An advertisement delivery method and system based on a marked set intersection

Abstract

The application discloses an advertisement delivery method and system based on marked privacy set intersection, and the method comprises a consensus stage, a preprocessing stage, an inadvertent transmission stage, a data encryption stage, a data sharing and delivery stage. The system comprises an advertiser end and an advertisement client end. The application can obtain the intersection of both parties without leaking excess data, protect the user privacy of the client, accurately deliver the advertisement, and improve the conversion rate.

Description

Technical Field

[0001] This invention relates to the field of data security, including secure multi-party computation and privacy computation. Specifically, it relates to an advertising delivery method and system based on intersection of marked privacy sets, which can achieve targeted advertising delivery in a closed state and avoid leaking unnecessary user privacy data. Background Technology

[0002] With the booming development of the digital economy and the continuous iteration of internet technology, advertising models have undergone tremendous changes. Currently, the most popular advertising delivery model is short video advertising. Short video platforms have a large user base, rapid data updates, and higher conversion rates for advertising. In the past, common advertising delivery technologies mainly included those based on big data analytics and those based on large-scale model recognition. In recent years, Privacy Set Intersection (PSI) technology has also been applied to some aspects of advertising delivery, but these technologies still suffer from inefficiencies and a lack of privacy protection when applied to short video advertising.

[0003] Advertising delivery methods based on big data analytics and large-scale model classification technologies are highly dependent on the accuracy of data tags and the model's analytical capabilities. However, in reality, tags typically only cover a general user group, resulting in coarse granularity. This leads to ads reaching a large number of users, but only a small portion being the client's target audience, resulting in low conversion rates. Similarity recommendation technologies based on collaborative filtering are highly dependent on positive feedback in the initial stages of delivery and are easily affected by outliers, making them unsuitable for the current short-video-dominated advertising model. Traditional Privacy Set Intersection (PSI, a secure multi-party computation technique used to calculate the intersection of two participating user sets without revealing non-intersecting elements in their sets) can calculate the intersection of advertisers' and clients' user bases and target ads accordingly. However, the intersection result is completely public to both parties. When the advertiser's user base is much larger than the client's, the intersection is almost equal to the client's entire user set, leading to complete exposure of the client's user privacy and limiting its application scenarios. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes an advertising delivery method and system based on intersection of marked privacy sets. Grounded in cryptographic principles and aided by large-scale model analysis, this method can determine the intersection of two sets without revealing redundant data, while simultaneously protecting customer privacy and enabling precise advertising delivery to improve conversion rates.

[0005] The method includes the following steps performed sequentially:

[0006] S1. Consensus Phase: The advertiser and the advertising client collaboratively initialize the public parameters of the 2HashDH protocol (Diffie-Hellman key exchange protocol). The advertising client provides the advertising content to be delivered to the advertiser, and the two parties negotiate to determine the matrix dimension parameters of the user element mapping. The advertiser uses a large model to parse the advertising content and extract content tags. Based on the content tags, the advertiser classifies and marks its own user set, resulting in a set of marked users and a set of unmarked users.

[0007] S2, Preprocessing stage: The advertiser generates a private selection string that is only visible to itself; the advertising client constructs two sets of initial key matrices, completes the matrix position mapping for its own target user set through the position mapping function, and corrects the key matrix parameters according to the mapping position to complete the initial construction of the encrypted mapping carrier;

[0008] S3, Unintentional Transmission Phase: The advertiser generates a dedicated matrix pair for the Unintentional Transmission Extension Protocol. Both parties generate private and non-disclosure selection strings. Secure selective interaction and parameter synchronization of matrix columns are completed through the Unintentional Transmission Extension Protocol. Both parties only obtain the matrix data required for their own calculations.

[0009] S4. Data Encryption Phase: Both parties complete key exchange and user data blinding processing based on the 2HashDH cryptographic protocol. Through a unified byte expansion rule, a unique and accurate mapping between user elements and matrix positions is achieved. The advertiser's end and the advertising client respectively complete the cryptographic conversion of their respective user sets. Data processing is carried out in cryptographic mode throughout the entire process.

[0010] S5. Data Sharing and Delivery Phase: The advertiser shares a set of encrypted labeled users and a set of encrypted unlabeled users with the advertising client; the advertising client solves for the intersection of the user sets in the encrypted environment, filters the labeled non-intersection users from the advertiser, combines the encrypted intersection users with the labeled non-intersection users to form an encrypted delivery set, and sends it back to the advertiser; the advertiser decrypts the encrypted delivery set, restores the real user identities, and completes targeted ad delivery.

[0011] During the data interaction process, advertisers and advertising clients cannot obtain each other's complete user data set.

[0012] The present invention also discloses an advertising delivery system based on the intersection of a tokenized privacy set, the system comprising an advertiser end and an advertising client end, the system performing the above-described method.

[0013] This invention achieves multiple technical benefits, including privacy protection, efficiency improvement, and accurate delivery, through core technology design, as detailed below:

[0014] 1. Enhanced Privacy Protection: Utilizing the 2HashDH protocol (a secure protocol based on the discrete logarithm problem for transmitting critical keys, ensuring the security of user element mapping and avoiding the risk of cracking due to both parties using the same hash and pseudo-random function) and Inadvertent Transmission (OT, a secure communication protocol that allows the receiver to obtain one or more messages from the sender without revealing its own choices, and the sender is unaware of which message the receiver obtained), encrypted processing of user data is achieved between the two parties. This avoids the privacy leakage problem caused by the public intersection in traditional PSI technology. Advertisers cannot obtain the complete target user set of the advertising client, and the advertising client cannot obtain the complete user set of the advertiser, ensuring that user privacy is not leaked.

[0015] 2. Lower computational overhead: A multi-point mapping encryption method is designed to map user elements onto a matrix and generate ciphertext by taking matrix values, replacing the traditional public-key encryption method. At the same time, the Unintentional Transfer Extension (OTE, an extension technology based on the basic OT protocol, generates a large number of OT instances based on symmetric encryption from a small number of public and private key OT instances, reducing the communication and computational overhead of matrix column transmission. Specifically, it can generate n OT instances based on symmetric encryption from m public-key OT instances) protocol reduces the computational overhead of large matrix transmission, making the solution more lightweight and suitable for scenarios with a large user base, such as short video platforms.

[0016] 3. Higher ad conversion rate: The solution includes both the overlapping users of the two user sets (the target users of the ad client) and the non-overlapping users marked by the advertiser who are highly relevant to the ad content, achieving precise targeting. At the same time, combined with large model analysis for user tagging, it further improves the matching degree between ads and users, effectively increasing the ad conversion rate.

[0017] 4. Enhanced adaptability: The solution is suitable for advertising scenarios on social media (especially short video platforms). It can achieve a balance between targeted delivery and privacy protection, given that advertisers have a large user base. This aligns with the current advertising delivery model dominated by short videos and has a wider range of applications. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a model for applying the method described in this invention.

[0019] Figure 2 This is a simplified diagram showing the mapping relationship between element x and the matrix.

[0020] Figure 3 for A schematic diagram of the output mapped onto a matrix. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the following description, in conjunction with specific embodiments and appendices, provides further details. Figure 1-3 The technical solution of the present invention will be further described in detail in this embodiment. This embodiment will fully explain all the steps of the advertising delivery method based on the intersection of marked privacy sets of the present invention, clarify the specific implementation process of each step, and verify the feasibility of the solution with examples.

[0022] 1. Application System Composition and Functions

[0023] like Figure 1 As shown, the system used in this method includes two core participants: the advertiser (participant P0) and the advertising client (participant P1). The two parties collaborate to complete targeted ad delivery. The components and functions are as follows:

[0024] 1.1 Advertiser's Side (Party P0)

[0025] The advertiser side is typically represented by social media companies such as short video companies and video companies. They are the main implementers of ad delivery, and their core functions include: possessing a large user base as a potential audience for ad delivery; extracting content tags through large-scale model analysis based on the ad content provided by the ad client, labeling their own user base, and dividing it into a labeled user set (users with high relevance to the ad content) and a non-labeled user set (users with low relevance to the ad content); generating a random 0-1 string S (selecting a string with a length consistent with the number of columns w in subsequent matrices M0 and M1, used to select matrix columns generated by the ad client during the inadvertent delivery phase); and participating in the inadvertent delivery extension (OTE) process to generate matrix M0. M1 (Two 0-1 matrices, with dimensions identical to matrices M0 and M1, are used for matrix swapping during unintentional transmission expansion, and M0...) With M1 (Saving specific row-level constraints), complete matrix exchange with the advertising client; map the set of marked users and the set of unmarked users onto the obtained mapping matrix D (matrix D is formed by the advertiser based on the selected bits of string S, through unintentional transmission expansion, and by selecting corresponding columns from M0 and M1 on the advertiser's side, used for encrypted mapping of user elements, with the same dimension as M0, n*w), generating the set of marked users and the set of unmarked users in the encrypted state; receive the delivery set sent by the advertising client (obtained by the advertising client, containing the intersection users in the encrypted state and the non-intersection users in the encrypted marked user set, which is the user basis for the advertiser's final ad delivery, and all users in the delivery set are users of advertiser P0), decrypt it using the mapping matrix D it possesses, restore the specific user identity, and complete the ad delivery. n is the number of client sets, w can be arbitrary, more is more secure, but the amount of computation is larger, so w should be as large as possible within the allowed computation time.

[0026] 1.2 Advertising Client (Partner P1)

[0027] The advertising client is a client with advertising delivery needs, possessing its own target user set SET (the advertising client's own target user set). Its core functions include: providing advertising content to advertisers as the basis for user tagging on the advertiser's end; generating two matrices M0 and M1 for mapping user elements (M0 is a random 0-1 matrix, M1 is initially a 0 matrix, both with dimensions n×w, where n is the number of rows and w is the number of columns, used to implement encrypted mapping of user elements), completing the mapping of its own user set onto the matrix; and generating a 0-1 selection string S. (Length and matrix M0) M1 The number of columns w is consistent, used to select M0 during the unintentional transmission extension phase. M1 (corresponding column in the table); participate in the 2HashDH protocol to generate blinding factors (two elements r0, r1 selected by the advertising client, belonging to the integer field Z*, where Z* represents the set of non-zero integers modulo m, and satisfies that r0 and r1 are inversely multiplied), calculate relevant parameters, and ensure the security of user element mapping; receive the set of densely labeled users and the set of densely unlabeled users sent by the advertiser, calculate the intersection with its own user set in the dense state, filter out the labeled users that do not intersect, form a delivery set, and send it to the advertiser.

[0028] The method is explained below.

[0029] Step 1: Consensus Phase

[0030] The core function of this step is to determine the core parameters for collaborative calculation by both parties, complete the user tagging on the advertiser's side, lay the foundation for the implementation of subsequent steps, and clarify the scope of candidate users and calculation standards for ad delivery.

[0031] S1.0: Both parties collaboratively complete the initialization of system common parameters: determine the common parameters of the 2HashDH protocol, including the cyclic group G (a common parameter of the 2HashDH protocol, which is a finite cyclic group), the generator g (a common parameter of the 2HashDH protocol, which is the generator element of the cyclic group G), the group order m (a common parameter of the 2HashDH protocol, which refers to the number of elements in the cyclic group G, i.e., the order of the group), and two collision-resistant hash functions. (The common parameters of the 2HashDH protocol are all collision-resistant hash functions, among which...) :{0,1}*→G, where the input is a string of arbitrary length and the output is the elements of the cyclic group G; :G→{0,1} λ The input is an element in a cyclic group G, and the output is a 0-1 string of length λ, where λ is a security parameter used to determine the ciphertext length.

[0032] S1.1: The advertising client (P1) provides the advertising content to be delivered to the advertiser (P0), clarifying the core appeal and target audience characteristics of the advertisement;

[0033] S1.2: Based on the size of their respective user sets, both parties negotiate and determine the matrix dimension parameters used for subsequent user element mapping, namely the number of matrix rows n and the number of columns w. n must be sufficient to cover the positional range required for user element mapping, and w must be consistent with the subsequently generated selection strings S and S'. Consistent length;

[0034] S1.3: The advertiser side (P0) uses a large model to analyze the advertising content provided by the advertising client (P1) and extract content tags that are highly relevant to the advertising content (e.g., if the advertisement is for a beauty product, the tags may include "women", "beauty lovers", "18-35 years old", etc.).

[0035] S1.4: The advertiser (P0) marks all its users based on the extracted content tags. Users who meet the tag requirements and are highly relevant to the advertising content are classified as marked users, while users who do not meet the tag requirements and are less relevant to the advertising content are classified as unmarked users. After completing the user marking, the advertiser saves both sets for subsequent encryption.

[0036] Step 2: Preprocessing Stage

[0037] The core function of this step is to generate the core matrix and selection string required for subsequent encrypted mapping and unintentional transmission between the two parties, complete the initial mapping of the advertising client user set on the matrix, and provide a carrier for subsequent secure transmission and data encryption.

[0038] S2.1: The advertiser (P0) randomly generates a 0-1 string S (selection string) of length w, where each bit of S is either 0 or 1. This string is used to select the corresponding column in the matrix M0 and M1 generated by the advertising client (P1) during the subsequent unintentional transmission phase. The string S is only stored by the advertiser (P0) and is not leaked to the advertising client (P1).

[0039] S2.2: The advertising client (P1) generates two matrices M0 and M1 with dimensions n×w, where M0 is a random 0-1 matrix and M1 is initially a 0 matrix (a 0 matrix means that each element in the matrix is ​​0); matrix M0 is the encryption key of P1.

[0040] S2.3: The advertising client (P1) maps each user element x in its target user set SET to a location using a location mapping function. (This is an unintentional pseudo-random function used to map user elements to specific positions in a matrix, achieving encrypted mapping of user elements, and outputting a sequence of position indices in the matrix.) Generates the corresponding position set P (the set of matrix positions corresponding to the position index sequence output by the position mapping function F(x), where each element is (i,j), representing the position in the i-th row and j-th column of the matrix).

[0041] Specific ideas Figure 2 For example. The output is 02315, corresponding to the positions of the 0th column and 0th row, the 1st column and 2nd row, the 2nd column and 3rd row, the 3rd column and 1st row, the 4th column and 5th row of the matrix, and so on.

[0042] This is used to extract the value mapped to each position on the matrix, which is 01110. Ultimately, 01110 is the encrypted value of that element. This process is based on a multi-point mapping encryption scheme, mapping elements onto a matrix and then extracting the encrypted value from the matrix. This significantly reduces computational overhead.

[0043] S2.4: The advertising client (P1) adjusts matrices M0 and M1 according to the location set P: For all positions (i,j)∈P (i.e., the positions of user element x mapped to in the matrix), let M1[i,j]=M0[i,j]; For all positions (i,j)∉P (i.e., the matrix positions not mapped to by user element x), let M1[i,j]=1-M0[i,j] (bitwise inversion), ensuring that the same user element is mapped to the same ciphertext on M0 and M1, and different user elements are mapped to different ciphertexts;

[0044] S2.5: The advertising client (P1) completes the mapping of its own user set SET onto matrices M0 and M1, and saves matrices M0 and M1 for subsequent unintentional transmission and encryption. P0 exchanges matrix D with P1 based on the unintentional transmission extension.

[0045] Step 3: Accidental Transmission Stage

[0046] The core function of this step is to enable secure exchange of matrices between the two parties through the Unintentional Transfer Extension (OTE) protocol, generate the mapping matrices and OT instances required by both parties, ensure the security of the matrix transmission process, and prevent both parties from leaking their own core matrix information.

[0047] S3.1: The advertiser's side (P0) generates two 0-1 matrices M0 with dimensions n×w. and M1 Both are parameters used for unintentional transmission of extended values. The generation rules are as follows: First, take the 0-1 string S generated in step two, copy it n times (n is the number of rows in the matrix), and concatenate them to form an n×w matrix M. (The n×w matrix generated by the advertiser is formed by copying the 0-1 string S from the preprocessing stage n times and concatenating them); then, an n×w 0-1 matrix M1 is randomly generated. Calculate M0 =M ⊕M1 (Bitwise XOR operation), where when M When the i-th row is all 0, M1 The i-th row has the same value as the i-th row of M0'; when M When the i-th row is all 1, M1 The i-th row and M0 The values ​​in the i-th row are inverses of each other;

[0048] S3.2: The advertising client (P1) randomly generates a 0-1 selection string S of length w. String S It is stored only by the advertising client (P1) and is not disclosed to the advertiser (P0);

[0049] S3.3: The advertising client (P1) selects the string S. Each value is used to generate a matrix M0 from the advertiser's (P0) side using a small number of public / private key OT instances. M1 Select the corresponding column, with the specific selection rule being: if S If [i] = 0 (the i-th element of S' is 0), then choose M0. The i-th column; if S [i]=1 (S If the i-th element is 1, then choose M1. The i-th column; the formula is as follows: ;

[0050] S3.4: The advertising client (P1) combines all selected columns into its own mapping matrix D. Complete the mapping matrix D The generation of matrix D, at this point, has already utilized the OTE protocol to generate a large number of OT instances based on symmetric encryption using a small number of public-key OT instances. Sum of matrix M0 Together they constitute the OT instances of both parties; matrix D Used for unintentional transmission extension; It is D The i-th row.

[0051] S3.5: The advertiser (P0) selects the corresponding column from the matrices M0 and M1 generated by the advertising client (P1) based on each value of the string S generated in step two. The specific selection rule is as follows: if S[i]=0 (the i-th element of S is 0), then select the i-th column of M0; if S[i]=1 (the i-th element of S is 1), then select the i-th column of M1. This completes the selection of the matrix columns required by the advertiser, ensuring that both parties can only obtain the matrix columns they need and cannot obtain the complete matrix M0, M1, or M2 of the other party. M1 Simultaneously satisfying the matrix row-level constraint relationship: ;

[0052] in, D The j-th row, M0 The j-th row, This represents the j-th selected position in string S. The length of S is w, and w is not necessarily equal to the number of rows n in the matrix.

[0053] Step 4: Data Encryption Stage

[0054] The core function of this step is to encrypt the user sets of both parties through the 2HashDH protocol and matrix mapping, generating a user set in a encrypted state, ensuring that user data is not leaked during transmission and processing, while reducing computational overhead.

[0055] S4.1: The advertising client (P1) selects two blinding factors from the integer field Z*. , (The two multiplications are inverses) For each user element x in its own user set SET, calculate And send a to the advertiser's end (P0);

[0056] S4.2: The advertiser (P0) randomly generates a large integer k (randomly generated by the advertiser and used for key calculation in the 2HashDH protocol), receives a from the advertising client (P1), calculates b = aᵏ (a raised to the power of k, group operation), and sends b to the advertising client (P1).

[0057] S4.3: After receiving b, the advertising client (P1) calculates c1 = The advertiser (P0) calculates the following for each user element y in its user set: If x = y (i.e., the intersection of the user sets of both parties), then c1 = c0, and at this time both parties pass through... After the post-mapping, the same sequence of position indices is obtained, ensuring that the intersection elements map to the same position in the matrix; where ;

[0058] S4.4: Both parties use the same pseudo-random number generator to... The output is padded to nw bits (n is the number of rows in the matrix (equal to the number of client sets), and w is the number of columns in the matrix). Each n bits corresponds to one column of the matrix, and the value of each n bits covers the number of rows in the matrix, achieving a complete mapping from user elements to matrix positions. Figure 3 For example, grouping bytes into twos allows us to represent four columns. The group 01 represents the first position in column 0 (starting from position 0). The group 10 represents the second position in column 1, and so on. This method maps elements to the matrix. Therefore, if both parties possess the matrix, only the intersection can yield the same ciphertext.

[0059] S4.5: The advertiser's side (P0) maps the marked user set and the unmarked user set from step one to each using the location mapping function. Map the data onto the matrix columns obtained in step three, and then use the matrix value encryption function. (Input position index sequence and matrix, output ciphertext string composed of corresponding position values, where k refers to the encryption key used in the encryption operation) Extract the values ​​of corresponding positions in the matrix to obtain the encrypted value of each user element, and then generate the set of marked users in the encrypted state and the set of unmarked users in the encrypted state;

[0060] The S4.6 advertising client (P1) generates a cryptic user set ESET (the cryptic set obtained by matrix mapping and encryption of the advertising client's user set SET for each user element x in its own user set SET) through the same mapping and encryption process described above.

[0061] S4.7 Both parties save their respective generated secret-state user sets for future data sharing and intersection calculations.

[0062] Step 5: Data Sharing Phase

[0063] The core function of this step is to achieve secure sharing of encrypted user sets, filter out the ad delivery sets under encrypted conditions, complete decryption and targeted ad delivery, and achieve the dual goals of accurate delivery and privacy protection.

[0064] S5.1 The advertiser (P0) sends the set of marked users and the set of unmarked users in the secret state generated in step four to the advertising client (P1). Since both parties are processed through secret state mapping, the advertising client (P1) cannot obtain the original user information of the advertiser (P0) from the secret state set.

[0065] After receiving the encrypted set, the advertising client (P1) calculates the intersection of its own encrypted user set ESET and the encrypted user set sent by the advertiser (P0) in encrypted form. This intersection is the encrypted intersection of the user sets of both parties (i.e., the group of target users of the advertising client (P1) that also belong to the unmarked users of the advertiser (P0)).

[0066] S5.3: The advertising client (P1) adds all the dense intersections to the delivery set, and then selects some non-intersection users from the densely marked user set sent by the advertiser (P0) to add to the delivery set (non-intersection users are those marked by the advertiser (P0) who are highly relevant to the advertisement but do not belong to the target users of the advertising client (P1)); the number of elements in the set depends on the advertiser's needs.

[0067] S5.4: The advertising client (P1) limits the size of the delivery set according to the delivery strategy and payment situation agreed upon by the advertising client (P1) and the advertiser (P0) during the consensus phase, to ensure that the users in the delivery set are all groups with a high relevance to the advertising content;

[0068] S5.5: The advertising client (P1) sends the filtered delivery set (in encrypted form) to the advertiser (P0).

[0069] S5.6: After receiving the delivery set, the advertiser (P0) uses the relevant parameters of the mapping matrix it has mastered to decrypt each encrypted user element in the delivery set and restore the specific real user identity.

[0070] S5.7: The advertiser (P0) delivers the advertising content provided by the advertising client (P1) to all users in the decrypted delivery set, completing the entire advertising delivery process.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An ad delivery method based on intersection of marked privacy sets, characterized in that, The method includes the following steps performed sequentially: S1. Consensus Phase: The advertiser and the advertising client work together to initialize the common parameters of the 2HashDH protocol. The advertising client provides the advertising content to be delivered to the advertiser. The two parties negotiate and determine the matrix dimension parameters for mapping user elements. The advertiser uses a large model to parse the advertising content and extract content tags. Based on the content tags, the advertiser classifies and marks its own user set, resulting in a set of marked users and a set of unmarked users. S2, Preprocessing stage: The advertiser generates a private selection string that is only visible to itself; the advertising client constructs two sets of initial key matrices, completes the matrix position mapping for its own target user set through the position mapping function, and corrects the key matrix parameters according to the mapping position to complete the initial construction of the encrypted mapping carrier; S3, Unintentional Transmission Phase: The advertiser generates a dedicated matrix pair for the Unintentional Transmission Extension Protocol. Both parties generate private and non-disclosure selection strings. Secure selective interaction and parameter synchronization of matrix columns are completed through the Unintentional Transmission Extension Protocol. Both parties only obtain the matrix data required for their own calculations. S4. Data Encryption Phase: Both parties complete key exchange and user data blinding processing based on the 2HashDH cryptographic protocol. Through a unified byte expansion rule, a unique and accurate mapping between user elements and matrix positions is achieved. The advertiser's end and the advertising client respectively complete the cryptographic conversion of their respective user sets. Data processing is carried out in cryptographic mode throughout the entire process. S5. Data Sharing and Delivery Phase: The advertiser shares a set of encrypted labeled users and a set of encrypted unlabeled users with the advertising client; the advertising client solves for the intersection of the user sets in the encrypted environment, filters the labeled non-intersection users from the advertiser, combines the encrypted intersection users with the labeled non-intersection users to form an encrypted delivery set, and sends it back to the advertiser; the advertiser decrypts the encrypted delivery set, restores the real user identities, and completes targeted ad delivery. During the data interaction process, advertisers and advertising clients cannot obtain each other's complete user data set.

2. The advertising delivery method based on the intersection of marked privacy sets according to claim 1, characterized in that, The S1 consensus phase specifically includes: S1.1 The advertiser and the advertising client collaboratively initialize the common parameters of the 2HashDH protocol. These common parameters include a finite cyclic group G, a cyclic group generator g, a group order m, a security parameter λ, and two collision-resistant hash functions, which respectively satisfy: :{0,1}*→G、 :G→{0,1} λ ; S1.2 The advertising client pushes the advertising content to be delivered to the advertiser, clarifying the target audience characteristics and delivery requirements of the advertising; S1.

3. Both parties negotiate and determine the number of rows n and columns w of the mapping matrix based on the size of their respective user sets. The number of rows n covers the mapping position range of all user elements, and the number of columns w is consistent with the length of the private selection string of both parties in the future. S1.4 Advertisers analyze ad content through a large model, extract refined user tags, and classify all their users based on user tags. Highly relevant users are divided into a set of labeled users, and lowly relevant users are divided into a set of unlabeled users, which are then stored locally for later use.

3. The advertising delivery method based on the intersection of marked privacy sets according to claim 2, characterized in that, The S2 preprocessing stage specifically includes: S2.1 The advertiser randomly generates a 0-1 private selection string S of length w. The string S is stored locally on the advertiser's end and is not disclosed to the advertising client. S2.2 The advertising client constructs a dual-key matrix with dimension n×w, including a random 0-1 matrix M0 and an initial zero matrix M1, wherein M0 and M1 are the core key matrices for the encryption mapping of user elements; S2.3, The advertising client calls the location mapping function. For each user element x in the target user set SET, perform position mapping to generate the corresponding matrix position index sequence and position set P; S2.4 The advertising client corrects the parameters of the dual matrix based on the location set P: For matrix positions (i,j) belonging to P, let M1[i,j]=M0[i,j]; for matrix positions (i,j) not belonging to P, let M1[i,j]=1-M0[i,j], so that the ciphertext of the same user element is consistent and the ciphertext of different user elements is unique. S2.5 The advertising client completes the matrix mapping and solidification of all user elements, and the advertiser obtains the exclusive user mapping matrix D through subsequent unintentional transmission of extended protocol interactions.

4. The advertising delivery method based on the intersection of marked privacy sets according to claim 3, characterized in that, The S3 unintentional transmission phase specifically includes: S3.1, The advertiser generates an n×w dimension transition matrix M based on the selected string S. Copy string S n times and concatenate them to get M. Randomly generate an n×w dimension 0-1 matrix M1 By performing a bitwise XOR operation M0 =M ⊕M1 Generating matrix M0 M0 M1 Satisfies the preset row-level constraints; S3.2 The advertising client randomly generates a private 0-1 selection string S of length w. The string S The ads are stored locally on the client side and will not be disclosed to external parties. S3.3, the advertising client is based on S Each value selects a column in the matrix: S When [i]=0, select the i-th column of M0', S When [i]=1, select M1. Column i; S3.4 The advertising client concatenates all selected matrix columns to generate its own unique mapping matrix D. By inadvertently transmitting an extended protocol, a large number of symmetric encrypted OT instances are generated from a small number of public-private key OT instances, matrix D. With M0 Together they constitute the OT instances of both parties; S3.

5. The advertiser selects each value of the string S according to its own private selection, and selects the corresponding matrix columns of the advertising client matrix M0 and M1: when S[i]=0, the i-th column of M0 is selected, and when S[i]=1, the i-th column of M1 is selected, and the combination generates a unique mapping matrix D; both parties only obtain the matrix data they need and cannot obtain the complete matrix of the other party; the interaction satisfies the row-level constraint.

5. The advertising delivery method based on intersection of marked privacy sets according to claim 4, characterized in that, The S4 data encryption phase specifically includes: S4.1 The advertising client selects a pair of multiplicative inverse blinding factors from the integer field Z*. , Calculate for each user element x And send parameter 'a' to the advertiser; S4.2 The advertiser generates a random encrypted large integer k, receives parameter a, calculates b=aᵏ through group operation, and sends parameter b back to the advertiser client. S4.3, The advertising client calculates the encrypted parameter c1= Advertisers calculate their own user element y. The user elements at the intersection of the two sides satisfy c1=c0 and are mapped to the same position in the matrix; S4.4 Both parties use a unified pseudo-random number generator to generate... (·) The output is expanded to nw bits, and according to the rule that every n bits corresponds to a column of the matrix, the precise and unique mapping between user elements and matrix positions is completed. S4.5, Advertisers use location mapping functions The set of labeled users and the set of unlabeled users are mapped to a dedicated mapping matrix. The ciphertext is extracted by the matrix value encryption function to generate a set of encrypted labeled users and a set of encrypted unlabeled users. S4.6 The advertising client encrypts its own target user set SET using the same mapping encryption rule, generates a encrypted user set ESET, and stores it locally.

6. The advertising delivery method based on intersection of marked privacy sets according to claim 5, characterized in that, The S5 data sharing and delivery phase specifically includes: S5.1 The advertiser sends the complete set of encrypted labeled users and the complete set of encrypted unlabeled users to the advertising client. S5.2 In the encrypted environment, the advertising client calculates the intersection of its own encrypted user set ESET and the advertiser's encrypted user set to obtain the target user encrypted set that overlaps between the two. S5.3 The advertising client will include all densely intersecting users into the delivery set, and will select highly relevant non-intersecting users from the densely marked user set on the advertiser's side and add them to the delivery set; S5.4 The advertising client limits the size of the delivery set according to the delivery strategy and payment rules preset by both parties; S5.5 The advertising client sends the filtered encrypted delivery set back to the advertiser's end; S5.6 Advertisers use local mapping matrix parameters and encryption keys to decrypt the encrypted delivery set and restore the real user identity information. S5.7 The advertiser pushes the decrypted target user group with the advertising content to complete the advertising delivery.

7. The advertising delivery method based on intersection of marked privacy sets according to claim 4, characterized in that, The row-level constraint formula for the S3 unintentional transmission phase is: ; Where, when r=0, q=t; when r=1, q=t⊕s; q is the mapping matrix D. The j-th row, t is matrix M0 The j-th row is where r is the j-th selection bit of the string S. The length of string S is w, and w is not necessarily equal to the number of rows n in the matrix.

8. The advertising delivery method based on intersection of marked privacy sets according to claim 1, characterized in that, The advertiser is a social media service provider on a short video platform, while the advertising client is an external entity with commercial advertising needs. The two parties only exchange and calculate encrypted data throughout the entire process.

9. The advertising delivery method based on intersection of marked privacy sets according to claim 5, characterized in that, The position mapping function is: .

10. An advertising delivery system based on intersection of marked privacy sets, characterized in that, The system includes an advertiser end and an advertising client end, and the system executes the advertising delivery method based on the intersection of a tokenized privacy set as described in any one of claims 1 to 9.

Images