Call privacy protection method, system, device, medium and product

Abstract

The invention discloses a call privacy protection method, system and device, a medium and a product, relates to the technical field of communication, ensures that the call privacy of a user is reliably protected in the whole analysis process through cooperation between a server and an operator node, and simultaneously ensures that the call privacy of the user is reliably protected through a federal aggregation mode constructed by a trusted cross-network communication relationship. The calculation load of the server can be effectively reduced, and the conversation privacy protection efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method, system, device, medium, and product for protecting call privacy. Background Technology

[0002] The call relationship network built on user call records is a core digital asset of operators, containing extremely high levels of personal privacy information, and requires strict privacy protection.

[0003] Existing technologies often employ a scheme combining differential privacy and federated learning for privacy protection. Specifically, each operator node independently calculates updates to its local call relationship network based on the Laplace mechanism and then directly uploads the noisy differential data to the server. After receiving the data from all nodes, the server aggregates the data using a federated averaging algorithm to obtain global privacy data.

[0004] However, in current differential privacy protection schemes, attackers or untrusted servers may infer sensitive information in the original call relationship by analyzing the amount of plaintext updates uploaded by each node. Summary of the Invention

[0005] The embodiments of the present invention aim to provide a method, system, device, medium and product for protecting call privacy. Through the collaboration between the server and the operator node, the user's call privacy is reliably protected throughout the analysis process. At the same time, the federated aggregation mode built through trusted cross-network communication relationships can effectively reduce the server's computing load and improve the efficiency of call privacy protection.

[0006] In a first aspect, embodiments of the present invention provide a method for protecting call privacy applied in a server, wherein the server is configured to communicate with an operator node, and the operator node stores a local call relationship network constructed based on local user call records, and the call privacy protection method includes: Initialize the global call relationship weight and distribute the global call relationship weight to each of the operator nodes so that each operator node can perform differential privacy protection on the local call relationship network according to the global call relationship weight; In response to the update command, the local aggregate share of each operator node is collected; the local aggregate share of an operator node is the sum of all secret shares received from other operator nodes; the secret share is obtained by the operator node based on the trusted cross-network communication relationship and using the secret sharing mechanism to divide the local update amount of the edge weight in its local call relationship network. Based on all the local aggregation shares, calculate the global update amount and update the global call relationship weight, and send the updated global call relationship weight to each of the operator nodes.

[0007] As an improvement to the above scheme, the initialization of global call relationship weights and the distribution of these weights to each operator node, enabling each operator node to perform differential privacy protection on its local call relationship network based on the global call relationship weights, includes: The global call relationship weights are initialized based on the preset initial weights of the business scenario. The global call relationship weights are distributed to each of the operator nodes, so that each operator node can calculate the sensitivity of the local call edge in the local call relationship network based on the global call relationship weights, and add adaptive first differential noise to the edge weight of the local call edge in the local call relationship network based on the sensitivity.

[0008] As an improvement to the above scheme, the update instructions include periodic update instructions from the server itself and local call relationship network update instructions from any operator node.

[0009] As an improvement to the above scheme, the step of calculating the global update amount and updating the global call relationship weight based on all the local aggregation shares, and sending the updated global call relationship weight to each of the operator nodes, includes: Based on all the aforementioned local aggregation shares, the global update amount of the global call relationship weight is reconstructed in plaintext; Calculate the average update volume based on the global update volume and the number of operator nodes; The global call relationship weight is updated based on the average update amount; The updated global call relationship weights are sent to each of the operator nodes so that the operator nodes can update their local call relationship networks based on the updated global call relationship weights.

[0010] As an improvement to the above scheme, updating the global call relationship weight based on the average update amount includes: Global differential privacy noise is added to the average update amount to obtain the average privacy update amount; The global call relationship weight is updated based on the average privacy update amount.

[0011] As an improvement to the above solution, the call privacy protection method further includes: In response to a privacy budget request, the periodic privacy budget for the current call service cycle is dynamically calculated based on the global privacy budget; the privacy budget request is sent to the server by the operator node or server when generating differential noise; Calculate the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario; If the remaining task-level budget meets the response limit corresponding to the privacy budget application, then the response limit will be deducted, and a successful reimbursement notification will be returned. Otherwise, a reconciliation failure notification will be returned.

[0012] As an improvement to the above solution, the step of responding to a privacy budget request and dynamically calculating the periodic privacy budget for the current call service cycle based on the global privacy budget includes: In response to the privacy budget request, calculate the remaining global budget based on the global privacy budget; Based on a preset time decay rate, the global remaining budget is dynamically allocated periodically to obtain the periodic privacy budget for the current call service cycle.

[0013] As an improvement to the above solution, the step of calculating the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario includes: Calculate the service demand level of the previous call service cycle, and calculate the priority score of the current service scenario in the current call service cycle based on the service demand level and the static priority of the current service scenario. Based on the priority score and the periodic privacy budget, calculate the remaining task-level budget corresponding to the current business scenario.

[0014] Secondly, embodiments of the present invention provide a method for protecting call privacy applied in a carrier node, wherein the carrier node is configured to communicate with a server and other carrier nodes, and the method for protecting call privacy includes: Construct a local call relationship network based on local user call records; Receive global call relationship weights from the server, and perform differential privacy protection on the local call relationship network based on the global call relationship weights; In response to the update of local user call records, the local update amount of the edge weights in the local call relationship network is calculated. Based on the operator nodes with trusted cross-network communication relationships, the local update amount is divided through a secret sharing mechanism. The divided secret shares are sent to the corresponding operator nodes, and an update instruction is sent to the server. In response to the local aggregated share collection command, all received secret shares are aggregated to obtain the local aggregated share, and the local aggregated share is returned to the server so that the server can update the global call relationship weight.

[0015] As an improvement to the above solution, the step of constructing a local call relationship network based on local user call records includes: Based on local user call records, we can obtain information about users, call relationships between users, and call characteristics between users. The user is used as a local call node, and a local call edge is constructed based on the call relationship between the users; Calculate the edge weight of the local call edge based on the call characteristics between the users; A local call relationship network is constructed based on the local call node, the local call edge, and the edge weight.

[0016] As an improvement to the above scheme, the call characteristics include call frequency, call duration, and call patterns; then, calculating the edge weight of the local call edge based on the call characteristics between users includes: Obtain the preset frequency weighting coefficient, duration weighting coefficient, and pattern weighting coefficient; The call frequencies are weighted according to the frequency weighting coefficients to obtain the weighted call frequencies; The call duration is weighted according to the duration weighting coefficient to obtain the weighted call duration; The call patterns are weighted according to the aforementioned pattern weighting coefficients to obtain weighted call patterns; The weight of the local call edge is obtained by calculating the sum of the weighted call frequency, the weighted call duration, and the weighted call pattern.

[0017] As an improvement to the above scheme, the step of receiving global call relationship weights from the server and performing differential privacy protection on the local call relationship network based on the global call relationship weights includes: Receive global call relationship weights from the server; Based on the global call relationship weight, calculate the sensitivity of the local call edge in the local call relationship network; Based on the sensitivity, an adaptive first differential noise is added to the edge weights of the local call edges in the local call relationship network.

[0018] As an improvement to the above scheme, the step of calculating the sensitivity of the local call edge in the local call relationship network based on the global call relationship weight includes: Based on the business scenario, obtain the preset basic sensitivity for the local call side in the local call relationship network; Based on the global call relationship weights, global call features are obtained; Calculate the feature adjustment coefficient of the basic sensitivity based on the global call characteristics; The sensitivity of the local call edge in the local call relationship network is calculated based on the feature adjustment coefficient and the basic sensitivity.

[0019] As an improvement to the above scheme, the step of adding adaptive first differential noise to the edge weights of local call edges in the local call relationship network based on the sensitivity includes: If the sensitivity is not greater than a preset first sensitivity threshold, then Laplace differential noise is generated based on the sensitivity to obtain the first noise; If the sensitivity is greater than a preset second sensitivity threshold, then Gaussian difference noise is generated based on the sensitivity to obtain the second noise; Otherwise, based on the aforementioned sensitivity, a mixture of Laplace difference noise and Gaussian difference noise is generated to obtain a third noise; The generated first noise, second noise, or third noise is added to the edge weight of the local call edge in the local call relationship network.

[0020] As an improvement to the above scheme, in response to the update of local user call records, the local update amount of the edge weights in the local call relationship network is calculated. Based on the operator nodes with trusted cross-network communication relationships, the local update amount is divided through a secret sharing mechanism. The divided secret shares are sent to the corresponding operator nodes, and an update instruction is sent to the server, including: In response to an update of the local user's call record, calculate the local update amount of the edge weights in the local call relationship network; A second differential noise is added to the local update amount to obtain the local privacy update amount; Based on trusted operator nodes with trusted cross-network communication relationships, the local privacy update amount is divided into N secret shares using a preset secret sharing algorithm; N is the number of trusted operator nodes. The secret share is sent to the corresponding trusted operator node; the secret share and the trusted operator node are in one-to-one correspondence. Send an update command to the server.

[0021] As an improvement to the above solution, the call privacy protection method further includes: Obtain the current cross-network communication relationship and cross-network communication frequency between the first operator node and other second operator nodes; the second operator node is configured to communicate with the server; If the first operator node and the second operator node have a cross-network communication relationship, and the cross-network communication frequency is greater than a preset frequency threshold, then the second operator node is regarded as a trusted operator node of the first operator node.

[0022] As an improvement to the above scheme, the step of responding to the local aggregated share collection command, aggregating all received secret shares to obtain a local aggregated share, and returning the local aggregated share to the server so that the server can update the global call relationship weight includes: In response to the local aggregated share collection instruction, all received secret shares are aggregated to obtain the local aggregated share; the local aggregated share collection instruction is issued by the server to all operator nodes after receiving the update instruction; The local aggregate share is returned to the server so that the server can update the global call relationship weight.

[0023] As an improvement to the above solution, the call privacy protection method further includes: In response to its own differential noise generation request, it generates a privacy budget request; The privacy budget request is sent to the server, and differential noise is generated based on the reconciliation success notification returned by the server.

[0024] Thirdly, embodiments of the present invention provide a call privacy protection system applied in a server. The server is configured to communicate with an operator node, and the operator node stores a local call relationship network constructed based on local user call records. The call privacy protection system includes: The global call relationship weight initialization module is used to initialize the global call relationship weight and distribute the global call relationship weight to each of the operator nodes, so that each operator node can perform differential privacy protection on the local call relationship network according to the global call relationship weight. The update command response module is used to respond to update commands and collect the local aggregate share of each operator node. The local aggregate share of an operator node is the sum of all secret shares received from other operator nodes. The secret share is obtained by the operator node based on trusted cross-network communication relationships and using a secret sharing mechanism to divide the local update amount of the edge weights in its local call relationship network. The global call relationship weight update module is used to calculate the global update amount and update the global call relationship weight based on all the local aggregation shares, and send the updated global call relationship weight to each of the operator nodes.

[0025] Fourthly, embodiments of the present invention provide a call privacy protection system applied in a carrier node, wherein the carrier node is configured to communicate with a server and other carrier nodes, and the call privacy protection system includes: The local call relationship network construction module is used to construct a local call relationship network based on local user call records. The differential privacy protection execution module is used to receive global call relationship weights from the server and perform differential privacy protection on the local call relationship network according to the global call relationship weights. The local user call record update response module is used to respond to the update of the local user call record, calculate the local update amount of the edge weight in the local call relationship network, divide the local update amount according to other operator nodes with cross-network communication relationship through a secret sharing mechanism, send the divided secret share to the corresponding operator node, and send an update instruction to the server. The local aggregation share collection instruction response module is used to respond to the local aggregation share collection instruction, aggregate all received secret shares to obtain the local aggregation share, and return the local aggregation share to the server so that the server can update the global call relationship weight.

[0026] Fifthly, embodiments of the present invention provide a call privacy protection device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the call privacy protection method as described above.

[0027] In a sixth aspect, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the call privacy protection method as described above.

[0028] In a seventh aspect, embodiments of the present invention provide a computer program product, the computer program product including a computer program or computer instructions, wherein when the computer program or computer instructions are executed by a processor, the call privacy protection method described above is performed.

[0029] Compared with existing technologies, the call privacy protection method, system, device, medium and product disclosed in this invention ensures that the user's call privacy is reliably protected throughout the analysis process through the collaboration between the server and the operator node. At the same time, the federated aggregation mode built through trusted cross-network communication relationships can effectively reduce the server's computing load and improve the efficiency of call privacy protection. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the steps of a call privacy protection method applied to a server, as provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the steps of a call privacy protection method applied to an operator node according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a call privacy protection system applied to a server, provided by an embodiment of the present invention; Figure 4This is a schematic diagram of a call privacy protection system applied to an operator node according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a call privacy protection device provided in an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] In the description and claims, it should be understood that the terms "first," "second," etc., used in the description and claims are only for the purpose of distinguishing the description of the same technical features, and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated, nor necessarily the order of description or chronological order. The terms are interchangeable where appropriate. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature.

[0033] In modern telecommunications networks, user call behavior constitutes a complex and dynamic network of relationships. The actual data storage and processing architecture of this network is highly distributed. During deployment, operators divide their networks into several operating regions based on geographical areas or functional logic. Within each operating region, operator nodes are set up to store and process the call records of users within their jurisdiction. These records form a local call relationship network. Generally, when users under the jurisdiction of different operator nodes make cross-network calls, the call record is simultaneously recorded in all of those operator nodes.

[0034] During the collaboration process, to protect the data privacy of each operator node and user, the direct sharing and aggregation of raw user data are strictly limited. Currently, mainstream differential privacy protection schemes include using a general Laplace mechanism to add differential noise to the data, combined with standard federated learning algorithms to solve the data silo problem between operator nodes.

[0035] However, an analysis of the current application process of the federated averaging algorithm revealed that attackers can still deduce sensitive information in the data by analyzing the weight updates uploaded by the operator nodes.

[0036] Based on the above considerations, this embodiment of the invention provides a method for protecting call privacy in a server, wherein the server is configured to communicate with an operator node, and the operator node stores a local call relationship network built based on local user call records.

[0037] Please see Figure 1 In this embodiment of the invention, the call privacy protection method is specifically executed through steps S1 to S3: S1. Initialize the global call relationship weight and distribute the global call relationship weight to each of the operator nodes so that each operator node can perform differential privacy protection on the local call relationship network according to the global call relationship weight. S2. In response to the update command, collect the local aggregate share of each operator node; the local aggregate share of an operator node is the sum of all secret shares received from other operator nodes; the secret share is obtained by the operator node based on the trusted cross-network communication relationship and using the secret sharing mechanism to divide the local update amount of the edge weight in its local call relationship network. S3. Calculate the global update amount based on all the local aggregation shares and update the global call relationship weight, then send the updated global call relationship weight to each of the operator nodes.

[0038] The global call relationship weight is a core parameter maintained by the server. It reflects the consensus knowledge about the call relationships of all users across the network under a specific business scenario, learned from data from all operator nodes. The server distributes the global call relationship weight to each operator node, which can serve as a unified benchmark for local calculations by the operator nodes.

[0039] It is worth noting that the communication relationship network has small-world and scale-free properties. The traditional federated averaging algorithm uses a star topology, which forces all operator nodes to communicate indirectly through the server, while ignoring the natural connection between operator nodes based on actual call services. Existing technologies directly apply the traditional federated averaging algorithm to the communication relationship network, without making full use of the characteristics of the communication relationship network, resulting in low aggregation efficiency.

[0040] In this embodiment of the invention, a secure communication channel is established between operator nodes that have feasible cross-network communication relationships to exchange secret shares, thereby overlaying the federated learning communication network on top of the real call network topology.

[0041] Through a secret sharing mechanism, operator nodes use the local update amount as the secret value and divide it into multiple secret shares using mathematical methods. A single secret share cannot reconstruct the secret information; only by aggregating a sufficient number of shares can the original secret value be reconstructed. For a single operator node, it receives secret shares from other operator nodes, and it adds these secret shares locally to obtain the local aggregated share. A single local aggregated share alone cannot reconstruct the secret.

[0042] The server can obtain the overall call relationship update by collecting the local aggregate share of all operator nodes without accessing any plaintext update data. Subsequently, the server calculates the global update data using a reconstruction algorithm. This embodiment of the invention cuts off the attacker's direct access to local update data, offering higher privacy and security compared to traditional federated averaging schemes that directly upload plaintext update data.

[0043] Under the iterative federated aggregation architecture provided in this embodiment of the invention, the global call relationship weights can be continuously learned and dynamically updated as new call data is generated, ensuring the timeliness and accuracy of the business task analysis results.

[0044] In the above scheme, the collaboration between the server and the operator node ensures that the user's call privacy is reliably protected throughout the analysis process. At the same time, the federated aggregation mode built through trusted cross-network communication relationships can effectively reduce the server's computational load and improve the efficiency of call privacy protection.

[0045] As a preferred implementation, step S1, initializing the global call relationship weights and distributing the global call relationship weights to each of the operator nodes, enables each operator node to perform differential privacy protection on the local call relationship network based on the global call relationship weights, including: The global call relationship weights are initialized based on the preset initial weights of the business scenario. The global call relationship weights are distributed to each of the operator nodes, so that each operator node can calculate the sensitivity of the local call edge in the local call relationship network based on the global call relationship weights, and add adaptive first differential noise to the edge weight of the local call edge in the local call relationship network based on the sensitivity.

[0046] In some preferred embodiments, the business scenarios include fraud network identification, user relationship strength analysis, abnormal behavior pattern discovery, and business recommendation optimization. The data features to be extracted and the privacy protection requirements differ under different business scenarios. For example, the business task in the fraud network identification scenario is to identify the call network structure of telecommunications fraud gangs. The data characteristics include high-frequency short-duration calls, one-to-many call patterns, and activity during abnormal time periods. The privacy protection requirements are to strictly protect the call privacy of normal users, while allowing for a moderate sacrifice of the privacy of fraudulent users.

[0047] The business task of the user relationship strength analysis scenario is to analyze the strength of real social relationships between users. Data characteristics include call duration, frequency, time regularity and bidirectionality. Privacy protection requirements include protecting the privacy of individual social relationships, while also supporting group relationship analysis.

[0048] The business task of the abnormal behavior pattern discovery scenario is to discover abnormal patterns in user call behavior. Data characteristics include sudden changes in call patterns, abnormal geographical locations, and abnormal device switching. The privacy protection requirement is to strongly protect user location and device information.

[0049] The business task in the business recommendation optimization scenario is to make personalized business recommendations based on call relationships. Data features include inferences about the occupation, age, and consumption habits of the call participants. The privacy protection requirement is to strictly protect the privacy of user profiles.

[0050] In this embodiment of the invention, different preset initial weights are set for each business scenario to initialize the global call relationship weights. The preset initial weights carry the prior knowledge of the business scenario. After the preset initial weights are sent to the operator nodes, the consistency of the local calculations of the operator nodes is ensured.

[0051] From the perspective of the operator nodes, after receiving the global call relationship weight, each operator node will use it as a benchmark to assess the privacy leakage risk, i.e., sensitivity, of its own local call edge. Then, based on the sensitivity, it will dynamically add differential noise to the edge weight of each call relationship edge to match it.

[0052] In a preferred embodiment of the present invention, the operator node applies stronger protection to highly sensitive call relationship edges and weaker protection to other highly sensitive call relationship edges. More preferably, the strength of privacy protection is achieved through different differential privacy algorithms.

[0053] In the above scheme, the server initializes and distributes global call relationship weights according to the business scenario, which ensures the consistency of goals and startup efficiency of each operator node from the top-level design. Furthermore, the operator node will calculate the sensitivity of the call relationship edge based on the global call relationship weight and adaptively perform differential privacy, which can achieve a fine-grained trade-off between privacy and utility.

[0054] In one preferred embodiment, the update instructions include periodic update instructions from the server itself and local call relationship network update instructions from any operator node.

[0055] The periodic update command is automatically generated by the server itself based on a preset time policy. For example, the preset time policy is to issue a periodic update command once every 24 hours.

[0056] In addition to periodic update commands, any operator node can also proactively send a local call relationship network update command to the server when it detects a change in its local call relationship network, which can effectively improve the response speed to local risks and threats.

[0057] In the above scheme, the update command is the command signal that triggers the entire federated aggregation process. By using periodic update commands and local call relationship network update commands, it is possible to capture and integrate normal call patterns and respond quickly to local time, thus ensuring the robustness and responsiveness of the federated aggregation process.

[0058] In a preferred implementation, step S3 involves calculating the global update amount based on all the local aggregation shares and updating the global call relationship weight. The updated global call relationship weight is then sent to each of the operator nodes, as executed through steps S31 to S34. S31. Based on all the local aggregation shares, reconstruct the global update amount of the global call relationship weight in plaintext; S32. Calculate the average update amount based on the global update amount and the number of operator nodes; S33. Update the global call relationship weight based on the average update amount; S34. The updated global call relationship weights are sent to each of the operator nodes so that the operator nodes can update the local call relationship network according to the updated global call relationship weights.

[0059] It should be noted that the local aggregated share uploaded by the operator node to the server is encrypted and contains global update information. The server can decrypt and synthesize the local aggregated share based on the same secret sharing mechanism as the operator node, and finally reconstruct and calculate the plaintext of the global update amount.

[0060] By averaging the reconstructed global update, the resulting average update can reflect the average direction and magnitude of changes in call relationships across the entire network.

[0061] In some preferred embodiments, the global update amount is obtained by plaintext reconstruction based on the local aggregate share at time t. Represented as: ; in, This represents the local aggregate share uploaded by operator node i to the server. This refers to the plaintext reconstruction algorithm under the aforementioned secret sharing mechanism. This refers to the number of operator nodes.

[0062] Based on this, the average update amount is expressed as: .

[0063] Using the average update amount to update the global call relationship weight can effectively mitigate the impact of abnormal data fluctuations from a single operator node on the global call relationship, ensuring the stability and reliability of global knowledge evolution.

[0064] The above scheme realizes the secure transformation and efficient application of distributed encrypted information to centralized reliable knowledge, enabling the global call relationship weight to continuously iterate and optimize itself based on decentralized local data under strict privacy protection.

[0065] Further, preferably, step S33, updating the global call relationship weight based on the average update amount, includes: Global differential privacy noise is added to the average update amount to obtain the average privacy update amount; The global call relationship weight is updated based on the average privacy update amount.

[0066] In this preferred embodiment, to protect the overall privacy of the aggregated call relationships, global differential privacy noise is further added to the average update amount.

[0067] It should be noted that the global differential privacy noise can be general Laplace differential noise, Gaussian differential noise, or mixed differential noise. The specific form of the differential noise does not affect the beneficial effects produced by the present invention.

[0068] In some preferred embodiments, the average number of privacy updates is expressed as: ; in, The mean is 0 and the variance is Global differential privacy noise obtained from Gaussian distribution sampling.

[0069] In this embodiment of the invention, the updated global call relationship weight is represented as follows: ; in, Let be the global call relationship weight at time t.

[0070] In the above scheme, noise is added to the average update amount to ensure that the updated global call relationship meets the strict differential privacy definition, ensuring that it cannot be used to reverse-engineer the original data of any individual participant, thereby further improving the performance of privacy protection.

[0071] In a preferred embodiment, the call privacy protection method further includes a method for responding to privacy budget requests, specifically executed through steps S41-S44: S41. In response to the privacy budget request, dynamically calculate the periodic privacy budget for the current call service cycle based on the global privacy budget; the privacy budget request is sent to the server by the operator node or server when generating differential noise; S42. Calculate the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario; S43. If the remaining task-level budget meets the response limit corresponding to the privacy budget application, then the response limit is deducted and a successful cancellation notification is returned. S44. Otherwise, return a reconciliation failure notification.

[0072] It should be noted that each differential privacy operation incurs a certain degree of privacy loss. As the accumulated loss increases, continuing to execute queries may result in insufficient actual privacy protection, compromising the security of differential privacy. Therefore, to ensure availability, the global privacy budget for the entire lifecycle is limited in the call privacy protection method provided in this invention.

[0073] A periodic privacy budget is a portion of the global privacy budget that is dynamically allocated for a specific call service period. For example, in some preferred embodiments, the runtime is divided into consecutive periods, such as "days" or "weeks." Each period receives a portion of the budget that is dynamically allocated from the global privacy budget.

[0074] In some preferred embodiments, when a call service period k begins, the periodic privacy budget for call service period k is calculated immediately, and the allocated portion is deducted from the global privacy budget. For example, this is represented as: ; in, The remaining global budget for the global privacy budget. The periodic privacy budget for call service period k.

[0075] Furthermore, if the periodic privacy budget is not fully consumed at the end of the call service period k, the remaining unused portion of the budget is reclaimed, processed, and added back to the global remaining budget pool.

[0076] By dynamically calculating the cyclical privacy budget, the system avoids exhausting all budget too quickly in the early stages of the lifecycle, ensuring that there are still available privacy resources to maintain operation at future points in time, thus providing continuous service assurance capabilities.

[0077] Within each call cycle, task-level budgets are set for different business scenarios to ensure the efficiency of configuring limited privacy resources and the output of business value across different services.

[0078] It should be noted that if the remaining task-level budget meets the response limit corresponding to the privacy budget application, the response limit can be allocated to the initiator of the privacy budget application, and the consumption of the global privacy budget, periodic privacy budget and remaining task-level budget can be updated synchronously.

[0079] In some preferred embodiments, the consumption of each call's service budget is recorded atomically to ensure that the total consumption does not exceed the global privacy budget.

[0080] In the above scheme, the optimal allocation of privacy resources is achieved through a dynamic allocation strategy, ensuring that the system maintains a provable privacy and security state while meeting long-term operational needs.

[0081] Further, preferably, step S41, in response to the privacy budget request, dynamically calculates the periodic privacy budget for the current call service cycle based on the global privacy budget, including: In response to the privacy budget request, calculate the remaining global budget based on the global privacy budget; Based on a preset time decay rate, the global remaining budget is dynamically allocated periodically to obtain the periodic privacy budget for the current call service cycle.

[0082] It should be noted that the preset time decay rate is a constant with a value in the range of (0,1), which is used to control the proportion of the budget allocated from the global remaining budget in each cycle.

[0083] In some preferred embodiments, the time decay rate is not entirely constant throughout the entire lifecycle; it is adjusted according to the execution of the call service. If the periodic privacy budget for the call service cycle is not fully consumed, the remaining budget is reinvested based on the call service status, and the time decay rate can be adjusted during the reinvestment process.

[0084] For example, if the system is in a low-load state for a long period of time, a portion of the recovered remaining budget can be returned to the global remaining budget pool and the time decay rate of subsequent cycles can be reduced to achieve longer-term planning.

[0085] Since the global remaining budget decreases with each cycle, the budget allocated from the global remaining budget for each cycle is also dynamically calculated. A time decay rate can transform a fixed global privacy budget into a sustainable, periodic budget stream that is smoothly released over time.

[0086] In some preferred embodiments, the periodic privacy budget for the call service period k is expressed as: ; in, For the remaining global budget, This is the preset time decay rate.

[0087] The time decay rate needs to be set according to the specific application scenario. The larger the preset time decay rate, the more budget is allocated in the early stages of the entire life cycle.

[0088] The above scheme enables the periodic privacy budget to decrease smoothly over time, thus reserving sufficient budget for future call relationship analysis and providing fundamental resource guarantees for long-term operation.

[0089] Preferably, step S42, calculating the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario, includes: Calculate the service demand level of the previous call service cycle, and calculate the priority score of the current service scenario in the current call service cycle based on the service demand level and the static priority of the current service scenario. Based on the priority score and the periodic privacy budget, calculate the remaining task-level budget corresponding to the current business scenario.

[0090] The business demand metric for the previous call cycle is the frequency or number of queries or computational tasks that require privacy budgets for a specific business scenario, reflecting the actual activity level of the business in history.

[0091] In some preferred embodiments, after normalizing the business demand degree of the previous call service cycle to obtain the demand weight, a priority score is calculated based on the demand weight and the priority of the current business scenario.

[0092] For example, the demand weight of business scenario i is represented as: ; in, The actual demand for business scenario i during the call service cycle k-1; Let $k$ be the set of tasks consisting of all call service scenarios that require budget allocation within the call service cycle $k$.

[0093] The priority score of business scenario i in call service cycle k is represented as follows: ; in, The preset balance coefficient, This represents the static priority of business scenario i.

[0094] The above solution introduces a dynamic demand awareness mechanism, which adaptively allocates the periodic privacy budget to different business tasks, thereby maximizing business support efficiency.

[0095] This invention also provides a method for protecting call privacy in operator nodes; please refer to [link to relevant documentation]. Figure 2 In this embodiment, the call privacy protection method is specifically executed through steps P1 to P4: P1. Construct a local call relationship network based on local user call records; P2. Receive the global call relationship weight from the server, and perform differential privacy protection on the local call relationship network according to the global call relationship weight; P3. In response to the update of local user call records, calculate the local update amount of edge weights in the local call relationship network, divide the local update amount according to the operator nodes with trusted cross-network communication relationships through a secret sharing mechanism, send the divided secret share to the corresponding operator node, and send an update instruction to the server. P4. In response to the local aggregated share collection command, aggregate all received secret shares to obtain the local aggregated share, and return the local aggregated share to the server so that the server can update the global call relationship weight.

[0096] The operator node stores the call records of users within its jurisdiction and builds a local call relationship network. The local call relationship network built in step P1 is the original sensitive data. The original local call relationship network will not be directly transmitted to other operator nodes or servers.

[0097] After receiving the global call relationship weights from the server, the operator node performs differential privacy processing on the local call relationship network to ensure that each operator node implements privacy protection from a unified global perspective. Moreover, this privacy protection process is executed locally, which can ensure the privacy of the original data.

[0098] When a local user's call record changes, the operator node calculates the local update amount of the edge weights and, based on trusted cross-network communication relationships, distributes the local update amount to multiple trusted operator nodes through secret sharing. During this process, no operator node or server can obtain the local update amount independently. At the same time, the distribution based on trusted relationships greatly optimizes communication efficiency.

[0099] In addition to sending secret shares, operator nodes can also receive secret shares from other operator nodes. They then aggregate all received secret shares and send them to the server. At this point, the local aggregated shares received by the server are intermediate aggregation results that cannot be used to deduce the specific information of the operator nodes, rather than the original updates from itself or other operator nodes, which greatly enhances the privacy of the entire aggregation process.

[0100] In the above scheme, the collaboration between the server and the operator node ensures that the user's call privacy is reliably protected throughout the analysis process. At the same time, the federated aggregation mode built through trusted cross-network communication relationships can effectively reduce the server's computational load and improve the efficiency of call privacy protection.

[0101] As a preferred implementation, step P1, constructing a local call relationship network based on local user call records, is executed through steps P11 to P14: P11. Based on local user call records, obtain the users, the call relationships between users, and the call characteristics between users; P12. Using the user as a local call node, construct a local call edge based on the call relationship between the users; P13. Calculate the edge weight of the local call edge based on the call characteristics between the users. P14. Construct a local call relationship network based on the local call node, the local call edge, and the edge weight.

[0102] In this embodiment of the invention, the elements of the local call relationship network include local call nodes, local call edges, and edge weights. Local call nodes are used to identify user entities; local call edges connect local call nodes and are used to define and record whether there is a direct call connection between users; edge weights are a numerical representation of the strength of the call relationship, providing a core quantitative basis for accurate data analysis.

[0103] In the above scheme, by extracting graph structure elements, the original local user call records can be transformed into a structured local call relationship network, which facilitates subsequent complex privacy-preserving collaborative calculations.

[0104] In some preferred embodiments, the call characteristics include call frequency, call duration, and call patterns.

[0105] Call frequency is the total number of calls between two local call nodes within a specific time window, and it is a core indicator of relationship activity. Call duration is the cumulative call duration or average word call duration between two local call nodes within a specific time window, and it reflects relationship depth. Call pattern is the regularity of the time distribution of call behavior between two local call nodes, and it reflects different call scenarios.

[0106] Further, preferably, step P13, calculating the edge weight of the local call edge based on the call characteristics between the users, includes: Obtain the preset frequency weighting coefficient, duration weighting coefficient, and pattern weighting coefficient; The call frequencies are weighted according to the frequency weighting coefficients to obtain the weighted call frequencies; The call duration is weighted according to the duration weighting coefficient to obtain the weighted call duration; The call patterns are weighted according to the aforementioned pattern weighting coefficients to obtain weighted call patterns; The weight of the local call edge is obtained by calculating the sum of the weighted call frequency, the weighted call duration, and the weighted call pattern.

[0107] It should be noted that in this embodiment of the invention, call frequency, call duration, and call pattern are all numerical variables. For example, the call time series is analyzed using the Fourier transform method to detect whether a periodic pattern exists, and the intensity of the corresponding periodic energy is used as the numerical value of the call pattern.

[0108] In some preferred embodiments, the edge weights between local call node i and local call node j are represented as follows: ; in, For frequency weighting coefficients, Let be the communication frequency between local call node i and local call node j. This is the duration weighting coefficient. Let be the duration of the call between local call node i and local call node j. For regularity weighting coefficients, The call pattern between local call node i and local call node j.

[0109] The above scheme calculates edge weights from three levels: call frequency, call duration, and call patterns, which enables the fusion of multi-dimensional features to build a more comprehensive relationship profile.

[0110] In a preferred implementation, step P2 involves receiving the global call relationship weights from the server and performing differential privacy protection on the local call relationship network based on these global call relationship weights, as executed through steps P21 to P23. P21. Receive the global call relationship weight from the server; P22. Calculate the sensitivity of the local call edge in the local call relationship network based on the global call relationship weight. P23. Based on the sensitivity, add adaptive first differential noise to the edge weights of the local call edges in the local call relationship network.

[0111] The sensitivity calculated by global call relationship weights can measure the potential privacy risk if the edge weights of local call edges change, and can quantify this risk as sensitivity, so as to determine how much noise needs to be added to the edge weights of local call edges based on the sensitivity.

[0112] It should be noted that the first differential noise is adaptively generated for the edge weight of each local call edge, which can provide stronger protection for edges with high privacy risks and weaker protection for edges with low privacy risks, thereby improving the efficiency of privacy budget utilization.

[0113] The above scheme constructs a refined local differential privacy protection mechanism based on global guidance, which enables operator nodes to complete the source privacy processing of local sensitive data in a non-uniform manner with maximum efficiency, while synchronizing with global knowledge. This provides high-quality and privacy-secure local update outputs for subsequent secure collaborative computing.

[0114] Further, preferably, step P22, calculating the sensitivity of the local call edge in the local call relationship network based on the global call relationship weight, includes: Based on the business scenario, obtain the preset basic sensitivity for the local call side in the local call relationship network; Based on the global call relationship weights, global call features are obtained; Calculate the feature adjustment coefficient of the basic sensitivity based on the global call characteristics; The sensitivity of the local call edge in the local call relationship network is calculated based on the feature adjustment coefficient and the basic sensitivity.

[0115] The basic sensitivity is related to the privacy protection requirements of the business scenario, reflecting the basic privacy risk level of the local call side under static business policies.

[0116] For example, in the scenario of fraud network identification, the basic sensitivity of normal users is 0.2, and the basic sensitivity of potential users is 0.1; in the scenario of user relationship strength analysis, the basic sensitivity of individual relationships is 0.2, and the basic sensitivity of group relationships is 0.15; in the scenario of abnormal behavior pattern discovery, the basic sensitivity of location-related data is 0.25, and the basic sensitivity of device-related data is 0.2; in the scenario of business recommendation optimization, the basic sensitivity of user profile data is 0.3.

[0117] It is understood that the above basic sensitivity settings for specific business scenarios are only examples. In actual applications, the value can be adjusted based on different business scenarios and business needs. The specific value of the basic sensitivity does not affect the beneficial effects produced by the embodiments of the present invention.

[0118] The global call relationship weight includes an indicator that reflects the macro-statistical characteristics of call relationships across the entire network, namely the global call characteristics. By adjusting the basic sensitivity through the global call characteristics, global statistical information can be incorporated into the calculation of local sensitivity.

[0119] In some preferred embodiments, the feature adjustment coefficients include call frequency adjustment coefficients, call duration adjustment coefficients, and time attenuation adjustment coefficients.

[0120] The call frequency adjustment factor is calculated based on the average value of the global call frequency and is expressed as: ; in, For local call nodes call frequency, For local call nodes call frequency, This represents the average global call frequency.

[0121] The call duration adjustment factor is calculated based on the average of the global call duration, and is expressed as follows: ; in, For local call nodes and local call nodes Average call duration between them This represents the average total call duration.

[0122] The time decay adjustment factor is calculated based on the times of adjacent calls and is expressed as follows: ; in, The current time of the call. This is the time of the last call.

[0123] The sensitivity of the local call side is represented as follows: .

[0124] In the above scheme, by calculating the sensitivity based on the global call relationship weight, the noise added to the local call relationship network can be adapted to the data distribution characteristics of the entire network, thus realizing refined and adaptive privacy protection under the global collaborative framework.

[0125] Preferably, step P23, adding adaptive first differential noise to the edge weights of local call edges in the local call relationship network according to the sensitivity, includes: If the sensitivity is not greater than a preset first sensitivity threshold, then Laplace differential noise is generated based on the sensitivity to obtain the first noise; If the sensitivity is greater than a preset second sensitivity threshold, then Gaussian difference noise is generated based on the sensitivity to obtain the second noise; Otherwise, based on the aforementioned sensitivity, a mixture of Laplace difference noise and Gaussian difference noise is generated to obtain a third noise; The generated first noise, second noise, or third noise is added to the edge weight of the local call edge in the local call relationship network.

[0126] In this embodiment of the invention, the sensitivity is divided into three intervals by using a first sensitivity threshold and a second sensitivity threshold, each corresponding to a different noise generation strategy.

[0127] Laplace differential noise is suitable for scenarios with extremely high privacy requirements and where large statistical errors are permissible. Therefore, when the sensitivity is no greater than a preset first sensitivity threshold, using Laplace differential noise may provide stronger privacy protection under the same privacy budget.

[0128] In some preferred embodiments, the first noise is represented as: ; in, Budget for privacy.

[0129] Difference-of-Gaussian noise typically provides better data utility while allowing for a very low probability of privacy failure, making it suitable for adding noise to highly sensitive data. Therefore, when the sensitivity exceeds a preset second sensitivity threshold, the overall statistical utility of the data is maintained while protecting privacy.

[0130] Preferably, the second noise is expressed as: ; in, For privacy failure probability.

[0131] When the sensitivity is greater than the preset first sensitivity threshold but not greater than the preset second sensitivity threshold, the sensitivity level is medium. Therefore, mixed noise is used to balance the advantages of Laplace difference noise and Gaussian difference noise to achieve a compromise protection effect.

[0132] Preferably, the third noise is represented as: ; in, This is the preset mixing coefficient.

[0133] In the above scheme, the privacy budget is allocated on demand by setting a sensitivity threshold for routing the edge weight noise-adding strategy.

[0134] In a preferred implementation, step P3, in response to the update of local user call records, calculates the local update amount of the edge weights in the local call relationship network, and, based on the operator nodes with trusted cross-network communication relationships, divides the local update amount through a secret sharing mechanism, sends the divided secret shares to the corresponding operator nodes, and sends an update instruction to the server, including: In response to an update of the local user's call record, calculate the local update amount of the edge weights in the local call relationship network; A second differential noise is added to the local update amount to obtain the local privacy update amount; Based on trusted operator nodes with trusted cross-network communication relationships, the local privacy update amount is divided into N secret shares using a preset secret sharing algorithm; N is the number of trusted operator nodes. The secret share is sent to the corresponding trusted operator node; the secret share and the trusted operator node are in one-to-one correspondence. Send an update command to the server.

[0135] When a local user's call record is updated, the operator's node automatically updates the local call relationship network, thereby obtaining the local update amount of the edge weights. The local update amount is an update of the original data information, so when synchronizing the local update amount to the server, a second differential noise needs to be added first to convert it into a local privacy update amount.

[0136] Under the federated aggregation architecture provided by this invention, the embodiments of this invention further use a secret sharing algorithm in cryptography to divide the local privacy update into several shares, and distribute them to each trusted operator node point-to-point through a secure channel, so as to achieve secure distribution and controllable propagation of information.

[0137] In the above scheme, when the operator node detects local changes, it can independently complete the calculation of the local privacy update. Based on a decentralized trust mechanism, the local privacy update is distributed to multiple operator nodes, and a secure calculation is completed together through a cryptographic protocol, which greatly enhances the robustness and security of the system.

[0138] As a preferred embodiment, the call privacy protection method further includes: Obtain the current cross-network communication relationship and cross-network communication frequency between the first operator node and other second operator nodes; the second operator node is configured to communicate with the server; If the first operator node and the second operator node have a cross-network communication relationship, and the cross-network communication frequency is greater than a preset frequency threshold, then the second operator node is regarded as a trusted operator node of the first operator node.

[0139] In the above scheme, the reliability of communication between operator nodes is quantified by communication frequency, and the secret share is divided only for trusted operator nodes, which improves the reliability of the federated aggregation architecture.

[0140] In a preferred implementation, step P4, in response to the local aggregated share collection instruction, aggregates all received secret shares to obtain a local aggregated share, and returns the local aggregated share to the server so that the server updates the global call relationship weight, includes: In response to the local aggregated share collection instruction, all received secret shares are aggregated to obtain the local aggregated share; the local aggregated share collection instruction is issued by the server to all operator nodes after receiving the update instruction; The local aggregate share is returned to the server so that the server can update the global call relationship weight.

[0141] After receiving the local aggregation share collection instruction from the server, the operator node immediately completes the pre-aggregation of the secret share locally, and then compresses the encrypted information from multiple operator nodes into a local aggregation share and sends it to the server. In this way, the server can perform global updates by obtaining the local aggregation shares of each operator node, while not being able to see the specific contribution of any individual operator node.

[0142] In the above scheme, the secret share of distributed storage is transformed into a local aggregate share that can be used by the server to securely reconstruct global knowledge. This can efficiently drive the update of global knowledge without leaking the individual privacy of a single operator node.

[0143] As a preferred embodiment, the call privacy protection method further includes: In response to its own differential noise generation request, it generates a privacy budget request; The privacy budget request is sent to the server, and differential noise is generated based on the reconciliation success notification returned by the server.

[0144] Because the global privacy budget is global and non-renewable, allowing operator nodes to consume the budget on their own will inevitably lead to its abuse, causing the overall privacy protection commitment to fail in the later stages.

[0145] In this embodiment of the invention, before the operator node generates the first differential noise and the second differential noise, it first generates a differential noise generation request internally, then creates a privacy budget application and sends it to the server for approval. When the server approves the application, it returns a successful cancellation notification. At this time, the operator node can generate differential noise based on the applied budget.

[0146] The above solution standardizes the management of differential noise addition, fundamentally preventing the overspending of the privacy budget and ensuring that the system always operates within a secure privacy boundary.

[0147] The call privacy protection method provided by this invention ensures reliable protection of user call privacy throughout the analysis process through collaboration between the server and operator nodes. At the same time, the federated aggregation mode built through trusted cross-network communication relationships can effectively reduce the server's computational load and improve the efficiency of call privacy protection.

[0148] This invention provides a call privacy protection system applied in a server. The server is configured to communicate with an operator node, which stores a local call relationship network built based on local user call records. See also... Figure 3 The call privacy protection system 1 includes a global call relationship weight initialization module 11, an update command response module 12, and a global call relationship weight update module 13, wherein: The global call relationship weight initialization module 11 is used to initialize the global call relationship weight and distribute the global call relationship weight to each of the operator nodes, so that each operator node can perform differential privacy protection on the local call relationship network according to the global call relationship weight. The update instruction response module 12 is used to respond to the update instruction and collect the local aggregate share of each operator node; the local aggregate share of the operator node is the sum of all secret shares received from other operator nodes; the secret share is obtained by the operator node based on the trusted cross-network communication relationship and using the secret sharing mechanism to divide the local update amount of the edge weight in its local call relationship network. The global call relationship weight update module 13 is used to calculate the global update amount and update the global call relationship weight based on all the local aggregation shares, and send the updated global call relationship weight to each of the operator nodes.

[0149] In a preferred embodiment, the global call relationship weight initialization module 11 is specifically used for: The global call relationship weights are initialized based on the preset initial weights of the business scenario. The global call relationship weights are distributed to each of the operator nodes, so that each operator node can calculate the sensitivity of the local call edge in the local call relationship network based on the global call relationship weights, and add adaptive first differential noise to the edge weight of the local call edge in the local call relationship network based on the sensitivity.

[0150] In one preferred embodiment, the update instructions include periodic update instructions from the server itself and local call relationship network update instructions from any operator node.

[0151] In a preferred embodiment, the global call relationship weight update module 13 includes: The global update quantity reconstruction unit is used to reconstruct the global update quantity of the global call relationship weight in plaintext based on all the local aggregation shares. The average update volume calculation unit is used to calculate the average update volume based on the global update volume and the number of operator nodes; A global call relationship weight update unit is used to update the global call relationship weight according to the average update amount; The global call relationship weight distribution unit is used to send the updated global call relationship weight to each of the operator nodes, so that the operator nodes can update the local call relationship network according to the updated global call relationship weight.

[0152] Further, preferably, the average update calculation unit is specifically used for: Global differential privacy noise is added to the average update amount to obtain the average privacy update amount; The global call relationship weight is updated based on the average privacy update amount.

[0153] In a preferred embodiment, the call privacy protection system 1 further includes a privacy budget request response module, used for: In response to a privacy budget request, the periodic privacy budget for the current call service cycle is dynamically calculated based on the global privacy budget; the privacy budget request is sent to the server by the operator node or server when generating differential noise; Calculate the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario; If the remaining task-level budget meets the response limit corresponding to the privacy budget application, then the response limit will be deducted, and a successful reimbursement notification will be returned. Otherwise, a reconciliation failure notification will be returned.

[0154] Further, preferably, the step of dynamically calculating the periodic privacy budget for the current call service cycle in response to a privacy budget request, based on the global privacy budget, includes: In response to the privacy budget request, calculate the remaining global budget based on the global privacy budget; Based on a preset time decay rate, the global remaining budget is dynamically allocated periodically to obtain the periodic privacy budget for the current call service cycle.

[0155] Preferably, calculating the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario includes: Calculate the service demand level of the previous call service cycle, and calculate the priority score of the current service scenario in the current call service cycle based on the service demand level and the static priority of the current service scenario. Based on the priority score and the periodic privacy budget, calculate the remaining task-level budget corresponding to the current business scenario.

[0156] This invention provides a call privacy protection system applied in a carrier node, wherein the carrier node is configured to communicate with a server and other carrier nodes. Please refer to... Figure 4 The call privacy protection system 2 includes a local call relationship network construction module 21, a differential privacy protection execution module 22, a local user call record update response module 23, and a local aggregation share collection instruction response module 24, wherein: The local call relationship network construction module 21 is used to construct a local call relationship network based on local user call records; The differential privacy protection execution module 22 is used to receive global call relationship weights from the server and perform differential privacy protection on the local call relationship network according to the global call relationship weights. The local user call record update response module 23 is used to respond to the update of the local user call record, calculate the local update amount of the edge weight in the local call relationship network, divide the local update amount according to other operator nodes with cross-network communication relationship through a secret sharing mechanism, send the divided secret share to the corresponding operator node, and send an update instruction to the server. The local aggregation share collection instruction response module 24 is used to respond to the local aggregation share collection instruction, aggregate all received secret shares to obtain the local aggregation share, and return the local aggregation share to the server so that the server can update the global call relationship weight.

[0157] In a preferred embodiment, the local call relationship network construction module 21 is specifically used for: Based on local user call records, we can obtain information about users, call relationships between users, and call characteristics between users. The user is used as a local call node, and a local call edge is constructed based on the call relationship between the users; Calculate the edge weight of the local call edge based on the call characteristics between the users; A local call relationship network is constructed based on the local call node, the local call edge, and the edge weight.

[0158] Further, preferably, the call characteristics include call frequency, call duration, and call patterns; then, calculating the edge weight of the local call edge based on the call characteristics between users includes: Obtain the preset frequency weighting coefficient, duration weighting coefficient, and pattern weighting coefficient; The call frequencies are weighted according to the frequency weighting coefficients to obtain the weighted call frequencies; The call duration is weighted according to the duration weighting coefficient to obtain the weighted call duration; The call patterns are weighted according to the aforementioned pattern weighting coefficients to obtain weighted call patterns; The weight of the local call edge is obtained by calculating the sum of the weighted call frequency, the weighted call duration, and the weighted call pattern.

[0159] In a preferred embodiment, the differential privacy protection execution module 22 includes: The global call relationship weight receiving unit is used to receive the global call relationship weight from the server. A sensitivity calculation unit is used to calculate the sensitivity of the local call edge in the local call relationship network based on the global call relationship weight. The first differential noise addition unit is used to add adaptive first differential noise to the edge weights of the local call edges in the local call relationship network according to the sensitivity.

[0160] Further, preferably, the sensitivity calculation unit is specifically used for: Based on the business scenario, obtain the preset basic sensitivity for the local call side in the local call relationship network; Based on the global call relationship weights, global call features are obtained; Calculate the feature adjustment coefficient of the basic sensitivity based on the global call characteristics; The sensitivity of the local call edge in the local call relationship network is calculated based on the feature adjustment coefficient and the basic sensitivity.

[0161] Preferably, the first differential noise enhancement unit is specifically used for: If the sensitivity is not greater than a preset first sensitivity threshold, then Laplace differential noise is generated based on the sensitivity to obtain the first noise; If the sensitivity is greater than a preset second sensitivity threshold, then Gaussian difference noise is generated based on the sensitivity to obtain the second noise; Otherwise, based on the aforementioned sensitivity, a mixture of Laplace difference noise and Gaussian difference noise is generated to obtain a third noise; The generated first noise, second noise, or third noise is added to the edge weight of the local call edge in the local call relationship network.

[0162] In a preferred embodiment, the local user call record update response module 23 is specifically used for: In response to an update of the local user's call record, calculate the local update amount of the edge weights in the local call relationship network; A second differential noise is added to the local update amount to obtain the local privacy update amount; Based on trusted operator nodes with trusted cross-network communication relationships, the local privacy update amount is divided into N secret shares using a preset secret sharing algorithm; N is the number of trusted operator nodes. The secret share is sent to the corresponding trusted operator node; the secret share and the trusted operator node are in one-to-one correspondence. Send an update command to the server.

[0163] In a preferred embodiment, the call privacy protection system 2 further includes a trusted operator node determination module, used for: Obtain the current cross-network communication relationship and cross-network communication frequency between the first operator node and other second operator nodes; the second operator node is configured to communicate with the server; If the first operator node and the second operator node have a cross-network communication relationship, and the cross-network communication frequency is greater than a preset frequency threshold, then the second operator node is regarded as a trusted operator node of the first operator node.

[0164] In a preferred embodiment, the local aggregation share acquisition instruction response module 24 includes: In response to the local aggregated share collection instruction, all received secret shares are aggregated to obtain the local aggregated share; the local aggregated share collection instruction is issued by the server to all operator nodes after receiving the update instruction; The local aggregate share is returned to the server so that the server can update the global call relationship weight.

[0165] In a preferred embodiment, the call privacy protection system 2 further includes a budget request module for: In response to its own differential noise generation request, it generates a privacy budget request; The privacy budget request is sent to the server, and differential noise is generated based on the reconciliation success notification returned by the server.

[0166] The call privacy protection system provided by this invention can reliably protect users' call privacy throughout the analysis process through cooperation between the server and operator nodes. At the same time, the federated aggregation mode built through trusted cross-network communication relationships can effectively reduce the server's computing load and improve the efficiency of call privacy protection.

[0167] Please see Figure 5 , Figure 5 This is a structural block diagram of a call privacy protection device provided in an embodiment of the present invention. The call privacy protection device includes a processor 31, a memory 32, and a computer program stored in the memory 32 and executable on the processor 31. When the processor 31 executes the computer program, it implements the steps in the above-described embodiments of the call privacy protection methods, such as steps S1-S3 or steps P1-P4.

[0168] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 31 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the call privacy protection device.

[0169] The call privacy protection device may include, but is not limited to, a processor 31 and a memory 32. Those skilled in the art will understand that the schematic diagram is merely an example of a call privacy protection device and does not constitute a limitation on the device. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the call privacy protection device may also include input / output devices, network access devices, buses, etc.

[0170] The processor 31 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 31 is the control center of the call privacy protection device, connecting various parts of the device via various interfaces and lines.

[0171] The memory 32 can be used to store the computer programs and / or modules. The processor 31 implements various functions of the call privacy protection device by running or executing the computer programs and / or modules stored in the memory 32 and calling the data stored in the memory 32. The memory 32 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 32 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0172] Wherein, if the modules / units integrated into the call privacy protection device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor 31, it can implement the steps of the various method embodiments described above. Wherein, the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0173] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for protecting call privacy, characterized in that, Applied in a server, the server is configured to communicate with an operator node, which stores a local call relationship network built based on local user call records. The call privacy protection method includes: Initialize the global call relationship weight and distribute the global call relationship weight to each of the operator nodes so that each operator node can perform differential privacy protection on the local call relationship network according to the global call relationship weight; In response to the update command, the local aggregate share of each operator node is collected; the local aggregate share of an operator node is the sum of all secret shares received from other operator nodes; the secret share is obtained by the operator node based on the trusted cross-network communication relationship and using the secret sharing mechanism to divide the local update amount of the edge weight in its local call relationship network. Based on all the local aggregation shares, calculate the global update amount and update the global call relationship weight, and send the updated global call relationship weight to each of the operator nodes.

2. The method for protecting call privacy as described in claim 1, characterized in that, The initialization of global call relationship weights, and the distribution of these global call relationship weights to each of the operator nodes, enable each operator node to perform differential privacy protection on its local call relationship network based on these global call relationship weights, including: The global call relationship weights are initialized based on the preset initial weights of the business scenario. The global call relationship weights are distributed to each of the operator nodes, so that each operator node can calculate the sensitivity of the local call edge in the local call relationship network based on the global call relationship weights, and add adaptive first differential noise to the edge weight of the local call edge in the local call relationship network based on the sensitivity.

3. The method for protecting call privacy as described in claim 1, characterized in that, The update instructions include periodic update instructions from the server itself and local call relationship network update instructions from any operator node.

4. The method for protecting call privacy as described in claim 1, characterized in that, The step of calculating the global update amount and updating the global call relationship weight based on all the local aggregation shares, and sending the updated global call relationship weight to each of the operator nodes includes: Based on all the aforementioned local aggregation shares, the global update amount of the global call relationship weight is reconstructed in plaintext; Calculate the average update volume based on the global update volume and the number of operator nodes; The global call relationship weight is updated based on the average update amount; The updated global call relationship weights are sent to each of the operator nodes so that the operator nodes can update their local call relationship networks based on the updated global call relationship weights.

5. A method for protecting call privacy as described in claim 4, characterized in that, The step of updating the global call relationship weight based on the average update amount includes: Global differential privacy noise is added to the average update amount to obtain the average privacy update amount; The global call relationship weight is updated based on the average privacy update amount.

6. The method for protecting call privacy as described in claim 1, characterized in that, The method for protecting call privacy also includes: In response to a privacy budget request, the periodic privacy budget for the current call service cycle is dynamically calculated based on the global privacy budget; the privacy budget request is sent to the server by the operator node or server when generating differential noise; Calculate the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario; If the remaining task-level budget meets the response limit corresponding to the privacy budget application, then the response limit will be deducted, and a successful reimbursement notification will be returned. Otherwise, a reconciliation failure notification will be returned.

7. A method for protecting call privacy as described in claim 6, characterized in that, In response to a privacy budget request, the periodic privacy budget for the current call service cycle is dynamically calculated based on the global privacy budget, including: In response to the privacy budget request, calculate the remaining global budget based on the global privacy budget; Based on a preset time decay rate, the global remaining budget is dynamically allocated periodically to obtain the periodic privacy budget for the current call service cycle.

8. A method for protecting call privacy as described in claim 6, characterized in that, The step of calculating the remaining task-level budget based on the periodic privacy budget and the static priority of the current business scenario includes: Calculate the service demand level of the previous call service cycle, and calculate the priority score of the current service scenario in the current call service cycle based on the service demand level and the static priority of the current service scenario. Based on the priority score and the periodic privacy budget, calculate the remaining task-level budget corresponding to the current business scenario.

9. A method for protecting call privacy, characterized in that, Applied to operator nodes, which are configured to communicate with servers and other operator nodes, the call privacy protection method includes: Construct a local call relationship network based on local user call records; Receive global call relationship weights from the server, and perform differential privacy protection on the local call relationship network based on the global call relationship weights; In response to the update of local user call records, the local update amount of the edge weights in the local call relationship network is calculated. Based on the operator nodes with trusted cross-network communication relationships, the local update amount is divided through a secret sharing mechanism. The divided secret shares are sent to the corresponding operator nodes, and an update instruction is sent to the server. In response to the local aggregated share collection command, all received secret shares are aggregated to obtain the local aggregated share, and the local aggregated share is returned to the server so that the server can update the global call relationship weight.

10. A method for protecting call privacy as described in claim 9, characterized in that, The step of constructing a local call relationship network based on local user call records includes: Based on local user call records, we can obtain information about users, call relationships between users, and call characteristics between users. The user is used as a local call node, and a local call edge is constructed based on the call relationship between the users; Calculate the edge weight of the local call edge based on the call characteristics between the users; A local call relationship network is constructed based on the local call node, the local call edge, and the edge weight.

11. A method for protecting call privacy as described in claim 10, characterized in that, The call characteristics include call frequency, call duration, and call patterns; The step of calculating the edge weight of the local call edge based on the call characteristics between the users includes: Obtain the preset frequency weighting coefficient, duration weighting coefficient, and pattern weighting coefficient; The call frequencies are weighted according to the frequency weighting coefficients to obtain the weighted call frequencies; The call duration is weighted according to the duration weighting coefficient to obtain the weighted call duration; The call patterns are weighted according to the aforementioned pattern weighting coefficients to obtain weighted call patterns; The weight of the local call edge is obtained by calculating the sum of the weighted call frequency, the weighted call duration, and the weighted call pattern.

12. A method for protecting call privacy as described in claim 9, characterized in that, The step of receiving global call relationship weights from the server and performing differential privacy protection on the local call relationship network based on the global call relationship weights includes: Receive global call relationship weights from the server; Based on the global call relationship weight, calculate the sensitivity of the local call edge in the local call relationship network; Based on the sensitivity, an adaptive first differential noise is added to the edge weights of the local call edges in the local call relationship network.

13. A method for protecting call privacy as described in claim 12, characterized in that, The step of calculating the sensitivity of the local call edge in the local call relationship network based on the global call relationship weight includes: Based on the business scenario, obtain the preset basic sensitivity for the local call side in the local call relationship network; Based on the global call relationship weights, global call features are obtained; Calculate the feature adjustment coefficient of the basic sensitivity based on the global call characteristics; The sensitivity of the local call edge in the local call relationship network is calculated based on the feature adjustment coefficient and the basic sensitivity.

14. A method for protecting call privacy as described in claim 12, characterized in that, The step of adding adaptive first differential noise to the edge weights of local call edges in the local call relationship network based on the sensitivity includes: If the sensitivity is not greater than a preset first sensitivity threshold, then Laplace differential noise is generated based on the sensitivity to obtain the first noise; If the sensitivity is greater than a preset second sensitivity threshold, then Gaussian difference noise is generated based on the sensitivity to obtain the second noise; Otherwise, based on the aforementioned sensitivity, a mixture of Laplace difference noise and Gaussian difference noise is generated to obtain a third noise; The generated first noise, second noise, or third noise is added to the edge weight of the local call edge in the local call relationship network.

15. A method for protecting call privacy as described in claim 9, characterized in that, In response to an update to the local user's call record, the local update amount of the edge weights in the local call relationship network is calculated. Based on the operator nodes with trusted cross-network communication relationships, the local update amount is divided through a secret sharing mechanism. The divided secret shares are sent to the corresponding operator nodes, and an update instruction is sent to the server, including: In response to an update of the local user's call record, calculate the local update amount of the edge weights in the local call relationship network; A second differential noise is added to the local update amount to obtain the local privacy update amount; Based on trusted operator nodes with trusted cross-network communication relationships, the local privacy update amount is divided into N secret shares using a preset secret sharing algorithm; N is the number of trusted operator nodes. The secret share is sent to the corresponding trusted operator node; the secret share and the trusted operator node are in one-to-one correspondence. Send an update command to the server.

16. A method for protecting call privacy as described in claim 9, characterized in that, The method for protecting call privacy also includes: Obtain the current cross-network communication relationship and cross-network communication frequency between the first operator node and other second operator nodes; the second operator node is configured to communicate with the server; If the first operator node and the second operator node have a cross-network communication relationship, and the cross-network communication frequency is greater than a preset frequency threshold, then the second operator node is regarded as a trusted operator node of the first operator node.

17. A method for protecting call privacy as described in claim 9, characterized in that, The step of responding to the local aggregated share collection command, aggregating all received secret shares to obtain a local aggregated share, and returning the local aggregated share to the server so that the server can update the global call relationship weight includes: In response to the local aggregated share collection instruction, all received secret shares are aggregated to obtain the local aggregated share; the local aggregated share collection instruction is issued by the server to all operator nodes after receiving the update instruction; The local aggregate share is returned to the server so that the server can update the global call relationship weight.

18. A method for protecting call privacy as described in claim 9, characterized in that, The method for protecting call privacy also includes: In response to its own differential noise generation request, it generates a privacy budget request; The privacy budget request is sent to the server, and differential noise is generated based on the reconciliation success notification returned by the server.

19. A call privacy protection system, characterized in that, The system is applied in a server configured to communicate with an operator node, which stores a local call relationship network built based on local user call records. The call privacy protection system includes: The global call relationship weight initialization module is used to initialize the global call relationship weight and distribute the global call relationship weight to each of the operator nodes, so that each operator node can perform differential privacy protection on the local call relationship network according to the global call relationship weight. The update command response module is used to respond to update commands and collect the local aggregate share of each operator node. The local aggregate share of an operator node is the sum of all secret shares received from other operator nodes. The secret share is obtained by the operator node based on trusted cross-network communication relationships and using a secret sharing mechanism to divide the local update amount of the edge weights in its local call relationship network. The global call relationship weight update module is used to calculate the global update amount and update the global call relationship weight based on all the local aggregation shares, and send the updated global call relationship weight to each of the operator nodes.

20. A call privacy protection system, characterized in that, The call privacy protection system is applied in operator nodes, which are configured to communicate with servers and other operator nodes. The system includes: The local call relationship network construction module is used to construct a local call relationship network based on local user call records. The differential privacy protection execution module is used to receive global call relationship weights from the server and perform differential privacy protection on the local call relationship network according to the global call relationship weights. The local user call record update response module is used to respond to the update of the local user call record, calculate the local update amount of the edge weight in the local call relationship network, divide the local update amount according to other operator nodes with cross-network communication relationship through a secret sharing mechanism, send the divided secret share to the corresponding operator node, and send an update instruction to the server. The local aggregation share collection instruction response module is used to respond to the local aggregation share collection instruction, aggregate all received secret shares to obtain the local aggregation share, and return the local aggregation share to the server so that the server can update the global call relationship weight.

21. A call privacy protection device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the call privacy protection method as described in any one of claims 1 to 8, or the call privacy protection method as described in any one of claims 9 to 18.

22. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the call privacy protection method as described in any one of claims 1 to 8, or the call privacy protection method as described in any one of claims 9 to 18.

23. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, perform the call privacy protection method as described in any one of claims 1 to 8, or the call privacy protection method as described in any one of claims 9 to 18.

Images