Decentralized trust relationship collaborative prediction method and related device

Abstract

The invention discloses a decentralized trust relationship collaborative prediction method and a related device, and the method comprises the steps: starting a block chain system, publishing a public safety parameter and an initialization model on a block chain, and loading a consensus algorithm at a consensus node; constructing a local complete graph by using local data, pre-training the local graph to obtain a user feature extraction and relationship prediction model, and obtaining features of user nodes; establishing a public graph based on the public knowledge, sending the public graph to a block chain consensus node, and performing consensus confirmation on the public graph; and outputting a final trust relationship prediction model. The invention discloses a decentralized trust relationship collaborative prediction method based on federated learning and a graph neural network, and aims to solve the limitation of a traditional trust evaluation method in a multi-source distributed scene and improve the accuracy of trust relationship prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of trust relationship prediction, and in particular relates to a decentralized trust relationship collaborative prediction method and related devices. Background Art

[0002] Trust assessment is the process of evaluating trust by quantifying the attributes that affect trust. It has many applications in sensor networks, social networks, mobile crowdsourcing, blockchain and other fields. Traditional trust assessment schemes are based on heuristic trust theory. Usually, based on user attributes and scores between users, the above information is integrated based on a predefined trust model to obtain a comprehensive trust score of the user or a trust relationship between users. The effectiveness of such methods depends greatly on the predefined model, has poor universality, and cannot be flexibly applied to different scenarios. In view of the limitations of traditional trust assessment algorithms, some scholars have applied machine learning algorithms to trust assessment, and automatically processed trust-related data through machine learning models, extracted user trust features and predicted trust relationships between users. Traditional trust assessment algorithms based on machine learning are usually based on matrix decomposition theory. By representing the trust relationship between users as a matrix and decomposing the matrix, the node trust features and the trusted features are obtained, and then the matrix multiplication is performed to obtain the complete trust relationship matrix. For example, Fazeli et al. explored the effectiveness of different measurement methods in trust relationship evaluation based on matrix decomposition and proposed a matrix decomposition trust evaluation algorithm based on implicit trust score. By incorporating implicit trust score into matrix decomposition instead of explicit trust relationship, the model was experimentally confirmed to achieve similar performance without explicit trust score. Similar work has also been applied in. The matrix decomposition-based method has shown good effectiveness, but this method is based on a shallow model and cannot mine the deep complex features of nodes, making it difficult to effectively discover the complex nonlinear mapping relationship between user trust features and trust relationships.

[0003] In response to the above problems, some scholars have introduced deep learning models for the extraction of trust features and the prediction of trust relationships. By modeling the trust relationship as a graph and processing it with a graph convolutional neural network, this type of model has achieved remarkable results. For example: Guardian uses a graph convolutional neural network to process and fuse the two trust features of the user separately, and builds a trust relationship prediction scheme based on a graph convolutional neural network; on this basis, KGTrust further integrates other social networks besides the trust relationship network to expand the trust relationship graph and achieve a certain performance improvement; GATTrust integrates trust features from different fields based on the attention mechanism to improve trust prediction performance; some work further considers the dynamic nature of trust and builds a dynamic prediction mechanism for trust relationships based on dynamic graph neural networks, such as TrustGuard. The above work has solved the problem of trust relationship prediction very well.

[0004] However, the above work has the following problems: 1. Existing neural network-based trust relationship prediction schemes are usually oriented towards centralized scenarios, but do not consider scenarios where trust relationship data is distributed from multiple sources, and cannot predict trust relationships in scenarios where different institutions are dispersed; 2. Existing trust relationship prediction schemes based on federated learning do not consider cross-client feature propagation, and the learned features are not sufficient, resulting in low accuracy of trust assessment. Summary of the invention

[0005] The purpose of the present invention is to provide a decentralized trust relationship collaborative prediction method and related devices to solve the above problems.

[0006] To achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a decentralized trust relationship collaborative prediction method, comprising: Start the blockchain system and publish the public security parameters and initialization model on the blockchain, and load the consensus algorithm on the consensus node; Use local data to build a local complete graph, pre-train on this map to obtain user feature extraction and relationship prediction models, and obtain the features of user nodes; A public graph is established based on public knowledge, and the public graph is sent to the blockchain consensus node for consensus confirmation of the public graph; By using the public graph as an intermediate node, the features of the user nodes are updated, the user feature extraction and relationship prediction models are trained, the trained models are sent to the blockchain, and the model fusion is completed based on the federated averaging algorithm until the objective function converges and the final trust relationship prediction model is output.

[0007] Furthermore, the blockchain system is started and the public security parameters and initialization model are published on the blockchain, and the consensus algorithm is loaded on the consensus node, including: A group of nodes are selected as blockchain consensus nodes, and the consensus nodes negotiate the public key and public security parameters of the encryption system; Establish a consensus mechanism and start the blockchain; The consensus nodes randomize the model parameters and integrate them using smart contracts, and publish the blockchain or public security parameters on the blockchain.

[0008] Furthermore, the construction of a local complete graph using local data includes: Each client Use local data to build a user social behavior graph and trust relationship diagram , user-centric, building a complete local graph ; For the local user behavior data of each client, other entity nodes other than users are introduced. With the user as the center, the interaction relationships between users and between users and other objects are collected. These interaction objects are regarded as nodes and the interaction relationships are regarded as edges to construct a social relationship graph from this perspective. ; Generate edge features based on edge information. For numerical data, standardize and discretize them, and then use one-hot encoding to construct edge features. For complex data such as text comments and image comments, use feature extractors to extract features. For user nodes, their personal public attribute information is used for modeling; for product, article, and movie nodes, their attribute information and public descriptions are comprehensively utilized, and a multimodal feature extractor is used for feature initialization.

[0009] Furthermore, the user feature extraction and relationship prediction model is obtained by pre-training in the map to obtain the features of the user nodes, including: Each client , for a given node , which is shown in Figure The neighbor set is ,node , and nodes The edge of the record , the feature is recorded as ,but The attention parameter The calculation is performed using the following formula:

[0010] Get attention score Afterwards, given The layer output is , then The output of the layer is:

[0011] Based on the attention mechanism, we further design a multi-head attention mechanism, which includes Independent attentions extract features and perform feature fusion respectively; assuming that The features extracted by the attention mechanism are , then The output of the layer is:

[0012] Feature output: When all are completed After extracting the features of the layer, the final features are output .

[0013] Furthermore, the public knowledge graph is established based on the public knowledge, and the public graph is sent to the blockchain consensus node to confirm the consensus of the public graph, including: For non-user nodes of products, movies, and music, search for relevant expected data for knowledge graph construction based on node type; Use named entity recognition and relationship extraction to extract knowledge from the corpus, extract the relationship between nodes, and form a public knowledge graph , It should cover non-user node entities contained in the user client's local data; Graph recording based on public knowledge , It is sent to the blockchain consensus node, which verifies the correctness of the knowledge graph content. After verification, the consensus algorithm is called to complete the consensus confirmed.

[0014] Furthermore, the public graph is used as an intermediate node to update the features of the user node, train the user feature extraction and relationship prediction model, send the trained model to the blockchain, complete the model fusion based on the federated average algorithm until the objective function converges, and output the final trust relationship prediction model, including: Each client passes As an intermediate node, it discovers neighbors across clients and propagates features to update user features and complete the model training, and will The features of the knowledge entities contained in are sent to the blockchain, which uses the graph attention mechanism to fuse the features of the knowledge entities sent by different clients and publish them on the blockchain; The client will use the locally trained model The data is sent to the blockchain, and the blockchain consensus node verifies the validity of the model. When a sufficient number of models are collected or the waiting time exceeds the preset time, the model fusion smart contract is loaded and the model fusion is completed based on the federated average algorithm. The fusion method is as follows:

[0015] The blockchain consensus mechanism is then called to complete the The consensus is reached and published on the blockchain; Each client and blockchain consensus node loops until the objective function converges and outputs the final trust relationship prediction model ; The client loads the final model Processing local data, given a user and ,use Extract its trust features , And the trust relationship prediction results .

[0016] Furthermore, joint learning and cross-client feature propagation includes the following steps: Local graph augmentation based on public knowledge graph: client side Map and Public Knowledge Graph To perform fusion, first extract the local map The set of non-user nodes in , and use this as a constraint in Extract subgraphs from , and and To merge

[0017] Using multi-head attention mechanism Process and output the features updated by the user and Characteristics of midpoints , and publish the feature set on the blockchain; Blockchain smart contracts use the pre-loaded GCN model to perform simple aggregation of node features. , whose features from different clients are represented as , and its comprehensive characteristics are , and publish the results on the blockchain.

[0018] In a second aspect, the present invention provides a decentralized trust relationship collaborative prediction system, comprising: The initialization module is used to start the blockchain system and publish the public security parameters and initialization model on the blockchain, and load the consensus algorithm on the consensus node; The graph construction module is used to construct a local complete graph using local data, and pre-train the user feature extraction and relationship prediction model in this map to obtain the features of user nodes; The public graph construction module is used to build a public graph based on public knowledge. The public graph is sent to the blockchain consensus node to confirm the consensus of the public graph. The training output module is used to update the features of user nodes through the public graph as an intermediate node, train the user feature extraction and relationship prediction model, send the trained model to the blockchain, complete the model fusion based on the federated average algorithm until the objective function converges, and output the final trust relationship prediction model.

[0019] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the decentralized trust relationship collaborative prediction method when executing the computer program.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the decentralized trust relationship collaborative prediction method are implemented.

[0021] Compared with the prior art, the present invention has the following technical effects: The present invention provides a decentralized trust relationship collaborative prediction method based on federated learning and graph neural network, which aims to address the limitations of traditional trust assessment methods in multi-source distributed scenarios and improve the accuracy of trust relationship prediction.

[0022] The present invention introduces a federated learning framework to achieve model training and feature extraction without leaving the local data. Each client (such as a social platform) can independently build a local social behavior graph and trust relationship graph, and perform collaborative training through a federated learning mechanism. This decentralized processing method not only protects data privacy, but also enables the model to make full use of multi-source data, improving the accuracy and generalization ability of trust relationship prediction.

[0023] The present invention breaks down the data barriers between different clients by introducing a public knowledge graph as an intermediate layer. The knowledge graph contains rich entity and relationship information, which can provide additional supplements and enhancements for the user's trust features. By propagating and updating features on the knowledge graph through a graph neural network, the user's trust features are richer and more comprehensive, thereby improving the accuracy of trust relationship prediction.

[0024] The present invention combines the advantages of federated learning and graph neural networks, enabling the model to fully utilize multi-source data and external knowledge to extract user trust features and predict trust relationships between users. This prediction method that integrates multi-faceted information is more accurate and reliable than the traditional single method.

[0025] In the fields of social networking platforms, mobile crowdsourcing, blockchain, etc., accurate trust relationship prediction is of great significance for improving user experience, reducing transaction risks, promoting social interaction, etc. The application of the present invention can significantly improve the trust assessment capabilities in these fields and provide strong support for the healthy development of related businesses.

[0026] The method of the present invention can be flexibly adjusted according to different application scenarios and requirements. For example, specific knowledge graphs and model parameters can be constructed for different social platforms or business fields to meet different trust assessment requirements.

[0027] The method of the present invention can be easily extended to more application scenarios and data types. For example, more external data sources can be introduced to enrich the content of the knowledge graph, or more advanced graph neural network models can be used to improve the capabilities of feature extraction and relationship prediction.

[0028] In summary, the present invention has created a decentralized trust relationship collaborative prediction method based on federated learning and graph neural network, which not only supports cross-source trust relationship prediction, but also enriches the user's trust characteristics by introducing external knowledge, thereby improving the precision and accuracy of trust evaluation. At the same time, the method also has good flexibility and scalability, providing strong support for the future development of trust evaluation technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0030] The present invention is further described below in conjunction with the accompanying drawings: Example 1, please refer to Figure 1 The present invention provides a decentralized trust relationship collaborative prediction method, comprising: Start the blockchain system and publish the public security parameters and initialization model on the blockchain, and load the consensus algorithm on the consensus node; Use local data to build a local complete graph, pre-train on this map to obtain user feature extraction and relationship prediction models, and obtain the features of user nodes; A public graph is established based on public knowledge, and the public graph is sent to the blockchain consensus node for consensus confirmation of the public graph; By using the public graph as an intermediate node, the features of the user nodes are updated, the user feature extraction and relationship prediction models are trained, the trained models are sent to the blockchain, and the model fusion is completed based on the federated averaging algorithm until the objective function converges and the final trust relationship prediction model is output.

[0031] By publishing public security parameters and initialization models on the blockchain and leveraging the distributed and tamper-proof characteristics of the blockchain, the security and integrity of the data are ensured.

[0032] User data is preprocessed and feature extracted locally, avoiding the direct uploading of raw data to the central server, thereby effectively protecting user privacy.

[0033] Use local data to build a local complete graph and perform pre-training so that the model can fully learn the characteristics of local data.

[0034] By using the public graph as an intermediate node, the features of the user nodes are updated, which further enriches the learning content of the model and improves the generalization ability of the model.

[0035] This method adopts a decentralized architecture, where each node independently trains and updates its local model, and then integrates the models through blockchain and federated averaging algorithm.

[0036] This decentralized approach avoids single point failures and data monopoly problems caused by centralization, and improves the stability and reliability of the system.

[0037] The introduction of public graphs provides the model with more background knowledge and related information, which helps to more accurately understand the relationships and trust between users.

[0038] The application of the federated averaging algorithm enables the models of each node to draw on and learn from each other, thereby continuously optimizing and improving the accuracy of predictions.

[0039] The establishment of a public graph and consensus confirmation mechanism promotes the sharing and updating of knowledge. Each node can jointly maintain and update the public graph to ensure its timeliness and accuracy.

[0040] This knowledge sharing approach helps improve the intelligence and predictive capabilities of the entire system.

[0041] This method is highly flexible and scalable. Model parameters and algorithms can be adjusted according to actual needs to adapt to different application scenarios and data sets.

[0042] At the same time, with the development of blockchain technology and federated learning, this method can also continuously introduce new technologies and algorithms to further improve its performance and effects.

[0043] In summary, this decentralized trust relationship collaborative prediction method achieves accurate prediction of user trust relationships by combining blockchain technology and federated learning principles, while protecting user privacy and data security. It has broad application prospects and important practical value.

[0044] Embodiment 2, the present invention provides a decentralized trust relationship collaborative prediction method, specifically comprising: System initialization: consensus nodes load consensus algorithms, agree on public security parameters, and initialize model parameters , start the blockchain system and publish the public security parameters and initialization model on the blockchain; Composition module: each client Use local data to build a user social behavior graph and trust relationship diagram , user-centric, building a complete local graph ; Public knowledge graph modeling: For specific application scenarios, public information is graphed to break through the data barriers of different clients. , It is sent to the blockchain consensus node, which verifies the correctness of the knowledge graph content. After verification, the consensus algorithm is called to complete the consensus confirmation; Local learning and feature propagation: each client In this map Pre-training to obtain user feature extraction and relationship prediction model , get the characteristics of the user node ; Joint learning and cross-client feature propagation: Each client passes As an intermediate node, it discovers neighbors across clients and propagates features to update user features and complete the model training, and will The features of the knowledge entities contained in are sent to the blockchain, which uses the graph attention mechanism to fuse the features of the knowledge entities sent by different clients and publish them on the blockchain; Model fusion based on weighted federated average algorithm: The client uses the locally trained model The data is sent to the blockchain, and the blockchain consensus node verifies the validity of the model. When a sufficient number of models are collected or the waiting time exceeds the preset time, the model fusion smart contract is loaded and the model fusion is completed based on the federated average algorithm. The fusion method is as follows:

[0045] The blockchain consensus mechanism is then called to complete the The consensus is reached and published on the blockchain; Model training: Each client and blockchain consensus node cyclically completes the above (4)-(6) until the objective function converges and outputs the final trust relationship prediction model. ; Model application: The client loads the final model Processing local data, given a user and ,use Extract its trust features , And the trust relationship prediction results .

[0046] The system initialization mainly includes the following steps: Select a group of suitable nodes as blockchain consensus nodes, and the consensus nodes negotiate public security parameters such as public keys and encryption systems; Establish consensus mechanisms, such as Practical Byzantine Fault Tolerance Algorithm (PBFT), Raft, Proof of Stake (PoS), Algorand, Dfinity, etc., and launch blockchain; The consensus node randomizes the model parameters and integrates them using smart contracts, and publishes the blockchain or public security parameters on the blockchain; The composition module mainly includes the following steps: For the local user behavior data of each client, other entity nodes other than users are introduced, such as users (such as ) as the center, collect the interactive relationships between users (such as attention, trust relationships, etc.) and between users and other objects (such as goods, articles, etc.), take these interactive objects as nodes, and the interactive relationships as edges, and construct a social relationship graph from this perspective. ; Edge feature extraction: Generate edge features for edge information. For numerical data (such as ratings), standardize and discretize them, and then use one-hot encoding to construct edge features. For complex data (such as text comments, image comments, etc.), use feature extractors such as CLIP, BERT, and Convolutional Neural Network (CNN) for feature extraction. Node feature extraction: For user nodes, their personal public attribute information is used for modeling; for other nodes (such as products, articles, movies, etc.), their attribute information, public descriptions, etc. are comprehensively utilized, and multimodal feature extractors are flexibly used for feature initialization.

[0047] The public knowledge graph modeling mainly includes the following steps: Corpus construction: For non-user nodes (such as products, movies, music, etc.), search for relevant expected data for node type for knowledge graph construction; Knowledge graph construction: Use named entity recognition and relationship extraction to extract knowledge from the corpus, extract the relationship between nodes, and form a knowledge graph , It should cover non-user node entities contained in the user client's local data; The local learning and feature propagation based on the multi-head graph attention network mainly includes the following steps: Feature propagation and fusion based on attention mechanism: each client , for a given node , which is shown in Figure The neighbor set is ,node , and nodes The edge of the record , the feature is recorded as ,but The attention parameter The calculation is performed using the following formula:

[0048] Get attention score Afterwards, given The layer output is , then The output of the layer is:

[0049] Multi-head attention mechanism: Based on the attention mechanism, a multi-head attention mechanism is further designed, which includes Each independent attention extracts features and performs feature fusion. Assume that The features extracted by the attention mechanism are , then The output of the layer is:

[0050] Feature output: When all are completed After extracting the features of the layer, the final features are output ; The joint learning and cross-client feature propagation module mainly includes the following steps: Local graph augmentation based on public knowledge graph: client side Map and Public Knowledge Graph To perform fusion, first extract the local map The set of non-user nodes in , and use this as a constraint in Extract subgraphs from , and and To merge

[0051] Feature propagation and updating: Using multi-head attention mechanism Process and output the features updated by the user and Characteristics of midpoints , and publish the feature set on the blockchain; Blockchain smart contracts use the pre-loaded GCN model to perform simple aggregation of node features. , whose features from different clients are represented as , and its comprehensive characteristics are , and publish the results on the blockchain.

[0052] Focusing on the trust evaluation problem in social network platforms including ubiquitous social networks and mobile crowdsourcing, the platform mainly includes social networking platforms and For ease of representation, this patent uses Indicates Different social platforms provide different online social services, such as instant messaging, information, shopping, etc. They are also responsible for predicting trust relationships based on users' trust relationships and social behaviors to better serve users. The algorithms involved in this patent are all deployed on social platforms. , the platform maintains a local social behavior graph and trust relationship diagram , these two graphs are further fused into a local complete graph with the user as the center ; In addition, we assume that different clients share a common knowledge graph Used to connect different social platform data; the algorithm designed by this patent is based on federated learning. In order to coordinate the collaborative training between different clients (i.e. social platforms), this patent introduces the blockchain system to build a decentralized learning platform, and uses blockchain smart contracts to integrate the models or features learned by different clients; the user set is represented as , user use The social services provided can form trust or distrust relationship information about other users based on personal preferences.

[0053] Taking graph neural network as the basic feature extraction model and combining it with the idea of ​​federated learning, a multi-source collaborative trust relationship evaluation scheme is constructed. By introducing external knowledge as an intermediate layer, the organic integration of node features distributed across clients is achieved, thus realizing accurate trust evaluation in distributed scenarios.

[0054] In yet another embodiment of the present invention, a decentralized trust relationship collaborative prediction system is provided, which can be used to implement the above-mentioned decentralized trust relationship collaborative prediction method. Specifically, the system includes: The initialization module is used to start the blockchain system and publish the public security parameters and initialization model on the blockchain, and load the consensus algorithm on the consensus node; The graph construction module is used to construct a local complete graph using local data, and pre-train the user feature extraction and relationship prediction model in this map to obtain the features of user nodes; The public graph construction module is used to build a public graph based on public knowledge. The public graph is sent to the blockchain consensus node to confirm the consensus of the public graph. The training output module is used to update the features of user nodes through the public graph as an intermediate node, train the user feature extraction and relationship prediction model, send the trained model to the blockchain, complete the model fusion based on the federated average algorithm until the objective function converges, and output the final trust relationship prediction model.

[0055] The division of modules in the embodiments of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention may be integrated into one processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0056] In another embodiment of the present invention, a computer device is provided, the computer device including a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute 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 (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of a decentralized trust relationship collaborative prediction method.

[0057] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in a computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of a decentralized trust relationship collaborative prediction method in the above embodiment.

[0058] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented 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.

[0059] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0060] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A decentralized trust relationship collaborative prediction method, characterized in that: include: Start the blockchain system and publish the public security parameters and initialization model on the blockchain, and load the consensus algorithm on the consensus node; Use local data to build a local complete graph, pre-train on this map to obtain user feature extraction and relationship prediction models, and obtain the features of user nodes; A public graph is established based on public knowledge, and the public graph is sent to the blockchain consensus node for consensus confirmation of the public graph; By using the public graph as an intermediate node, the features of the user nodes are updated, the user feature extraction and relationship prediction models are trained, the trained models are sent to the blockchain, and the model fusion is completed based on the federated averaging algorithm until the objective function converges and the final trust relationship prediction model is output.

2. A decentralized trust relationship collaborative prediction method according to claim 1, characterized in that: The said starting the blockchain system and publishing the public security parameters and the initialization model on the blockchain, and loading the consensus algorithm on the consensus node, include: A group of nodes are selected as blockchain consensus nodes, and the consensus nodes negotiate the public key and public security parameters of the encryption system; Establish a consensus mechanism and start the blockchain; The consensus nodes randomize the model parameters and integrate them using smart contracts, and publish the blockchain or public security parameters on the blockchain.

3. A decentralized trust relationship collaborative prediction method according to claim 1, characterized in that: The method of using local data to construct a local complete graph includes: Each client Use local data to build a user social behavior graph and trust relationship diagram , user-centric, building a complete local graph ; For the local user behavior data of each client, other entity nodes other than users are introduced. With the user as the center, the interaction relationships between users and between users and other objects are collected. These interaction objects are regarded as nodes and the interaction relationships are regarded as edges to construct a social relationship graph from this perspective. ; Generate edge features based on edge information. For numerical data, standardize and discretize them, and then use one-hot encoding to construct edge features. For complex data such as text comments and image comments, use feature extractors to extract features in a targeted manner. For user nodes, their personal public attribute information is used for modeling; for product, article, and movie nodes, their attribute information and public descriptions are comprehensively utilized, and a multimodal feature extractor is used for feature initialization.

4. The decentralized trust relationship collaborative prediction method according to claim 1, characterized in that: The user feature extraction and relationship prediction model is obtained by pre-training in this map, and the features of the user nodes are obtained, including: Each client , for a given node , which is shown in Figure The neighbor set is ,node , and nodes The edge of the record , the feature is recorded as ,but The attention parameter The calculation is performed using the following formula: Get attention score Afterwards, given The layer output is , then The output of the layer is: Based on the attention mechanism, we further design a multi-head attention mechanism, which includes Independent attentions extract features and perform feature fusion respectively; assuming that The features extracted by the attention mechanism are , then The output of the layer is: Feature output: When all are completed After extracting the features of the layer, the final features are output .

5. The decentralized trust relationship collaborative prediction method according to claim 1, characterized in that: The public knowledge graph is established based on public knowledge, and the public graph is sent to the blockchain consensus node to confirm the consensus of the public graph, including: For non-user nodes of products, movies, and music, search for relevant expected data for knowledge graph construction based on node type; Use named entity recognition and relationship extraction to extract knowledge from the corpus, extract the relationship between nodes, and form a public knowledge graph , It should cover non-user node entities contained in the local data of the user client; Graph recording based on public knowledge , It is sent to the blockchain consensus node, which verifies the correctness of the knowledge graph content. After verification, the consensus algorithm is called to complete the consensus confirmed.

6. A decentralized trust relationship collaborative prediction method according to claim 1, characterized in that: The public graph is used as an intermediate node to update the features of the user node, train the user feature extraction and relationship prediction model, send the trained model to the blockchain, complete the model fusion based on the federated average algorithm until the objective function converges, and output the final trust relationship prediction model, including: Each client passes As an intermediate node, it discovers neighbors across clients and propagates features to update user features and complete the model training, and will The features of the knowledge entities contained in are sent to the blockchain, which uses the graph attention mechanism to fuse the features of the knowledge entities sent by different clients and publish them on the blockchain; The client will use the locally trained model The data is sent to the blockchain, and the blockchain consensus node verifies the validity of the model. When a sufficient number of models are collected or the waiting time exceeds the preset time, the model fusion smart contract is loaded and the model fusion is completed based on the federated average algorithm. The fusion method is as follows: The blockchain consensus mechanism is then called to complete the The consensus is reached and published on the blockchain; Each client and blockchain consensus node loops until the objective function converges and outputs the final trust relationship prediction model ; The client loads the final model Processing local data, given a user and ,use Extract its trust features , And the trust relationship prediction results .

7. A decentralized trust relationship collaborative prediction method according to claim 6, characterized in that: Federated learning and cross-client feature propagation includes the following steps: Local graph augmentation based on public knowledge graph: client side Map and Public Knowledge Graph To perform fusion, first extract the local map The set of non-user nodes in , and use this as a constraint in Extract subgraphs from , and and To merge Using multi-head attention mechanism Process and output the features updated by the user and Characteristics of midpoints , and publish the feature set on the blockchain; Blockchain smart contracts use the pre-loaded GCN model to perform simple aggregation of node features. , whose features from different clients are represented as , and its comprehensive characteristics are , and publish the results on the blockchain.

8. A decentralized trust relationship collaborative prediction system, characterized in that: include: The initialization module is used to start the blockchain system and publish the public security parameters and initialization model on the blockchain, and load the consensus algorithm on the consensus node; The graph construction module is used to construct a local complete graph using local data, and pre-train the user feature extraction and relationship prediction model in this map to obtain the features of user nodes; The public graph construction module is used to build a public graph based on public knowledge. The public graph is sent to the blockchain consensus node to confirm the consensus of the public graph. The training output module is used to update the features of user nodes through the public graph as an intermediate node, train the user feature extraction and relationship prediction model, send the trained model to the blockchain, complete the model fusion based on the federated average algorithm until the objective function converges, and output the final trust relationship prediction model.

9. 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, the steps of a decentralized trust relationship collaborative prediction method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of a decentralized trust relationship collaborative prediction method as described in any one of claims 1 to 7 are implemented.

Images