Label determination method

Abstract

The application relates to a label determination method. The method comprises the following steps: determining a participant probability propagation matrix of participant unlabeled data according to a participant weight matrix, and determining a participant matrix transmission factor according to the participant weight matrix; determining an initiator weight matrix according to an initiator topological structure, determining an initiator joint matrix factor and an initiator matrix and vector according to the initiator weight matrix; determining a joint matrix and vector, determining an initiator block probability matrix according to the joint matrix and vector and the initiator weight matrix; determining a participant update probability matrix according to the participant probability propagation matrix and the participant block probability matrix; determining an initiator update probability matrix according to the participant transmission probability matrix and the initiator probability propagation matrix; and determining a target label of the unlabeled data according to the initiator update probability matrix and the participant update probability matrix. In the above scheme, the risk of leaking data privacy is reduced in the process of label information calculation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a label determination method. Background Technology

[0002] In the graph domain, community detection is a very popular and widely discussed topic. In real social networks, some users are closely connected, while others are sparsely connected. Closely connected user groups can be considered a community. Community detection algorithms are a data mining technique primarily used to detect relevant communities in graph or network structures. They determine the labels of each node within a community using a label propagation algorithm. The basic idea of ​​the label propagation algorithm is to use the label that appears most frequently among a node's neighboring nodes as that node's own label. Each node is labeled to represent its community, and this label propagation forms a community structure with the same label. After initializing each node with a unique label, the algorithm repeatedly community-izes a node's label with the most frequent label among its neighboring nodes. When each node's label appears most frequently among its neighboring nodes, the label propagation algorithm stops calculating and determines the label information of each node in the community.

[0003] Currently, determining the label information of nodes in different communities often requires an additional server as an intermediary to coordinate and calculate the results. If plaintext is used to calculate node label information, there is a risk of data privacy leakage, and the label calculation efficiency is low while the cost is high. Therefore, how to calculate label information for unlabeled nodes in two communities where identical nodes exist, while minimizing the risk of data privacy leakage, improving the efficiency of label determination, and reducing the computational cost, is a problem that needs to be solved. Summary of the Invention

[0004] Therefore, it is necessary to provide a label determination method that can calculate label information for unlabeled nodes in two communities when the same nodes exist in both communities, and reduce the risk of data privacy leakage, improve the efficiency of label determination, and reduce the computation cost of label information in the process of label information calculation.

[0005] Firstly, this application provides a tag determination method, wherein the tag determination system includes a communication initiator and communication participants, and the tag determination method includes:

[0006] The business participants determine the participant weight matrix based on the participant topology, and determine the participant probability propagation matrix for the unlabeled data based on the participant weight matrix. They also determine the participant matrix transmission factor based on the participant weight matrix and send the participant matrix transmission factor to the business initiator.

[0007] The business initiator obtains the transmission factor of the participant matrix, determines the initiator weight matrix based on the initiator topology, calculates the propagation factor of the initiator's unlabeled data based on the initiator weight matrix, determines the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix.

[0008] The business participants determine the participant matrix and vector based on the participant weight matrix, and send the participant matrix and vector to the business initiator;

[0009] The business initiator obtains the participant matrix and vector, determines the joint matrix and vector based on the initiator matrix and vector and the participant matrix and vector, sends the joint matrix and vector to the business participants, and determines the initiator block probability matrix based on the joint matrix and vector and the initiator weight matrix.

[0010] The business participants obtain the joint matrix and vector, determine the participant block probability matrix based on the joint matrix and vector and the participant weight matrix, and send the participant transmission probability matrix in the participant block probability matrix to the business initiator. They also determine the participant update probability matrix based on the participant probability propagation matrix and the participant block probability matrix.

[0011] The business initiator obtains the transmission probability matrix of the participants and determines the initiator's update probability matrix based on the transmission probability matrix of the participants and the probability propagation matrix of the initiator.

[0012] The business initiator determines the target label for the initiator's unlabeled data based on the initiator's updated probability matrix and label data matrix, and the business participant determines the target label for the participant's unlabeled data based on the participant's updated probability matrix and label data matrix.

[0013] In one embodiment, the service initiator determines the target label for which the initiator has no label data based on the initiator's updated probability matrix and the label data matrix, including:

[0014] The business initiator determines the initiator's first transmission parameter based on the initiator's updated probability matrix, sends the initiator's first transmission parameter to the business participant, and obtains the participant's first transmission parameter sent by the business participant.

[0015] The business initiator determines the initiator's second transmission parameters based on the participant's first transmission parameters, sends the initiator's second transmission parameters to the business participant, and obtains the participant's second transmission parameters sent by the business participant.

[0016] The business initiator determines the initiator label parameters and the initiator third transmission parameters based on the participant's second transmission parameters and the initiator update probability matrix, and sends the initiator third transmission parameters to the business participants;

[0017] The business initiator determines the initiator target parameters based on the initiator tag parameters and the tag data matrix, sends the initiator transmission tag in the initiator target parameters to the business participants, and obtains the participant transmission tag sent by the business participants.

[0018] The business initiator determines the target label for the initiator's unlabeled data based on the initiator's target parameters and the tags transmitted by the participants.

[0019] In one embodiment, the service initiator determines the target label for the initiator's unlabeled data based on the initiator's target parameters and the participant's transmitted labels, including:

[0020] The business initiator determines candidate tags for which the initiator has no data based on the initiator's target parameters and the data of the tags transmitted by the participants, and determines the probability of the initiator's candidate tags for candidate tags for which the initiator has no data.

[0021] The business initiator determines the target label for the initiator's unlabeled data from the candidate labels of the initiator's unlabeled data based on the probability of the initiator's candidate labels.

[0022] In one embodiment, the business participant determines the target label for the participant's unlabeled data based on the participant's updated probability matrix and the label data matrix, including:

[0023] The business participant obtains the first transmission parameters from the initiator, determines the first transmission parameters of the participant based on the first transmission parameters of the initiator, and sends the transmission parameters of the participant to the business initiator;

[0024] The business participant obtains the second transmission parameter from the initiator, determines the second transmission parameter of the participant based on the second transmission parameter of the initiator and the participant update probability matrix, and sends the second transmission parameter of the participant to the business initiator.

[0025] The business participants obtain the third-party transmission parameters from the initiator, and determine the participant's label parameters based on the third-party transmission parameters from the initiator and the participant's updated probability matrix.

[0026] The business participants determine the participant target parameters based on the participant tag parameters and the tag data matrix, determine the participant transmission tag from the participant target parameters, and send the participant transmission tag to the business initiator.

[0027] The business participants obtain the initiator's transmission tag and determine the target tag for the participant's untagged data based on the participant's target parameters and the initiator's transmission tag.

[0028] In one embodiment, a business participant obtains the initiator's transmission tag and determines the target tag for the participant's untagged data based on the participant's target parameters and the initiator's transmission tag, including:

[0029] Based on the participant's target parameters and the data of the participant's transmitted tags, the business participants determine the candidate tags for the participants without tag data, and determine the participant candidate tag probability of the candidate tags for the participants without tag data.

[0030] Business participants determine the target label for the unlabeled data of the participant from the candidate labels of the unlabeled data of the participant based on the probability of the participant's candidate labels.

[0031] In one embodiment, a business participant obtains the initiator's second transmission parameters, determines its own second transmission parameters based on the initiator's second transmission parameters and the participant's updated probability matrix, and sends the participant's second transmission parameters to the business initiator, including:

[0032] The business participants obtain the second transmission parameters from the initiator and determine the initiator parameter calculation factor based on the second transmission parameters from the initiator and the participant update probability matrix.

[0033] The second transmission parameters of the participants are determined based on the initiator's parameter calculation factor and the participant's updated probability matrix, and then sent to the business initiator.

[0034] In one embodiment, the business participants determine the participant weight matrix based on the participant topology, including:

[0035] Business participants determine the participant topology based on their local data relationship matrix.

[0036] Business participants determine the length of the topological edge of the participant based on the participant topology structure, and determine the weight of the topological edge of the participant based on the weight coefficient of the topological edge of the participant and the weight of the topological edge of the participant, and determine the weight matrix of the participant based on the weight of the topological edge of the participant.

[0037] In one embodiment, determining the initiator weight matrix based on the initiator topology includes:

[0038] The business initiator determines the initiator's topology based on the initiator's local data relationship matrix;

[0039] The business initiator determines the length of the initiator's topology edge based on the initiator's topology structure, and determines the weight of the initiator's topology edge based on the coefficients calculated by the initiator's topology edge length and weight, and determines the initiator's weight matrix based on the initiator's topology edge weight.

[0040] In one embodiment, before the business initiator determines the target label for which the initiator has no label data based on the initiator's updated probability matrix and the label data matrix, and before the business participant determines the target label for which the participant has no label data based on the participant's updated probability matrix and the label data matrix, the method further includes:

[0041] The business initiator performs a privacy-preserving union on the initiator's dataset and the participant's dataset to determine the tag data sets for the business initiator and the business participants;

[0042] The label data matrix is ​​determined based on the amount of data in the initiator's dataset, the amount of data in the participant's dataset, and the label data set.

[0043] In one embodiment, the business initiator obtains the participant matrix transmission factor, determines the initiator weight matrix based on the initiator's topology, calculates the propagation factor for the initiator's unlabeled data based on the initiator weight matrix to determine the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix, including:

[0044] The business initiator obtains the transmission factor of the participant matrix, determines the initiator weight matrix based on the initiator's topology, and calculates the propagation factor for the initiator's unlabeled data based on the initiator's weight matrix to determine the initiator's probability propagation matrix.

[0045] The business initiator determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator summation matrix from the initiator weight matrix;

[0046] The business initiator determines the initiator matrix and vector by summing the initiator's summation matrix row by row.

[0047] The above label determination method involves the following steps: The business initiator determines the communication data to be sent to the business initiator based on the initiator dataset and the communication data sent from the business participants. The business participants determine their communication data based on their participant datasets and the communication data sent from the business initiator. The business initiator then determines its own block probability matrix based on the initiator dataset and the participant communication data, updates the initiator block probability matrix, and determines the updated initiator probability matrix. Finally, the target label for the initiator's unlabeled data is determined based on the updated initiator probability matrix and the label data matrix. Similarly, the business participants determine their own block probability matrix based on their participant datasets and the initiator communication data, updates the participant block probability matrix, and determine the updated participant probability matrix. The target label for the participant's unlabeled data is then determined based on the updated participant probability matrix and the label data matrix. This solution addresses the issues of low efficiency and high cost in calculating tag information for unlabeled data from both the initiator and participant. Previously, calculating tag information for unlabeled data required an additional intermediate server to coordinate and calculate the information, posing a risk of data privacy leakage. Furthermore, it necessitated multiple calls to homomorphic encryption methods to calculate the tag information for unlabeled data, and offline negotiation between the initiator and participant regarding data relationships. The proposed solution, however, enables tag information calculation for unlabeled nodes in both communities when there are shared nodes between the initiator and participant. This solution reduces the risk of data privacy leakage during the tag information calculation process, improves the efficiency of tag information determination, and lowers the computational cost. Attached Figure Description

[0048] Figure 1 This is a diagram illustrating the application environment of the label determination method in one embodiment;

[0049] Figure 2 This is a flowchart illustrating a label determination method in one embodiment;

[0050] Figure 3 This is an example diagram of the initiator topology of the business initiator in one embodiment;

[0051] Figure 4 This is an example diagram of the participant topology of the business participants in one embodiment;

[0052] Figure 5 This is an example diagram showing the node connections between the business initiator and the business participants in one embodiment;

[0053] Figure 6 This is a flowchart illustrating the label determination method in another embodiment;

[0054] Figure 7 This is a flowchart illustrating the label determination method in another embodiment;

[0055] Figure 8 This is a flowchart illustrating the label determination method in another embodiment. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] The label determination method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with tag identification system 104 via a network. A data storage system can store the data that tag identification system 104 needs to process. The data storage system can be integrated into tag identification system 104, or it can be located in the cloud or on another network server. In the label determination system 104, the business participant determines the participant weight matrix based on the participant topology, and determines the participant probability propagation matrix for the unlabeled data based on the participant weight matrix. It also determines the participant matrix transmission factor based on the participant weight matrix and sends the participant matrix transmission factor to the business initiator. The business initiator obtains the participant matrix transmission factor, determines the initiator weight matrix based on the initiator topology, calculates the propagation factor for the initiator's unlabeled data based on the initiator weight matrix, determines the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix. The business participant determines the participant matrix and vector based on the participant weight matrix and sends the participant matrix and vector to the business initiator. The business initiator obtains the participant matrix and vector, determines the joint matrix and vector based on the initiator matrix and vector and the participant matrix and vector, and sends the joint matrix and vector to the business initiator. The matrix and vectors are sent to the business participants, and the initiator block probability matrix is ​​determined based on the joint matrix and vectors and the initiator weight matrix. The business participants obtain the joint matrix and vectors, determine the participant block probability matrix based on the joint matrix and vectors and the participant weight matrix, and send the participant transmission probability matrix from the participant block probability matrix to the business initiator. They also determine the participant update probability matrix based on the participant probability propagation matrix and the participant block probability matrix. The business initiator obtains the participant transmission probability matrix and determines the initiator update probability matrix based on the participant transmission probability matrix and the initiator probability propagation matrix. The business initiator determines the target label for the initiator's unlabeled data based on the initiator update probability matrix and the label data matrix. The business participants determine the target label for the participant's unlabeled data based on the participant update probability matrix and the label data matrix, and send the target labels for the initiator's unlabeled data and the participant's unlabeled data to terminal 102 via the communication network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The tag identification system 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0058] In one embodiment, such as Figure 2As shown, a label determination method is provided. The label determination method is executed by a label determination system, which includes a business initiator and business participants.

[0059] It should be noted that the initiator dataset of the business initiator and the participant dataset of the business participants in the label determination system have the exact same feature space. The business initiator maintains the nodes in the initiator community, and the business participants maintain the nodes in the participant community. The initiator dataset is D. A ,and Where num A Indicates the amount of data from the business initiator, l A The initiator representing the business initiator has a defined amount of data, u A The amount of unlabeled data representing the initiator of the business, C A This represents the range of initiator tags for the business initiator. The participant dataset is D. B ,and Where num B Indicates the amount of data from business participants, l B The amount of tagged data representing the participants in the business is u B The amount of unlabeled data representing the business stakeholders, C B This indicates the range of participant labels for business participants. There is at least one intersecting data node between the business initiator and the business participants. The relationship between the data nodes of the data initiator and the data participants cannot be known through offline negotiation and organization. Therefore, all node relationships are stored in the original data topology graph. The business initiator and the business participants do not expose the specific information of the data nodes to each other. The specific information includes feature values ​​and label values.

[0060] In this embodiment, the method includes the following steps:

[0061] S210. The business participant determines the participant weight matrix based on the participant topology, determines the participant probability propagation matrix for the unlabeled data based on the participant weight matrix, and determines the participant matrix transmission factor based on the participant weight matrix, and sends the participant matrix transmission factor to the business initiator.

[0062] It should be noted that business participants can execute the label determination method through their participant servers, and business initiators can execute the label determination method through their initiator servers. When a business initiator needs to determine the label information of unlabeled data in its initiator dataset, it can collaborate with business participants to perform a privacy-preserving intersection calculation on the data of both the business initiator and the business participants to determine the joint data intersection D of the business initiator and the business participants. P D P =DAB In the full dataset of the business initiator, excluding the intersection D of the joint dataset, P The data unique to the initiator other than D is D. AP In the full dataset of the business participants, excluding the dataset intersection D P Data unique to participants other than D BP Among them, D A =D AP ∪D AB and D B =D BP ∪D AB and The business initiator and business participants jointly access the joint data intersection D. P The system performs privacy-preserving intersection calculations on the labeled data to determine the intersection of the labeled data and the label data numbers within the intersection. The business initiator generates an initiator's labeled data list based on the labeled data intersection and the initiator's labeled data; similarly, business participants generate their own labeled data lists based on the labeled data intersection and their own labeled data. The business initiator and participants then jointly perform data rearrangement, sorting the full initiator data in the initiator's dataset and the full participant data in the participant's dataset by their respective numbers to generate a sorted data list. Furthermore, the data ranking list is shared between the business initiator and the business participants, n A n represents the total amount of data from the initiator in the initiator dataset. B This represents the complete data of all participants in the participant dataset. When sorting by sequence number, the order of the initiator's complete data and the participants' complete data is as follows: business initiator-specific tag data. D P Unique tag data of the business initiator D P Both the business initiator and the business participants share tag data. D P Unique tag data of business participants Unique tag data of business participants Business initiator unlabeled data D P Initiator unlabeled data and D P Unlabeled data of participants and unlabeled data from participating parties in, Right now: In this context, A represents the business initiator, and B represents the business participant.

[0063] The data in the data sorting list is arranged in the following order: business initiator, and unique initiator tag data from the full dataset of initiators. Dataset intersection D P Unique initiator tag data for business initiators Dataset intersection D P Tag data shared by business initiators and participants Dataset intersection D P Unique participant tag data for business participants Unique tag data from the full dataset of business participants Unlabeled data of the initiator in the full dataset Dataset intersection D P Unlabeled data of business initiators and business participants Unlabeled participant data in the full participant data set The business initiator obtains the initiator's local data relationship matrix M A Business participants obtain the participant's local data relationship matrix M B .

[0064] For example, the initiator topology of the business initiator is as follows: Figure 3 As shown, the participant topology of the business participants is as follows: Figure 4 As shown in the diagram, the node connection relationship between the business initiator and the business participants is as follows: Figure 5 As shown, the business initiator and business participants only possess their own data relationships and tag data, and are unaware of each other's data types and data relationships. Compared to a centralized relationship graph, nodes 2 and 3 of the business initiator, and nodes 5, 9, and 10 of the business participants are overlapping points and interconnected. Node 2 is the only tag data point for the business initiator, node 5 is a data point solely owned by the business participant, and node 3 is a data tag owned by both the business initiator and the business participant. Through privacy-preserving computation, a global community network for both the business initiator and the business participants can be constructed without exposing the initiator's data, the participant's data, the internal data relationships within the business initiator, or the internal data relationships within the business participants.

[0065] Business participants, based on the participant's local data relationship matrix M B Determine the participant topology, and then determine the participant weight matrix based on the participant topology. For example, the participant weight matrix may include: and in, l PC For the joint data intersection DP The business participants and business initiators in the process share tag data, l PB For the joint data intersection D P The participants in the data have tagged data, u C For the joint data intersection D P Unlabeled data in the data. Based on the participant weight matrix, the unlabeled data of the participants is analyzed. The propagation factor is calculated for the corresponding data point p to determine the participant probability propagation matrix T for unlabeled data. B ,and Determine the transfer factor of the participant matrix from the participant weight matrix. and Send the participant matrix transmission factor to the business initiator.

[0066] S220. The business initiator obtains the transmission factor of the participant matrix, determines the initiator weight matrix based on the initiator topology, calculates the propagation factor of the initiator's unlabeled data based on the initiator weight matrix, determines the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix.

[0067] Specifically, the business initiator obtains the participant matrix transmission factor sent by the participants. and The initiator weight matrix is ​​determined based on the initiator's topology, and the initiator weight matrix may include: and in l PA For the joint data intersection D P The initiator has labeled data. The initiator's unlabeled data is analyzed based on the initiator's weight matrix. Perform propagation factor calculation to generate the initiator probability propagation matrix T. A ,and in,

[0068] The business initiator transfers factors based on the participant matrix. Initiator matrix factors in the initiator weight matrix Generate initiator joint matrix factor and And determine the initiator matrix and vector S based on the initiator weight matrix. A .in If all positions are 0, the value is 0; if all are non-zero, or only one value is 0 and the other is non-zero, then the non-zero value remains unchanged.

[0069] S230. The business participants determine the participant matrix and vector based on the participant weight matrix, and send the participant matrix and vector to the business initiator.

[0070] Specifically, the business participants determine the participant summation matrix from the participant weight matrix. Then, the participant matrix and vector S are determined by summing the participant summation matrix row by row. B .in, And the participating matrix and vector S B Send to the business initiator.

[0071] S240. The business initiator obtains the participant matrix and vector, determines the joint matrix and vector based on the initiator matrix and vector and the participant matrix and vector, sends the joint matrix and vector to the business participants, and determines the initiator block probability matrix based on the joint matrix and vector and the initiator weight matrix.

[0072] Specifically, the business initiator obtains the participant matrix and vector, and then uses the initiator matrix and vector S... A And the participating matrix and vector S B Determine the joint matrix and vector S, i.e., S = S A +S B The joint matrix and vectors are then sent to the business participants. The business initiator determines its block probability matrix based on the joint matrix and vectors S and the initiator's weight matrix. The matrix elements in the block probability matrix are the ratios of the matrix elements in the joint matrix and the joint matrix and vector S, based on the business initiator.

[0073] S250. The business participant obtains the joint matrix and vector, determines the participant block probability matrix based on the joint matrix and vector and the participant weight matrix, and sends the participant transmission probability matrix in the participant block probability matrix to the business initiator. The participant update probability matrix is ​​determined based on the participant probability propagation matrix and the participant block probability matrix.

[0074] Specifically, business participants determine their target matrix from the participant weight matrix. and And based on the joint matrix and vectors and the target matrix of the participants and The participant block probability moments are determined as follows: and Furthermore, the matrix elements in the participant block probability matrix are the ratios of the matrix elements in the participant's target matrix and the joint matrix and vector. The business participant will use the participant transmission probability matrix from the participant block probability matrix. Send to the business initiator, and propagate the participant probability matrix T according to the participant block probability matrix. B Update the probability matrix to determine the participants' update probability matrix. and in,

[0075] S260. The business initiator obtains the participant transmission probability matrix and determines the initiator update probability matrix based on the participant transmission probability matrix and the initiator probability propagation matrix.

[0076] Specifically, the business initiator obtains the transmission probability matrix of the participating parties. Based on the transmission probability matrix of the participants Update the initiator's probability propagation matrix TA to determine the initiator's updated probability matrix. and in,

[0077] S270. The business initiator determines the target label of the initiator's unlabeled data based on the initiator's updated probability matrix and label data matrix, and the business participant determines the target label of the participant's unlabeled data based on the participant's updated probability matrix and label data matrix.

[0078] The tag data matrix refers to a matrix determined based on the tag data of the business initiator and business participants.

[0079] In the above label determination method, the business initiator determines the communication data sent to the business initiator based on the initiator dataset and the communication data sent from the business participants to the business initiator. The business participants determine their communication data based on their participant dataset and the communication data sent from the business initiator to the business participants. The business initiator determines the initiator block probability matrix based on the initiator dataset and the participant communication data, updates the initiator block probability matrix, determines the updated initiator probability matrix, and determines the target label for the initiator's unlabeled data based on the updated initiator probability matrix and the label data matrix. Similarly, the business participants determine the participant block probability matrix based on their participant dataset and the initiator communication data, updates the participant block probability matrix, determines the updated participant probability matrix, and determines the target label for the participant's unlabeled data based on the updated participant probability matrix and the label data matrix. This solution addresses the issues of low efficiency and high cost in calculating tag information for unlabeled data from both the initiator and participant. Previously, calculating tag information for unlabeled data required an additional intermediate server to coordinate and calculate the information, posing a risk of data privacy leakage. Furthermore, it necessitated multiple calls to homomorphic encryption methods to calculate the tag information for unlabeled data, and offline negotiation between the initiator and participant regarding data relationships. The proposed solution, however, enables tag information calculation for unlabeled nodes in both communities when there are shared nodes between the initiator and participant. This solution reduces the risk of data privacy leakage during the tag information calculation process, improves the efficiency of tag information determination, and lowers the computational cost.

[0080] In one embodiment, such as Figure 6 As shown, the business initiator determines the target label for which the initiator has no label data based on the initiator's updated probability matrix and label data matrix, including:

[0081] S310. The service initiator determines the initiator's first transmission parameter based on the initiator's updated probability matrix, sends the initiator's first transmission parameter to the service participant, and obtains the participant's first transmission parameter sent by the service participant.

[0082] The business initiator determines a random parameter γ, and updates the probability matrix based on the random parameter γ and the initiator's own information. Determine the initiator's first transmission parameter R0, and send the initiator's first transmission parameter R0 to the service participant, and obtain the participant's first transmission parameter E sent by the service participant. B .in,

[0083] S320. The service initiator determines the initiator's second transmission parameters based on the participant's first transmission parameters, sends the initiator's second transmission parameters to the service participant, and obtains the participant's second transmission parameters sent by the service participant.

[0084] The service initiator determines its second transmission parameter R1 based on the participant's transmission parameters and the random parameter γ, where R1 = E. B / γ, and send the initiator's second transmission parameter R1 to the service participant, and obtain the participant's second transmission parameters R2, R3, R4, and R5 sent by the service participant.

[0085] S330. The business initiator determines the initiator label parameters and the initiator third transmission parameters based on the participant's second transmission parameters and the initiator update probability matrix, and sends the initiator third transmission parameters to the business participant.

[0086] Specifically, the service initiator updates the probability matrix based on the second transmission parameters R2, R3, R4, and R5 from the participants. Determine the initiator's tag parameter R AA and R AC The initiator's third transmission parameters R6, R7, and R8 are sent to the service participants. AA R AC The calculation formulas for R6, R7 and R8 are shown in formulas (1), (2), (3), (4) and (5).

[0087]

[0088]

[0089]

[0090]

[0091]

[0092] S340. The business initiator determines the initiator target parameter based on the initiator tag parameter and the tag data matrix, sends the initiator transmission tag in the initiator target parameter to the business participant, and obtains the participant transmission tag sent by the business participant.

[0093] Specifically, the business initiator determines the scope of the business based on the initiator tag parameter R. AA and R AC and the label data matrix Y AB Determine the initiator's target parameter Y AA and Y AC Determine the initiator's transmission tag Y from the initiator's target parameters. AC The initiator will transmit tag Y AC Send to the business participant and obtain the participant transmission tag Y sent by the business participant.BB .

[0094] S350. The business initiator determines the target label for the initiator's unlabeled data based on the initiator's target parameters and the participant's transmitted labels.

[0095] Specifically, the business initiator determines candidate tags for the initiator's unlabeled data based on the sum of the initiator's target parameters and the tags transmitted by the participants. Based on the confidence level of the candidate tags for the initiator's unlabeled data, the target tag for the initiator's unlabeled data is determined from the candidate tags for the initiator's unlabeled data.

[0096] The above scheme determines the initiator's first transmission parameters based on the initiator's update probability matrix, the initiator's second transmission parameters based on the participant's first transmission parameters, the initiator's tag parameters and third transmission parameters based on the obtained participant's second transmission parameters and the initiator's update probability matrix, the initiator's target parameters based on the initiator's tag parameters and tag data matrix, and the target tag for the initiator's untagged data based on the initiator's target parameters and participant's transmitted tags. While protecting the security of the data sets of both the business initiator and the business participants, the scheme determines the target tag for the initiator's untagged data based on data exchange between the business initiator and the business participants. This improves data security when calculating tag information for untagged data, reduces the computational cost of tag information, and avoids resource waste.

[0097] In one embodiment, the service initiator determines the target label for the initiator's unlabeled data based on the initiator's target parameters and the participant's transmitted labels, including:

[0098] The business initiator determines candidate tags for the initiator's unlabeled data based on the initiator's target parameters and the data transmitted by the participants, and determines the initiator's candidate tag probability for the candidate tags for the initiator's unlabeled data; the business initiator determines the target tag for the initiator's unlabeled data from the candidate tags for the initiator's unlabeled data based on the initiator's candidate tag probability.

[0099] The label probability refers to the ratio of the number of times a candidate label appears to the total number of times all candidate labels appear.

[0100] Specifically, the business initiator determines the target parameter Y based on the initiator's criteria. AA Transmit tag Y with participating parties BB Data and candidate labels Y for identifying unlabeled data from the initiator uA Y uA =Y AA +Y BBThe probability of the candidate tags for the unlabeled data of the initiator is determined. Based on this probability, the business initiator determines the target tag for the unlabeled data from the candidate tags of the unlabeled data. Where Y... AB =[Y 1A Y 1B Y uA Y uB ] T Y AA =R AA ×Y 1A Y AC =R AC ×Y 1A .

[0101] Determining the target label of the unlabeled data from the candidate labels of the unlabeled data can improve the computational efficiency and reliability of the target label of the unlabeled data.

[0102] In one embodiment, such as Figure 7 As shown, the business participants determine the target labels for participants without labeled data based on the participant update probability matrix and the label data matrix, including:

[0103] S410. The business participant obtains the first transmission parameters from the initiator, determines the first transmission parameters of the participant based on the first transmission parameters of the initiator, and sends the transmission parameters of the participant to the business initiator.

[0104] Specifically, the business participant obtains the initiator's first transmission parameter R0, and determines the participant's first transmission parameter E based on the initiator's first transmission parameter R0. B , Transmit parameter E of the participants B Send to the business initiator.

[0105] S420. The business participant obtains the second transmission parameters from the initiator, determines the second transmission parameters of the participant based on the second transmission parameters of the initiator and the participant update probability matrix, and sends the second transmission parameters of the participant to the business initiator.

[0106] Specifically, the business participant obtains the initiator's second transmission parameter R1, determines the participant's second transmission parameters R2, R3, R4, and R5 based on the initiator's second transmission parameter R1 and the participant's update probability matrix, and sends the participant's second transmission parameters R2, R3, R4, and R5 to the business initiator.

[0107] S430. The business participant obtains the third transmission parameters from the initiator and determines the participant's label parameters based on the third transmission parameters from the initiator and the participant's update probability matrix.

[0108] Specifically, the business participants obtain the initiator's third transmission parameters R6, R7, and R8, and determine the participant's label parameter R based on the initiator's third transmission parameters R6, R7, and R8, and the participant's update probability matrix. BB and R BD Among them, R BB and R BD The calculation formulas are shown in formulas (6) and (7):

[0109]

[0110]

[0111] S440. The business participant determines the participant target parameters based on the participant tag parameters and the tag data matrix, determines the participant transmission tag from the participant target parameters, and sends the participant transmission tag to the business initiator.

[0112] Specifically, business participants determine the participation based on the participant label parameter R. BB and R BD and the label data matrix Y AB Determine the target parameter Y of the participants BB and Y BD Determine the participant's transmission tag Y from the participant's target parameters. BB The participants will transmit the tag Y. BB Send to the service initiator. Among them, Y AB =[Y lA Y lB Y uA Y uB ] T Y BB =R BB ×Y lB Y BD =R BD ×Y lB .

[0113] S450. The business participant obtains the initiator's transmission tag and determines the target tag for the participant's untagged data based on the participant's target parameters and the initiator's transmission tag.

[0114] Specifically, the business participants obtain the initiator's transmitted tag Y. AC According to the target parameter Y of the participants BD Transmit tag Y to the initiator AC The sum determines the target label for the unlabeled data of the participating parties.

[0115] The above solution, while protecting the security of the data sets of both the business initiator and the business participants, determines the target labels for the unlabeled data of the participants based on the data exchange between the business initiator and the business participants. This improves data security when calculating the label information of the unlabeled data of the participants, reduces the calculation cost of the label information, and avoids the problem of wasting resources.

[0116] In one embodiment, a business participant obtains the initiator's transmission tag and determines the target tag for the participant's untagged data based on the participant's target parameters and the initiator's transmission tag, including:

[0117] The business participant determines the candidate labels for the unlabeled data of the participant based on the participant's target parameters and the data of the participant's transmitted labels, and determines the participant's candidate label probability for the candidate labels of the unlabeled data of the participant; the business participant determines the target label for the unlabeled data of the participant from the candidate labels of the unlabeled data of the participant based on the participant's candidate label probability.

[0118] Specifically, the business participants determine the target parameter Y based on the participant's objective. BD Transmit tag Y to the initiator AC Data and candidate labels Y for identifying unlabeled data from participating parties uB Y uB =Y AC +Y BD The probability of the initiator's candidate label for the candidate labels of the unlabeled data of the participating party is determined. Based on the probability of the participant's candidate label, the business participant determines the target label of the unlabeled data of the participating party from the candidate labels of the unlabeled data of the participating party.

[0119] The above scheme determines the target label of the unlabeled data from the candidate labels of the unlabeled data based on the label probability of the candidate labels of the unlabeled data, which can improve the computational efficiency and reliability of the target label of the unlabeled data.

[0120] In one embodiment, a business participant obtains the initiator's second transmission parameters, determines its own second transmission parameters based on the initiator's second transmission parameters and the participant's updated probability matrix, and sends the participant's second transmission parameters to the business initiator, including:

[0121] The business participant obtains the initiator's second transmission parameters, determines the initiator's parameter calculation factor based on the initiator's second transmission parameters and the participant's update probability matrix, determines the participant's second transmission parameters based on the initiator's parameter calculation factor and the participant's update probability matrix, and sends the participant's second transmission parameters to the business initiator.

[0122] Specifically, the business participants obtain the initiator's second transmission parameter R1, and determine the initiator parameter calculation factor E based on the initiator's second transmission parameter R1 and the participant's updated probability matrix. The second transmission parameters R2, R3, R4, and R5 of the participants are determined based on the initiator parameter calculation factor E and the participant update probability matrix. Among them,

[0123] The above scheme determines the initiator parameter calculation factor based on the initiator's second transmission parameter and the participant's update probability matrix, and determines the participant's second transmission parameter based on the initiator parameter calculation factor and the participant's update probability matrix. This allows the participant's second transmission parameter to contain both initiator-related data and participant-related data, enabling the business initiator to obtain a highly accurate target label for the participant's unlabeled data when determining the target label for the participant's unlabeled data based on the participant's second transmission parameter.

[0124] In one embodiment, the business participants determine the participant weight matrix based on the participant topology, including:

[0125] Business participants determine the participant topology based on the participant's local data relationship matrix; business participants determine the participant topology edge length based on the participant topology structure, and determine the participant topology edge weight based on the participant topology edge length and weight coefficient, and determine the participant weight matrix based on the participant topology edge weight.

[0126] The above scheme determines the length of the topological edge of the participant based on the topological structure of the participant, and determines the weight of the topological edge of the participant based on the length of the topological edge of the participant and the weight calculation coefficient. The participant weight matrix is ​​then determined based on the weight of the topological edge of the participant, which can improve the calculation efficiency and accuracy of the weight of the topological edge of the participant, thereby further improving the reliability of the participant weight matrix.

[0127] In one embodiment, determining the initiator weight matrix based on the initiator topology includes:

[0128] The business initiator determines the initiator topology based on the initiator's local data relationship matrix; the business initiator determines the length of the initiator topology edge based on the initiator topology structure, and determines the weight of the initiator topology edge based on the initiator topology edge length and weight calculation coefficient, and determines the initiator weight matrix based on the initiator topology edge weight.

[0129] The above scheme determines the length of the initiator's topological edge based on the initiator's topological structure, determines the weight of the initiator's topological edge based on the length of the initiator's topological edge and the weight calculation coefficient, and determines the initiator's weight matrix based on the weight of the initiator's topological edge. This can improve the calculation efficiency and accuracy of the initiator's topological edge weight, thereby further improving the reliability of the initiator's weight matrix.

[0130] In one embodiment, such as Figure 8 As shown, the business initiator updates the probability matrix and label data matrix Y based on the initiator's update. AB The initiator determines the target label for which there is no labeled data. Business participants then update the probability matrix and label data matrix Y based on this information. AB Before determining the target label for unlabeled data from participating parties, the following steps are also included:

[0131] S510. The business initiator performs a privacy-preserving union on the initiator's dataset and the participant's dataset to determine the tag data sets of the business initiator and the business participants.

[0132] The tag dataset can be used in C. AB express.

[0133] S520. Determine the label data matrix based on the data volume of the initiator's dataset, the data volume of the participant's dataset, and the label data set.

[0134] Specifically, based on the amount of data n in the initiator's dataset A The amount of data in the participant's dataset n B and tag data set C AB Determine the label data matrix Y AB Y AB ∈(n A +n B )×C AB And Y AB =[Y lA Y lB Y uA Y uB ] T .

[0135] The above scheme determines the label data matrix based on the amount of data in the initiator's dataset, the amount of data in the participant's dataset, and the label data set, which can improve the computational efficiency and accuracy of the label data matrix.

[0136] In one embodiment, the business initiator obtains the participant matrix transmission factor, determines the initiator weight matrix based on the initiator's topology, calculates the propagation factor for the initiator's unlabeled data based on the initiator weight matrix to determine the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix, including:

[0137] The business initiator obtains the participant matrix transmission factor, determines the initiator weight matrix based on the initiator topology, calculates the propagation factor for the initiator's unlabeled data based on the initiator weight matrix, and determines the initiator probability propagation matrix; the business initiator determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator summation matrix from the initiator weight matrix; the business initiator sums the initiator summation matrix row by row to determine the initiator matrix and vector.

[0138] Specifically, the business initiator obtains the transmission factor of the participant matrix. and Based on the transfer factor of the participant matrix Initiator matrix factors in the initiator weight matrix Generate initiator joint matrix factor and The business initiator obtains the initiator weight matrix based on the initiator's topology. Based on this weight matrix, propagation factors are calculated for the initiator's unlabeled data to determine the initiator probability propagation matrix. From the initiator weight matrix, the initiator summation matrix is ​​determined. The initiator summation matrix includes... The initiator matrix and vector S are determined by summing the rows of the initiator matrix. A ,Right now

[0139] The above scheme provides a method for calculating the initiator matrix and vector. It determines the initiator joint matrix factor based on the participant matrix transfer factor and the initiator weight matrix, and determines the initiator summation matrix from the initiator weight matrix. The initiator matrix and vector are determined by summing the initiator summation matrix row by row, which improves the calculation efficiency of the initiator joint matrix factor.

[0140] For example, based on the above embodiments, the label determination method includes:

[0141] The business initiator and business participants jointly perform privacy intersection calculations on the initiator's data and the participants' data to determine the joint data intersection D between the business initiator and business participants. P In the full dataset of the business initiator, excluding the intersection D of the joint dataset, P The data unique to the initiator other than D is D. AP In the full dataset of the business participants, excluding the dataset intersection D P Data unique to participants other than D BP .

[0142] The business initiator and business participants jointly access the joint data intersection D. PPrivacy-preserving intersection is performed on the labeled data to determine the intersection of the labeled data and the label data number in the intersection. The business initiator generates the initiator's label data list based on the intersection of the labeled data and the initiator's labeled data; the business participants generate the participant's label data list based on the intersection of the labeled data and the participant's labeled data.

[0143] The business initiator, in conjunction with the business participants, reorders the data, sorting the data from both the initiator's and participants' full data by sequence number, and generating a data sorting list. Furthermore, the data ranking list is shared between the business initiator and the business participants.

[0144] The data in the data sorting list is arranged in the following order: business initiator, and unique initiator tag data from the full dataset of initiators. Dataset intersection D P Unique initiator tag data for business initiators Dataset intersection D P Tag data shared by business initiators and participants Dataset intersection D P Unique participant tag data for business participants Unique tag data from the full dataset of business participants Unlabeled data of the initiator in the full dataset Dataset intersection D P Unlabeled data of business initiators and business participants Unlabeled participant data in the full participant data set The business initiator and business participants perform privacy-preserving union operations on their respective participant and initiator datasets to determine their respective label datasets, and then generate a label data matrix based on these datasets. The business initiator determines the initiator's topological edge weights based on its topological structure, and the business participants determine their own topological edge weights based on their respective topological structures. The business initiator organizes the initiator's topological edge weights according to the data ranking list to determine the initiator's weight matrix; similarly, the business participants organize their own topological edge weights according to the data ranking list to determine their own weight matrix.

[0145] The business initiator assigns unlabeled data to the initiator based on the initiator's topological edge weights. Perform propagation factor calculation to generate the initiator probability propagation matrix. Business participants then assign unlabeled data to their respective participants based on the participant's topological edge weights. Perform propagation factor calculation to generate the participant probability propagation matrix.

[0146] The business participants determine the participant matrix transmission factor based on the participant weight matrix and send the matrix transmission factor to the business initiator. The business initiator generates the initiator joint matrix factor based on the participant matrix transmission factor and the initiator matrix factor in the initiator weight matrix. The business initiator determines the initiator summation matrix from the initiator weight matrix and performs row-wise summation on the initiator summation matrix to determine the initiator matrix and vector.

[0147] The business participant determines the participant summation matrix from the participant weight matrix, sums the participant summation matrix row by row to determine the participant matrix and vector, and sends the participant matrix and vector to the business initiator. The business initiator obtains the participant matrix and vector, sums the initiator matrix and vector with the participant matrix and vector, determines the joint matrix and vector, and sends the joint matrix and vector to the business participant.

[0148] The business initiator determines its target matrix from the initiator weight matrix and its block probability matrix based on the joint matrix and vectors and the target matrix. The business participants determine their target matrices from the participant weight matrices and their block probability matrices based on the joint matrix and vectors and the target matrix. The participants then determine their transmission probability matrices from their block probability matrices and send them to the business initiator. The business initiator determines its update probability matrix based on the initiator probability propagation matrix and the participant transmission probability matrix; the business participants determine their update probability matrices based on their probability propagation matrices and their block probability matrices.

[0149] The service initiator determines random parameters, and based on these random parameters and the initiator's updated probability matrix, determines the initiator's first transmission parameter and sends it to the service participants. The service participants then determine their own first transmission parameters based on these parameters and send them back to the service initiator. The service initiator then determines its second transmission parameter based on the participant's transmission parameters and the random parameters, and sends it to the service participants. The service participants then determine their own second transmission parameter based on the initiator's second transmission parameter and the participant's updated probability matrix, and send it to the service initiator. Finally, the service initiator determines its own tag parameter and its third transmission parameter based on the participant's second transmission parameter and the initiator's updated probability matrix, and sends these third transmission parameters to the service participants. The service participants then determine their own tag parameter based on the initiator's third transmission parameter. The business initiator determines its target parameters based on its initiator tag parameters and tag data matrix, identifies its transmission tag from these target parameters, and sends the transmission tag to the business participants. The business participants determine their target parameters based on their tag parameters and tag data matrix, identify their transmission tags from these target parameters, and send the transmission tags to the business initiator. The business initiator obtains the participant transmission tags and determines the target tag for any untagged data from the initiator based on its target parameters and transmission tags. Similarly, the business participants obtain the initiator transmission tags and determine the target tag for any untagged data from the participants based on their target parameters and transmission tags.

[0150] In the above label determination method, the business initiator determines the communication data to be sent to the business initiator based on the initiator dataset and the communication data sent from the business participants to the business initiator. The business participants determine their communication data based on their participant datasets and the communication data sent from the business initiator to the business participants. The business initiator determines the initiator block probability matrix based on the initiator dataset and the participant communication data, updates the initiator block probability matrix, determines the updated initiator probability matrix, and determines the target label for the initiator's unlabeled data based on the updated initiator probability matrix and the label data matrix. Similarly, the business participants determine the participant block probability matrix based on their participant datasets and the initiator communication data, updates the participant block probability matrix, determines the updated participant probability matrix, and determines the target label for the participant's unlabeled data based on the updated participant probability matrix and the label data matrix. This solution addresses the issues of low efficiency and high cost in calculating tag information for unlabeled data from both the initiator and participant. Previously, calculating tag information for unlabeled data required an additional intermediate server to coordinate and calculate the information, posing a risk of data privacy leakage. Furthermore, it necessitated multiple calls to homomorphic encryption methods to calculate the tag information for unlabeled data, and offline negotiation between the initiator and participant regarding data relationships. The proposed solution, however, enables tag information calculation for unlabeled nodes in both communities when there are shared nodes between the initiator and participant. This solution reduces the risk of data privacy leakage during the tag information calculation process, improves the efficiency of tag information determination, and lowers the computational cost.

[0151] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0152] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0153] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A label determination method, characterized in that, The tagging is performed by a tagging system, which includes an initiator and participants in the communication. The tagging method includes: The business participants determine the participant weight matrix based on the participant topology, and determine the participant probability propagation matrix for the unlabeled data based on the participant weight matrix. They also determine the participant matrix transmission factor based on the participant weight matrix and send the participant matrix transmission factor to the business initiator. The business initiator obtains the transmission factor of the participant matrix, determines the initiator weight matrix based on the initiator topology, calculates the propagation factor of the initiator's unlabeled data based on the initiator weight matrix, determines the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix. The business participants determine the participant matrix and vector based on the participant weight matrix, and send the participant matrix and vector to the business initiator; The business initiator obtains the participant matrix and vector, determines the joint matrix and vector based on the initiator matrix and vector and the participant matrix and vector, sends the joint matrix and vector to the business participants, and determines the initiator block probability matrix based on the joint matrix and vector and the initiator weight matrix. The business participants obtain the joint matrix and vector, determine the participant block probability matrix based on the joint matrix and vector and the participant weight matrix, and send the participant transmission probability matrix in the participant block probability matrix to the business initiator. They also determine the participant update probability matrix based on the participant probability propagation matrix and the participant block probability matrix. The business initiator obtains the transmission probability matrix of the participants and determines the initiator's update probability matrix based on the transmission probability matrix of the participants and the probability propagation matrix of the initiator. The business initiator determines the target label for the initiator's unlabeled data based on the initiator's updated probability matrix and label data matrix, and the business participant determines the target label for the participant's unlabeled data based on the participant's updated probability matrix and label data matrix.

2. The method according to claim 1, characterized in that, The business initiator determines the target tags for which the initiator has no tag data based on the initiator's updated probability matrix and tag data matrix, including: The business initiator determines the initiator's first transmission parameter based on the initiator's updated probability matrix, sends the initiator's first transmission parameter to the business participant, and obtains the participant's first transmission parameter sent by the business participant. The business initiator determines the initiator's second transmission parameters based on the participant's first transmission parameters, sends the initiator's second transmission parameters to the business participant, and obtains the participant's second transmission parameters sent by the business participant. The business initiator determines the initiator label parameters and the initiator third transmission parameters based on the participant's second transmission parameters and the initiator update probability matrix, and sends the initiator third transmission parameters to the business participants; The business initiator determines the initiator target parameters based on the initiator tag parameters and the tag data matrix, sends the initiator transmission tag in the initiator target parameters to the business participants, and obtains the participant transmission tag sent by the business participants. The business initiator determines the target label for the initiator's unlabeled data based on the initiator's target parameters and the tags transmitted by the participants.

3. The method according to claim 2, characterized in that, The business initiator determines the target tags for the initiator's untagged data based on the initiator's target parameters and the tags transmitted by the participants, including: The business initiator determines candidate tags for which the initiator has no data based on the initiator's target parameters and the data of the tags transmitted by the participants, and determines the probability of the initiator's candidate tags for candidate tags for which the initiator has no data. The business initiator determines the target label for the initiator's unlabeled data from the candidate labels of the initiator's unlabeled data based on the probability of the initiator's candidate labels.

4. The method according to claim 2, characterized in that, Business participants determine the target labels for unlabeled data based on the participant update probability matrix and the label data matrix, including: The business participant obtains the first transmission parameters from the initiator, determines the first transmission parameters of the participant based on the first transmission parameters of the initiator, and sends the transmission parameters of the participant to the business initiator; The business participant obtains the second transmission parameter from the initiator, determines the second transmission parameter of the participant based on the second transmission parameter of the initiator and the participant update probability matrix, and sends the second transmission parameter of the participant to the business initiator. The business participants obtain the third-party transmission parameters from the initiator, and determine the participant's label parameters based on the third-party transmission parameters from the initiator and the participant's updated probability matrix. The business participants determine the participant target parameters based on the participant tag parameters and the tag data matrix, determine the participant transmission tag from the participant target parameters, and send the participant transmission tag to the business initiator. The business participants obtain the initiator's transmission tag and determine the target tag for the participant's untagged data based on the participant's target parameters and the initiator's transmission tag.

5. The method according to claim 4, characterized in that, Business participants obtain the initiator's transmission tag, and determine the target tag for the participant's untagged data based on the participant's target parameters and the initiator's transmission tag, including: Based on the participant's target parameters and the data of the participant's transmitted tags, the business participants determine the candidate tags for the participants without tag data, and determine the participant candidate tag probability of the candidate tags for the participants without tag data. Business participants determine the target label for the unlabeled data of the participant from the candidate labels of the unlabeled data of the participant based on the probability of the participant's candidate labels.

6. The method according to claim 4, characterized in that, The business participant obtains the initiator's second transmission parameters, determines its own second transmission parameters based on the initiator's second transmission parameters and the participant's updated probability matrix, and sends the participant's second transmission parameters to the business initiator, including: The business participants obtain the second transmission parameters from the initiator and determine the initiator parameter calculation factor based on the second transmission parameters from the initiator and the participant update probability matrix. The second transmission parameters of the participants are determined based on the initiator's parameter calculation factor and the participant's updated probability matrix, and then sent to the business initiator.

7. The method according to claim 1, characterized in that, The business participants determine the participant weight matrix based on the participant topology, including: Business participants determine the participant topology based on their local data relationship matrix. Business participants determine the length of the topological edge of the participant based on the participant topology structure, and determine the weight of the topological edge of the participant based on the weight coefficient of the topological edge of the participant and the weight of the topological edge of the participant, and determine the weight matrix of the participant based on the weight of the topological edge of the participant.

8. The method according to claim 1, characterized in that, The initiator weight matrix is ​​determined based on the initiator's topology, including: The business initiator determines the initiator's topology based on the initiator's local data relationship matrix; The business initiator determines the length of the initiator's topology edge based on the initiator's topology structure, and determines the weight of the initiator's topology edge based on the coefficients calculated by the initiator's topology edge length and weight, and determines the initiator's weight matrix based on the initiator's topology edge weight.

9. The method according to claim 1, characterized in that, Before the business initiator determines the target label for which the initiator has no labeled data based on the initiator's updated probability matrix and label data matrix, and before the business participants determine the target label for which they have no labeled data based on the participant's updated probability matrix and label data matrix, the following steps are also included: The business initiator performs a privacy-preserving union on the initiator's dataset and the participant's dataset to determine the tag data sets for the business initiator and the business participants; The label data matrix is ​​determined based on the amount of data in the initiator's dataset, the amount of data in the participant's dataset, and the label data set.

10. The method according to claim 1, characterized in that, The business initiator obtains the participant matrix transmission factor, determines the initiator weight matrix based on the initiator's topology, calculates the propagation factor for the initiator's unlabeled data based on the initiator weight matrix, determines the initiator probability propagation matrix, determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator matrix and vector based on the initiator weight matrix, including: The business initiator obtains the transmission factor of the participant matrix, determines the initiator weight matrix based on the initiator's topology, and calculates the propagation factor for the initiator's unlabeled data based on the initiator's weight matrix to determine the initiator's probability propagation matrix. The business initiator determines the initiator joint matrix factor based on the participant matrix transmission factor and the initiator weight matrix, and determines the initiator summation matrix from the initiator weight matrix; The business initiator determines the initiator matrix and vector by summing the initiator's summation matrix row by row.

Images