Multi-party information interaction monitoring system and method based on federal learning
By building a decision tree model and information management platform privacy processing, the security and delay problems of information interaction between information flow nodes in federated learning are solved, data privacy and information interaction efficiency are improved, and the load of the information management center is reduced.
Patent Information
- Application Number
- CN202411226134.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The intermediate parameters of model training in existing federated learning have problems such as leakage, delay in information interaction process and object selection, which affect data privacy and security.
By collecting the characteristic values of information flow nodes, building a decision tree model, obtaining characteristic privacy values, and performing privacy processing and management on the information management platform, establishing information interaction links, and improving privacy management and data security between information flow nodes.
It improves the security and privacy of information data in the information flow node, reduces information leakage, optimizes the delay and object selection during information interaction, and reduces the workload of the information management center.
Smart Images

Figure CN119026172A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information interaction, and in particular to a multi-party information interaction monitoring system and method based on federated learning. Background Art
[0002] Federated learning is a distributed machine learning technology. Its core idea is to build a global model based on virtual fused data by performing distributed model training among multiple data sources with local data, without the need to exchange local individual or sample data, only by exchanging model parameters or intermediate results, thereby achieving a balance between data privacy protection and data sharing computing, that is, a new application paradigm of "data available but invisible" and "data not moving but model moving".
[0003] In the prior art, federated learning usually uploads the intermediate parameters of model training to the parameter server to achieve model aggregation. This process has some problems: there are security issues such as leakage of intermediate parameters, the intermediate parameters will change continuously, and there will be delays in locating the interaction objects and selection problems during the information interaction process. Therefore, a multi-party information interaction monitoring method based on federated learning is provided. Summary of the invention
[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a multi-party information interaction monitoring method based on federated learning, comprising the following steps: Step S1: Collect the characteristic values of the information flow nodes, and perform characteristic processing on the characteristic values to obtain the characteristic privacy values of the information flow nodes; Step S2: Setting up an information management platform, and transmitting the feature privacy value difference to the information management platform for processing; Step S3: Construct interactive links between information flow nodes, and perform privacy management on information interactions between information flow nodes through the management platform.
[0005] Furthermore, the process of collecting characteristic values of information flow nodes includes: Get each information sending end of the information interaction process, recorded as information flow node; Obtain node information of the information flow node, record it as a root feature, and set a root feature value corresponding to the root feature; Obtaining node parameters of the information flow node, recording the node parameters of the information flow node as point features, and setting point feature values corresponding to the point features; Obtaining interactive links between adjacent information flow nodes and information flow nodes, recording them as branch features, and setting branch feature values corresponding to the branch features; Obtaining interactive links between inter-phase information flow nodes and information flow nodes, recording them as fork features, and setting fork feature values corresponding to the fork features; A decision tree model is constructed based on root features, branch features, fork features and point features, and the eigenvalues corresponding to each feature in the decision tree model are obtained.
[0006] Furthermore, the process of performing feature processing on the feature value to obtain the feature privacy value of the information flow node includes: Obtain corresponding characteristic coefficients according to the information interaction process; Mark the information flow nodes and corresponding interactive links that the information sender passes through during the interaction process, and obtain the corresponding total eigenvalue according to the eigenvalues corresponding to the marked information flow nodes and interactive links; Divide the total eigenvalue obtained into n feature segments; Set a k*k blank matrix; Perform base conversion on n feature fragments to obtain dynamic matrices, and associate each dynamic matrix with the corresponding information flow node; Perform feature transformation on the dynamic matrix to obtain the privacy value; According to the obtained privacy value and characteristic coefficient, the characteristic privacy value of the interaction process is obtained.
[0007] Furthermore, an information management platform is set up, and the process of transmitting the feature privacy value difference to the information management platform for processing includes: Construct an information intersection transmission link composed of several information flow nodes, obtain the characteristic privacy value of the information interaction transmission link in real time, and convert the characteristic privacy value into a string composed of several characters, which are recorded as privacy characters; When the privacy characters are transmitted sequentially through the information interaction transmission link, the number of transmission times of the information flow node is recorded; Set the transfer rate; According to the transmission times and transmission rates of the information flow nodes, the intersection transmission rate of the information flow nodes is obtained; The privacy characters of information flow nodes are transmitted according to the intersection transmission rate of information flow nodes; When the privacy character intersects with the privacy character on the information intersection transmission link, the continuous identical character parts are recorded as overlapping characters, and it is determined whether the overlapping characters need to be overlapped; Set the standard length of overlapping characters; When the length of the overlapping characters is less than or equal to the standard length of the overlapping characters, there is no need to perform overlapping processing on the privacy characters; When the length of the overlapping characters is greater than the standard length of the overlapping characters, the overlapping characters are separated from the private characters and removed, and the private characters after the overlapping processing are recorded as the private deduplication characters; The intersection transmission rate of the privacy-deduplicated characters is restored to the transmission rate until they are transmitted to the information management platform, and the information management platform processes the privacy-deduplicated characters.
[0008] Furthermore, the process of the information management platform processing the privacy deduplication characters includes: Set the iterative cycle; According to the iterative difference period, the deduplicated privacy characters of each information flow node are aggregated, and the deduplicated privacy characters of the information flow nodes are combined to obtain the iterative feature privacy value of the information flow node, and the iterative feature privacy value of each information flow node is stored; The information management platform aggregates the iterative feature privacy values of each information flow node to obtain the global parameters of the global model, which include optimal parameters and interaction parameters.
[0009] Furthermore, the process of constructing interactive links between information flow nodes and performing privacy management on information interactions between information flow nodes through the management platform includes: Set information flow node a; Information flow node a sends an information interaction request signal to the information management platform; The information interaction request signal includes information interaction object parameters and a characteristic privacy value calling instruction of information flow node a; The information interaction object parameters are parameter data obtained through the local model of the information flow node a; The parameter calling instruction of the information flow node a refers to the information management center calling the iterative feature privacy value of the information flow node a; When the information management platform receives the information interaction request from information flow node a, it retrieves the iterative feature privacy value of information flow node a; According to the interaction parameters of the information management platform, the information interaction object parameters of the information flow node a are matched with the privacy values of each iterative feature of each information interaction request signal sent by other information flow nodes to the information management platform; Perform character conversion on the information interaction object parameters of the information flow node a and each iterative feature privacy value of each information interaction request signal sent by other information flow nodes to the management platform to obtain the required character and the supplied character; Set the same frequency character length threshold; According to the interaction parameter and the same-frequency character length threshold, an interaction character length threshold is obtained; According to the interactive character length threshold, obtain the information flow node whose length of overlapping characters between the required character and the supplied character is greater than or equal to the interactive character length threshold, record it as the node to be interacted, and associate it with the overlapping character length; According to the optimal parameters of the information interaction platform, the iterative feature privacy value of the information flow node a and the parameters of each information interaction object of each information interaction request signal sent by each node to be interacted to the information management platform are matched at the same frequency.
[0010] Furthermore, the process of matching the iterative feature privacy value of the information flow node a with each information interaction object parameter of each information interaction request signal sent by each to-be-interacted node to the information management platform according to the optimal parameters of the information interaction platform includes: According to the optimal parameters and the same-frequency character length threshold, the optimal mutual character length threshold is obtained; Perform character conversion on the privacy value of the iterated feature of the information flow node a and the parameters of each information interaction object of the node to be interacted with to obtain the transfer character and the withdrawal character; According to the preferred interactive character length threshold, obtain the information flow node whose overlapping character length between the transfer character and the referral character is greater than or equal to the interactive character length threshold, record it as an equal interactive node, and associate it with the overlapping character length. According to the equal interactive node, obtain the interactive object of the information flow node a.
[0011] Furthermore, the process of obtaining the interaction object of the information flow node a according to the equal interaction node includes: According to the maximum product of the lengths of two overlapping characters associated with the equal interaction nodes, the interaction object of the information flow node a is obtained and recorded as the interaction node; The information management platform sends the information interaction request signal to the interaction node, and the interaction node interacts with the information flow node; In another embodiment of the present invention, the present invention further discloses a multi-party information interaction monitoring system based on federated learning, including a management center, wherein the management center is communicatively connected to an information collection module, an information transmission module, and an information interaction module; The information collection module is used to collect characteristic values of information flow nodes, and perform characteristic processing on the characteristic values to obtain characteristic privacy values of the information flow nodes; The information transmission module is used to set up an information management platform and transmit the feature privacy value difference to the information management platform for processing; The information interaction module is used to build an interactive link between each information flow node, and to perform privacy management on the information interaction between each information flow node through the management platform.
[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: the parameter characteristics, eigenvalues and eigencoefficients of the eigenvalues of the information flow nodes are obtained through the decision tree model, the dynamic matrix is feature processed according to the eigenvalues and eigencoefficients to obtain the privacy matrix, and the privacy matrix is further processed to obtain the characteristic privacy value of the information flow node, so as to improve the security and privacy of the information data in the information flow node, reduce the leakage of the information data of the information flow node, transmit the characteristic privacy value to the information management platform in an iterative manner, record the number of transmissions of the information flow node, and specify the transmission rate each time according to the number of transmissions. During the transmission process, an iterative period is set, and there are overlapping parts when the characteristic privacy values intersect, and the overlapping parts are deduplicated. The characteristic privacy values uploaded by the information flow nodes are regularly updated to reduce the workload of the information management center. During the information interaction process of the information flow nodes, the information flow nodes send information interaction requests to the information management platform, and the information management platform performs the best match on the interaction objects of the information flow nodes according to the information interaction requests, and sends the information interaction requests to the interaction objects, so as to improve the demand for information interaction. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 This is a schematic diagram of a multi-party information interaction monitoring system based on federated learning according to an embodiment of the present application.
[0014] Figure 2 This is a schematic diagram of a multi-party information interaction monitoring method based on federated learning in an embodiment of the present application. DETAILED DESCRIPTION
[0015] like Figure 1 As shown, a multi-party information interaction monitoring system based on federated learning includes the following steps: Step S1: Collect the characteristic values of the information flow nodes, and perform characteristic processing on the characteristic values to obtain the characteristic privacy values of the information flow nodes; Step S2: Setting up an information management platform, and transmitting the feature privacy value difference to the information management platform for processing; Step S3: Construct interactive links between information flow nodes, and perform privacy management on information interactions between information flow nodes through the management platform.
[0016] It should be further explained that, in the specific implementation process, the specific process of collecting the characteristic values of the information flow nodes and performing characteristic processing on the characteristic values to obtain the characteristic privacy values of the information flow nodes includes: Get each information sending end of the information interaction process, recorded as information flow node; Get the node parameters of the information flow node; The node parameters are related parameters of information data in the information flow node; Get each information sending end of the information interaction process, recorded as information flow node; Obtain node information of the information flow node, record it as a root feature, and set a root feature value corresponding to the root feature; Obtaining node parameters of the information flow node, recording the node parameters of the information flow node as point features, and setting point feature values corresponding to the point features; Obtaining interactive links between adjacent information flow nodes and information flow nodes, recording them as branch features, and setting branch feature values corresponding to the branch features; Obtaining interactive links between inter-phase information flow nodes and information flow nodes, recording them as fork features, and setting fork feature values corresponding to the fork features; Construct a decision tree model based on root features, branch features, fork features and point features, and obtain the eigenvalues corresponding to each feature in the decision tree model; Perform interactive operations on the eigenvalues according to the information interaction process to obtain the corresponding eigencoefficients; The eigenvalues include root eigenvalues, branch eigenvalues, fork eigenvalues and point eigenvalues; It should be further explained that, in the specific implementation process, the process of obtaining the characteristic coefficients in the information interaction process includes: for example, in the information interaction process, when the information interaction passes through the information flow node and the interactive link, the root characteristic value is set to 00, the point characteristic value is 01, the branch characteristic value is 10, and the fork characteristic value is 11, and the root characteristic value corresponding to the node information of the information flow node through which the information interaction passes, the point characteristic value corresponding to the node parameter of the information flow node, the interactive link between adjacent information flow nodes and the branch characteristic corresponding to the information flow node, and the interactive link between alternate information flow nodes and the fork characteristic value corresponding to the information flow node are obtained, and the root characteristic value, point characteristic value, branch characteristic value and fork characteristic value of the information interaction process are interactively calculated. For example, the characteristic values passed by the information interaction process are: 00, 01, 10, 11, 01, 11, 10, 11, and the characteristic coefficient obtained by the interactive calculation is 0001101101111011; It should be further explained that, in the specific implementation process, the root eigenvalue, branch eigenvalue, fork eigenvalue, point eigenvalue and eigencoefficient are binary sequences; Mark the information flow nodes and corresponding interactive links that the information sender passes through during the interaction process, and obtain the corresponding total eigenvalue according to the eigenvalues corresponding to the marked information flow nodes and interactive links; Divide the total eigenvalue obtained into n feature segments; It should be further explained that, in the specific implementation process, the lengths of the n feature segments always need to be consistent. If the lengths of the n feature segments are inconsistent, the last feature segment is padded with 0 to make the lengths of the n feature segments consistent. Set a k*k blank matrix; Perform base conversion on n feature fragments, place the base-converted feature fragments in the items of the blank matrix, generate a dynamic matrix, and associate each dynamic matrix with the corresponding information flow node; It should be further explained that, in the specific implementation process, if there are empty items in the dynamic matrix, the dynamic matrix is padded with 1 so that there are no empty items in the dynamic matrix; Set the dynamic matrix set, denoted as ; The dynamic matrix set is composed of a plurality of dynamic matrices; It should be further explained that, in the specific implementation process, the dynamic matrix is unique, and the dynamic matrix of each information flow node is different; Perform column-row calculations on the dynamic matrix to obtain the privacy value, denoted as Q; It should be further explained that, in the specific implementation process, the specific process of column-row calculation is: multiply the first column of the dynamic matrix by the matrix item of the same row of the second column to obtain a new matrix column, multiply the new matrix column by the matrix item of the same row of the third column, and so on, the result obtained is a single matrix column, and all items of the single matrix column are summed to obtain the privacy value; According to the obtained privacy value and characteristic coefficient, the characteristic privacy value of the interaction process is obtained.
[0017] It should be further explained that, in the specific implementation process, the specific process of setting up an information management platform and transmitting the feature privacy value difference to the information management platform for processing includes: Set the iterative cycle; It should be further explained that, in the specific implementation process, the characteristic privacy value sent by the information flow node is analyzed every other iteration period; Construct an information intersection transmission link composed of several information flow nodes, and transmit the feature privacy value through the information flow nodes in an iterative manner; Obtain the characteristic privacy value of the interactive transmission link in real time, and convert the characteristic privacy value into a string consisting of several characters, which are recorded as privacy characters; When the privacy characters are transmitted sequentially through the information intersection transmission link, the number of transmission times of the information flow node is recorded, which is recorded as ; Set the transmission rate, denoted as Sv; According to the transmission times and transmission rate of the information flow node, the intersection transmission rate of the information flow node is obtained, which is recorded as Wu; but ; in, is the intersection transmission coefficient; The privacy characters of information flow nodes are transmitted according to the intersection transmission rate of information flow nodes; When the privacy character intersects with the privacy character on the information intersection transmission link, the continuous identical character parts are recorded as overlapping characters, and it is determined whether the overlapping characters need to be overlapped; It should be further explained that, in the specific implementation process, the privacy character intersects with the privacy character on the information intersection transmission link. For example, the information intersection transmission link a and the information intersection transmission link b of the information flow node, the information intersection transmission link b is used for the current transmission of the privacy character, the information intersection transmission link a is used for the last transmission of the privacy character, the privacy character being transmitted by the information intersection transmission link b and the privacy character being transmitted by the information intersection transmission link a have part or all of the character string in the same position; Set the standard length of overlapping characters, denoted as G; When the length of the overlapping characters is less than or equal to the standard length G of the overlapping characters, there is no repeated privacy data in the privacy characters, and there is no need to perform overlapping processing on the privacy characters; When the length of the overlapping characters is greater than the overlapping character standard length G, there are repeated privacy data in the privacy characters, and the privacy characters are overlapped, the overlapping characters are decomposed from the privacy characters and cleared, and the privacy characters after the overlapping processing are recorded as privacy deduplication characters; The intersection transmission rate of the privacy deduplicated characters is restored to the transmission rate until it is transmitted to the information management platform; According to the iterative difference period, the deduplicated privacy characters transmitted by each informatics flow node to the information management platform are aggregated, and the deduplicated privacy characters of the information flow nodes are combined according to the iterative difference transmission order of the deduplicated privacy characters to obtain the iterative feature privacy value of the information flow node, and the iterative feature privacy value of each information flow node is stored; The information management platform aggregates the iterative feature privacy values of each information flow node, generates a global model formed by the information flow node and the information management platform, and then obtains the global parameters of the global model, which include optimal parameters and interaction parameters; The optimal parameter is recorded as ZY; The interaction parameter is denoted as JH.
[0018] It should be further explained that, in the specific real-time process, the interactive links between the information flow nodes are constructed, and the specific process of performing privacy management on the information interaction between the information flow nodes through the information management platform includes: Set information flow node a; When information flow node a performs information interaction, information flow node a sends an information interaction request signal to the information management platform; The information interaction request signal includes information interaction object parameters and a characteristic privacy value calling instruction of information flow node a; The information interaction object parameters are parameter data obtained through the local model of the information flow node a; The parameter calling instruction of the information flow node a refers to the information management center calling the iterative feature privacy value of the information flow node a; When the information management platform receives the information interaction request from information flow node a, it retrieves the iterative feature privacy value of information flow node a; According to the interaction parameters of the information management platform, the information interaction object parameters of the information flow node a are matched with the privacy values of each iterative feature of each information interaction request signal sent by other information flow nodes to the information management platform; Perform character conversion on the information interaction object parameter of the information flow node a, and record the result of the character conversion as the required character; Convert the privacy values of each iterative feature of each information interaction request signal sent by other information flow nodes to the management platform into characters, and record the result of the character conversion as the iterative character; Set the same-frequency character length threshold, denoted as Tp; According to the interaction parameters and the same-frequency character length threshold, the interaction character length threshold is obtained, which is recorded as Jz; but ; in, is the length conversion coefficient; When the length of the overlapping characters of the demand character and the supply character is less than the interactive character length threshold Jz, the information flow node corresponding to the supply character does not meet the demand of the information flow node a for the information interaction object; When the length of the overlapping characters of the demand character and the supply character is greater than or equal to the interactive character length threshold Jz, the information flow node corresponding to the supply character is recorded as the node to be interacted, the length of the overlapping characters is obtained, and the length of the overlapping characters is associated with the node to be interacted; According to the optimal parameters of the information interaction platform, the iterative feature privacy value of the information flow node a is matched with the parameters of each information interaction object of each information interaction request signal sent by each node to be interacted to the information management platform; According to the optimal parameters and the same-frequency character length threshold, the optimal mutual character length threshold is obtained, denoted as Zf; but ; Perform character conversion on the privacy value of the iterated feature of the information flow node a, and record the result of the character conversion as the conversion character; Performing character conversion on each information interaction object parameter of each information interaction request signal sent by the node to be interacted to the information management platform, and recording the result of the character conversion as the collection character; When the length of the overlapping characters of the transfer character and the transfer character is less than the optimal character length threshold Zf, the information flow node corresponding to the transfer character does not meet the information flow node a's demand for the information interaction object; When the length of the overlapping characters of the transfer character and the transfer character is greater than or equal to the optimal interactive character length threshold Zf, the information flow node corresponding to the transfer character is recorded as an equal interactive node, the length of the overlapping characters is obtained, and the length of the overlapping characters is associated with the equal interactive node; If there are two overlapping characters associated with the equal interaction node, the lengths of the two overlapping characters associated with the equal interaction node are multiplied to obtain the equal interaction node corresponding to the maximum value, which is recorded as the interaction node; The information management platform sends an information interaction request signal to the interaction node, and the interaction node interacts with the information flow node.
[0019] like Figure 2 As shown, in another embodiment of the present invention, the present invention further discloses a multi-party information interaction monitoring system based on federated learning, including a management center, characterized in that the management center is communicatively connected with an information collection module, an information transmission module and an information interaction module; The information collection module is used to collect characteristic values of information flow nodes, and perform characteristic processing on the characteristic values to obtain characteristic privacy values of the information flow nodes; The information transmission module is used to set up an information management platform and transmit the feature privacy value difference to the information management platform for processing; The information interaction module is used to build an interactive link between each information flow node, and to perform privacy management on the information interaction between each information flow node through the management platform.
[0020] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.
Claims
1. A multi-party information interaction monitoring method based on federated learning, characterized in that: The following steps are involved: Step S1: Collect the characteristic values of the information flow nodes, and perform characteristic processing on the characteristic values to obtain the characteristic privacy values of the information flow nodes; Step S2: Setting up an information management platform, and transmitting the feature privacy value difference to the information management platform for processing; Step S3: Construct interactive links between information flow nodes, and perform privacy management on information interactions between information flow nodes through the management platform.
2. According to claim 1, a multi-party information interaction monitoring method based on federated learning is characterized in that: The process of collecting the characteristic values of information flow nodes includes: Get each information sending end of the information interaction process, recorded as information flow node; Obtain node information of the information flow node, record it as a root feature, and set a root feature value corresponding to the root feature; Obtaining node parameters of the information flow node, recording the node parameters of the information flow node as point features, and setting point feature values corresponding to the point features; Obtaining interactive links between adjacent information flow nodes and information flow nodes, recording them as branch features, and setting branch feature values corresponding to the branch features; Obtaining interactive links between inter-phase information flow nodes and information flow nodes, recording them as fork features, and setting fork feature values corresponding to the fork features; A decision tree model is constructed based on root features, branch features, fork features and point features, and the eigenvalues corresponding to each feature in the decision tree model are obtained.
3. According to claim 2, a method for monitoring multi-party information interaction based on federated learning is characterized in that: The process of performing feature processing on the feature value to obtain the feature privacy value of the information flow node includes: Obtain corresponding characteristic coefficients according to the information interaction process; Mark the information flow nodes and corresponding interactive links that the information sender passes through during the interaction process, and obtain the corresponding total eigenvalue according to the eigenvalues corresponding to the marked information flow nodes and interactive links; Divide the total eigenvalue obtained into n feature segments; Set a k*k blank matrix; Perform base conversion on n feature fragments to obtain dynamic matrices, and associate each dynamic matrix with the corresponding information flow node; Perform feature transformation on the dynamic matrix to obtain the privacy value; According to the obtained privacy value and characteristic coefficient, the characteristic privacy value of the interaction process is obtained.
4. According to claim 3, a multi-party information interaction monitoring method based on federated learning is characterized in that: The process of setting up an information management platform and transmitting the feature privacy value difference to the information management platform for processing includes: Construct an information intersection transmission link composed of several information flow nodes, obtain the characteristic privacy value of the information interaction transmission link in real time, and convert the characteristic privacy value into a string composed of several characters, which are recorded as privacy characters; When the privacy characters are transmitted sequentially through the information interaction transmission link, the number of transmission times of the information flow node is recorded; Set the transfer rate; According to the transmission times and transmission rates of the information flow nodes, the intersection transmission rate of the information flow nodes is obtained; The privacy characters of information flow nodes are transmitted according to the intersection transmission rate of information flow nodes; When the privacy character intersects with the privacy character on the information intersection transmission link, the continuous identical character parts are recorded as overlapping characters, and it is determined whether the overlapping characters need to be overlapped; Set the standard length of overlapping characters; When the length of the overlapping characters is less than or equal to the standard length of the overlapping characters, there is no need to perform overlapping processing on the privacy characters; When the length of the overlapping characters is greater than the standard length of the overlapping characters, the overlapping characters are separated from the private characters and removed, and the private characters after the overlapping processing are recorded as the private deduplication characters; The intersection transmission rate of the privacy-deduplicated characters is restored to the transmission rate until they are transmitted to the information management platform, and the information management platform processes the privacy-deduplicated characters.
5. According to claim 4, a method for monitoring multi-party information interaction based on federated learning is characterized in that: The process of the information management platform processing privacy deduplication characters includes: Set the iterative cycle; According to the iterative difference period, the deduplicated privacy characters of each information flow node are aggregated, and the deduplicated privacy characters of the information flow nodes are combined to obtain the iterative feature privacy value of the information flow node, and the iterative feature privacy value of each information flow node is stored; The information management platform aggregates the iterative feature privacy values of each information flow node to obtain the global parameters of the global model, which include optimal parameters and interaction parameters.
6. A method for monitoring multi-party information interaction based on federated learning according to claim 5, characterized in that: The process of building interactive links between information flow nodes and performing privacy management on information interactions between information flow nodes through the management platform includes: Set information flow node a; Information flow node a sends an information interaction request signal to the information management platform; The information interaction request signal includes information interaction object parameters and a characteristic privacy value calling instruction of information flow node a; The information interaction object parameters are parameter data obtained through the local model of the information flow node a; The parameter calling instruction of the information flow node a refers to the information management center calling the iterative feature privacy value of the information flow node a; When the information management platform receives the information interaction request from information flow node a, it retrieves the iterative feature privacy value of information flow node a; According to the interaction parameters of the information management platform, the information interaction object parameters of the information flow node a are matched with the privacy values of each iterative feature of each information interaction request signal sent by other information flow nodes to the information management platform; Perform character conversion on the information interaction object parameters of the information flow node a and each iterative feature privacy value of each information interaction request signal sent by other information flow nodes to the management platform to obtain the required character and the supplied character; Set the same frequency character length threshold; According to the interaction parameter and the same-frequency character length threshold, an interaction character length threshold is obtained; According to the interactive character length threshold, obtain the information flow node whose length of overlapping characters between the required character and the supplied character is greater than or equal to the interactive character length threshold, record it as the node to be interacted, and associate it with the overlapping character length; According to the optimal parameters of the information interaction platform, the iterative feature privacy value of the information flow node a and the parameters of each information interaction object of each information interaction request signal sent by each node to be interacted to the information management platform are matched at the same frequency.
7. The method for monitoring multi-party information interaction based on federated learning according to claim 6, characterized in that: The process of matching the iterative feature privacy value of the information flow node a with the parameters of each information interaction object of each information interaction request signal sent by each to-be-interacted node to the information management platform according to the optimal parameters of the information interaction platform includes: According to the optimal parameters and the same-frequency character length threshold, the optimal mutual character length threshold is obtained; Perform character conversion on the privacy value of the iterated feature of the information flow node a and the parameters of each information interaction object of the node to be interacted with to obtain the transfer character and the withdrawal character; According to the preferred interactive character length threshold, obtain the information flow node whose overlapping character length between the transfer character and the referral character is greater than or equal to the interactive character length threshold, record it as an equal interactive node, and associate it with the overlapping character length. According to the equal interactive node, obtain the interactive object of the information flow node a.
8. The method for monitoring multi-party information interaction based on federated learning according to claim 7, characterized in that: The process of obtaining the interaction object of information flow node a according to the equal interaction node includes: According to the maximum product of the lengths of two overlapping characters associated with the equal interaction nodes, the interaction object of the information flow node a is obtained and recorded as the interaction node; The information management platform sends an information interaction request signal to the interaction node, and the interaction node interacts with the information flow node.
9. A multi-party information interaction monitoring system based on federated learning, specifically applied to a multi-party information interaction monitoring method based on federated learning as claimed in any one of claims 1 to 8, comprising a management center, characterized in that: The management center is communicatively connected with an information collection module, an information transmission module and an information interaction module; The information collection module is used to collect characteristic values of information flow nodes, and perform characteristic processing on the characteristic values to obtain characteristic privacy values of the information flow nodes; The information transmission module is used to set up an information management platform and transmit the feature privacy value difference to the information management platform for processing; The information interaction module is used to build an interactive link between each information flow node, and to perform privacy management on the information interaction between each information flow node through the management platform.
