Universal lossless privacy protection recommendation method and device, equipment and medium

Abstract

The invention relates to the technical field of recommendation systems. The invention discloses a universal lossless privacy protection recommendation method and device, equipment and a medium, which can protect the privacy of a user in a recommendation system, improve the accuracy of a prediction result and improve the universality of the privacy protection recommendation method. The method comprises the following steps: receiving first grouping information sent by a server; performing data exchange processing on all public keys in the first grouping information by adopting an article exchange mechanism based on random response to obtain target disturbance data; performing coding processing on each element in the target disturbance data by adopting a common item marking mechanism based on PSI to obtain a target coding mark; sending the target code mark to a server; receiving first to-be-forwarded information sent by the server, and analyzing the first to-be-forwarded information to obtain first to-be-reasoned data; and performing reasoning processing on the first to-be-reasoned data by adopting an inverse process of an article exchange mechanism based on random response to obtain a target recommended article.

Description

Technical Field

[0001] The present application relates to the technical field of recommendation systems. More specifically, the present application relates to a universal lossless privacy protection recommendation method, device, equipment and medium. Background Art

[0002] In today's digital age, recommendation systems have become an indispensable key component of various online platforms, such as e-commerce and social media. It can greatly alleviate the problem of information overload and help users (clients) accurately find items that match their interests and preferences. However, traditional recommendation systems mainly rely on servers to collect and organize users' historical data, such as browsing records, click records, and purchase behavior records, in order to train a centralized recommendation model to predict the final recommended items. However, these historical data often involve user privacy, which raises privacy and security issues.

[0003] In order to solve this privacy security problem, the traditional privacy protection recommendation method adopts the privacy protection framework of cross-domain sequence recommendation and the privacy protection framework of cross-user sequence recommendation to achieve the purpose of privacy protection while recommending items. Specifically, the privacy protection framework using cross-domain sequence recommendation disturbs the data of the source domain by designing a differential privacy algorithm, and then sends the disturbed data to the target domain. The target domain protects the user privacy of the source domain during the training or prediction of the disturbed data; while the privacy protection framework of user sequence recommendation operates by using the data collected before the implementation of GDPR (General Data Protection Regulation) to pre-train a recommendation model on the server. Subsequently, when GDPR takes effect, the framework uses personal privacy data to fine-tune the pre-trained model locally. In this way, the purpose of recommending items and protecting user privacy can be achieved. However, these two frameworks have two common shortcomings. First, compared with traditional recommendation models, due to the addition of noise or the implementation of model pruning and other operations, these two frameworks will suffer from loss, resulting in a decrease in prediction performance and inaccurate prediction results. Secondly, both frameworks are only applicable to sequential recommendation models, but are not applicable to other recommendation models, such as single-behavior, multi-behavior, and multi-behavior sequence recommendation models. Therefore, the versatility of traditional privacy-preserving recommendation methods needs to be improved. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a universal and lossless privacy-preserving recommendation method, apparatus, device and medium, which can protect user privacy in the recommendation system while improving the accuracy of the prediction results and the versatility of the privacy-preserving recommendation method. The embodiments of the present application are mainly implemented through the following technical solutions: In a first aspect of an embodiment of the present application, a universal lossless privacy protection recommendation method is provided, which is applied to a first client. The universal lossless privacy protection recommendation method includes: Receiving first group information sent by the server; Using a random response-based item exchange mechanism to perform data exchange processing on all public keys in the first group information to obtain target disrupted data; Using a common item labeling mechanism based on PSI to encode each element in the target disturbed data to obtain a target coding label; Sending the target coding mark to the server so that the server predicts the target coding mark using the recommendation model to obtain a prediction result; Receiving first information to be forwarded sent by the server, and parsing and processing the first information to be forwarded to obtain first data to be inferred; The first data to be inferred is processed by inference using the inverse process of the random response-based item exchange mechanism to obtain target recommended items.

[0005] According to an embodiment of the present application, the step of performing data exchange processing on all public keys in the first group information by adopting an item exchange mechanism based on random response to obtain target disturbed data includes: Sending a plurality of encrypted exchange request information to the server based on the first group information, wherein the server generates an ordered request set based on all the exchange request information; receiving response information corresponding to each of the exchange request information sent by the server; Parsing response information corresponding to first target exchange request information to obtain target parsing data, wherein the first target exchange request information is any one of all the exchange request information; Encrypting the first item interaction record to obtain first encrypted data; In the case where the target parsing data is received, first encrypted data is sent to the server, and second encrypted data corresponding to the first target exchange request information sent by the server is received, and the interaction mark corresponding to the first target exchange request information is set to a first preset value; parsing the second encrypted data to obtain a second item interaction record corresponding to the first target exchange request information; The second item interaction record is used as the target disrupting data.

[0006] According to an embodiment of the present application, each element in the target disturbed data is encoded using a common item marking mechanism based on PSI, and the step of obtaining the target encoding mark includes: Calculate the common intersection between the target disturbed data and the data to be encoded of the target second client by using the PSI protocol, where the target second client is any second client among the specified multiple second clients; encoding a first target element in the target disturbed data that belongs to the common intersection and is not encoded using an item marking rule to obtain a coding mark corresponding to the first target element, and adding the coding mark corresponding to the first target element to a set of coding marks to be processed, wherein the first target element is any element in the target disturbed data; After calculating the common intersection between the target disturbed data and the data to be encoded of all the specified plurality of second clients, extracting the unencoded data from the target disturbed data to obtain the remaining data to be encoded; Traversing the remaining data to be encoded, encoding a second target element in the remaining data to be encoded using the item marking rule, obtaining a coding mark corresponding to the second target element, and adding the coding mark corresponding to the second target element to the set of coding marks to be processed, wherein the second target element is any one element in the remaining data to be encoded; After traversing the remaining data to be encoded, the set of encoding tags to be processed is used as the target encoding tags.

[0007] According to an embodiment of the present application, the step of performing reasoning processing on the first to-be-reasoned data by adopting the inverse process of the random response-based item exchange mechanism to obtain the target recommended item includes: receiving a target request set sent by the server, wherein the target request set is obtained by the server performing a reverse operation on the ordered request set; Traversing the target request set, when the interaction flag corresponding to the second target exchange request information is a first preset value, executing the following steps: Encrypting the first data to be inferred to obtain third encrypted data; sending the third encrypted data to the server, and receiving from the server fourth encrypted data of the second client corresponding to the second target exchange request information; Parsing the fourth encrypted data to obtain second data to be inferred; Replacing the data content of the first data to be inferred with the data content of the second data to be inferred; After traversing the target request set, using the first data to be inferred as the target recommended item; The second target exchange request information is any element in the target request set.

[0008] According to an embodiment of the present application, before the step of receiving the first group information sent by the server, the universal lossless privacy protection recommendation method further includes: Generate a public key of the first client and a private key of the first client by using an asymmetric encryption method; Sending the public key of the first client to the server, so that the server sends the public key of the first client to the second client, thereby realizing encrypted communication between the first client and the second client; Receive the public key of the second client sent by the server.

[0009] According to an embodiment of the present application, after receiving the first information to be forwarded sent by the server and parsing the first information to be forwarded to obtain the first data to be inferred, the universal lossless privacy protection recommendation method further includes: Receiving first request information sent by the server; Responding to the first request information using a request-based recommendation mechanism to obtain first real name information of a first recommended item; The first real name information is encrypted into second information to be forwarded and then sent to the server.

[0010] A second aspect of an embodiment of the present application provides a universal lossless privacy protection recommendation method, which is applied to a server. The universal lossless privacy protection recommendation method includes: Sending first group information to a first client; Receiving a target encoding mark sent by the first client; Using the recommendation model to predict the target coding mark to obtain a prediction result; generating second request information based on the prediction result; Sending the second request information to the second client; Receiving first information to be forwarded generated by the second client in response to the second request information; The first information to be forwarded is forwarded to the first client, so that the first client obtains a target recommended item according to the first information to be forwarded.

[0011] According to a third aspect of the embodiments of the present application, a universal and lossless privacy protection recommendation device is provided, comprising: A first receiving module, used for receiving first grouping information sent by a server; A first obtaining module, configured to perform data exchange processing on all public keys in the first group information by adopting an item exchange mechanism based on random response to obtain target disturbed data; A second acquisition module is used to encode each element in the target disturbed data using a common item marking mechanism based on PSI to obtain a target coding mark; A first sending module, used for sending the target coding mark to the server, so that the server predicts the target coding mark using a recommendation model to obtain a prediction result; A second receiving module is used to receive the first information to be forwarded sent by the server, and parse and process the first information to be forwarded to obtain first data to be inferred; The reasoning module is used to perform reasoning processing on the first data to be inferred by adopting the inverse process of the item exchange mechanism based on random response to obtain the target recommended item.

[0012] In a fourth aspect of an embodiment of the present application, a terminal device is provided, comprising: a processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory, and executing the steps of the general lossless privacy protection recommendation method provided in the first aspect of the embodiment of the present application and / or the second aspect of the embodiment of the present application.

[0013] In a fifth aspect of an embodiment of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium is used to store a computer program, wherein the computer program enables a computer to execute the steps of the general lossless privacy protection recommendation method provided in the first aspect of the embodiment of the present application and / or the second aspect of the embodiment of the present application.

[0014] The beneficial effects of the embodiments of the present application include: The embodiment of the present application designs an item exchange mechanism based on random response to protect the privacy of the client (ie, the user); the embodiment of the present application also designs a common item tagging mechanism based on PSI to deal with de-anonymization attacks and further enhance the strength of privacy protection. The combination of these two mechanisms constitutes a completely non-intrusive privacy protection framework, which can be flexibly applied to existing recommendation models. Therefore, the entire framework is universal and has good versatility. Specifically, the embodiment of the present application is to receive the first group information sent by the server; use the item exchange mechanism based on random response to perform data exchange processing on all public keys in the first group information to obtain target disturbed data; use the common item tagging mechanism based on PSI to encode each element in the target disturbed data to obtain the target coding tag; send the target coding tag to the server so that the server uses the recommendation model to predict the target coding tag to obtain the prediction result; receive the first information to be forwarded sent by the server, and parse the first information to be forwarded to obtain the first data to be inferred; use the inverse process of the item exchange mechanism based on random response to infer the first data to be inferred to obtain the target recommended item. Compared with the prior art, the embodiments of the present application do not require the addition of noise or the implementation of operations such as model pruning, nor do they require a specific type of recommendation model. Therefore, the embodiments of the present application can protect user privacy in the recommendation system while improving the accuracy of the prediction results and the versatility of the privacy protection recommendation method.

[0015] In addition, the embodiment of the present application also designs a request-based recommendation mechanism, which is also used to protect the privacy of the client. The random response-based item exchange mechanism, the request-based recommendation mechanism, and the PSI-based common item tagging mechanism constitute a generic and lossless privacy-preserving recommendation framework (Generic and Lossless Privacy-preserving Framework for Recommendation, GLPFR), which is also a completely non-invasive privacy protection framework that can be flexibly applied to existing recommendation models. Therefore, the entire framework is universal. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1A flowchart of a general lossless privacy protection recommendation method of the present application in some embodiments; Figure 2 A reference diagram of the communication relationship between the first client, the second client and the server in this application; Figure 3 A flowchart of a general lossless privacy protection recommendation method of the present application in some embodiments; Figure 4 A flowchart of a general lossless privacy protection recommendation method of the present application in some other embodiments; Figure 5 This is a principle block diagram of the universal lossless privacy protection recommendation device of the present application in some embodiments; Figure 6 This is a functional block diagram of the terminal device of the present application in some embodiments. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0019] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0020] The terms "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0021] The terms "comprises," "includes," or any other variations thereof are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or elements is not necessarily limited to those steps or elements explicitly listed but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0022] The term "recommendation system" has a wide range of applications, especially in e-commerce platforms and short video sharing platforms. Recommendation systems usually do not consider privacy protection in user behavior modeling. Recommendation systems are used to collect all user interaction data to the server, then centrally train the recommendation model, and then return the prediction results of the recommendation model to the user.

[0023] The term "privacy" means that everything a client interacts with cannot be leaked to the server or other clients.

[0024] Unless otherwise defined, all technical and scientific terms used in the specification of this application have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in the specification of this application includes any and all combinations of one or more related listed items.

[0025] In today's digital age, recommendation systems have become an indispensable key component of various online platforms, such as e-commerce and social media. It can greatly alleviate the problem of information overload and help users (clients) accurately find items that match their interests and preferences. However, traditional recommendation systems mainly rely on servers to collect and organize users' historical data, such as browsing records, click records, and purchase behavior records, in order to train a centralized recommendation model to predict the final recommended items. However, these historical data often involve user privacy, which raises privacy and security issues.

[0026] In order to solve this privacy security problem, the traditional privacy protection recommendation method adopts the privacy protection framework of cross-domain sequence recommendation and the privacy protection framework of cross-user sequence recommendation to achieve the purpose of privacy protection while recommending items. Specifically, the privacy protection framework using cross-domain sequence recommendation disturbs the data of the source domain by designing a differential privacy algorithm, and then sends the disturbed data to the target domain. The target domain protects the user privacy of the source domain during the training or prediction of the disturbed data; while the privacy protection framework of user sequence recommendation operates by using the data collected before the implementation of GDPR (General Data Protection Regulation) to pre-train a recommendation model on the server. Subsequently, when GDPR takes effect, the framework uses personal privacy data to fine-tune the pre-trained model locally. In this way, the purpose of recommending items and protecting user privacy can be achieved. However, these two frameworks have two common shortcomings. First, compared with traditional recommendation models, due to the addition of noise or the implementation of model pruning and other operations, these two frameworks will suffer from loss, resulting in a decrease in prediction performance and inaccurate prediction results. Secondly, both frameworks are only applicable to sequential recommendation models, but are not applicable to other recommendation models, such as single-behavior, multi-behavior, and multi-behavior sequence recommendation models. Therefore, the versatility of traditional privacy-preserving recommendation methods needs to be improved.

[0027] In addition, traditional privacy-preserving recommendation methods have also introduced federated recommendation methods, such as FedMF (FedMF is a matrix decomposition method under the federated learning framework), FedRec (federated recommendation system), FedeRank for implicit feedback modeling (FedeRank is a sorting algorithm under the federated learning framework), FedNCF (Federated Neural Collaborative Filtering), FedVAE (Federated Variational AutoEncoder), FR-FMSS (Federated Recommendation- FakeMarks and Secret Sharing, a federated recommendation framework based on false marks and secret sharing), FedPerGNN (a federated recommendation framework based on graph neural networks (GNN)), or SemiDFEG (Semi-decentralized Federated EgoGraph Learning for Recommendation). Compared with traditional centralized recommendation models, federated recommendation methods also face two challenges. First, the versatility still needs to be improved. The above-mentioned lossless federated frameworks are all suitable for single-behavior recommendations, and cannot be applied to multi-behavior and sequence recommendations. Second, the communication cost is high. This problem is determined by the characteristics of federated recommendation itself, because it needs to transmit necessary model parameters during model training. Although many studies have reduced the communication cost efficiency of federated recommendation by discretizing and compressing transmitted data, its communication cost is much greater than that of the corresponding centralized version model.

[0028] Therefore, in existing research, there is no privacy-preserving recommendation method that uses a universal, efficient, and lossless privacy-preserving algorithm framework for recommendation systems. To this end, the present application embodiment proposes a universal and lossless privacy-preserving recommendation method to solve the above problems. The specific implementation methods of the present application are further described below in conjunction with the accompanying drawings.

[0029] like Figure 1 , which is a flowchart of a universal lossless privacy protection recommendation method provided by the first aspect of an embodiment of the present application. The universal lossless privacy protection recommendation method is applied to a first client. Figure 1 In the above, the general lossless privacy protection recommendation method includes: S101: Receive first grouping information sent by a server.

[0030] The first group information includes a public key of the first client and a public key of at least one second client. The first group information is a set.

[0031] Each of the second clients is a terminal device that communicates with the first client through the server. Each of the second clients also communicates with other second clients in the at least one second client through the server. The communication relationship between the first client, the second client and the server can be referred to Figure 2 shown.

[0032] S102: Use a random response-based item exchange mechanism to perform data exchange processing on all public keys in the first group information to obtain target disrupted data.

[0033] It should be understood that the function implemented by the random response-based item exchange mechanism is that the item interaction record of the first client (i.e., the first user) is randomly exchanged with the item interaction record of the second client (i.e., the second user), and then the first client uploads the exchanged item interaction record to the server as the personal item interaction record of the first client. Since the item interaction record uploaded by the first client to the server is not its original item interaction record, even if the server has improper communication with some of the second clients and obtains the specific codes of some items, the server still cannot decipher the specific items interacted by these first clients or second clients.

[0034] Furthermore, the step S102 includes: S1021. Send multiple encrypted exchange request information to the server based on the first group information, and the server generates an ordered request set based on all the exchange request information.

[0035] Further, the step of sending a plurality of encrypted exchange request messages to the server based on the first group information includes: S10211. Remove the public key of the first client from the first group information to obtain second group information.

[0036] Exemplarily, the first grouping information may be expressed as [u, ut1, ut2, ut3, ut4], and the second grouping information may be expressed as [ut1, ut2, ut3, ut4].

[0037] Each element in the second grouping information corresponds to an exchange request information.

[0038] S10212. Traverse the second grouping information to generate request information to be processed corresponding to a third target element, where the third target element is any element in the second grouping information.

[0039] S10213. Use the third target element to encrypt the request information to be processed to obtain exchange request information corresponding to the third target element.

[0040] Exemplarily, when the third target element is the public key ut1 of the first second client, the exchange request information corresponding to the third target element can be expressed as u→ut1; when the third target element is the public key ut2 of the second second client, the exchange request information corresponding to the third target element can be expressed as u→ut2; when the third target element is the public key ut3 of the third second client, the exchange request information corresponding to the third target element can be expressed as u→ut3; when the third target element is the public key ut4 of the fourth second client, the exchange request information corresponding to the third target element can be expressed as u→ut4.

[0041] S1022: Receive response information corresponding to each of the exchange request information sent by the server.

[0042] Each of the response messages is generated by the second client corresponding to each of the exchange request messages. Exemplarily, when one of the exchange request messages is u→ut1, the response message is generated by the first second client ut1.

[0043] S1023: Parse response information corresponding to the first target exchange request information to obtain target parsing data, where the first target exchange request information is any one of all the exchange request information.

[0044] It should be understood that no matter whether the response information is sent to the server by the first second client, the second second client, the third second client or the Nth second client and then forwarded to the first client by the server, the response information is parsed and processed by the private key of the first client.

[0045] The first target exchange request information may also be understood as the i-th exchange request information among all the exchange request information.

[0046] S1024: Encrypt the first item interaction record to obtain first encrypted data.

[0047] In the embodiment of the present application, the first item interaction record is encrypted using the public key corresponding to the first target exchange request information. Exemplarily, when the first target exchange request information is an exchange request between the first client and the first second client, the first item interaction record is encrypted using the public key of the first second client; when the first target exchange request information is an exchange request between the first client and the second second client, the first item interaction record is encrypted using the public key of the second second client; and so on, when the first target exchange request information is an exchange request between the first client and the Nth second client, the first item interaction record is encrypted using the public key of the Nth second client.

[0048] It should also be understood that after the first second client receives the first encrypted data, the first encrypted data is parsed and processed using the private key of the first second client; after the second second client receives the first encrypted data, the first encrypted data is parsed and processed using the private key of the second second client; and so on, after the Nth second client receives the first encrypted data, the first encrypted data is parsed and processed using the private key of the Nth second client.

[0049] The first item interaction record is an item interaction record of the first client. The first item interaction record includes the names of the interacted items, the names of the browsed items, the names of the favorite items, and the names of the purchased items. In other embodiments, the first item interaction record may also be other information, which may be specifically set by those skilled in the art according to actual needs.

[0050] The first item interaction record can refer to Figure 3 in the “Original Interaction Record”.

[0051] S1025. When the target parsing data is received, first encrypted data is sent to the server, and second encrypted data corresponding to the first target exchange request information sent by the server is received, and the interaction mark corresponding to the first target exchange request information is set to a first preset value.

[0052] The interaction flag corresponding to the first target exchange request information can be expressed as flag (u, ut i ), where u is the first client, ut i is the i-th second client ut, (u, ut i ) can be understood as the i-th request sent by the first client to the server.

[0053] The first preset value is 1, which indicates that the first client and the corresponding second client have exchanged item interaction records.

[0054] It should also be understood that when executing the step of "sending the first encrypted data to the server", the first client does not know to which second client the server will send the first encrypted data, and the server determines the sending object based on the public key corresponding to the first encrypted data.

[0055] When the first target exchange request information is an exchange request between the first client and the first second client, the second encrypted data is sent by the first second client to the server, and then forwarded by the server to the first client; when the first target exchange request information is an exchange request between the first client and the second second client, the second encrypted data is sent by the second second client to the server, and then forwarded by the server to the first client; and so on, when the first target exchange request information is an exchange request between the first client and the Nth second client, the second encrypted data is sent by the Nth second client to the server, and then forwarded by the server to the first client.

[0056] S1026. Parse the second encrypted data to obtain a second item interaction record corresponding to the first target exchange request information.

[0057] It should be understood that no matter which second client sends the second encrypted data, in the first client, the second encrypted data is parsed by the private key of the first client.

[0058] When the second encrypted data is sent by the first second client, the second item interaction record is the item interaction record of the first second client; when the second encrypted data is sent by the second second client, the second item interaction record is the item interaction record of the second second client; and so on, when the second encrypted data is sent by the Nth second client, the second item interaction record is the item interaction record of the Nth second client.

[0059] S1027. Use the second object interaction record as the target disturbance data.

[0060] It should be understood that after processing from step S1021 to step S1027, the second item interaction record corresponding to the last exchange request information in all the exchange request information is used as the target disrupted data, and at this time, all item interaction records of the client are disrupted.

[0061] The target disturbance data can be referenced Figure 3 "Exchanging Raw Interaction Data" in .

[0062] In other implementations, the step S102 further includes: S1028. When the target parsing data indicates rejection, the first encrypted data is not sent to the server, and the second encrypted data corresponding to the first target exchange request information sent by the server is not received.

[0063] The step S1028 may be understood as the first client not exchanging item interaction records with the second client corresponding to the first target exchange request information.

[0064] S1029: Set the interaction mark corresponding to the first target exchange request information to a second preset value.

[0065] The second preset value is 0, which indicates that the first client and the corresponding second client have not exchanged item interaction records.

[0066] It should also be understood that, when the target resolution data is rejected, although the first client and the second client corresponding to the first target exchange request information have not exchanged item interaction records, the two parties will still conduct normal communication interaction. Since the communication interaction between the two parties is encrypted, the server cannot know the specific content of the interaction between the two parties, nor can it infer whether the two parties have actually exchanged item interaction records.

[0067] S103, using the PSI-based common item tagging mechanism to encode each element in the target disturbed data to obtain a target coding tag. The step S103 can be understood as Figure 3 Item Number Matching step in .

[0068] The function of the common item marking mechanism based on PSI (Private Set Intersection) is that all clients share a set of item marking rules without leaking the privacy of the client (i.e., user), but the server does not know this set of item marking rules. Specifically, for the same item, the corresponding marking symbols (i.e., encoding) are the same in different clients, so the subsequent upload to the server can be applied to the training of the current existing recommendation model. For example, if the first client u and the first second client ut1 have interacted with the item "apple", then in the local of the first client and the first second client, apple will be numbered as "1", and then "1" will be uploaded to the server. Since the server cannot infer "apple" from "1", the privacy of the client (i.e., user) is protected. However, this numbering can be used for training or prediction in the existing recommendation model.

[0069] The technology that has emerged at present is similar to the common item marking mechanism based on PSI. It selects any client to generate a key S by symmetric encryption, and then the client sends it to other clients by asymmetric encryption. However, the server does not know the key S. After the same item is encrypted by the key S, the content obtained is the same, so the same function is achieved. However, the difference between it and the common item marking mechanism based on PSI is that if the server is improperly communicating with some clients, the server only needs to obtain the key S through one user, then the privacy of all users will be leaked. However, the common item marking mechanism based on PSI is that when the server is improperly communicating with a client, other clients will only leak the names of items that interact with this client. Due to the sparsity of data in the recommendation field, the privacy leaked is only a few items. It is worth noting that after combining the common item marking mechanism based on PSI with the item exchange mechanism based on random response, the problem of privacy leakage is completely solved, because after the item interaction records between clients are exchanged, the server cannot use this small number of leaked items to perform de-anonymization attacks to obtain the ultimate source of any item interaction record. From this perspective, the common item marking mechanism based on PSI proposed in the embodiment of the present application has higher security performance. In addition, the PSI (Private Set Intersection) described in this article is a private item intersection technology, which is a cryptographic protocol designed to allow two or more participants (clients) to safely calculate the intersection of sets without exposing the specific content of their own sets. The core goal of PSI is that the participants (clients) can know the elements of the intersection part, but will not disclose the information of other elements. In the embodiment of the present application, it is to enable two clients (that is, users) to know which items the other party has interacted with through PSI operation, but cannot know other items interacted with by the other party. Exemplarily, the first client u and the first second client ut1 have interacted with the item "apple", but the first second client ut1 has also interacted with "Snow Pear", while the first client u has not interacted with "Snow Pear". Therefore, when the first client u performs PSI operation with the first second client ut1, the first client u cannot know that the first second client ut1 has interacted with the item "Snow Pear".

[0070] In the embodiment of the present application, before performing the PSI operation, the server will number the first client and N second clients, and then select one of the clients to perform the PSI operation with the other clients. Exemplarily, the first client is numbered 1, the first second client is numbered 2, the second second client is numbered 3, and so on, the Nth second client is numbered N+1, then the first client numbered 1 is selected to perform the PSI operation with all the second clients (that is, the clients numbered 2 to N+1), after the PSI operation with the first client numbered 1 is completed, the first second client numbered 2 is selected to perform the PSI operation with the second clients numbered 3 to N+1, and so on, until all any two clients have performed the PSI operation.

[0071] Furthermore, the step S103 includes: S1031. Calculate a common intersection between the target disturbed data and the to-be-encoded data of the target second client by using the PSI protocol, where the target second client is any second client among the specified multiple second clients.

[0072] The PSI (Private Set Intersection) protocol is a secure multi-party computing protocol that is used by multiple parties to jointly calculate the intersection of these sets without revealing their respective private sets.

[0073] The specified multiple second clients are specified by the server. Exemplarily, the server may set the specified multiple second clients to be all the second clients, or to be any several of the second clients.

[0074] The data to be encoded is target disturbed data of the target second client.

[0075] Exemplarily, the step S1031 may include: S10311. Apply a hash function to perform calculation processing on each element in the target disturbed data to generate a corresponding first hash value set.

[0076] S10312. Secretly share each element in the first hash value set to generate a first shared value.

[0077] S10313: Receive a second shared value forwarded by the server, where the second shared value is sent from the target second client to the server.

[0078] The second shared value is generated by the target second client applying a hash function to each element in the data to be encoded to perform calculation processing to generate a corresponding second hash value set; then each element in the second hash value set is secretly shared to generate the second shared value.

[0079] S10314. Calculate the intersection of the first shared value and the second shared value to obtain a processed intersection.

[0080] S10315. According to the PSI protocol, reconstruct each original intersection element from the processed intersection to obtain the common intersection.

[0081] It should be understood that the implementation of step S1031 is not limited to the above steps S10311 to S10315, and those skilled in the art may also implement step S1031 in other ways, which is not further limited herein.

[0082] S1032. Encode the first target element in the target disturbed data that belongs to the common intersection and is not encoded using an item marking rule to obtain a coding mark corresponding to the first target element, and add the coding mark corresponding to the first target element to a set of coding marks to be processed, wherein the first target element is any element in the target disturbed data.

[0083] For example, the item marking rule is that the code mark of "apple" is 1, the code mark of "orange" is 2, the code mark of "tangerine" is 3, and the code mark of "pear" is 4, etc. In other embodiments, the item marking rule can be set by those skilled in the art according to actual needs, and this document does not make further restrictions on this.

[0084] S1033: After calculating the common intersection between the target disturbed data and the data to be encoded of all the specified second clients, extract the unencoded data from the target disturbed data to obtain the remaining data to be encoded.

[0085] Each item in the remaining data to be encoded is an item that has not been interacted with by any other second client.

[0086] S1034, traverse the remaining data to be encoded, use the item marking rule to encode the second target element in the remaining data to be encoded, obtain the encoding tag corresponding to the second target element, and add the encoding tag corresponding to the second target element to the set of encoding tags to be processed, the second target element is any element in the remaining data to be encoded.

[0087] S1035. After traversing the remaining data to be encoded, use the set of encoding tags to be processed as the target encoding tags.

[0088] The target coding mark can be understood as Figure 3 "Encoded raw interaction data" in.

[0089] The embodiment of the present application takes the PSI operation of the first client u and the first second client ut1 as an example, and the specific PSI operation process is as follows: Assume that the target disturbing data of the first client u is [apple, orange, tangerine], and the data to be encoded of the first second client ut1 is [apple, pear, orange], then the common intersection between the target disturbing data and the data to be encoded is [apple, orange]; then, encode "apple" and "orange" respectively, and the specific encoding process is to determine whether "apple" already has an encoding mark. If "apple" already has an encoding mark, and the encoding mark is 1, then the first client will feedback the information "the encoding mark of apple is 1" to the first second client. If the "apple" in the first client has not been encoded, and the "apple" in the first second client has not been encoded, then the first client and the first second client jointly agree on the encoding mark. After encoding "apple", the encoding mark of "apple" is added to the set of encoding marks to be processed. The encoding process for "orange" is the same as that for "apple", so it will not be repeated.

[0090] It is worth noting that since any client will only allow the next client to perform PSI operations with other clients after performing PSI operations with all other clients that have not performed PSI operations with it, there will be no situation where the first client and any second client have different PSI numbers (i.e., coding marks) for the same item.

[0091] S104, sending the target coding mark to the server, so that the server uses the recommended model to predict the target coding mark and obtain a prediction result. The step S104 can be understood as Figure 3 The "Training" or "Prediction" step in the "Server" in .

[0092] S105: Receive the first information to be forwarded sent by the server, and parse the first information to be forwarded to obtain the first data to be inferred. The first data to be inferred can be understood as Figure 3 "Recommended item 1" in.

[0093] It should be understood that after the step S104, the server uses the recommendation model to predict the target coding mark and obtains a prediction result. The prediction result is the coding mark (a label) of the recommended item. Then, the server will generate a second request message based on the prediction result and send the second request message to one of the second clients. The selection of this second client must meet the preset requirements, and the preset requirement is that the prediction result belongs to the target coding mark of this second client. Then, this second client responds to the second request message using a request-based recommendation mechanism to obtain the second real name information of the second recommended item; then, this second client uses the public key of the first client to encrypt the second real name information to obtain the first information to be forwarded, and this second client sends the first information to be forwarded to the server, which is then forwarded to the first client by the server.

[0094] The step of "parsing and processing the first information to be forwarded" is to use the private key of the first client to perform the parsing process.

[0095] The content of the first data to be inferred is the same as the content of the second real name information.

[0096] It should be understood that after the request-based recommendation mechanism, the first client obtains the real name of the recommended item (that is, the first data to be inferred), but because the item interaction records between the first client and all second clients are randomly exchanged and disrupted, the real name obtained by the first client at this time is not the real name of the item it needs to recommend, but the real name of the item recommended by other second clients after being disrupted. Therefore, the reverse process of the disruption needs to be used to restore the recommended items.

[0097] S106: Use the inverse process of the random response-based item exchange mechanism to perform inference processing on the first data to be inferred to obtain target recommended items.

[0098] The step S106 can be understood as Figure 3 The target recommended item can be understood as Figure 3 "Recommended Items 2" in the .

[0099] Furthermore, the step S106 includes: S1061. Receive a target request set sent by the server, where the target request set is obtained by the server performing a reverse operation on the ordered request set.

[0100] Exemplarily, the ordered request set can be expressed as g-list=[(u→ut1), (u→ut2), (u→ut3), . . . , (u→ut N )], where (u→ut1) represents the first request in g-list, which comes from the first client u and is a request for the first client u to exchange with the second client ut1; (u→ut2) represents the second request in g-list, which comes from the first client u and is a request for the first client u to exchange with the second client ut2; (u→ut3) represents the third request in g-list, which comes from the first client u and is a request for the first client u to exchange with the second client ut3; (u→ut N ) indicates the Nth request in g-list, which comes from the first client u and is made by the first client u to communicate with the second client ut N The target request set can be expressed as g-list-reversed=[(u→ut N )、···、(u→ut3)、(u→ut2)、(u→ut1)],(u→ut N ), (u→ut3), (u→ut2) and (u→ut1) are the same as those mentioned above.

[0101] S1062, traversing the target request set, and executing the following steps S1063 to S1067 when the interaction mark corresponding to the second target exchange request information is a first preset value, and the second target exchange request information is any element in the target request set.

[0102] S1063. Encrypt the first data to be inferred to obtain third encrypted data, and the second target exchange request information is any element in the target request set.

[0103] In the embodiment of the present application, the first reasoning data is encrypted using the public key of the interactive object. N ), the public key of the Nth second client is used to encrypt the first inference data; when the second target exchange request information is (u→ut3), the public key of the third second client is used to encrypt the first inference data, and so on.

[0104] S1064: Send the third encrypted data to the server, and receive from the server fourth encrypted data of the second client corresponding to the second target exchange request information.

[0105] When the second target exchange request information is (u→ut N ), the server forwards the third encrypted data to the Nth second client after receiving the third encrypted data; when the second target exchange request information is (u→ut3), the server forwards the third encrypted data to the third second client after receiving the third encrypted data; and so on.

[0106] When the second target exchange request information is (u→ut N ), the fourth encrypted data is sent by the Nth second client to the server, and then forwarded by the server to the first client; when the second target exchange request information is (u→ut3), the fourth encrypted data is sent by the third second client to the server, and then forwarded by the server to the first client; and so on.

[0107] The fourth encrypted data is obtained by encrypting the public key of the first client.

[0108] The step S1064 may be understood as the first client exchanging the obtained recommended items with one of the second clients.

[0109] S1065. Parse the fourth encrypted data to obtain second data to be inferred.

[0110] The embodiment of the present application uses the private key of the first client for parsing.

[0111] When the second target exchange request information is (u→ut N ), the second inference data is the second real name information of the Nth second client; when the second target exchange request information is (u→ut3), the second inference data is the second real name information of the third second client; and so on.

[0112] S1066. Replace the data content of the first data to be inferred with the data content of the second data to be inferred.

[0113] S1067. After traversing the target request set, use the first data to be inferred as the target recommended item.

[0114] The target recommended item is the actual recommended item in this application.

[0115] In some embodiments, the step S1062 also includes that when the interaction mark corresponding to the second target exchange request information is a second preset value, the first client does not send the third encrypted data to the server, nor does it receive the fourth encrypted data sent by the server. That is, the first client does not exchange the recommended items with any second client. However, the first client still maintains communication with all second clients. The purpose of this setting is to prevent the server from determining whether the client has exchanged the recommended items.

[0116] After executing step S106, the first client can obtain the real recommended items.

[0117] Through the above implementation methods, the embodiment of the present application designs an item exchange mechanism based on random responses to protect the privacy of the client (ie, the user); the embodiment of the present application also designs a common item tagging mechanism based on PSI to deal with de-anonymization attacks and further enhance the strength of privacy protection. The combination of these two mechanisms constitutes a completely non-invasive privacy protection framework that can be flexibly applied to existing recommendation models. Therefore, the entire framework is universal and has good versatility. Compared with the prior art, the embodiment of the present application does not require the addition of noise or the implementation of model pruning and other operations, nor does it require a specific type of recommendation model. Therefore, the embodiment of the present application can protect the privacy of users (ie, clients) in the recommendation system while improving the accuracy of prediction results and the versatility of privacy protection recommendation methods.

[0118] It should be understood that the universal and lossless privacy-preserving recommendation method proposed in the embodiment of the present application can be directly applied to the existing recommendation model and will not cause any changes to the existing recommendation model itself.

[0119] In addition, the embodiment of the present application also designs a request-based recommendation mechanism, which is also used to protect the privacy of the client. The random response-based item exchange mechanism, the request-based recommendation mechanism, and the PSI-based common item marking mechanism constitute a generic and lossless privacy-preserving recommendation framework (Generic and Lossless Privacy-preserving Framework for Recommendation, GLPFR), which is also a completely non-invasive privacy protection framework that can be flexibly applied to existing recommendation models. Therefore, the entire framework is universal. The generic and lossless privacy-preserving recommendation framework is also safe when there is improper communication between the server and some clients.

[0120] In some implementations, before step S101, the universal lossless privacy protection recommendation method further includes: S1071. Generate a public key of the first client and a private key of the first client using an asymmetric encryption method.

[0121] The setting of the public key and the private key can not only realize encrypted communication, but also realize tamper-proof authentication of the content.

[0122] Regarding encrypted communication, it can be understood as follows: if the first client wants to send a content "a" to the first second client, but does not want other second clients to know it, then the first client can use the public key of the first second client to encrypt this "a", and then send it to the server, and then the server forwards it to the first second client, and then the first second client uses its own private key to decrypt and obtain the content.

[0123] Regarding the tamper-proof authentication of content, it can be understood as follows: assuming that the first client wants to send a content "I want an apple", it can use the private key of the first client to digitally sign the content and then broadcast it; any second client can use the public key of the first client to digitally authenticate the content, because the public key of the first client can be shared by all second clients and the server. Once the digital signature authentication is passed, it can be considered that the information is authentic and has not been tampered with by the sender.

[0124] S1072: Send the public key of the first client to the server, so that the server sends the public key of the first client to the second client, thereby realizing encrypted communication between the first client and the second client.

[0125] S1073. Receive the public key of the second client sent by the server.

[0126] In some implementations, after step S105, the universal lossless privacy protection recommendation method further includes: S1081. Receive first request information sent by the server.

[0127] The first request information is generated by the server according to the prediction result, and is used to generate real name information of a recommended item for one of the second clients.

[0128] Exemplarily, for a prediction result numbered T(i), the server generates the first request information based on the prediction result and sends the first request information to the first client. However, it should be understood that the server sends the first request information to the first client on the premise that the first client has the real name information of the item corresponding to the prediction result (i.e., numbered T(i)).

[0129] S1082: Using a request-based recommendation mechanism to respond to the first request information, and obtain first real name information of the first recommended item.

[0130] Since the server training is completed and the prediction should be the number of the item recommended to each second client, rather than the real name of the recommended item, the request-based recommendation mechanism is designed. The function implemented by the request-based recommendation mechanism is that the server can return the real name of the recommended item to the corresponding second client while protecting the privacy of the first client (that is, the user).

[0131] The request-based recommendation mechanism enables the server to recommend the real name of the item to each client, which is equivalent to completing the de-coding process from an item label (that is, the encoding mark of the item) to the item name.

[0132] Furthermore, the step S1082 includes: S10821. Calculate a first usage frequency of the public key of the first client.

[0133] Exemplarily, for the first client, any behavior that needs to use the public key of the first client for encryption is counted as one use. More specifically, when the public key of the first client is needed to encrypt "apple", the first usage frequency is counted as one time; when the public key of the first client is needed to encrypt "apple" and "Sydney", the first usage frequency is counted as two times.

[0134] The first usage frequency may be understood as the number of times the public key of the first client is used.

[0135] S10822. Use the private key of the first client to digitally sign the first usage frequency to obtain fifth encrypted data.

[0136] S10823. Send the fifth encrypted data to the server.

[0137] S10824. Receive multiple sixth encrypted data sent by the server.

[0138] Each of the sixth encrypted data is an encrypted second usage frequency of the public key of one of the second clients.

[0139] The second usage frequency may be understood as the number of times a public key of one of the second clients is used.

[0140] S10825. Traverse the multiple sixth encrypted data, use the public key corresponding to the target sixth encrypted data to verify the target sixth encrypted data, and obtain a verification result. The target sixth encrypted data is any one of the multiple sixth encrypted data.

[0141] The execution of step S10825 can prevent the server from tampering with the usage frequency corresponding to each client.

[0142] S10826. parse the target sixth encrypted data to obtain a second usage frequency.

[0143] S10827. When the verification result is passed and the second usage frequency does not exceed the preset number of recommended items, the first client responds to the first request information and uses the item name corresponding to the first request information as the first real name information.

[0144] The preset number of recommended items can be set by those skilled in the art according to actual needs.

[0145] In some implementations, the step S1082 further includes: S10828: When the verification result is failure to pass and / or the second usage frequency exceeds the preset number of recommended items, reject the first request information.

[0146] When the verification result is failed, it means that the second usage frequency has been modified, thereby making the target sixth encrypted data unreliable.

[0147] Secondly, if a client is recommended k items, then the server should also request k items using its public key. If the server requests more than k items using its public key, there is reason to suspect that the server is colluding with the client to crack the rules for encoding item names. Therefore, the client can reject the server's request and terminate the recommendation service. k is the preset number of recommended items.

[0148] In step S10828, it can be considered that the server is illegally accessing the privacy of the client.

[0149] S1083: Encrypt the first real name information into second information to be forwarded and send the second information to the server.

[0150] Specifically, the first real name information is encrypted using the public key of the second client corresponding to the target sixth encrypted data.

[0151] When the server receives the second information to be forwarded, it sends the second information to be forwarded to a corresponding second client. Exemplarily, when the second information to be forwarded corresponds to the first second client, the server sends the second information to be forwarded to the first second client. Then, the first second client uses its private key to parse the second information to be forwarded and obtains the third real name information, which is the same as the first real name information.

[0152] It should be noted that the embodiments of the present application are equivalent to the corresponding traditional centralized recommendation model and are therefore lossless in performance.

[0153] The universal and lossless privacy-preserving recommendation framework of the embodiment of the present application is applicable to single-behavior, multi-behavior, sequence, and multi-behavior sequence modeling in existing recommendation systems. The universal and lossless privacy-preserving recommendation framework is a model- and data-independent and non-intrusive method that can be directly applied to existing recommendation models without causing any changes to the recommendation models themselves.

[0154] The universal and lossless privacy protection recommendation method proposed in the embodiment of the present application is not limited to the e-commerce recommendation system scenario, but can also be applied to the POI (Point of Interest) recommendation field, the financial field, or the IoT device recommendation. In the POI recommendation field, different place names can be used as processing objects; in the financial field, different service names can be used as processing objects; in the IoT device recommendation, different device names can be used as processing objects.

[0155] The universal lossless privacy protection recommendation method proposed in the embodiment of the present application is to first execute the random response-based item exchange mechanism, and then execute the PSI-based common item marking mechanism. In other implementations, the order of using these two mechanisms can also be interchanged.

[0156] refer to Figure 4 As shown, in the second aspect of the embodiment of the present application, a universal lossless privacy protection recommendation method is provided, which is applied to a server. The universal lossless privacy protection recommendation method includes: S201. Send first group information to a first client.

[0157] S202: Receive a target coding mark sent by the first client.

[0158] S203: Use the recommendation model to predict the target coding mark to obtain a prediction result.

[0159] S204: Generate second request information based on the prediction result.

[0160] S205: Send the second request information to the second client.

[0161] S206: Receive first information to be forwarded generated by the second client in response to the second request information.

[0162] S207: Forward the first information to be forwarded to the first client, so that the first client obtains a target recommended item according to the first information to be forwarded.

[0163] In some implementations, before step S201, the universal lossless privacy protection recommendation method further includes: S211. Obtain a public key of the first client and at least one public key of the second client.

[0164] S212: Randomly divide all public keys to obtain at least one group information, where the at least one group information includes the first group information.

[0165] The grouping information may be K groups, see Figure 3 As shown, they are divided into Group 1 and Group K.

[0166] In some implementations, after step S202, the universal lossless privacy protection recommendation method further includes: S221. Receive multiple exchange request information sent by the first client.

[0167] It should be understood that the multiple exchange request messages are transmitted to the corresponding second clients through the server. Since the server does not have the private key of the second client, it is impossible to know the specific content of each exchange request message, thereby protecting the privacy of all clients (i.e., the first client and all second clients). In the embodiment of the present application, the client can be understood as a user.

[0168] S222: Generate an ordered request set based on all the exchange request information.

[0169] Exemplarily, the ordered request set can be expressed as g-list=[(u→ut1), (u→ut2), (u→ut3), . . . , (u→ut N)], where (u→ut1) represents the first request in g-list, which comes from the first client u and is a request for the first client u to exchange with the second client ut1; (u→ut2) represents the second request in g-list, which comes from the first client u and is a request for the first client u to exchange with the second client ut2; (u→ut3) represents the third request in g-list, which comes from the first client u and is a request for the first client u to exchange with the second client ut3; (u→ut N ) indicates the Nth request in g-list, which comes from the first client u and is made by the first client u to communicate with the second client ut N A request to make an exchange.

[0170] The server processes all requests in the ordered request set in sequence.

[0171] In some implementations, after step S222, the universal lossless privacy protection recommendation method further includes: S223, performing a reverse operation on the ordered request set to obtain a target request set. In some implementations, after the step of forwarding the first information to be forwarded to the first client, the universal lossless privacy protection recommendation method further includes: The target request set is sent to the first client.

[0172] In some implementations, after step S205, the universal lossless privacy protection recommendation method further includes: receiving fifth encrypted data sent by the first client; The fifth encrypted data is broadcasted to all second clients, and the sixth encrypted data corresponding to each second client is forwarded to the first client.

[0173] refer to Figure 5 , which is a principle block diagram of a universal lossless privacy protection recommendation device provided by the third aspect of the embodiment of the present application. Figure 5 In the example, the universal lossless privacy protection recommendation device 10 comprises: A first receiving module 11, used for receiving first grouping information sent by a server; A first obtaining module 12 is used to perform data exchange processing on all public keys in the first group information by adopting an item exchange mechanism based on random response to obtain target disturbed data; A second obtaining module 13 is used to encode each element in the target disturbed data using a common item marking mechanism based on PSI to obtain a target coding mark; A first sending module 14 is used to send the target coding mark to the server, so that the server predicts the target coding mark using a recommendation model to obtain a prediction result; The second receiving module 15 is used to receive the first information to be forwarded sent by the server, and parse the first information to be forwarded to obtain the first data to be inferred; The reasoning module 16 is used to perform reasoning processing on the first data to be inferred by adopting the inverse process of the item exchange mechanism based on random response to obtain the target recommended item.

[0174] A fourth aspect of the embodiments of the present application provides a terminal device. The principle block diagram of the terminal device can be as follows: Figure 6 As shown. The terminal device includes a processor, a memory, a network interface, a display screen and a temperature sensor connected via a system bus. The processor is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a universal and lossless privacy protection recommendation method is implemented. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor is pre-set inside the terminal device to detect the operating temperature of the internal device.

[0175] Those skilled in the art will understand that Figure 6 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal device to which the scheme of the present invention is applied. The specific terminal device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0176] In some embodiments, an embodiment of the present application provides a terminal device, which includes a processor and a memory, the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the steps of the general lossless privacy protection recommendation method provided in the first aspect of the embodiment of the present application and / or the second aspect of the embodiment of the present application.

[0177] In a fifth aspect of an embodiment of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium is used to store a computer program, wherein the computer program enables a computer to execute the steps of the general lossless privacy protection recommendation method provided in the first aspect of the embodiment of the present application and / or the second aspect of the embodiment of the present application.

[0178] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0179] A sixth aspect of the embodiments of the present application provides a universal and lossless privacy protection recommendation device, including: The second sending module is used to send the first group information to the first client.

[0180] The third receiving module is used to receive the target coding mark sent by the first client.

[0181] The prediction module is used to predict the target coding mark using the recommendation model to obtain a prediction result.

[0182] A generating module is used to generate second request information based on the prediction result.

[0183] The third sending module is used to send the second request information to the second client.

[0184] The fourth receiving module is used to receive the first to-be-forwarded information generated by the second client in response to the second request information.

[0185] A forwarding module is used to forward the first information to be forwarded to the first client, so that the first client obtains a target recommended item according to the first information to be forwarded.

[0186] The technical features of the above embodiments can be combined without changing the basic principles of the present application. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0187] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the scope of patent protection of the present application shall be subject to the attached claims.

Claims

1. A universal and lossless privacy protection recommendation method, applied to a first client, characterized in that: include: Receiving first group information sent by the server; Using a random response-based item exchange mechanism to perform data exchange processing on all public keys in the first group information to obtain target disrupted data; Using a common item labeling mechanism based on PSI to encode each element in the target disturbed data to obtain a target coding label; Sending the target coding mark to the server so that the server predicts the target coding mark using the recommendation model to obtain a prediction result; Receiving first information to be forwarded sent by the server, and parsing and processing the first information to be forwarded to obtain first data to be inferred; The first data to be inferred is processed by inference using the inverse process of the random response-based item exchange mechanism to obtain target recommended items.

2. The universal lossless privacy-preserving recommendation method according to claim 1, characterized in that: The step of performing data exchange processing on all public keys in the first group information by adopting the random response-based item exchange mechanism to obtain target disturbed data includes: Sending a plurality of encrypted exchange request information to the server based on the first group information, wherein the server generates an ordered request set based on all the exchange request information; receiving response information corresponding to each of the exchange request information sent by the server; Parsing response information corresponding to first target exchange request information to obtain target parsing data, wherein the first target exchange request information is any one of all the exchange request information; Encrypting the first item interaction record to obtain first encrypted data; In the case where the target parsing data is received, first encrypted data is sent to the server, and second encrypted data corresponding to the first target exchange request information sent by the server is received, and the interaction mark corresponding to the first target exchange request information is set to a first preset value; parsing the second encrypted data to obtain a second item interaction record corresponding to the first target exchange request information; The second item interaction record is used as the target disrupting data.

3. The universal lossless privacy-preserving recommendation method according to claim 1, characterized in that: Each element in the target disturbed data is encoded using a common item marking mechanism based on PSI, and the steps of obtaining the target encoding mark include: Calculate the common intersection between the target disturbed data and the data to be encoded of the target second client by using the PSI protocol, where the target second client is any second client among the specified multiple second clients; encoding a first target element in the target disturbed data that belongs to the common intersection and is not encoded using an item marking rule to obtain a coding mark corresponding to the first target element, and adding the coding mark corresponding to the first target element to a set of coding marks to be processed, wherein the first target element is any element in the target disturbed data; After calculating the common intersection between the target disturbed data and the data to be encoded of all the specified plurality of second clients, extracting the unencoded data from the target disturbed data to obtain the remaining data to be encoded; Traversing the remaining data to be encoded, encoding a second target element in the remaining data to be encoded using the item marking rule, obtaining a coding mark corresponding to the second target element, and adding the coding mark corresponding to the second target element to the set of coding marks to be processed, wherein the second target element is any one element in the remaining data to be encoded; After traversing the remaining data to be encoded, the set of encoding tags to be processed is used as the target encoding tags.

4. The universal lossless privacy-preserving recommendation method according to claim 2, characterized in that: The step of performing reasoning processing on the first to-be-reasoned data by adopting the inverse process of the random response-based item exchange mechanism to obtain the target recommended item includes: receiving a target request set sent by the server, wherein the target request set is obtained by the server performing a reverse operation on the ordered request set; Traversing the target request set, when the interaction flag corresponding to the second target exchange request information is a first preset value, executing the following steps: Encrypting the first data to be inferred to obtain third encrypted data; sending the third encrypted data to the server, and receiving from the server fourth encrypted data of the second client corresponding to the second target exchange request information; Parsing the fourth encrypted data to obtain second data to be inferred; Replacing the data content of the first data to be inferred with the data content of the second data to be inferred; After traversing the target request set, using the first data to be inferred as the target recommended item; The second target exchange request information is any element in the target request set.

5. The universal lossless privacy-preserving recommendation method according to claim 1, characterized in that: Before the step of receiving the first group information sent by the server, the universal lossless privacy protection recommendation method further includes: Generate a public key of the first client and a private key of the first client by using an asymmetric encryption method; Sending the public key of the first client to the server, so that the server sends the public key of the first client to the second client, thereby realizing encrypted communication between the first client and the second client; Receive the public key of the second client sent by the server.

6. The universal lossless privacy-preserving recommendation method according to claim 1, characterized in that: After receiving the first information to be forwarded sent by the server and parsing the first information to be forwarded to obtain the first data to be inferred, the universal lossless privacy protection recommendation method further includes: Receiving first request information sent by the server; Responding to the first request information using a request-based recommendation mechanism to obtain first real name information of a first recommended item; The first real name information is encrypted into second information to be forwarded and then sent to the server.

7. A universal and lossless privacy-preserving recommendation method, applied to a server, characterized in that: include: Sending first group information to a first client; Receiving a target encoding mark sent by the first client; Using the recommendation model to predict the target coding mark to obtain a prediction result; generating second request information based on the prediction result; Sending the second request information to the second client; Receiving first information to be forwarded generated by the second client in response to the second request information; The first information to be forwarded is forwarded to the first client, so that the first client obtains a target recommended item according to the first information to be forwarded.

8. A universal and lossless privacy protection recommendation device, characterized in that: include: A first receiving module, used for receiving first grouping information sent by a server; A first obtaining module, configured to perform data exchange processing on all public keys in the first group information by adopting an item exchange mechanism based on random response to obtain target disturbed data; A second acquisition module is used to encode each element in the target disturbed data using a common item marking mechanism based on PSI to obtain a target coding mark; A first sending module, used for sending the target coding mark to the server, so that the server predicts the target coding mark using a recommendation model to obtain a prediction result; A second receiving module is used to receive the first information to be forwarded sent by the server, and parse and process the first information to be forwarded to obtain first data to be inferred; The reasoning module is used to perform reasoning processing on the first data to be inferred by adopting the inverse process of the item exchange mechanism based on random response to obtain the target recommended item.

9. A terminal device, characterized in that: include: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory to execute the steps of the universal lossless privacy protection recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein the computer program enables a computer to execute the steps of the universal lossless privacy protection recommendation method as described in any one of claims 1 to 7.

Images