model is trained based on the updated samples, and the parameters of the security assessment model are updated.
[0103] This invention acquires the social information of the target to be evaluated, constructs a heterogeneous social relationship graph with different types of social relationships based on the social information, determines the attention weight corresponding to the target to be evaluated, and inputs the attention weight vector into a preset security assessment model to calculate the trust level of the target to be evaluated. This solves the problem that when there is little user data, it is impossible to predict user trust level based on user characteristics or the prediction results are inaccurate. Based on the predicted user trust level, timely response strategies can be taken before providing services to users to avoid losses and improve service security.
[0104] S204. Based on the comparison results between the trust level and the trust level trend strategy, determine the service strategy of the target to be evaluated and push it.
[0105] For example, a trust trend strategy could be based on preset trust thresholds, extracted from a large amount of historical user data to generate multiple preset trust intervals. Different service strategies could then be developed for different trust intervals. Alternatively, a trust trend strategy could set trust thresholds based on the changing trend of trust over time. For instance, multiple different baseline trust thresholds and a unit trust threshold could be set. When the trend of social relationships changing over time shows that the trust of the target being identified is gradually decreasing, a change curve is generated based on the trust trend. The derivative of the change curve is used to obtain the rate of decrease. Subtracting the rate of decrease from the multiple different baseline trust thresholds and multiplying it by the unit trust threshold yields the trust threshold corresponding to the current trust trend. The final service strategy is then determined based on these multiple trust thresholds. Similarly, when the trend of trust changing over time shows that trust is gradually increasing, the baseline trust threshold is added to the growth rate multiplied by the unit trust threshold. The final service strategy is determined based on these multiple trust thresholds. Ultimately, different trust thresholds can be set for different trust trends to improve the accuracy of push notifications.
[0106] Those skilled in the art will understand that all or part of the steps of the above embodiments are implemented as a program (computer program) executed by a computer data processing device. When the computer program is executed, the method provided by the present invention can be implemented. Moreover, the computer program can be stored in a computer-readable storage medium, which can be a disk, optical disk, ROM, RAM, or other readable storage medium, or a storage array composed of multiple storage media, such as a disk or magnetic tape storage array. The storage medium is not limited to centralized storage; it can also be distributed storage, such as cloud storage based on cloud computing.
[0107] The following describes an embodiment of the apparatus of the present invention, which can be used to perform the method embodiments of the present invention. The details described in the apparatus embodiments of the present invention should be considered as supplements to the above method embodiments; details not disclosed in the apparatus embodiments of the present invention can be implemented with reference to the above method embodiments.
[0108] Figure 3 This is a schematic diagram of a target service determination device based on a heterogeneous graph neural network, as one embodiment of the present invention. Figure 3 As shown, the device 300 includes:
[0109] Information acquisition module 301 is used to acquire social information of the target to be evaluated;
[0110] Attention weight calculation module 302 is used to construct a social relationship heterogeneous graph with different types of social relationships based on the social information, and determine the attention weight corresponding to the target to be evaluated;
[0111] The evaluation module 303 is used to input the attention weights into a preset security evaluation model to calculate the trust level of the target to be evaluated;
[0112] The service determination module 304 is used to determine the service strategy of the target to be evaluated and push it based on the comparison result between the trust level and the trust level trend strategy.
[0113] The information acquisition module 301 further includes:
[0114] Social Relationship Acquisition Unit: Used to acquire social relationship information of the target to be evaluated, including call information, contact information, and geographical location information.
[0115] According to a preferred embodiment of the present invention, the device further includes a model training module, which is used to use the social information of historical users as training samples; construct a social relationship heterogeneous graph based on the training samples, calculate the attention weight vector of the historical users; and input the attention weight vector into a convolutional neural network model for training to obtain the security assessment model.
[0116] According to a preferred embodiment of the present invention, the model training module is further configured to: input the attention weight vector of historical users into the neural network model, compare the output security score with the actual security score of the historical users; adjust the model parameters through the loss function to minimize the error between the security score output by the model and the actual security score of the historical users, and finally obtain the security assessment model.
[0117] According to a preferred embodiment of the present invention, the attention weight calculation module 302 further includes: a social relationship heterogeneous graph creation unit, used to construct a social relationship heterogeneous graph by taking targets that have social relationships with the target to be evaluated as neighbor nodes and social information between targets as edges;
[0118] The node attention weight calculation unit is used to calculate the node attention weights between different neighbor nodes and the target to be evaluated based on the heterogeneous social relationship graph.
[0119] The edge attention weight calculation unit is used to calculate the edge attention weight between the target to be evaluated and each neighboring node based on the node attention weight.
[0120] According to a preferred embodiment of the present invention, the evaluation module 303 further includes: a node attention weight vector calculation unit, used to calculate the node attention weights by weighted summation to obtain a node attention weight vector;
[0121] The edge attention weight vector calculation unit is used to calculate the edge attention weights by weighted summation to obtain the edge attention weight vector.
[0122] The evaluation unit is used to input the node attention weight vector and edge attention weight vector of the target to be evaluated into the security evaluation model to obtain the trust level of the target to be evaluated.
[0123] According to a preferred embodiment of the present invention, the service determination module 304 further includes: a trust threshold range setting unit, used to set different trust threshold ranges; a service strategy formulation unit, used to set corresponding service strategies for different trust threshold ranges; and a service determination unit, used to provide corresponding service strategies according to the trust threshold range in which the trust level of the target to be evaluated is located.
[0124] According to a preferred embodiment of the present invention, the device further includes an update module for periodically updating the training samples and updating the security assessment model based on the updated training samples.
[0125] The information acquisition module 301 acquires different types of social information from users. The attention weight calculation module 302 constructs a heterogeneous social relationship graph with different types of social relationships based on the social information, determines the attention weight corresponding to the target to be evaluated, and the evaluation module 303 calculates the attention weight vector based on the calculated attention weight and inputs it into a preset security evaluation model to calculate the trust level of the target to be evaluated. The service determination module 304 determines the service strategy for the target to be evaluated and pushes it based on the comparison result between the evaluated trust level and the trust level trend strategy. This solves the problem that when there is limited user data, it is impossible to predict user trust level based on user characteristics or the prediction result is inaccurate. By making timely response strategies based on the predicted user trust level before providing services to users, losses can be avoided and data security of services can be ensured.
[0126] An electronic device according to an embodiment of the present invention includes a processor and a memory, the memory being used to store a computer-executable program, wherein when the computer program is executed by the processor, the processor executes a target service determination method based on a heterogeneous graph neural network.
[0127] The electronic device is manifested in the form of a general-purpose computing device. It may contain one or more processors that work collaboratively. This invention also does not preclude distributed processing, meaning that processors may be distributed across different physical devices. The electronic device of this invention is not limited to a single entity, but may also be the sum of multiple physical devices.
[0128] The memory stores a computer-executable program, typically machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least some steps of the method.
[0129] The memory includes volatile memory, such as random access memory (RAM) and / or cache memory, and may also be non-volatile memory, such as read-only memory (ROM).
[0130] Optionally, in this embodiment, the electronic device further includes an I / O interface for exchanging data with external devices. The I / O interface can represent one or more of several bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.
[0131] It should be understood that the electronic device of the present invention may also include elements or components not shown in the examples above. For example, some electronic devices also include display units such as a display screen, and some electronic devices also include human-computer interaction elements such as buttons and keyboards. Any electronic device capable of executing a computer-readable program in its memory to implement the method of the present invention or at least some steps of the method can be considered as an electronic device covered by the present invention.
[0132] An embodiment of the present invention provides a computer-readable recording medium. The computer-readable recording medium stores a computer-executable program, which, when executed, implements the target service determination method based on heterogeneous graph neural networks described above. The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.
[0133] The aforementioned computer-readable medium carries one or more programs, which, when executed by a device, enable the computer-readable medium to perform the following functions: acquire social information of the target to be evaluated; construct a heterogeneous social relationship graph with different types of social relationships based on the social information, and determine the attention weight corresponding to the target to be evaluated; input the attention weight into a preset security assessment model to calculate the trust level of the target to be evaluated.
[0134] From the above description of the embodiments, those skilled in the art will readily understand that the present invention can be implemented by hardware capable of executing specific computer programs, such as the system of the present invention, and the electronic processing unit, server, client, mobile phone, control unit, processor, etc. included in the system. The present invention can also be implemented by computer software that executes the methods of the present invention. However, it should be noted that the computer software executing the methods of the present invention is not limited to execution in one or a specific set of hardware entities; it can also be implemented in a distributed manner by unspecified hardware. For example, some method steps executed by the computer program can be executed in a mobile client, while others can be executed in a smart watch, smart pen, etc. For computer software, the software product can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or distributed across a network, as long as it enables electronic devices to execute the methods according to the present invention.
[0135] In summary, this invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that in practice, general-purpose data processing devices such as microprocessors or digital signal processors (DSPs) can be used to implement some or all of the functions of some or all of the components according to the embodiments of the invention. The invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the invention can be stored on a computer-readable medium or can take the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.
[0136] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the present invention is not inherently related to any specific computer, virtual device, or electronic device, and various general-purpose devices can also implement the present invention. The above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for determining target services based on heterogeneous graph neural networks, characterized in that, include: The social information of the target to be evaluated is obtained, wherein the target to be evaluated is an individual user or an enterprise user; wherein, when the target to be evaluated is an individual user, the social information includes basic information and behavioral information authorized by the user. The basic information authorized by the user includes call information, telephone address book information, social software address book, and the user's geographical location information. The behavioral information includes transaction information between users and communication information between users. When the target to be evaluated is an enterprise user, the social information includes equity penetration information between enterprises, news information between enterprises, and social information of managers within the enterprise. Based on the social information, a heterogeneous social relationship graph with different types of social relationships is constructed. The attention weights corresponding to the target to be evaluated are determined by the social relationships of the nodes in the heterogeneous social relationship graph. This includes: calculating the node attention weights of the target to be evaluated in terms of the node dimension, and comparing the influence between neighboring nodes and the target to be evaluated in the same type of social relationship; and calculating the edge attention weights that are different from the social relationships of the target to be evaluated based on the node attention weights in terms of the view dimension, and comparing the influence of nodes in different target social relationships. The trust level of the target to be evaluated is calculated by inputting the attention weights into a preset security assessment model, including: inputting the node attention weight vector and edge attention weight vector of the target to be evaluated into the trained security assessment model respectively, to obtain the trust level of the target to be evaluated between the same type of social relationship and the trust level of the target to be evaluated between different types of social relationships, and comprehensively evaluating and calculating the trust level of the target to be evaluated. Based on the comparison results of the trust level and the trust level trend strategy, the service strategy of the target to be evaluated is determined and pushed. The trust level trend strategy includes: setting a trust level threshold based on multiple preset trust level intervals generated by statistical analysis values of the target to be evaluated extracted from a large amount of historical user data, or setting a trust level threshold based on the trend of trust level changes over time.
2. The method according to claim 1, characterized in that, Based on the social information, a heterogeneous social relationship graph with different types of social relationships is constructed, and the attention weights corresponding to the target to be evaluated are determined, further including: Targets that have social relationships with the target to be evaluated are designated as neighbor nodes, and the social information between targets is designated as edges to construct a heterogeneous graph of social relationships. Calculate the node attention weights between different neighbor nodes and the target to be evaluated based on the heterogeneous social relationship graph; The edge attention weights between the target to be evaluated and each of its neighboring nodes are calculated based on the node attention weights.
3. The method according to claim 2, characterized in that, The attention weights are input into a preset security assessment model to calculate the trust level of the target to be assessed, further including: The node attention weights are weighted and summed to obtain the node attention weight vector; The edge attention weight vector is obtained by weighted summation of the edge attention weights.
4. The method according to claim 1, characterized in that, Based on the comparison results of the trust level and trust trend strategy, determining the service strategy for the target to be evaluated and pushing it further includes: Set corresponding service policies for different trust threshold ranges; Provide corresponding service strategies based on the trust level range in which the target to be evaluated falls.
5. The method according to claim 1, characterized in that, Before constructing a heterogeneous social relationship graph with different types of social relationships based on the social information, and determining the attention weight corresponding to the target to be evaluated, the process further includes: Use historical users' social information as training samples; A heterogeneous social relationship graph is constructed based on the training samples, and the attention weight vector of the historical user is calculated. The attention weight vector is input into the convolutional neural network model for training to obtain the security assessment model.
6. The method according to claim 5, characterized in that, The attention weight vector is input into a convolutional neural network model for training to obtain the security assessment model, which further includes: The attention weight vector of historical users is input into the neural network model, and the output security score is compared with the actual security score of the historical users. By adjusting the model parameters using a loss function, the error between the security score output by the model and the actual security score of the historical user is minimized, thus obtaining the security assessment model.
7. The method according to claim 6, characterized in that, The method further includes: The training samples are updated periodically, and the security assessment model is updated based on the updated training samples.
8. A target service determination device based on heterogeneous graph neural networks, characterized in that, include: The information acquisition module is used to acquire the social information of the target to be evaluated, which includes individual users or enterprise users. When the target to be evaluated is an individual user, the social information includes basic information and behavioral information authorized by the user. The basic information authorized by the user includes call information, telephone address book information, social software address book, and the user's geographical location information. The behavioral information includes transaction information between users and communication information between users. When the target to be evaluated is an enterprise user, the social information includes equity penetration information between enterprises, news information between enterprises, and social information of managers within the enterprise. The attention weight calculation module is used to construct a heterogeneous graph of social relationships with different types of social relationships based on the social information, and determine the attention weight corresponding to the target to be evaluated. This includes: calculating the node attention weight of the target to be evaluated in terms of the node dimension, and comparing the influence between neighboring nodes of the same type of target social relationship and the target to be evaluated; and calculating the edge attention weight of the social relationship different from that of the target to be evaluated based on the node attention weight in terms of the view dimension, and comparing the influence of nodes between different target social relationships. The evaluation module is used to input the attention weights into a preset security evaluation model to calculate the trust level of the target to be evaluated. It includes an evaluation unit that inputs the node attention weight vector and edge attention weight vector of the target to be evaluated into the trained security evaluation model to obtain the trust level of the target to be evaluated between the same type of social relationship and the trust level of the target to be evaluated between different types of social relationships, and comprehensively evaluates and calculates the trust level of the target to be evaluated. The service determination module is used to determine the service strategy of the target to be evaluated and push it based on the comparison result of the trust level and the trust level trend strategy. The trust level trend strategy includes: setting a trust level threshold based on multiple preset trust level intervals generated by statistical analysis values for the target to be evaluated extracted from a large amount of historical user data, or setting a trust level threshold based on the trend of trust level changes over time.
9. The apparatus according to claim 8, characterized in that, The attention weight calculation module further includes: The social relationship heterogeneous graph creation unit is used to construct a social relationship heterogeneous graph by taking targets that have social relationships with the target to be evaluated as neighbor nodes and the social information between targets as edges. The node attention weight calculation unit is used to calculate the node attention weights between different neighbor nodes and the target to be evaluated based on the heterogeneous social relationship graph. The edge attention weight calculation unit is used to calculate the edge attention weight between the target to be evaluated and each neighboring node based on the node attention weight.
10. The apparatus according to claim 9, characterized in that, The evaluation module further includes: The node attention weight vector calculation unit is used to calculate the node attention weights by weighted summation to obtain the node attention weight vector. The edge attention weight vector calculation unit is used to sum the edge attention weights to obtain the edge attention weight vector.
11. The apparatus according to claim 8, characterized in that, The service determination module further includes: The trust threshold range setting unit is used to set different trust ranges; The service strategy formulation unit is used to set corresponding service strategies for different trust level ranges. The service determination unit is used to provide a corresponding service strategy based on the trust level range in which the trust level of the target to be evaluated falls.
12. An electronic device comprising a processor and a memory, the memory being used to store a computer-executable program, characterized in that: When the computer executable program is executed by the processor, the processor performs the method as described in any one of claims 1-7.
13. A computer-readable medium storing a computer-executable program, characterized in that, When the computer executable program is executed, it implements the method as described in any one of claims 1-7.