Trust feature completion and trust relationship prediction method and system in the case of missing trust information

Abstract

The application provides a trust feature completion and trust relationship prediction method and system for missing trust information, comprising: constructing a social relationship graph and a trust relationship graph to obtain social attribute features and trust attribute features respectively; compressing the social attribute features and the trust attribute features, and constructing a feature converter to convert the compressed social attribute features; constructing a user trust relationship prediction model, taking the compressed features of the user as input, and outputting a prediction label; taking the converted compressed social attribute features as input, and outputting a prediction label; taking a real label as supervision information, combining a reconstruction loss and an information bottleneck loss, training the model, and outputting a trust relationship. The application constructs a social behavior graph and a trust relationship graph, and learns the social attribute features and the trust attribute features of the user respectively. This multi-view learning method can more comprehensively capture the feature information of the user, and improve the accuracy of trust relationship prediction.

Description

Technical Field

[0001] This invention belongs to the field of social relationship prediction technology, and specifically relates to a method and system for completing trust features and predicting trust relationships when trust information is missing. Background Technology

[0002] Trust assessment is the process of evaluating trust by quantifying attributes that influence trust. It has numerous applications in sensor networks, social networks, mobile crowdsourcing, and blockchain. Traditional trust assessment schemes are based on heuristic trust theory, typically fusing information from user attributes and user ratings using a predefined trust model to obtain a comprehensive user trust score or trust relationships between users. The effectiveness of such methods heavily relies on the predefined model, lacks universality, and cannot flexibly apply to different scenarios. To address the limitations of traditional trust assessment algorithms, some researchers have applied machine learning algorithms to trust assessment. Machine learning models automate the processing of trust-related data, extracting user trust features and predicting trust relationships between users. Traditional machine learning-based trust assessment algorithms typically rely on matrix factorization theory. They represent the trust relationships between users as a matrix, decompose the matrix to obtain node trust features and trusted features, and then perform matrix multiplication to obtain the complete trust relationship matrix. For example, Fazeli et al. explored the effectiveness of different metric methods in matrix factorization-based trust relationship evaluation and proposed a matrix factorization-based trust evaluation algorithm based on implicit trust scores. This algorithm incorporates implicit trust scores into matrix factorization instead of explicit trust relationships, and experiments demonstrated that the model achieves similar performance even without explicit trust scores. Similar work has also been applied in [the context of the algorithm]. While matrix factorization-based methods have shown good effectiveness, they are based on shallow models and cannot uncover the deep, complex features of nodes, making it difficult to effectively discover the complex nonlinear mapping between user trust features and trust relationships.

[0003] To address the aforementioned problems, some scholars have introduced deep learning models for trust feature extraction and trust relationship prediction. By modeling trust relationships as graphs and processing them using graph convolutional neural networks (GCNNs), these models have achieved remarkable results. For example, Guardian uses GCNNs to process and fuse two types of user trust features separately, constructing a trust relationship prediction scheme based on GCNNs. Building upon this, KGTrust further integrates other social networks besides trust relationship networks, expanding the trust relationship graph and achieving certain performance improvements. GATTrust integrates trust features from different domains based on an attention mechanism to improve trust prediction performance. Some works further consider the dynamic nature of trust, constructing dynamic trust relationship prediction mechanisms based on dynamic graph neural networks, such as TrustGuard. These works have effectively solved the trust relationship prediction problem.

[0004] However, the above work has the following problems:

[0005] 1. Existing trust relationship prediction methods based on graph neural networks usually require obtaining sufficient trust relationship interactions between nodes. However, in practical applications, it is unrealistic to ensure that all nodes have sufficient trust interactions, which makes it difficult for existing trust relationship prediction algorithms based on graph neural networks to play a role.

[0006] 2. Although some algorithms (such as KGTrust) have introduced external knowledge to supplement the trust relationship, these methods still require the predicted nodes to have trust interactions, and thus still struggle to overcome the problem of missing trust relationships. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for completing trust features and predicting trust relationships when trust information is missing, so as to solve the above-mentioned problems.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a method for completing trust features and predicting trust relationships when trust information is missing, including:

[0010] Based on users' social behavior and the trust relationships between users, social relationship graphs and trust relationship graphs are constructed respectively. Based on the construction of social relationship graphs and trust relationship graphs, social attribute features and trust attribute features are obtained respectively.

[0011] Social attribute features and trust attribute features are compressed, and a feature converter is constructed to transform the compressed social attribute features;

[0012] Construct a user trust relationship prediction model, taking the compressed features of users as input and outputting predicted labels; then take the transformed compressed social attribute features as input and output predicted labels.

[0013] Using real labels as supervisory information, and combining reconstruction loss and information bottleneck loss, the model is trained and outputs trust relationships.

[0014] Furthermore, the construction of social relationship graphs and trust relationship graphs based on users' social behavior and trust relationships between users includes:

[0015] By preprocessing user behavior information on social networks, a user social behavior network graph is constructed for different types of user social behaviors. ;

[0016] Construct a user trust relationship network diagram based on the trust relationships between users. And construct initial trust features for users based on their trust relationships. Specifically:

[0017] Standardize the trust characteristics between users to form The system identifies discrete trust levels and models the trust relationships between users as a directed weighted graph.

[0018] With users For example, based on its trust relationships with other users and other users' trust relationships with it, out-degree trust features are initialized respectively. and in-degree trust features This forms a complete trust relationship network diagram. .

[0019] Furthermore, the social attribute features and trust attribute features obtained based on the construction of social relationship graphs and trust relationship graphs respectively include:

[0020] exist A graph neural network model based on meta-paths is constructed for feature extraction. Features from different neighbor nodes and semantic features extracted from different meta-paths are fused using intra-semantic attention and inter-semantic attention mechanisms, respectively, ultimately forming a set of social attribute features for user nodes. Specifically:

[0021] according to The relation type definition meta-path set ;

[0022] Based on the defined metapath, Divided into Image ;

[0023] Using graph convolutional neural networks and fusing multi-head attention mechanisms to Process and extract user features ;

[0024] For different graphs Extracted user features Establish an attention mechanism to integrate it into the user's social characteristics. ;

[0025] Build initial user characteristics based on node trust relationships. and trust relationship network By leveraging the attention mechanism to fuse neighbor information of nodes, trust features of user nodes are formed. Specifically:

[0026] Set up A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For the in-degree neighbors, the attention score is calculated as follows:

[0027]

[0028] Based on this score, the fused features are calculated:

[0029]

[0030] go through After training the layer, output its in-degree features. ;

[0031] Set up A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For out-degree neighbors, the attention score is calculated as follows:

[0032]

[0033] Based on this score, the fused features are calculated:

[0034]

[0035] go through After training the layer, output its in-degree features. ;

[0036] Will and After concatenation, input a fully connected network and an activation layer, and output... Trust characteristics .

[0037] Furthermore, the compression of social attribute features and trust attribute features includes:

[0038] Given and Build encoders respectively and right and Processing to obtain and Constructing information bottleneck loss based on information bottleneck theory as follows:

[0039] .

[0040] Furthermore, given Further construct the conversion encoder right Processed for learning arrive The mapping relationship, based on the feature transformation loss, is as follows:

[0041] .

[0042] Furthermore, the construction of the user trust relationship prediction model, using the user's compressed features as input, outputs predicted labels; and then using the transformed compressed social attribute features as input, outputs predicted labels, including:

[0043] A user trust relationship prediction model is built based on multilayer perceptron (MLP), with user... and Compression characteristics and Input: predict label At the same time and Input: predict label .

[0044] Furthermore, the model is trained using real labels as supervision information, combined with reconstruction loss and information bottleneck loss, and outputs trust relationships, including:

[0045] The feature extraction module, feature compression module, and feature transformation module are used in the training dataset respectively. and The above process the graph data to ultimately obtain the predicted relationships. Combined with real trust relationships Constructing cross-entropy loss ;

[0046] Combination , and The overall loss is constructed as follows:

[0047]

[0048] Based on training data and Train the model and guide its convergence.

[0049] Secondly, the present invention provides a system for completing trust features and predicting trust relationships when trust information is missing, comprising:

[0050] The graph construction module is used to construct social relationship graphs and trust relationship graphs based on users' social behavior and trust relationships between users, respectively, and to obtain social attribute features and trust attribute features based on the constructed social relationship graphs and trust relationship graphs, respectively;

[0051] The compression and conversion module is used to compress social attribute features and trust attribute features, and to build a feature converter to convert the compressed social attribute features.

[0052] The predicted label output module is used to build a user trust relationship prediction model. It takes the user's compressed features as input and outputs predicted labels; then it takes the transformed compressed social attribute features as input and outputs predicted labels.

[0053] The training output module is used to train the model using real labels as supervision information, combined with reconstruction loss and information bottleneck loss, and outputs trust relationships.

[0054] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a method for completing trust features and predicting trust relationships when the trust information is missing.

[0055] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the trust feature completion and trust relationship prediction method for missing trust information.

[0056] Compared with the prior art, the present invention has the following technical effects:

[0057] This invention introduces a multi-view node social relationship graph as a feature supplement, learning a mapper from social attribute features to trust features on a trust relationship graph composed of a small number of nodes with trust interactions. Then, on a large dataset lacking trust relationships, the learned feature mapper is used to complete the conversion from social attribute features to trust features. In this way, even if nodes lack trust relationship data, their trust features can be effectively learned and trust relationship prediction can be performed.

[0058] This invention constructs social behavior graphs and trust relationship graphs, and learns users' social attribute features and trust attribute features respectively. This multi-view learning approach can more comprehensively capture users' feature information and improve the accuracy of trust relationship prediction.

[0059] Through feature compression and feature transformation modules, the technical solution can compress irrelevant information in the original features, extract consistent information between social attribute features and trust attribute features, and further learn the mapping relationship from social attribute features to trust attribute features. These processing steps help reduce noise interference, improve feature effectiveness, and thus enhance the accuracy of trust relationship prediction.

[0060] This invention employs a meta-path-based heterogeneous graph neural network, which is effectively applicable to modeling and feature learning of different types of social relationships. This allows the model to handle different types of social data and trust relationships more flexibly, enhancing its universality.

[0061] This invention adopts a modular design approach, including a graphing module, a feature learning module, a feature compression and transformation module, and a relationship prediction module. This design allows the model to be flexibly combined and adjusted according to the needs of specific application scenarios, enhancing the model's flexibility and scalability.

[0062] This invention introduces the information bottleneck theory to construct an information bottleneck loss mechanism. This technical solution extracts as much useful information as possible during feature compression while discarding irrelevant information. This helps reduce model complexity and improve training efficiency and convergence.

[0063] This invention constructs an overall loss function by combining information bottleneck loss, feature transformation loss, and cross-entropy loss, and then trains the model using this function. This multi-objective optimization approach can more comprehensively consider the model's performance metrics, promoting model convergence and performance improvement.

[0064] In summary, this technical solution effectively solves the problem of missing trust relationship information by introducing a multi-view node social relationship graph as a feature supplement, using a heterogeneous graph neural network based on meta-paths for feature learning, applying information bottleneck theory for feature compression and transformation, and constructing a modular model design. This improves the accuracy of trust relationship prediction, enhances the universality and flexibility of the model, and improves the training efficiency and convergence of the model. Attached Figure Description

[0065] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings:

[0067] Example 1, please refer to Figure 1 This invention provides a method for completing trust features and predicting trust relationships when trust information is missing, including:

[0068] Based on users' social behavior and the trust relationships between users, social relationship graphs and trust relationship graphs are constructed respectively. Based on the construction of social relationship graphs and trust relationship graphs, social attribute features and trust attribute features are obtained respectively.

[0069] Social attribute features and trust attribute features are compressed, and a feature converter is constructed to transform the compressed social attribute features;

[0070] Construct a user trust relationship prediction model, taking the compressed features of users as input and outputting predicted labels; then take the transformed compressed social attribute features as input and output predicted labels.

[0071] Using real labels as supervisory information, and combining reconstruction loss and information bottleneck loss, the model is trained and outputs trust relationships.

[0072] The trust feature completion and trust relationship prediction method proposed in this invention achieves trust relationship prediction in the case of incomplete or missing trust information through a series of meticulously designed steps.

[0073] By constructing social relationship graphs and trust relationship graphs, this invention effectively extracts rich feature information from users' social behaviors and the trust relationships between users. This graph-structured data representation can capture the complex relationships and interaction patterns between users, providing a solid foundation for subsequent trust relationship prediction.

[0074] Compression of social and trust attribute features helps reduce data redundancy and noise, improving the model's computational efficiency and generalization ability. Furthermore, constructing a feature transformer to transform the compressed social attribute features allows for further refinement and optimization, making them more suitable for predicting trust relationships.

[0075] This invention constructs a user trust relationship prediction model and employs a dual-path prediction approach. First, prediction is performed using the user's compressed features as input; second, prediction is performed again using the transformed compressed social attribute features as input. This dual-path design not only improves prediction accuracy but also enhances the model's sensitivity and robustness to different feature combinations.

[0076] Using real labels as supervision information and combining reconstruction loss and information bottleneck loss to train the model helps to maximize the retention of information related to trust relationships while maintaining data compression efficiency. This loss function design ensures the model's predictive performance while avoiding overfitting and underfitting.

[0077] Through the combined effect of the above steps, this invention can more accurately predict trust relationships between users. Even when trust information is missing or incomplete, reliable predictions can be made by supplementing trust features and utilizing social relationship information. This has significant application value in fields such as social network analysis, recommendation systems, and credit assessment.

[0078] The method of this invention is not dependent on a specific dataset or scenario and can be flexibly applied to different social network and trust relationship prediction tasks. Furthermore, with the increase in data volume and continuous model training, the predictive performance of the model is expected to further improve, demonstrating good scalability and adaptability.

[0079] In summary, the trust feature completion and trust relationship prediction method for missing trust information proposed in this invention has significant technical effects in enhancing data representation capabilities, improving prediction accuracy, and optimizing model performance, providing new ideas and methods for social network analysis and trust relationship prediction.

[0080] Example 2: The invention provides a method for completing trust features and predicting trust relationships when trust information is missing, specifically including:

[0081] Social relationship graphs are constructed based on users' social behavior and the trust relationships between users. Trust Relationship Diagram ;

[0082] To each and The data is processed to obtain the user's social attribute characteristics. and ;

[0083] Based on the information bottleneck theory and encoder network, a feature compressor is constructed to obtain compressed social attribute features and trust attribute features. and ; Construct feature converter pairs The transformation is performed, and the transformed features are guaranteed based on the reconstruction loss. Approaching ;

[0084] A user trust relationship prediction module is built based on Multi-Layer Perception (MLP) to predict user trust relationships. and Compression characteristics and Input: predict label At the same time and Input: predict label ;

[0085] With real labels As supervisory information, the model is trained by combining reconstruction loss and information bottleneck loss;

[0086] The model is applied to the target graph. The target scenario only contains a social relationship graph and lacks a trust relationship graph. The aforementioned feature learning module, feature compression and transformation module, and relationship prediction module are used to process the graph and output the trust relationship.

[0087] The composition module mainly includes the following parts:

[0088] Social Behavior Graph Construction: By preprocessing information on user behavior on social networks (such as likes, purchases, comments, etc.), a user social behavior network graph is constructed for different types of user social behaviors. ;

[0089] Trust Relationship Graph Construction: Trust Relationship Network Graph Construction Module: Based on the trust relationships between users, construct a user trust relationship network graph. And construct initial trust features for users based on their trust relationships. ;

[0090] The feature learning module mainly includes the following steps:

[0091] The user social attribute feature extraction module based on meta-path: In A graph neural network model based on meta-paths is constructed for feature extraction. Features from different neighbor nodes and semantic features extracted from different meta-paths are fused using intra-semantic attention and inter-semantic attention mechanisms, respectively, ultimately forming a set of social attribute features for user nodes.

[0092] The user trust attribute feature extraction module based on graph attention mechanism constructs initial user features based on the trust relationships between nodes. and trust relationship network By leveraging the attention mechanism to fuse neighbor information of nodes, trust features of user nodes are formed. ;

[0093] The feature compression and conversion module mainly includes the following steps:

[0094] Feature compression module: given and Build encoders respectively and right and Processing to obtain and Constructing information bottleneck loss based on information bottleneck theory as follows:

[0095]

[0096] The main purpose of this module is to compress irrelevant information in the original features and extract as much consistent information as possible between social attribute features and trust attribute features for trust relationship prediction.

[0097] Feature transformation module: given Further construct the conversion encoder right Processed for learning arrive The mapping relationship, based on the feature transformation loss, is as follows:

[0098]

[0099] The model training module mainly includes the following steps:

[0100] The feature extraction module, feature compression module, and feature transformation module are used in the training dataset respectively. and The above process the graph data to ultimately obtain the predicted relationships. Combined with real trust relationships Constructing cross-entropy loss ;

[0101] Combination , and The overall loss is constructed as follows:

[0102]

[0103] Based on training data and Train the model and guide its convergence.

[0104] The construction of the trust relationship graph mainly includes the following steps:

[0105] User trust relationship extraction: Standardize the trust characteristics between users to form... The system identifies discrete trust levels and models the trust relationships between users as a directed weighted graph.

[0106] User trust feature initialization: based on user For example, based on its trust relationships with other users and other users' trust relationships with it, out-degree trust features are initialized respectively. and in-degree trust features This forms a complete trust relationship network diagram. ;

[0107] The user social attribute feature extraction module based on meta-path mainly includes the following steps:

[0108] Metapath definition: based on The relationship types (such as likes, follows, purchases, etc.) define a set of metapaths. For example, "user-product-user" can form a meta-path;

[0109] Isomorphic graph extraction: Based on the defined metapath, extract the following: Divided into Image ;

[0110] User feature extraction based on graph attention mechanism: Utilizing graph convolutional neural networks and incorporating multi-head attention mechanisms to extract user features... Process and extract user features ;

[0111] User social feature fusion based on semantic inter-attention mechanism: targeting different graphs Extracted user features Establish an attention mechanism to integrate it into the user's social characteristics. ;

[0112] The user trust attribute feature extraction module based on graph attention mechanism mainly includes the following steps:

[0113] In-degree trust feature embedding module based on multi-layer graph convolutional network: Construction A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For the in-degree neighbors, the attention score is calculated as follows:

[0114]

[0115] Based on this score, the fused features are calculated:

[0116]

[0117] go through After training the layer, output its in-degree features.

[0118] Out-degree trust feature embedding module based on multi-layer graph convolutional network: Construction A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For out-degree neighbors, the attention score is calculated as follows:

[0119]

[0120] Based on this score, the fused features are calculated:

[0121]

[0122] go through After training the layer, output its in-degree features.

[0123] Trust Feature Fusion and Output: and After concatenation, input a fully connected network and an activation layer, and output... Trust characteristics .

[0124] This invention addresses the trust assessment problem in social networking platforms, including ubiquitous social networks and mobile crowdsourcing. The platform primarily comprises a social networking platform and... It consists of individual users. For ease of representation, this patent uses... This refers to a social platform that primarily provides online social services such as instant messaging, information sharing, and shopping. It is also responsible for predicting trust relationships based on users' trust patterns and social behaviors to better serve them. The algorithms involved in this patent are all deployed on this platform. middle, Maintaining user social behavior graphs in different social scenarios And user trust relationship diagram The user set is represented as User use It provides social services and forms trust relationships with other users based on personal preferences, indicating either trust or distrust.

[0125] This invention learns on a trust relationship graph composed of a small number of nodes with trust interactions. By introducing a multi-view node social relationship graph as a feature supplement, a mapper from social attribute features to trust features is learned on this small dataset. Then, it processes a large dataset with missing trust relationships, learning its social attribute features through the rich social relationships of the nodes in the dataset, and using the learned feature mapper to complete the conversion of social attribute features to trust features. This ensures that even when trust relationships are missing, trust features can still be effectively learned and trust relationship prediction can be performed, effectively making up for the shortcomings of existing works.

[0126] In another embodiment of the present invention, a trust feature completion and trust relationship prediction system for trust information missing is provided, which can be used to implement the above-mentioned trust feature completion and trust relationship prediction method for trust information missing. Specifically, the system includes:

[0127] The graph construction module is used to construct social relationship graphs and trust relationship graphs based on users' social behavior and trust relationships between users, respectively, and to obtain social attribute features and trust attribute features based on the constructed social relationship graphs and trust relationship graphs, respectively;

[0128] The compression and conversion module is used to compress social attribute features and trust attribute features, and to build a feature converter to convert the compressed social attribute features.

[0129] The predicted label output module is used to build a user trust relationship prediction model. It takes the user's compressed features as input and outputs predicted labels; then it takes the transformed compressed social attribute features as input and outputs predicted labels.

[0130] The training output module is used to train the model using real labels as supervision information, combined with reconstruction loss and information bottleneck loss, and outputs trust relationships.

[0131] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0132] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve corresponding method flows or corresponding functions. The processor described in this embodiment of the present invention can be used for the operation of trust feature completion and trust relationship prediction methods for missing trust information.

[0133] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the trust feature completion and trust relationship prediction methods related to missing trust information in the above embodiments.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for completing trust features and predicting trust relationships when trust information is missing, characterized in that, include: Based on users' social behavior and the trust relationships between users, social relationship graphs and trust relationship graphs are constructed respectively. Based on the construction of social relationship graphs and trust relationship graphs, social attribute features and trust attribute features are obtained respectively. Social attribute features and trust attribute features are compressed, and a feature converter is constructed to transform the compressed social attribute features. Construct a user trust relationship prediction model, taking the compressed features of users as input and outputting predicted labels; then take the transformed compressed social attribute features as input and output predicted labels. Using real labels as supervision information, the model is trained by combining reconstruction loss and information bottleneck loss, and the trust relationship is output. The construction of social relationship graphs and trust relationship graphs based on users' social behavior and trust relationships between users includes: By preprocessing user behavior information on social networks, a user social behavior network graph is constructed for different types of user social behaviors. ; Construct a user trust relationship network diagram based on the trust relationships between users. And construct initial trust features for users based on their trust relationships. Specifically: Standardize the trust characteristics between users to form The system identifies discrete trust levels and models the trust relationships between users as a directed weighted graph. With users For example, based on its trust relationships with other users and other users' trust relationships with it, out-degree trust features are initialized respectively. and in-degree trust features This forms a complete trust relationship network diagram. ; The social attribute features and trust attribute features obtained based on the construction of social relationship graphs and trust relationship graphs, respectively, include: exist A graph neural network model based on meta-paths is constructed for feature extraction. Based on intra-semantic attention and inter-semantic attention mechanisms, features from different neighbor nodes and inter-semantic features extracted under different meta-paths are fused to form a set of social attribute features of user nodes. Specifically: according to The relation type definition meta-path set ; Based on the defined metapath, Divided into Image ; Using graph convolutional neural networks and fusing multi-head attention mechanisms to Process and extract user features ; For different graphs Extracted user features Establish an attention mechanism to integrate it into the user's social characteristics. ; Build initial user characteristics based on node trust relationships. and trust relationship network By leveraging the attention mechanism to fuse neighbor information of nodes, trust features of user nodes are formed. Specifically: Set up A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For the in-degree neighbors, the attention score is calculated as follows: Based on this score, the fused features are calculated: go through After training the layer, output its in-degree features. ; Set up A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For out-degree neighbors, the attention score is calculated as follows: Based on this score, the fused features are calculated: go through After training the layer, output its in-degree features. ; Will and After concatenation, input a fully connected network and an activation layer, and output... Trust characteristics .

2. The method for completing trust features and predicting trust relationships when trust information is missing, as described in claim 1, is characterized in that, The compression of social attribute features and trust attribute features includes: Given and Build encoders respectively and right and Processing to obtain and Constructing information bottleneck loss based on information bottleneck theory as follows: 。 3. The method for completing trust features and predicting trust relationships when trust information is missing, as described in claim 2, is characterized in that... Given Further construct the conversion encoder right Processed for learning arrive The mapping relationship, based on the feature transformation loss, is as follows: 。 4. The method for completing trust features and predicting trust relationships when trust information is missing, as described in claim 1, is characterized in that... The method for constructing a user trust relationship prediction model takes the user's compressed features as input and outputs a predicted label; then, it takes the transformed compressed social attribute features as input and outputs a predicted label, including: A user trust relationship prediction model is built based on multilayer perceptron (MLP), with user... and Compression characteristics and Input: predict label At the same time and Input: predict label .

5. The method for completing trust features and predicting trust relationships when trust information is missing, as described in claim 1, is characterized in that... The method uses real labels as supervision information, combines reconstruction loss and information bottleneck loss to train the model, and outputs trust relationships, including: The feature extraction module, feature compression module, and feature transformation module are used in the training dataset respectively. and The above process the graph data to ultimately obtain the predicted relationships. Combined with real trust relationships Constructing cross-entropy loss ; Combination , and The overall loss is constructed as follows: Based on training data and Train the model and guide its convergence.

6. A system for completing trust features and predicting trust relationships when trust information is missing, characterized in that: include: The graph construction module is used to construct social relationship graphs and trust relationship graphs based on users' social behavior and trust relationships between users, respectively, and to obtain social attribute features and trust attribute features based on the constructed social relationship graphs and trust relationship graphs, respectively; The compression and conversion module is used to compress social attribute features and trust attribute features, and to build a feature converter to convert the compressed social attribute features. The predicted label output module is used to build a user trust relationship prediction model. It takes the user's compressed features as input and outputs predicted labels; then it takes the transformed compressed social attribute features as input and outputs predicted labels. The training output module is used to train the model using real labels as supervision information, combined with reconstruction loss and information bottleneck loss, and output trust relationships. The construction of social relationship graphs and trust relationship graphs based on users' social behavior and trust relationships between users includes: By preprocessing user behavior information on social networks, a user social behavior network graph is constructed for different types of user social behaviors. ; Construct a user trust relationship network diagram based on the trust relationships between users. And construct initial trust features for users based on their trust relationships. Specifically: Standardize the trust characteristics between users to form The system identifies discrete trust levels and models the trust relationships between users as a directed weighted graph. With users For example, based on its trust relationships with other users and other users' trust relationships with it, out-degree trust features are initialized respectively. and in-degree trust features This forms a complete trust relationship network diagram. ; The social attribute features and trust attribute features obtained based on the construction of social relationship graphs and trust relationship graphs, respectively, include: exist A graph neural network model based on meta-paths is constructed for feature extraction. Based on intra-semantic attention and inter-semantic attention mechanisms, features from different neighbor nodes and inter-semantic features extracted under different meta-paths are fused to form a set of social attribute features of user nodes. Specifically: according to The relation type definition meta-path set ; Based on the defined metapath, Divided into Image ; Using graph convolutional neural networks and fusing multi-head attention mechanisms to Process and extract user features ; For different graphs Extracted user features Establish an attention mechanism to integrate it into the user's social characteristics. ; Build initial user characteristics based on node trust relationships. and trust relationship network By leveraging the attention mechanism to fuse neighbor information of nodes, trust features of user nodes are formed. Specifically: Set up A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For the in-degree neighbors, the attention score is calculated as follows: Based on this score, the fused features are calculated: go through After training the layer, output its in-degree features. ; Set up A graph convolutional neural network with layers of user initial in-degree trust features. Initial features Assuming the first The output of the layer is ,if There exists right Trust edge Then it is believed yes For out-degree neighbors, the attention score is calculated as follows: Based on this score, the fused features are calculated: go through After training the layer, output its in-degree features. ; Will and After concatenation, input a fully connected network and an activation layer, and output... Trust characteristics .

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the trust feature completion and trust relationship prediction method for missing trust information as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the trust feature completion and trust relationship prediction method for missing trust information as described in any one of claims 1 to 5.

Images