Multi-party participated advertisement effect evaluation method, system and device and storage medium

Abstract

The invention relates to the technical field of data processing, and particularly provides a multi-party participation advertisement effect evaluation method, system and device and a storage medium, and the method comprises the steps: obtaining exposure feature data from a media platform, and predicting a click rate according to the exposure feature data through a locally deployed prediction model; comparing the predicted click rate with a preset expected click rate, and outputting a comparison result as an evaluation result; the prediction model is updated by adopting a multi-party participated federated learning mechanism, in the model updating process, participants perform cooperative training based on statistical data of the participants, and encryption protection is performed on transmitted gradient information by applying a homomorphic encryption scheme; the statistical data comprises the encrypted exposure of the plurality of participants within a specified period. According to the invention, unification of cross-platform accuracy and privacy security of advertisement effect evaluation is realized.

Description

Technical Field

[0001] The present invention belongs to the field of data processing, and in particular relates to a method, system, device and storage medium for evaluating advertising effects involving multiple parties. Background Art

[0002] Current advertising effectiveness evaluation primarily relies on centralized data analysis provided by media platforms, aggregating user behavior data to calculate metrics like click-through rates. Traditional methods pose privacy risks, as raw data transmission is susceptible to interception, and data aggregation can expose sensitive user information. While existing federated learning solutions can ensure data remains within the domain, they lack fine-grained privacy protection mechanisms tailored to advertising scenarios, and the gradient exchange process still presents the risk of model inversion attacks. Summary of the Invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method, system, device and storage medium for advertising effectiveness evaluation with multi-party participation to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides a method for evaluating advertising effectiveness with multi-party participation, comprising: Obtain exposure feature data from the media platform, and use a locally deployed prediction model to predict click-through rate based on the exposure feature data; Comparing the predicted click-through rate with a preset expected click-through rate, and outputting the comparison result as an evaluation result; The prediction model is updated using a federated learning mechanism involving multiple parties. During the model update process, each participant conducts collaborative training based on their statistical data, and uses a homomorphic encryption scheme to encrypt and protect the transmitted gradient information. The statistical data includes the encrypted exposure of multiple participants within a specified period. In the process of generating the statistical data, differential privacy protection technology is used to inject noise into the original data, and the data after noise injection is anonymized.

[0005] In an optional embodiment, the federated learning mechanism includes: Advertisers use the gradient descent algorithm to train the locally deployed prediction model based on locally stored statistical data, calculating the loss value and the gradient of the model parameters in each training iteration. The advertiser homomorphically encrypts the gradient and uploads the encrypted gradient to the parameter server; The parameter server receives encrypted gradients from multiple advertisers, aggregates the encrypted gradients, and obtains the encrypted global gradient; The parameter server distributes the encrypted global gradient to advertisers; The advertiser decrypts the encrypted global gradient and uses the decrypted gradient to update the parameters of the locally deployed prediction model.

[0006] In an optional embodiment, the prediction model adopts a deep neural network model, which includes an input layer, multiple hidden layers and an output layer.

[0007] In an optional embodiment, the parameter server receives encrypted gradients from multiple advertisers, aggregates the encrypted gradients, and obtains an encrypted global gradient, including: The parameter server obtains the credit distribution ratio of each advertiser uploaded by the trusted third party; The parameter server calculates the advertiser's weight according to the credit distribution ratio, and calculates the probability of the advertiser participating in the training based on the advertiser's weight; The parameter server calculates the encrypted global gradient based on the probability of the advertiser participating in training and the encrypted gradient uploaded by the advertiser.

[0008] In an optional embodiment, the method for calculating the credit distribution ratio includes: The trusted third party obtains the encrypted exposure timestamp from the advertiser and the desensitized conversion time from the media platform; The trusted third party calculates the time decay factor based on the encrypted exposure timestamp and the desensitized conversion time through a secure multi-party computing protocol without decryption, and calculates the credit distribution ratio of each advertiser based on the time decay factor.

[0009] In an optional embodiment, the method further comprises: Each participant periodically aggregates exposures locally to obtain aggregated data; Each participant performs differential privacy protection and anonymization on the aggregated data to obtain encrypted aggregated data; Each participant uploads the encrypted summary data generated by themselves to a trusted third party; A trusted third party uses a secret sharing protocol to securely aggregate the encrypted summary data to obtain statistical data.

[0010] In an optional embodiment, a trusted third party securely aggregates the encrypted summary data using a secret sharing protocol to obtain statistical data, including: The encrypted summary data is hierarchically aggregated based on the historical reputation scores of each party.

[0011] In a second aspect, the present invention provides an advertising effectiveness evaluation system involving multiple parties, comprising: A prediction module, configured to obtain exposure feature data from a media platform and predict click-through rate based on the exposure feature data using a locally deployed prediction model; A comparison module, configured to compare the predicted click-through rate with a preset expected click-through rate, and output the comparison result as an evaluation result; The prediction model is updated using a federated learning mechanism involving multiple parties. During the model update process, each participant conducts collaborative training based on their statistical data, and uses a homomorphic encryption scheme to encrypt and protect the transmitted gradient information. The statistical data includes the encrypted exposure of multiple participants within a specified period. In the process of generating the statistical data, differential privacy protection technology is used to inject noise into the original data, and the data after noise injection is anonymized.

[0012] According to a third aspect, a device is provided, comprising: A memory for storing an advertising effect evaluation program involving multiple parties; The processor is configured to implement the steps of the multi-party advertising effectiveness evaluation method provided in the first aspect when executing the multi-party advertising effectiveness evaluation program.

[0013] In a fourth aspect, a computer-readable storage medium is provided, on which a multi-party advertising effectiveness evaluation program is stored. When the multi-party advertising effectiveness evaluation program is executed by a processor, the steps of the multi-party advertising effectiveness evaluation method provided in the first aspect are implemented.

[0014] The beneficial effects of the present invention are that the multi-party advertising effect evaluation method, system, device and storage medium provided by the present invention integrate the encrypted feature data of each participant through federated learning, so that the prediction model obtains more comprehensive user behavior characteristics and breaks through the limitations of data islands. Secondly, the dual protection of homomorphic encryption and differential privacy is adopted to prevent the leakage of model parameters in the gradient transmission stage, and the original data is made untraceable through noise injection in the data statistics stage, and the privacy protection strength reaches the financial-level security standard. By comparing the predicted click-through rate with the advertiser's expected click-through rate, an evaluation result is obtained. If the evaluation result is that the predicted click-through rate does not reach the expected click-through rate, it means that the advertising delivery strategy needs to be changed. The evaluation system has dynamic optimization capabilities. Each participant participates in the model iteration by continuously contributing encrypted data, so that the evaluation indicators reflect market changes in real time. Compared with the traditional periodic evaluation model, it can improve the timeliness of abnormal advertising identification. Under the premise of ensuring the sovereignty of commercial data, this method has achieved the unification of cross-platform accuracy and privacy security of advertising effect evaluation for the first time.

[0015] In addition, the present invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention.

[0018] Figure 2 FIG. 4 is a schematic block diagram of a system according to an embodiment of the present invention.

[0019] Figure 3 A schematic structural diagram of a device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0022] The advertising effect evaluation method with multi-party participation provided in the embodiment of the present invention is executed by a computer device. Accordingly, the advertising effect evaluation system with multi-party participation runs in the computer device.

[0023] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention. Figure 1 The execution entity can be a multi-party advertising effect evaluation system. According to different needs, the order of the steps in the flowchart can be changed, and some steps can be omitted.

[0024] like Figure 1 As shown, the method includes: S1. Obtain exposure feature data from the media platform and use a locally deployed prediction model to predict click-through rate based on the exposure feature data; S2. Compare the predicted click-through rate with the preset expected click-through rate and output the comparison result as an evaluation result; The prediction model is updated using a federated learning mechanism involving multiple parties. During the model update process, each participant conducts collaborative training based on their statistical data, and uses a homomorphic encryption scheme to encrypt and protect the transmitted gradient information. The statistical data includes the encrypted exposure of multiple participants within a specified period. In the process of generating the statistical data, differential privacy protection technology is used to inject noise into the original data, and the data after noise injection is anonymized.

[0025] In an embodiment of the present invention, based on step S1, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0026] During the data collection phase, this study established a data transmission channel through the media platform's open API interface and utilized the OAuth 2.0 protocol for security authentication to ensure legal data acquisition. HTTPS was used for encrypted transmission to prevent data leakage and tampering. The acquired data encompassed multiple dimensions of exposure characteristics, including user profile characteristics (such as age, gender, region, and spending habits), ad content characteristics (such as ad copy, image material, and video length), and delivery environment characteristics (such as delivery time, channel, and device type). The acquired data was stored in a local distributed file system (such as Hadoop Distributed File System, HDFS), leveraging its high fault tolerance and scalability to ensure stable storage and efficient access.

[0027] For click-through rate prediction, the locally deployed prediction model is built using a deep learning framework (such as TensorFlow or PyTorch). The model utilizes a deep neural network (DNN) architecture, with the input layer receiving preprocessed exposure feature data. This preprocessing includes normalizing numerical data (e.g., using Min-Max normalization to map data to the [0, 1] range) and one-hot encoding (One-Hot Encoding) of categorical data to eliminate dimensional differences between data features and improve model training. The hidden layer uses multiple layers of neurons for feature extraction and nonlinear transformation. The ReLU (Rectified Linear Unit) activation function mitigates the vanishing gradient problem and enhances the model's expressiveness. The output layer uses a sigmoid function to map the model output to the [0, 1] range; the output value represents the predicted click-through rate. During the model training phase, stochastic gradient descent (SGD) and its variants (such as Adagrad, Adadelta, and Adam) are used for parameter optimization. The cross-entropy loss function is used as the optimization target. Parallel computing technology is used on a local computing cluster to accelerate the model training process. Through multiple iterative training, the model reaches the optimal state, thereby realizing the click-through rate prediction of the input exposure feature data.

[0028] In an embodiment of the present invention, based on step S2, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0029] When comparing a predicted CTR with a preset expected CTR, first clarify the basis for setting the preset expected CTR. The expected CTR can be set based on statistical analysis of historical advertising data (e.g., calculating the average CTR of a certain type of ad under specific delivery conditions), industry benchmarks, or business objectives. During the comparison process, use error metrics such as mean absolute error (MAE), mean relative error (MRE), or mean squared error (MSE) to quantify the difference between the predicted and expected CTRs.

[0030] Based on the calculated error metric, the comparison results are categorized and output as evaluation results. If the error between the predicted and expected CTRs falls within a preset acceptable threshold, the evaluation is deemed "meeting the criteria." If the error exceeds the threshold, further analysis of the cause based on the magnitude and direction of the error results in either "overestimate" (the predicted CTR is significantly higher than the expected CTR) or "underestimate" (the predicted CTR is significantly lower than the expected CTR), accompanied by specific error values ​​and analysis, providing data support for subsequent ad placement strategy optimization and model adjustments.

[0031] In a specific implementation, the federated learning mechanism includes: (1) Advertisers use the gradient descent algorithm to train the locally deployed prediction model based on locally stored statistical data, and calculate the loss value and the gradient of the model parameters in each training iteration.

[0032] During model training, advertisers build a training dataset based on locally stored statistical data. This dataset contains historical ad placement records, covering multiple dimensions, including ad impressions, clicks, user behavior (such as dwell time and pageview depth), and creative attributes (copy length and image resolution). To ensure data quality, data cleaning is required, including handling missing values ​​(using methods such as mean filling and multiple imputation), removing duplicate records, and identifying and correcting outliers.

[0033] A gradient descent algorithm is used to iteratively optimize the locally deployed prediction model. Prediction models are typically deep learning architectures (such as multilayer perceptrons and deep neural networks) or machine learning models (such as logistic regression and random forests). In each training iteration, the model receives input data and computes a predicted output through forward propagation. Based on the predicted output and the true label (such as whether the ad was clicked), a loss function (such as cross-entropy loss for classification tasks and mean squared error loss for regression tasks) is used to calculate a loss value, measuring the degree of discrepancy between the model's prediction and the true reality. The backpropagation algorithm is then used to calculate the gradient of the loss function with respect to the model parameters (such as weights and biases in a neural network). This algorithm uses the chain rule to propagate gradient information backward from the output layer to the input layer, efficiently calculating the gradient value of each parameter and providing the direction and step size for model parameter updates.

[0034] (2) The advertiser homomorphically encrypts the gradient and uploads the encrypted gradient to the parameter server.

[0035] To ensure data privacy, advertisers use homomorphic encryption to encrypt the calculated gradients. Homomorphic encryption uses algorithms that support additive or multiplicative homomorphism, such as the Paillier homomorphic encryption algorithm. This algorithm allows specific operations (such as addition and multiplication) to be performed directly on ciphertext. The decrypted results are consistent with the plaintext results.

[0036] During the encryption process, the advertiser first generates an encryption key pair (public and private keys) and uses the public key to encrypt the gradient data. The encrypted gradient data is stored in ciphertext and uploaded to the parameter server via a secure communication protocol (such as TLS / SSL). During transmission, digital signature technology is used to ensure data integrity and source authenticity, preventing data tampering or malicious forgery.

[0037] (3) The parameter server receives encrypted gradients from multiple advertisers, aggregates the encrypted gradients, and obtains the encrypted global gradient.

[0038] The parameter server, the core node of distributed training, receives encrypted gradient data from multiple advertisers. Before aggregating the encrypted gradients, the parameter server obtains each advertiser's credit allocation from a trusted third party. This allocation is determined based on a comprehensive assessment of factors such as the quality of the advertiser's data (e.g., sample size and diversity) and their computing resource contributions (e.g., local computing power and training duration).

[0039] Based on the credit distribution ratio, the parameter server determines the advertiser's weight through normalization and other calculation methods. The advertiser's weight reflects its relative importance in global model training. Based on the advertiser's weight, a probabilistic generative model (such as multinomial distribution sampling) is used to calculate the probability of the advertiser participating in training. Advertisers with higher probability values ​​receive a greater weight in the global gradient calculation.

[0040] When calculating the encrypted global gradient, the parameter server performs a weighted sum of the encrypted gradients uploaded by each advertiser based on their probability of participating in training. Due to the use of homomorphic encryption, this aggregation operation can be performed directly on the ciphertext without decrypting the gradient data, thereby protecting advertiser data privacy. The result of this aggregation is the encrypted global gradient, which represents the combined update direction of the gradients of all participating advertisers.

[0041] In one example, a method of aggregating encrypted gradients includes: 1) The parameter server obtains the credit distribution ratio of each advertiser uploaded by the trusted third party.

[0042] 2) The parameter server calculates the advertiser's weight according to the credit distribution ratio, and calculates the probability of the advertiser participating in the training based on the advertiser's weight.

[0043] For advertiser k, the weight of its t-th round of training is determined by both its historical contribution and recent contribution:

[0044] Where γ represents the historical weight attenuation coefficient (for example, γ = 0.9, retaining 90% of the historical contribution); M k represents the set of impressions contributed by advertiser k in period t; Indicates the percentage of credit for the mth impression (from the most recent conversion event).

[0045] In each round of federated learning, the parameter server selects advertisers to participate in training according to the following probabilities:

[0046] Rep(k) represents the reputation score of advertiser k (initial 1, drops to 0 if advertiser quits midway); K represents the total number of currently active advertisers.

[0047] 3) The parameter server calculates the encrypted global gradient based on the probability of the advertiser participating in training and the encrypted gradient uploaded by the advertiser.

[0048] When the parameter server aggregates the gradients, it performs a weighted average on the encrypted gradients:

[0049] ∇ k represents the encrypted gradient uploaded by advertiser k; ∇ agg Represents the aggregated encrypted global gradient.

[0050] By adjusting the weights of each participant, high-contribution advertisers (p j The weight of advertisers with high contributions (high) is increased, thereby increasing their probability of participating in training and participating in federated learning more frequently. Advertisers with low contributions or frequent withdrawals are gradually eliminated.

[0051] (4) The parameter server distributes the encrypted global gradient to the advertiser.

[0052] After the parameter server completes the encrypted global gradient calculation, it distributes it to each advertiser via a secure and reliable communication mechanism. To ensure the accuracy and integrity of data transmission, checksums (such as CRC) or hash value verification are used to verify the transmitted data. During the distribution process, data transmission strategies (such as resuming transmission and flow control) are dynamically adjusted based on the advertiser's network environment and computing power to prevent transmission failures due to network failures or data congestion.

[0053] (5) The advertiser decrypts the encrypted global gradient and uses the decrypted gradient to update the parameters of the locally deployed prediction model.

[0054] After receiving the encrypted global gradient, the advertiser decrypts it using the locally stored private key to recover the plaintext global gradient. Based on the decrypted global gradient, the locally deployed prediction model parameters are updated using a parameter update strategy (such as stochastic gradient descent, the adaptive gradient algorithm Adagrad, or the Adam algorithm).

[0055] During the update process, the model parameter update amount is calculated based on the learning rate (either a fixed learning rate or a dynamically adjusted learning rate strategy) and the global gradient. This update is then applied to the original model parameters to complete iterative optimization of the model parameters. By repeating this training, encrypted upload, aggregated distribution, and decryption update process multiple times, the predictive model can be collaboratively trained and continuously optimized in a distributed environment, improving the model's accuracy in predicting business metrics such as ad click-through rates.

[0056] In one embodiment, a trusted third party is introduced. The trusted third party uses a secret sharing protocol to securely aggregate the encrypted summary data uploaded by each participant to obtain statistical data, and calculates the credit distribution ratio of each advertiser based on the statistical data.

[0057] 1. Symbol definition.

[0058] Participants: P1, P2, …, P N , each participant P i After preprocessing, the local data contains: k-anonymized equivalence class data:

[0059] x ij : The exposure of the jth equivalence class; c ij : The number of transformations of the jth equivalence class; k ij : Equivalence class size (satisfying k ij ≥k); Noise parameters: Laplace noise L(λ i ), where λ i =ϵ i / 2Δf; ϵ i :Participant P i allocated privacy budget; Δf=1 (sum sensitivity); Trusted Third Party (TTP): Coordinates secret sharing and aggregation without accessing the original data.

[0060] 2. Hierarchical dynamic secret sharing and aggregation.

[0061] Step 1: Local processing by participating parties. Each participant P i For each equivalence class j, perform the following operations: Impressions and conversions after noise addition:

[0062] Noise intensity and k ij Inversely proportional to the size of the group, k-anonymization is used to reduce the impact of noise. For example, if k ij =100, the noise amount is only 1 / 100 of the traditional method (k=1).

[0063] Dynamic weight allocation:

[0064] Among them, Repi Represents participant P i Historical reputation scores (e.g., data quality assessments).

[0065] Weighted Secret Sharing: Divide the noise data into t parts according to the weight (t <N):

[0066] Per Serving and Assigned to different parties or TTPs.

[0067] Step 2: TTP Security Aggregation Layered aggregate exposure, the calculation formula for aggregate exposure is:

[0068] TTP only sums the shares without decrypting the original data.

[0069] Layered polymerization conversion rate, the calculation formula of polymerization conversion rate is:

[0070] δ=1, which means the smoothing factor prevents division by zero errors.

[0071] The dynamic privacy budget allocation formula is:

[0072] DataVol i Represents participant P i The amount of data (such as the total number of equivalence classes M i ), participants with high reputation and large data volumes are allocated more privacy budget to improve global statistical accuracy.

[0073] Step 3: Calculate the credit distribution ratio.

[0074] The trusted third party obtains the encrypted exposure timestamp from the advertiser and the desensitized conversion time from the media platform. Using a secure multi-party computing protocol, the trusted third party calculates a time decay factor based on the encrypted exposure timestamp and desensitized conversion time without decrypting the information. This factor is then used to calculate the credit distribution ratio for each advertiser.

[0075] Time decay factor calculation: For the jth ad exposure event:

[0076] is the attenuation coefficient (needs to be adjusted according to the business scenario, for example, λ=0.01 indicates attenuation strength). The time of occurrence of the jth exposure, t conversion Indicates the time when a conversion event occurred.

[0077] Calculate the credit distribution ratio. For a certain conversion event, the credit ratio of all associated exposures {j} is:

[0078] Among them, C represents the set of all exposure events that may affect the conversion, p j Indicates the contribution ratio of the j-th exposure to conversion.

[0079] In some embodiments, the multi-party advertising effectiveness evaluation system may include multiple functional modules composed of computer program segments. The computer program of each program segment in the multi-party advertising effectiveness evaluation system may be stored in a memory of a computer device and executed by at least one processor to perform (see Figure 1 Description) The function of advertising effectiveness evaluation with multi-party participation.

[0080] In this embodiment, the multi-party advertising effect evaluation system can be divided into multiple functional modules according to the functions it performs, such as Figure 2 As shown. The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can perform fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0081] A prediction module, configured to obtain exposure feature data from a media platform and predict click-through rate based on the exposure feature data using a locally deployed prediction model; A comparison module, configured to compare the predicted click-through rate with a preset expected click-through rate, and output the comparison result as an evaluation result; The prediction model is updated using a federated learning mechanism involving multiple parties. During the model update process, each participant conducts collaborative training based on their statistical data, and uses a homomorphic encryption scheme to encrypt and protect the transmitted gradient information. The statistical data includes the encrypted exposure of multiple participants within a specified period. In the process of generating the statistical data, differential privacy protection technology is used to inject noise into the original data, and the data after noise injection is anonymized.

[0082] Figure 3The multi-party advertising effect evaluation method provided for the embodiment of the present application can be applied to a device. Those skilled in the art will understand that the device structure involved in the embodiment of the present invention does not constitute a limitation on the device, and the device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. In the embodiment of the present invention, the device includes but is not limited to a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or required herein.

[0083] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will appreciate that the server structure shown in the figure does not limit the present invention. The server structure may be a bus structure or a star structure, and may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0084] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 can perform some or all of the steps in the above-described method embodiments.

[0085] The processor 310 is the control center of the storage device, which uses various interfaces and lines to connect various parts of the entire electronic device. It executes various functions of the electronic device and / or processes data by running or executing software programs and / or modules stored in the memory 320, and calling data stored in the memory. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 310 can only include a central processing unit (CPU). In an embodiment of the present invention, the CPU can be a single computing core or multiple computing cores.

[0086] The communication unit 330 is configured to establish a communication channel so that the storage device can communicate with other devices, receive user data sent by other devices, or send user data to other devices.

[0087] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program that, when executed, may include some or all of the steps of each embodiment provided herein. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0088] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software and a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code, and includes instructions for causing a computer device (which can be a personal computer, a server, or a second device, a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0089] In this specification, the same or similar parts between the various embodiments can be referred to each other. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

[0090] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or modules, and can be electrical, mechanical or other forms.

[0091] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0092] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0093] Although the present invention has been described in detail with reference to the accompanying drawings and in conjunction with preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, persons of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and such modifications or substitutions shall be within the scope of the present invention. Any changes or substitutions that can be easily conceived by persons skilled in the art within the technical scope disclosed in the present invention shall be within the scope of protection of the present invention.

Claims

1. A multi-party participation advertising effect evaluation method, characterized in that: include: Obtain exposure feature data from the media platform, and use a locally deployed prediction model to predict click-through rate based on the exposure feature data; Comparing the predicted click-through rate with a preset expected click-through rate, and outputting the comparison result as an evaluation result; The prediction model is updated using a federated learning mechanism involving multiple parties. During the model update process, each participant conducts collaborative training based on their statistical data, and uses a homomorphic encryption scheme to encrypt and protect the transmitted gradient information. The statistical data includes the encrypted exposure of multiple participants within a specified period. In the process of generating the statistical data, differential privacy protection technology is used to inject noise into the original data, and the data after noise injection is anonymized.

2. The method according to claim 1, characterized in that The federated learning mechanism includes: Advertisers use the gradient descent algorithm to train the locally deployed prediction model based on locally stored statistical data, calculating the loss value and the gradient of the model parameters in each training iteration. The advertiser homomorphically encrypts the gradient and uploads the encrypted gradient to the parameter server; The parameter server receives encrypted gradients from multiple advertisers, aggregates the encrypted gradients, and obtains the encrypted global gradient; The parameter server distributes the encrypted global gradient to advertisers; The advertiser decrypts the encrypted global gradient and uses the decrypted gradient to update the parameters of the locally deployed prediction model.

3. The method according to claim 2, characterized in that The prediction model adopts a deep neural network model, which includes an input layer, multiple hidden layers and an output layer.

4. The method according to claim 2, characterized in that The parameter server receives encrypted gradients from multiple advertisers, aggregates them, and obtains an encrypted global gradient, including: The parameter server obtains the credit distribution ratio of each advertiser uploaded by the trusted third party; The parameter server calculates the advertiser's weight according to the credit distribution ratio, and calculates the probability of the advertiser participating in the training based on the advertiser's weight; The parameter server calculates the encrypted global gradient based on the probability of the advertiser participating in training and the encrypted gradient uploaded by the advertiser.

5. The method according to claim 4, characterized in that The calculation method of the credit distribution ratio includes: The trusted third party obtains the encrypted exposure timestamp from the advertiser and the desensitized conversion time from the media platform; The trusted third party calculates the time decay factor based on the encrypted exposure timestamp and the desensitized conversion time through a secure multi-party computing protocol without decryption, and calculates the credit distribution ratio of each advertiser based on the time decay factor.

6. The method according to claim 1, characterized in that The method further comprises: Each participant periodically aggregates exposures locally to obtain aggregated data; Each participant performs differential privacy protection and anonymization on the aggregated data to obtain encrypted aggregated data; Each participant uploads the encrypted summary data generated by themselves to a trusted third party; A trusted third party uses a secret sharing protocol to securely aggregate the encrypted summary data to obtain statistical data.

7. The method according to claim 6, characterized in that A trusted third party uses a secret sharing protocol to securely aggregate the encrypted summary data to obtain statistical data, including: The encrypted summary data is hierarchically aggregated based on the historical reputation scores of each party.

8. A multi-party advertising effect evaluation system, characterized in that: include: A prediction module, configured to obtain exposure feature data from a media platform and predict click-through rate based on the exposure feature data using a locally deployed prediction model; A comparison module, configured to compare the predicted click-through rate with a preset expected click-through rate, and output the comparison result as an evaluation result; The prediction model is updated using a federated learning mechanism involving multiple parties. During the model update process, each participant conducts collaborative training based on their statistical data, and uses a homomorphic encryption scheme to encrypt and protect the transmitted gradient information. The statistical data includes the encrypted exposure of multiple participants within a specified period. In the process of generating the statistical data, differential privacy protection technology is used to inject noise into the original data, and the data after noise injection is anonymized.

9. A device, characterized in that include: A memory for storing an advertising effect evaluation program involving multiple parties; A processor is configured to implement the steps of the multi-party advertising effectiveness evaluation method as described in any one of claims 1 to 7 when executing the multi-party advertising effectiveness evaluation program.

10. A computer-readable storage medium storing a computer program, characterized in that: The readable storage medium stores an advertising effect evaluation program involving multiple parties, and when the advertising effect evaluation program involving multiple parties is executed by a processor, the steps of the advertising effect evaluation method involving multiple parties are implemented as described in any one of claims 1-7.

Images