A federated learning content reach method and device based on data aggregation

Abstract

The application discloses a kind of based on data aggregation's federal learning content reach method and device, the method includes: each data party based on common key agreement executes first encryption operation and generates shared key;Second encryption operation is executed to the user identification information encryption;Using shared key, the third encryption operation is executed to the user identification information after encryption, obtains each ciphertext data set, third party data platform is the union set of each data party's ciphertext data set Output model training data set and carry out federal learning model training, based on federal learning model carries out content reach.This application makes each data party complete three encryption operations to respective user identification information locally, to ensure that the privacy security of each party data set when union set;Again through third party data platform, according to the union set of each data party's ciphertext data set, it is guaranteed that the scale of model training data set after alignment does not decrease sharply, to improve the accuracy of content reach, improve user experience.

Description

Technical Field

[0001] This invention relates to the field of federated learning technology, and more specifically, to a method, apparatus, and system for reaching federated learning content based on data aggregation. Background Technology

[0002] In machine learning modeling, sample data often involves data from various domains, such as personal identification information, financial consumption information, internet behavior information, and geolocation information. To protect user privacy, federated learning is typically used to aggregate data from these different domains for modeling, achieving the effect of "data usable but not visible."

[0003] In existing federated learning, the Private Set Intersection (PSI) method is commonly used to aggregate data. Two-party private intersection means that each data party possesses its own dataset; after federated computation, their intersection is obtained, but neither party knows the other's data beyond this intersection. Multi-party private intersection means that multiple data parties can obtain the common intersection of all their datasets, but no additional information beyond this common intersection is revealed. Since private intersection takes the intersection of all data parties' datasets, the dataset size will inevitably decrease after private intersection. Furthermore, the more data parties involved, the smaller the dataset size after intersection. This could result in a final dataset too small to train the model; or, even if a model is trained on a small dataset, its generalization ability will be poor, failing to achieve accurate predictions. Moreover, if the model's test set is also obtained through private intersection, there is no guarantee that a corresponding model score will be generated for the entire test set. This severely limits the model's application scope in deployment and affects content reach. Summary of the Invention

[0004] In view of this, the main objective of the present invention is to propose a method and apparatus for reaching federated learning content based on data aggregation, in order to at least partially solve at least one of the above-mentioned technical problems.

[0005] To address the aforementioned technical problems, the first aspect of this invention proposes a federated learning content outreach method based on data aggregation, the method comprising:

[0006] Each data party performs the first encryption operation based on a common key protocol to generate a shared key;

[0007] Each data provider performs a second encryption operation to encrypt its own user identification information;

[0008] Each data party uses the shared key to perform a third encryption operation on the encrypted user identification information to obtain the encrypted dataset of each data party, and sends the encrypted dataset to a third-party data platform;

[0009] The third-party data platform calculates the union of the encrypted datasets from each data party to obtain the union data aggregation result of each data party, and returns the union data aggregation result to the corresponding data party.

[0010] Each data source extracts its own data based on its own union data aggregation results and outputs the model training dataset.

[0011] The federated learning model is trained based on the model training dataset, and content is delivered based on the federated learning model.

[0012] According to a preferred embodiment of the present invention, the generation of a shared key by the various data parties through a first encryption operation based on a common key protocol includes:

[0013] Each data party agrees on a common key protocol;

[0014] The third-party data platform distributes common key parameters to various data parties;

[0015] After each data party performs the first encryption operation on the common key parameters, a shared key is generated according to the common key protocol.

[0016] According to a preferred embodiment of the present invention, each data party maps the common key parameters using a first hash function, and generates a shared key based on the mapping result according to the common key protocol.

[0017] According to a preferred embodiment of the present invention, each data party uses the shared key to perform symmetric encryption on the encrypted user identification information to obtain the ciphertext dataset of each data party.

[0018] According to a preferred embodiment of the present invention, a third-party data platform calculates the union of the encrypted datasets from each data party, sorts the elements in the union, and obtains the union data aggregation result of each data party.

[0019] According to a preferred embodiment of the present invention, the method further includes:

[0020] The third-party data platform calculates the intersection of the encrypted datasets from each data party, obtains the intersection result of each data party, and returns the intersection result to the corresponding data party.

[0021] Each data source extracts its own data based on the intersection result and outputs the intersection.

[0022] To address the aforementioned technical problems, a second aspect of the present invention provides a federated learning content delivery device based on data aggregation, the device comprising:

[0023] The first encryption module is used to perform a first encryption operation based on a common key protocol to generate a shared key;

[0024] The second encryption module is used to perform a second encryption operation to encrypt the user identification information;

[0025] The third encryption module is used to perform a third encryption operation on the encrypted user identification information using the shared key to obtain a ciphertext dataset, and then send the ciphertext dataset to a third-party data platform.

[0026] The extraction module is used to receive the union data aggregation results sent by the third-party data platform, extract the respective data according to the union data aggregation results, and output the model training dataset.

[0027] The training module is used to train federated learning models based on the model training dataset.

[0028] The content outreach module is used to outreach content based on the trained federated learning model.

[0029] According to a preferred embodiment of the present invention, the first encryption module is specifically used for:

[0030] Each data party agrees on a common key protocol;

[0031] The third-party data platform distributes common key parameters to various data parties;

[0032] After each data party performs the first encryption operation on the common key parameters, a shared key is generated according to the common key protocol.

[0033] According to a preferred embodiment of the present invention, the first encryption module is specifically used for: each data party mapping the common key parameters using a first hash function, and generating a shared key based on the mapping result according to the common key protocol.

[0034] According to a preferred embodiment of the present invention, the third encryption module is specifically used for: each data party using the shared key to perform symmetric encryption on the encrypted user identification information to obtain the ciphertext dataset of each data party.

[0035] In this invention, each data party performs a first encryption operation based on a common key protocol to generate a shared key, and then performs a second encryption operation to encrypt their respective user identification information. Next, the shared key is used to perform a third encryption operation on the encrypted user identification information to obtain the encrypted datasets of each data party. This allows each data party to complete the encryption of their own user identification information locally, thus ensuring the privacy and security of each party's dataset during the aggregation process. While ensuring data privacy and security, a third-party data platform calculates the union of the encrypted datasets of each data party to obtain the union data aggregation result. This union data aggregation result is used to align the datasets of each data party, ensuring that the size of the aligned dataset does not drastically decrease. This guarantees sufficient training data for federated learning model training, testing, and evaluation, effectively avoiding the problem of a sharp reduction in the size of model training data caused by data privacy intersection calculations, which could lead to untrainable models or poor generalization performance. This improves the accuracy of content delivery, enhances user experience, and significantly improves content delivery effectiveness. Attached Figure Description

[0036] To make the technical problems solved by this invention, the technical means adopted, and the technical effects achieved clearer, specific embodiments of this invention will be described in detail below with reference to the accompanying drawings. However, it should be noted that the drawings described below are merely drawings of exemplary embodiments of this invention. Those skilled in the art can obtain drawings of other embodiments based on these drawings without any creative effort.

[0037] Figure 1 This is a flowchart illustrating a federated learning content outreach method based on data aggregation, according to an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram illustrating the training of a federated learning model based on data aggregation by two data parties in an embodiment of the present invention.

[0039] Figure 3 This is a schematic diagram of the structural framework of a federated learning model training data extraction device based on data aggregation according to an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the structural framework of a federated learning model training device based on data aggregation according to an embodiment of the present invention. Detailed Implementation

[0041] Subject to the inventive concept, the structures, performance, effects or other features described in a particular embodiment may be combined in any suitable manner with one or more other embodiments.

[0042] In the description of specific embodiments, detailed descriptions of structures, performance, effects, or other features are provided to enable those skilled in the art to fully understand the embodiments. However, it is not excluded that those skilled in the art can implement the present invention with technical solutions that do not contain the above-described structures, performance, effects, or other features under specific circumstances. The flowcharts in the accompanying drawings are merely exemplary flowcharts and do not imply that the solution of the present invention must include all the content, operations, and steps in the flowcharts, nor do they imply that the execution must be carried out in the order shown in the figures.

[0043] In this embodiment of the invention, data privacy aggregation refers to a situation where each data party possesses its own data set, and the new set information obtained by each data party through union or intersection methods is based on security and privacy requirements. Therefore, the original sets cannot be directly transmitted for element matching or union / intersection operations. Thus, it is necessary to explore privacy-preserving data sharing and aggregation schemes that do not leak user privacy, achieving completely random and invisible shared data between institutions (i.e., after sharing data, institutions still cannot locate, associate, or obtain any user information), achieving a high level of privacy, laying the technical foundation for stricter data control in the future, and providing feasible security solutions.

[0044] Privacy-preserving union refers to a system where each data party has its own set. After the data parties undergo a privacy-preserving union algorithm, multiple data parties can obtain the union of all their sets, but they do not know the contents of any other data party's set besides the union and their own set. This achieves the effect of protecting the privacy of the data party's set contents during the union process.

[0045] Privacy intersection refers to a process where each data party has a set. After the data parties pass through a privacy intersection algorithm, multiple data parties can obtain the common intersection of all data parties without revealing any additional information beyond the common intersection. This achieves the effect of protecting the privacy of the data party set content during the intersection process.

[0046] This invention, based on the principles of secure cryptography, employs a three-stage encryption process to ensure the privacy and security of each data party's user identification information during aggregation. Specifically: the first encryption operation generates a shared key for each data party based on a common key protocol, guaranteeing that each party can decrypt its own dataset while preventing privacy leaks caused by other parties decrypting data. The second encryption operation further encrypts the user identification information of each data party, protecting data privacy. Finally, the encrypted user identification information is subjected to a third encryption operation using the shared key, resulting in a ciphertext dataset encrypted multiple times by each data party. Third-party data platforms cannot access the multiple encryption operations and the key, thus preventing them from decrypting the user identification information and ensuring data privacy and security.

[0047] While ensuring the privacy and security of the datasets from all parties during the aggregation process, a third-party data platform is used to find the union of the encrypted datasets from each data party. This ensures that the size of the aligned dataset does not decrease drastically, guaranteeing a sufficiently large dataset to complete subsequent model training, testing, and evaluation. This effectively avoids the problem of a sharp decrease in dataset size caused by data privacy-preserving intersection calculations, which could lead to the inability to train the model or poor generalization performance of the trained model.

[0048] While ensuring the privacy and security of the datasets from all parties during the aggregation process, the intersection of the encrypted datasets from each data party is obtained through a third-party data platform, thereby satisfying application scenarios such as querying, advertising recommendation, electronic watermarking, anonymous fingerprinting, and anonymous data mining.

[0049] Please see Figure 1 , Figure 1 This invention provides a federated learning content outreach method based on data aggregation, such as... Figure 1 As shown, the method includes:

[0050] S1. Each data party performs the first encryption operation based on a common key protocol to generate a shared key;

[0051] In this embodiment, the data parties can be two or more, depending on actual needs. Specifically, the data parties can be clients or servers; this invention does not impose any specific limitations.

[0052] For example, each data party can process the common key parameters through a first encryption operation to obtain encrypted key parameters, and then generate a shared key from the encrypted key parameters through a common key protocol. This step can then include:

[0053] S11. Each data party determines a common key protocol;

[0054] Each data party can determine a common key protocol by exchanging key protocols with each other. For example, initiator A sends its key protocol 1 to participant B, and participant B sends key protocol 1 to participant C; or, initiator A sends its key protocol 1 to participants B and C respectively. In this case, A, B, and C will use key protocol 1 as their common key protocol. Each data party can also pre-agree to use the same key protocol as their common key protocol. In this embodiment, the common key protocol enables multiple parties (commonly two parties) to obtain a shared key through interaction, while other nodes on the network cannot obtain any information about this shared key. The common key protocol can adopt: an elliptic curve-based Diffie-Hellman (DH) key exchange protocol, a pre-shared key (PSK) key negotiation protocol, etc.

[0055] S12. The third-party data platform distributes common key parameters to each data party.

[0056] In this embodiment, the third-party data platform, independent of other data providers, is used to aggregate the data sets of each data provider. It can be a client or a server. The key parameter is used to generate a key according to a key protocol. For example, the key parameter can be a random number generated by the random function rand(). The third-party data platform generates the random number using rand() and distributes this random data to each data provider, giving each data provider a shared random number.

[0057] S13. After each data party performs the first encryption operation on the common key parameters, a shared key is generated according to the common key protocol.

[0058] The first encryption operation can be implemented using hash mapping, symmetric encryption, or asymmetric encryption. In a preferred embodiment, to ensure that all data parties obtain the same shared key after inputting common key parameters, the first encryption operation is implemented using hash mapping. Each data party maps the common key parameters using a first hash function, and generates a shared key based on the mapping result according to the common key protocol. The first hash function can be the SHA256 function.

[0059] S2. Each data party performs a second encryption operation to encrypt its own user identification information;

[0060] In this embodiment, each data party has a data set containing user identification information and other information based on the user identification information. The user identification information uniquely identifies the user and can be a user ID. Other information based on the user identification information can be other user information associated with the user identification information, such as age, gender, geographical location, and other privacy information. This invention ensures data privacy and security by encrypting the user identification information in the data set multiple times. Furthermore, this invention aggregates the multiple encrypted user identification information, thereby completing the aggregation of the entire set based on the user identification information, without needing to aggregate all data within the set, thus reducing computational load.

[0061] The second encryption operation can be implemented using hash mapping, symmetric encryption, or asymmetric encryption. Furthermore, to enhance data privacy, the first, second, and third encryption operations are all different. This difference can be due to different encryption types or different specific encryption methods. The encryption type can include hash mapping, symmetric encryption, or asymmetric encryption, and the encryption method refers to the specific encryption function or algorithm used within that encryption type. In other words, the first, second, and third encryption operations can use different encryption types, different encryption functions or algorithms within the same encryption type, or different encryption functions or algorithms across different encryption types.

[0062] In a preferred example, the second encryption operation is implemented using a second hash function mapping; optionally, the second hash function can be the MD5 function.

[0063] S3. Each data party uses the shared key to perform a third encryption operation on the encrypted user identification information to obtain the ciphertext dataset of each data party, and sends the ciphertext dataset to a third-party data platform;

[0064] To facilitate the aggregation of encrypted datasets from various data providers by a third-party data platform, the third encryption operation employs symmetric encryption. Each data provider then uses the shared key to perform symmetric encryption on the encrypted user identification information, thereby obtaining the encrypted datasets from each data provider.

[0065] S4. The third-party data platform calculates the union of the encrypted datasets of each data party to obtain the union data aggregation result of each data party, and returns the union data aggregation result to the corresponding data party.

[0066] In this embodiment, since the encrypted datasets of each data party are encrypted using a shared key, the elements in each encrypted dataset can be directly compared. The result of the direct comparison is used to determine whether the encrypted elements in different encrypted datasets are the same. If they are different, both elements are put into the set; if they are the same, one of them is selected to be put into the set.

[0067] Furthermore, to facilitate subsequent data use, the third-party data platform can sort the elements in the union of the encrypted datasets from each data provider to ensure data regularity and order. For example, the elements in the union can be sorted according to their data IDs, which are used to uniquely identify the data. Here, the element is the user identification information that has undergone the three encryption operations described above. Finally, the union data aggregation result of each data provider is obtained based on the sorted union. In this embodiment, the union data aggregation result of each data provider is the same, which is the result of the union operation of all elements from each data provider.

[0068] S5. Each data source extracts its own data based on its union data aggregation result and outputs the model training dataset.

[0069] For example, each data provider extracts its own elements based on its own union data aggregation result and fills in any elements not present in the union data aggregation result. For instance, it extracts its own elements based on the data ID in the union data aggregation result and fills in any elements not present in the union data aggregation result based on the data ID.

[0070] It should be noted that this embodiment only extracts the user identification information column (ID column) from the entire dataset of each data party for aggregation (i.e., encryption and union). However, the entire dataset contains other columns besides the user identification information column (such as columns X1, X2, and X3). After each data party completes the union of the dataset by ID column through steps S1-S5, the corresponding X1, X2, and X3 columns also need to be filled with data, thus completing the aggregation of the entire dataset. Finally, the aggregated dataset is used as the model training dataset for training or testing the model.

[0071] S6. Train the federated learning model based on the model training dataset, and deliver content based on the trained federated learning model.

[0072] Federated learning enables participating parties to obtain a global federated learning model without sharing their private data. This step utilizes model training datasets distributed across individual data party devices for local computation, and then updates the federated learning model by aggregating these local computations. Each data party (which can be a mobile device or an independent organization or institution) collaboratively trains the federated learning model under the coordination of a central server, while each data party maintains control over its local private data without uploading or exchanging data.

[0073] In the specific prediction process, steps S1 to S5 are performed on the prediction data of each participant to obtain the model prediction dataset. This dataset is then input into the trained federated learning model to output the prediction results. Specifically, the trained federated learning model can make predictions based on task type. For example, for product preference tasks, it can predict end-user preferences and then target content to end-users based on those predictions. Similarly, for purchasing power tasks, it can predict end-user purchasing power and then target content to end-users based on that prediction, and so on. This invention does not impose specific limitations.

[0074] Figure 2 This diagram illustrates the training of a federated learning model based on data aggregation by two data parties, A and B, with a third-party data platform, C. First, data parties A and B exchange key protocols to establish a common key protocol. Third-party data platform C generates a shared random number r using the random function rand() and distributes r to data parties A and B. Data parties A and B then map r using the first hash function H1 to obtain an encrypted random number H1(r), and generate a shared key e from H1(r) according to the common key protocol. Subsequently, data parties A and B retrieve data from their respective datasets D. A and D B Extract user identification information u id And use the second hash function H2 to hash u id Perform mapping to obtain H2(u) id Then, for H2(u) id Using the shared key e, symmetric encryption is performed to obtain the ciphertext data set S. A and S B and S A and S B Send to third-party data platform C; third-party data platform C uses shared key e to pair with S. A and S B Find the union of data from data A, and aggregate the union results into U. A Returning the result U to data provider A, which aggregates the union of data from data provider B. B Returned to data provider B, data provider A, based on U A Extract the element itself and fill in the missing elements in U based on the data ID. A Output the union of the elements in the set S. * Data provider B, based on U B Extract the element itself and fill in the missing elements in U based on the data ID. B Output the union of the elements in the set S. * For example: S A ={1,2,3},S B ={2,3,4}, then UA =U B ={1,2,3,4}, data party A extracts its own elements 1, 2, 3, and then follows U... A The data IDs in the set are padded with 4s to obtain the union {1,2,3,4}. Data element B extracts its own elements 2, 3, and 4, and then follows the sequence U... B The data IDs in the set are padded with 1s to obtain the union {1,2,3,4}.

[0075] As can be seen from steps S1 to S6 above, each data party cannot directly access the user identification information of other data parties during data exchange, thus preventing the leakage of their data information. Meanwhile, although the third-party data platform can access the data, this data is ciphertext that has undergone multiple encryption operations locally by each data party. The third-party data platform does not know the specific encryption operations and keys, nor can it decipher the user identification information, thus ensuring data privacy and security. Furthermore, the shared key fulfills the requirement of finding the union of data privacy features.

[0076] Furthermore, based on the fact that each data party performs three encryption operations on the data set locally in this invention, the intersection of the encrypted data can be calculated, thereby meeting the data privacy requirements of application scenarios such as querying, advertising recommendation, electronic watermarking, anonymous fingerprinting, and anonymous data mining. Therefore, after step S3 above, the following steps can also be performed:

[0077] S7. The third-party data platform calculates the intersection of the encrypted datasets of each data party to obtain the intersection result of each data party, and returns the intersection result to the corresponding data party.

[0078] Since all the encrypted datasets of each data party are encrypted using a shared key, the elements in each encrypted dataset can be directly compared to determine whether the encrypted elements in different encrypted datasets are the same. If a certain element is in all encrypted datasets, the element is placed into the intersection.

[0079] S8. Each data party extracts its own data based on the intersection result and outputs the intersection.

[0080] For example, you can extract your own elements from the data IDs in the intersection results, output the intersection, and provide it for applications such as querying, advertising recommendation, electronic watermarking, anonymous fingerprinting, and anonymous data mining.

[0081] Figure 3 This invention relates to a device for extracting training data for a federated learning model based on data aggregation, such as... Figure 3 As shown, the device includes:

[0082] The first encryption module 31 is used to perform a first encryption operation based on a common key protocol to generate a shared key;

[0083] The second encryption module 32 is used to encrypt the respective user identification information through the second encryption operation;

[0084] The third encryption module 33 is used to perform a third encryption operation on the encrypted user identification information using the shared key to obtain the ciphertext dataset of each data party, and send the ciphertext dataset to a third-party data platform;

[0085] Extraction module 34 is used to receive the union data aggregation result sent by the federated learning model training device based on data aggregation, extract the respective data according to the union data aggregation result, and output the model training dataset.

[0086] In one example, the first encryption module 31 includes:

[0087] The determination module is used to determine the common key protocol;

[0088] The receiving module is used to receive the common key parameters issued by the federated learning model training device based on data aggregation;

[0089] An encryption module is configured to perform a first encryption operation on the common key parameters and then generate a shared key according to a common key protocol. Optionally, the encryption module may map the common key parameters using a first hash function and generate a shared key from the mapping result according to the common key protocol.

[0090] The third encryption module 33 is used to perform symmetric encryption on the encrypted user identification information using the shared key to obtain the ciphertext dataset of each data party.

[0091] This invention also provides a federated learning model training device based on data aggregation, such as... Figure 4 The device includes:

[0092] The aggregation module 41 is used to receive the encrypted datasets of each federated learning model training data extraction device based on data aggregation, calculate the union of the encrypted datasets, and return the union data aggregation result of each federated learning model training data extraction device based on data aggregation to the corresponding federated learning model training data extraction device based on data aggregation.

[0093] Training module 42 is used to train the federated learning model based on the model training dataset output by the data aggregation-based federated learning model training data extraction device.

[0094] Furthermore, the aggregation module 41 also calculates the union of the encrypted datasets of each federated learning model training data extraction device based on data aggregation, sorts the elements in the union, and obtains the union data aggregation result of each data party.

[0095] The present invention also provides a federated learning model training system based on data aggregation, comprising: the above-mentioned at least two federated learning model training data extraction devices based on data aggregation and the above-mentioned training data extraction devices based on data aggregation.

[0096] The present invention also provides a federated learning content delivery device based on data aggregation, the device comprising:

[0097] The first encryption module is used to perform a first encryption operation based on a common key protocol to generate a shared key;

[0098] The second encryption module is used to perform a second encryption operation to encrypt the user identification information;

[0099] The third encryption module is used to perform a third encryption operation on the encrypted user identification information using the shared key to obtain a ciphertext dataset, and then send the ciphertext dataset to a third-party data platform.

[0100] The extraction module is used to receive the union data aggregation results sent by the third-party data platform, extract the respective data according to the union data aggregation results, and output the model training dataset.

[0101] The training module is used to train the federated learning model based on the model training dataset.

[0102] The content delivery module is used to deliver content based on the federated learning model.

[0103] Wherein: the first encryption module is specifically used for:

[0104] Each data party agrees on a common key protocol;

[0105] The third-party data platform distributes common key parameters to various data parties;

[0106] After each data party performs the first encryption operation on the common key parameters, a shared key is generated according to the common key protocol.

[0107] Specifically, the first encryption module is used to: map the common key parameters using a first hash function, and generate a shared key based on the mapping result according to the common key protocol.

[0108] The third encryption module is specifically used for: each data party to perform symmetric encryption on the encrypted user identification information using the shared key, to obtain the ciphertext dataset of each data party.

[0109] Those skilled in the art will understand that the modules in the above-described device embodiments can be distributed throughout the device as described, or they can be modified accordingly and distributed in one or more devices different from the above embodiments. The modules in the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0110] The methods of the present invention described above can be stored as computer programs on one or more computer-readable media. When the computer program is executed by one or more data processing devices, the computer-readable media is enabled to implement the methods of the present invention. Furthermore, the methods and apparatus embodiments of the present invention described above can be implemented in the physical form of an electronic device.

[0111] In summary, the present invention can be implemented by methods, apparatus, electronic devices, or computer-readable media that execute computer programs. In practice, some or all of the functions of the present invention can be implemented using general-purpose data processing devices such as microprocessors or digital signal processors (DSPs).

[0112] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the present invention is not inherently related to any specific computer, virtual device, or electronic device, and various general-purpose devices can also implement the present invention. The above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A federated learning content delivery method based on data aggregation, characterized in that, The method includes: Each data party determines a common key protocol; a third-party data platform distributes common key parameters to each data party; after each data party performs a first encryption operation on the common key parameters, it generates a shared key according to the common key protocol. Each data provider performs a second encryption operation to encrypt its own user identification information; Each data party uses the shared key to perform a third encryption operation on the encrypted user identification information to obtain the encrypted dataset of each data party, and sends the encrypted dataset to a third-party data platform; The third-party data platform calculates the union of the encrypted datasets from each data party to obtain the union data aggregation result of each data party, and returns the union data aggregation result to the corresponding data party. Each data source extracts its own data based on its own union data aggregation result, and fills in the elements that are not in the union data aggregation result, outputting the model training dataset; The federated learning model is trained based on the model training dataset, and content is delivered based on the federated learning model.

2. The method according to claim 1, characterized in that, Each data party maps the common key parameters using a first hash function, and generates a shared key based on the mapping result according to the common key protocol.

3. The method according to claim 1, characterized in that, Each data party uses the shared key to perform symmetric encryption on the encrypted user identification information to obtain the ciphertext dataset of each data party.

4. The method of claim 1, wherein, The third-party data platform calculates the union of the encrypted datasets from each data provider, sorts the elements in the union, and obtains the aggregated union data from each data provider.

5. The method of claim 1, wherein, The method further includes: The third-party data platform calculates the intersection of the encrypted datasets from each data party, obtains the intersection result of each data party, and returns the intersection result to the corresponding data party. Each data source extracts its own data based on the intersection result and outputs the intersection.

6. A federated learning content engagement device based on data aggregation, characterized in that, The device includes: The first encryption module is used for each data party to determine a common key protocol; the third-party data platform sends common key parameters to each data party; after each data party performs a first encryption operation on the common key parameters, a shared key is generated according to the common key protocol. The second encryption module is used to perform a second encryption operation to encrypt the user identification information; The third encryption module is used to perform a third encryption operation on the encrypted user identification information using the shared key to obtain a ciphertext dataset, and then send the ciphertext dataset to a third-party data platform. The extraction module is used to receive the union data aggregation result sent by the third-party data platform, extract the data of each entity according to the union data aggregation result, fill in the elements that are not in the union data aggregation result, and output the model training dataset. The training module is used to train federated learning models based on the model training dataset. The content delivery module is used to deliver content based on the federated learning model.

7. The apparatus of claim 6, wherein, The first encryption module is specifically used for: each data party to map the common key parameters using a first hash function, and to generate a shared key based on the mapping result according to the common key protocol.

8. The apparatus of claim 6, wherein, The third encryption module is specifically configured to perform symmetric encryption on the encrypted user identifier information by each data party using the shared key to obtain a ciphertext data set of each data party.

Images