Target user prediction method and device, storage medium, and electronic device

Abstract

The present disclosure relates to a target user prediction method and device, storage medium and electronic equipment, and relates to the technical field of big data processing. The scheme is initiated by a first participant, a first model score of a local first prediction model on a first selectable user group is obtained, a second model score of a second prediction model of a second participant on a second selectable user group is obtained, a first fusion score ciphertext based on homomorphic encryption is obtained, and the target user is determined in the first selectable user group based on the obtained first fusion model score after decryption. Wherein, the first selectable user group and the second selectable user group are aligned, the first prediction model and the second prediction model are calculated by the first participant and the second participant based on secret sharing to evaluate the correlation coefficient and meet the first fusion condition, and the model discrimination degree is calculated based on homomorphic encryption to evaluate the second fusion condition. On the basis of ensuring the model fusion effect, the scheme realizes the privacy protection of model fusion calculation, improves the information delivery security, accuracy and conversion rate.

Description

Technical Field

[0001] This disclosure relates to the field of big data processing technology, and more specifically, to a method for predicting target users, a device for predicting target users, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In information delivery services, different participants can merge their respective model predictions to enrich the dimensions considered when selecting target users for information delivery, avoid the biases and omissions that may exist in single model predictions, and improve the accuracy and conversion rate of information delivery.

[0003] However, the models of each participating party are private and cannot be directly shared for security reasons. This limits the computation of model fusion and makes it difficult to predict whether the effect of model fusion meets business needs, thus affecting the efficiency and effectiveness of model fusion.

[0004] It should be noted that the information in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this disclosure is to provide a method, device, computer-readable storage medium, and electronic device for predicting target users, thereby overcoming, to some extent, the problems of low privacy data security and insufficient efficiency and effectiveness of model fusion in multi-party target user prediction due to limitations and defects in related technologies.

[0006] According to one aspect of this disclosure, a method for predicting target users is provided, applied to a first participant. The method includes: obtaining a first model score based on a first prediction model on a first selectable user group; obtaining a first fusion score ciphertext of the first model score and a second model score based on homomorphic encryption, wherein the second model score is obtained by a second participant on a second selectable user group based on the second prediction model; aligning the first selectable user group with the second selectable user group; decrypting the first fusion score ciphertext to obtain a first fusion model score; and determining the target user within the first selectable user group based on the first fusion model score. The first prediction model and the second prediction model are determined by the first and second participants during the evaluation process, based on secret sharing to calculate a federated correlation coefficient that meets a first fusion condition, and based on homomorphic encryption to calculate a model discrimination that meets a second fusion condition. The first fusion condition includes a federated correlation coefficient less than or equal to a correlation threshold; the second fusion condition includes a model discrimination greater than or equal to a discrimination threshold.

[0007] In one exemplary embodiment of this disclosure, before obtaining the first model score on the first selectable user group based on the first prediction model, the method further includes: obtaining a third model score on the first evaluation sample based on the first prediction model; calculating the federated correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing; the fourth model score being obtained by the second participant on the second evaluation sample based on the second prediction model; when the federated correlation coefficient between the first evaluation sample and the second evaluation sample meets the first fusion condition, calculating the model discrimination degree after fusing the third model score and the fourth model score based on homomorphic encryption; and determining to perform federated prediction for the target user with the second participant when the model discrimination degree meets the second fusion condition.

[0008] In one exemplary embodiment of this disclosure, calculating the federated correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing includes: performing secret sharing with the second participant based on the third model score and the fourth model score to obtain the product model score of the third model score and the fourth model score; determining the expected value of the product model score, the expected value of the third model score, and the standard deviation of the third model score; obtaining the expected value of the fourth model score and the standard deviation of the fourth model score provided by the second participant; and calculating the federated correlation coefficient using the expected value of the third model score, the expected value of the fourth model score, the standard deviation of the third model score, the standard deviation of the fourth model score, and the expected value of the product model score.

[0009] In one exemplary embodiment of this disclosure, calculating the model discriminability of the fused third model score and fourth model score based on homomorphic encryption includes: generating a homomorphic public key and a homomorphic private key; encrypting the third model score with the homomorphic public key to obtain a first encryption result; providing the first encryption result and the homomorphic public key to a second participant; obtaining a first ciphertext sorting index provided by the second participant, wherein the first ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the ciphertext of the second fused score, and the second fused score ciphertext is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, and the fourth model score; decrypting the first ciphertext sorting index with the homomorphic private key to obtain a first sample sorting index of the first evaluation sample; and calculating the model discriminability based on the first sample sorting index.

[0010] In one exemplary embodiment of this disclosure, before encrypting the third model score with a homomorphic public key to obtain the first encryption result, the method further includes: weighting the third model score based on a first business weight; the first business weight being determined by negotiation between the first participant and the second participant; and the second fused score ciphertext being obtained by the second participant through homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the first business weight, and the fourth model score.

[0011] In one exemplary embodiment of this disclosure, after calculating the federated correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing, the method further includes: calculating the federated correlation coefficient between the third model score and the fifth model score provided by another second participant based on secret sharing when the federated correlation coefficient does not meet the first fusion condition; calculating the model discrimination degree after fusing the third model score and the fifth model score based on homomorphic encryption when the federated correlation coefficient meets the first fusion condition; and determining the prediction of the target user with another second participant when the model discrimination degree meets the second fusion condition.

[0012] In one exemplary embodiment of this disclosure, after calculating the model distinguishability of the fused third model score and fourth model score based on homomorphic encryption, the method further includes: when the model distinguishability does not meet the second fusion condition, calculating the federated correlation coefficient of the third model score and the sixth model score provided by another second participant based on secret sharing; when the federated correlation coefficient meets the first fusion condition, calculating the model distinguishability of the fused third model score and sixth model score based on homomorphic encryption; and when the model distinguishability meets the second fusion condition, determining the prediction of the target user with another second participant.

[0013] In one exemplary embodiment of this disclosure, after calculating the model distinguishability of the fused third model score and fourth model score based on homomorphic encryption, the method further includes: when the model distinguishability does not meet the second fusion condition, negotiating with the second participant to adjust the first business weight to obtain the second business weight; weighting the third model score based on the second business weight; generating a homomorphic public key and a homomorphic private key; encrypting the third model score with the homomorphic public key to obtain a first encryption result; providing the first encryption result and the homomorphic public key to the second participant; obtaining the second ciphertext sorting index provided by the second participant, wherein the second ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the ciphertext of the third fused score, and the ciphertext of the third fused score is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the second business weight, and the fourth model score; decrypting the second ciphertext sorting index with the homomorphic private key to obtain the second sample sorting index of the first evaluation sample; calculating the model distinguishability based on the second sample sorting index; and determining to predict the target user with the second participant when the model distinguishability meets the second fusion condition.

[0014] According to one aspect of this disclosure, a target user prediction device is provided, applied to a first participant. The target user prediction device includes: a model score prediction module, used to obtain a first model score on a first selectable user group based on a first prediction model; a homomorphic encryption calculation module, used to obtain a first fusion score ciphertext of the first model score and a second model score based on homomorphic encryption, wherein the second model score is obtained by a second participant on a second selectable user group based on a second prediction model; the first selectable user group and the second selectable user group are aligned; the homomorphic encryption calculation module is further used to decrypt the first fusion score ciphertext to obtain a first fusion model score; and a target user determination module, used to determine a target user in the first selectable user group based on the first fusion model score; wherein the first prediction model and the second prediction model are calculated by the first participant and the second participant during the evaluation process based on secret sharing to obtain a federated correlation coefficient that meets a first fusion condition, and based on homomorphic encryption to calculate a model discrimination that meets a second fusion condition; the first fusion condition includes a federated correlation coefficient less than or equal to a correlation threshold; and the second fusion condition includes a model discrimination greater than or equal to a discrimination threshold.

[0015] According to one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the prediction method for a target user as described in any of the preceding claims.

[0016] According to one aspect of this disclosure, an electronic device is provided, comprising:

[0017] Processor; and

[0018] Memory for storing the executable instructions of the processor;

[0019] The processor is configured to execute the target user prediction method described above by executing the executable instructions.

[0020] This disclosure provides a method, apparatus, computer-readable storage medium, and electronic device for predicting target users. The scheme is initiated by a first participant, who obtains a first model score of a local first prediction model on a first selectable user group, and a second model score of a second participant's second prediction model on the second selectable user group, based on homomorphic encryption. After decryption, the obtained first fusion model score is used to determine the target user within the first selectable user group. The first and second selectable user groups are aligned, and the first and second prediction models are calculated by the first and second participants based on secret sharing to obtain a federated correlation coefficient that meets a first fusion condition, and based on homomorphic encryption to calculate model discrimination that meets a second fusion condition. In this scheme, the first and second prediction models, based on secret sharing and homomorphic encryption privacy computing techniques, undergo thorough evaluation of indicators such as federated correlation coefficient and model discrimination, ensuring secure and effective model fusion results. Based on this, the first and second model scores can be fused using homomorphic encryption to obtain the first fusion model score and determine the target user, achieving privacy protection in the model fusion calculation process and improving the security, accuracy, and conversion rate of information delivery to target users.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0023] Figure 1 One of the flowcharts illustrating a method for predicting a target user according to an embodiment of the present disclosure is shown schematically.

[0024] Figure 2 The second schematic diagram illustrates a flowchart example of a method for predicting a target user according to an embodiment of the present disclosure.

[0025] Figure 3 The diagram schematically illustrates a structural example of a target user prediction device according to an embodiment of the present disclosure.

[0026] Figure 4 The diagram schematically illustrates an example structure of an electronic device for implementing a prediction method for a target user according to an embodiment of the present disclosure. Detailed Implementation

[0027] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0028] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0029] In collaborative data analytics initiatives, such as marketing and risk control, different participants can jointly predict the target users based on their respective user bases and predictive models. This reduces the bias that a single model might introduce and improves prediction accuracy. For example, in a credit marketing scenario, one participant's predictive model might focus more on credit risk identification, while another's focuses more on marketing. In this scenario, both parties can merge their predictions based on their respective models to enhance the overall capability of credit decision-making from multiple dimensions. Similarly, other scenarios such as credit risk control, advertising marketing, and e-commerce marketing can also be implemented.

[0030] Currently, due to privacy protection restrictions, different participants are usually unable to directly share models for security reasons. This limits the computation of model fusion and makes it difficult to predict whether the effect of model fusion meets business needs, thus affecting the efficiency and effectiveness of model fusion.

[0031] Based on this, this disclosure provides a method for predicting target users. This method can be applied to a first participant, which can be implemented as a mobile device, a server, or a server cluster. Those skilled in the art can also run the method on other platforms as needed, and this disclosure does not impose any special limitations on this. The first participant can first evaluate the fusion effect of their respective prediction models with other participants, and select other participants whose model fusion effects meet expectations as optional second participants. Based on this, during the prediction of target users, the first participant can initiate target user prediction for model fusion analysis to the second participants. Privacy-preserving computation techniques are used throughout the evaluation and prediction processes, ensuring that no plaintext data is disclosed between participants to guarantee the security of privacy data.

[0032] The following will provide a detailed explanation and description of the target user prediction method described in the embodiments of this disclosure, in conjunction with the accompanying drawings.

[0033] Figure 1 This schematically illustrates one example flowchart of a target user prediction method according to an embodiment of the present disclosure, with reference to... Figure 1 As shown, the target user prediction method can be implemented based on the evaluation by the first and second participants on whether the first and second prediction models meet the fusion conditions. Based on the determination by the first and second participants that the first and second prediction models meet the first fusion condition using a secret-sharing calculation of the relevant federated coefficients, and the determination based on the homomorphic encryption calculation model's discriminative ability that it meets the second fusion condition, the first participant can jointly predict the target user based on the first prediction model and the second participant's second prediction model.

[0034] In this embodiment, the first participant and the second participant are distinguished only by their roles as initiator and co-participant in the method implementation process. In different evaluation processes, the first participant acting as the initiator can be the same or different; and the second participant being co-participated can be the same or different. There can be one or more second participants being co-participated. In different prediction processes, the first participant acting as the initiator can be the same or different; and the second participants being co-participated can be the same or different, provided they meet the aforementioned fusion conditions. The first participant and the second participant can communicate via a wired or wireless network.

[0035] The predictive model can predict and score whether different users meet business needs based on user information provided by participants. The predictive model can be a cross-federated predictive model; this cross-federated predictive model can be a deep neural network predictive model or a decision tree model, or other models as specified in this disclosure. A first predictive model can be used by the first participant, and a second predictive model can be used by the second participant; the first and second predictive models can have different emphases. Different second predictive models can exist among different second participants, and these different second predictive models can also have different emphases.

[0036] In this embodiment of the disclosure, the federated correlation coefficient can characterize the correlation between the prediction results of the first prediction model and the second prediction model on the same sample. The first fusion condition can be used to select a second prediction model and its second participant that meet the fusion requirements based on the correlation characterized by the federated correlation coefficient for the first prediction model of the first participant. The model discrimination can characterize the performance of the model in distinguishing positive and negative samples. The second fusion condition can be used to select a second prediction model and its second participant that meet the requirements after fusion for the first prediction model of the first participant.

[0037] In an optional method embodiment of this disclosure, the first fusion condition includes a federated correlation coefficient less than or equal to a correlation threshold.

[0038] For example, the first fusion condition may be that the federated correlation coefficient is less than or equal to the correlation threshold, so as to assess that the correlation between the first prediction model and the second prediction model is moderate or weak, so that the fusion effect meets business requirements; it may also include less than the correlation threshold, less than or equal to the maximum correlation threshold and greater than or equal to the minimum correlation threshold, etc., and this disclosure does not impose specific limitations on this.

[0039] For example, if the correlation threshold is 0.7, the first fusion condition can be that the absolute value of the federated correlation coefficient is less than 0.7. Those skilled in the art can choose an appropriate correlation threshold according to actual needs, which can be 0.7, 0.8, or any value within the range of (0,1), such as 0.6 or 0.5. This disclosure does not impose specific limitations on this.

[0040] In an optional embodiment of the method disclosed herein, the second fusion condition includes a model discrimination degree greater than or equal to a discrimination degree threshold.

[0041] For example, the second fusion condition may be that the model discrimination is greater than or equal to the discrimination threshold, so as to evaluate whether the second prediction model has the discrimination capability that meets business needs after fusion; it may also include being greater than the discrimination threshold, etc., and this disclosure does not impose specific limitations on this.

[0042] For example, if the discrimination threshold is 0.4, the second fusion condition can be that the model discrimination is greater than 0.4. Those skilled in the art can choose an appropriate discrimination threshold according to actual needs. It can be 0.4, or 0.2, 0.3, or 0.5, 0.6, etc., any value within the range of (0,1). This disclosure does not impose specific limitations on this.

[0043] In this embodiment of the disclosure, the aforementioned evaluation can be initiated periodically by the first participant, or before each prediction, or when user information, model parameters, etc. are updated. This embodiment of the disclosure does not impose any specific restrictions on this.

[0044] In this embodiment of the disclosure, based on the evaluation by the first participant and the second participant that the first prediction model and the second prediction model meet the fusion conditions, the prediction method for the target user is as follows:

[0045] Step 101: Obtain the first model score based on the first prediction model on the first selectable user group.

[0046] In this embodiment of the disclosure, the first prediction model is a local prediction model of the first participant, and it meets the aforementioned fusion conditions after evaluation with the second prediction model of at least one second participant. The optional user group may include the range of users to be selected by the participants based on business needs and user information. The first optional user group may be determined by the first participant based on local user information and the business needs predicted by the target users. For example, the first participant may select at least one user from the same or different ranges to form the first optional user group based on different business needs such as risk control and marketing from the user information it holds. Preliminary screening can be performed through statistical filtering, collaborative filtering, rule filtering, model recall, etc.

[0047] In this embodiment of the disclosure, the first participant can input a first selectable user group into a first prediction model for relevant feature extraction and analysis, thereby obtaining a first model score. The first model score characterizes the degree to which each user in the first selectable user group meets business requirements in the dimension emphasized by the first prediction model.

[0048] Step 102: Obtain the first fused ciphertext of the first model score and the second model score based on homomorphic encryption. The second model score is obtained by the second participant based on the second prediction model on the second optional user group; the first optional user group and the second optional user group are aligned.

[0049] In this embodiment, the second selectable user group can be determined by the second participant based on local user information and the business needs predicted by the target users. The second participant can input the second selectable user group into the second prediction model for relevant feature extraction and analysis to obtain a second model score. The second model score characterizes the degree to which each user in the second selectable user group meets the business needs in the dimension emphasized by the second prediction model. Furthermore, the first and second participants can first align the first and second selectable user groups, so that the first and second participants perform predictions on a consistent selectable user group, providing a basis for fusion of the first and second model scores. The first and second participants can achieve alignment through intersection of privacy sets to protect the security of user information.

[0050] In this embodiment, after obtaining the first model score, the first participant and the second participant can calculate a first fused ciphertext based on homomorphic encryption of the first and second model scores. Homomorphic encryption (HE) enables homomorphic computation of ciphertext, ensuring that after the first and second participants encrypt the first and second model scores respectively, the second fused ciphertext obtained by performing fusion computation on the encrypted basis is equivalent to the ciphertext obtained by performing the same fusion computation on the plaintext basis. Therefore, computation can be performed directly on the encrypted basis without revealing the actual plaintext data, achieving secure computation that is computationally achievable but not visible.

[0051] Step 103: Decrypt the first fusion score ciphertext to obtain the first fusion model score.

[0052] In this embodiment, the first participant can decrypt the ciphertext of the first fusion score obtained by fusion calculation in encrypted state to obtain the first fusion model score. The first fusion model score is equivalent to the result of fusion calculation of the first model score and the second model score on the basis of plaintext, but there is no exchange of plaintext data during the calculation process, which improves data security.

[0053] Step 104: Determine the target user from the first selectable user group based on the first fusion model.

[0054] In this embodiment, the first fusion model score represents the degree to which each user in the first selectable user group meets business requirements in the dimensions emphasized by the first prediction model and the second prediction model. Based on this, target users can be determined from the first selectable user group according to actual business needs. For example, users with a first fusion model score greater than or equal to a score threshold in the first selectable user group can be identified as target users. Alternatively, a predetermined number of target users can be determined by sorting the first selectable user group based on the first fusion model score. Furthermore, different levels of target users can be determined by assigning the first fusion model score to different numerical ranges. Those skilled in the art can choose the method for determining target users according to specific business types and needs, and provide corresponding services to target users, such as targeted marketing to a predetermined number of target users, or tiered marketing to target users of different levels. This embodiment does not impose specific limitations in this regard.

[0055] During the aforementioned prediction and evaluation phases, neither the first nor the second participant discloses any plaintext data, which effectively enhances the security of the privacy data of each participant.

[0056] Figure 2 This is a second schematic flowchart illustrating a method for predicting a target user according to an embodiment of the present disclosure. (See also...) Figure 2 This method can be applied to the prediction method for the first participant, the target user. The evaluation stage is shown in steps 201 to 204, and the prediction stage is shown in steps 205 to 208.

[0057] Step 201: Obtain the third model score based on the first prediction model on the first evaluation sample.

[0058] In this embodiment of the disclosure, the first evaluation sample may be a user sample held by the first participant, with sample tags distinguished based on business needs. For example, it can be labeled based on the historical behavior data of the user sample; for marketing business, the sample tags include converted and non-converted, and for risk control business, the sample tags include overdue and not overdue. When the first participant enters the evaluation stage, it can input the first evaluation sample into the first prediction model, and the first prediction model outputs the predicted sample tags in the form of a third model score, which serve as the prediction result for the dimension emphasized by the first prediction model.

[0059] For example, the first participant is represented by A, and the first evaluation sample S （A） This can be expressed as the following formula (1):

[0060] (1)

[0061] in, x This represents a first evaluation sample, totaling... m strip;l The sample label represents the annotation of each first evaluation sample; f Represents the first prediction model, f ( x () represents the third model score output by the first prediction model on a first evaluation sample.

[0062] Based on this, the first evaluation sample S （A） The corresponding third model sub-matrix F It can be expressed as the following formula (2):

[0063] (2)

[0064] Step 202: Calculate the federal correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing; the fourth model score is obtained by the second participant on the second evaluation sample based on the second prediction model; the first evaluation sample is aligned with the second evaluation sample.

[0065] In this embodiment, the second evaluation sample can be a user sample held by the second participant, with sample tags differentiated based on business needs. For example, it can be labeled based on the historical behavior data of the user sample; for marketing business, the sample tags include "converted" and "non-converted," and for risk control business, the sample tags include "overdue" and "not overdue." When the second participant enters the evaluation stage, it can input the second evaluation sample into the second prediction model, which outputs the predicted sample tags in the form of a fourth model score, serving as the prediction result for the dimension emphasized by the second prediction model. The first and second evaluation samples are already aligned.

[0066] For example, the second participant is represented by B, and the second evaluation sample S （B） This can be expressed as the following formula (3):

[0067] (3)

[0068] in, x This represents a second evaluation sample, totaling... m strip; l The sample label represents the annotation of each second evaluation sample; g This represents the second prediction model. g ( x () represents the fourth model score output by the second prediction model on a second evaluation sample. Wherein, the first evaluation sample... S （A） Compared with the second evaluation sample S （B） Aligned.

[0069] Based on this, the second evaluation sample S （B） The corresponding fourth model submatrix G It can be expressed as the following formula (4):

[0070] (4)

[0071] Furthermore, taking the user identifier as the evaluation sample as an example, on the basis of alignment, the number of user identifiers in the first evaluation sample is consistent with the number of user identifiers in the second evaluation sample, and the user identifiers also correspond one-to-one; on this basis, the third model score and the labeled sample label corresponding to each user identifier in the first participant, and the fourth model score and the labeled sample label corresponding to each user identifier in the second participant can be consistent, inconsistent, or partially consistent, and this disclosure does not impose any special restrictions on this.

[0072] In this embodiment of the disclosure, based on the first participant obtaining the third model score and the second participant obtaining the fourth model score, the first participant and the second participant can secretly share and calculate the federated correlation coefficient between the third model score and the fourth model score. This federated correlation coefficient is as described above. Figure 1 The relevant descriptions in the text are omitted here to avoid repetition. Different types of federal correlation coefficients can characterize different types of correlation between the third model score and the fourth model score. Depending on business needs, calculation conditions, etc., the Pearson correlation coefficient or other types of correlation coefficients can be selected. This disclosure does not impose specific limitations on this.

[0073] In an optional embodiment of the method disclosed herein, the federal correlation coefficient is the Pearson correlation coefficient, which can be calculated as shown in the following formula (5):

[0074] (5)

[0075] in, For federal correlation coefficient, The covariance between the third model score and the fourth model score; The standard deviation of the third model, The standard deviation of the fourth model; The expected value of the product of the third model score and the fourth model score; For the third model, the expectation is calculated. The expected value of the fourth model is calculated.

[0076] Based on the above formula (5), the first participant and the second participant can execute the Secret Sharing Pearson Correlation Coefficient (SS-CORR) protocol, and the aforementioned step 202 may include the following steps A1 to A4.

[0077] Step A1: Perform secret sharing with the second participant based on the third model score and the fourth model score to obtain the product model score of the third model score and the fourth model score.

[0078] In this embodiment of the disclosure, the product model score can be calculated through secret sharing, obtaining the product without exposing plaintext data. For example, the third model score is represented as... F The fourth model is represented as G The first participant is based on F With the second participating party G The secret-sharing matrix multiplication protocol is executed, and the first and second participants secretly share the model. F The second participant secretly shares the model with the first participant. G The two parties collaborate to pre-construct Beaver's multiplication triples. The first participant then calculates the multiplication triples based on the shared two-party model fragments. FG The secret fragment one, the second participant is based on the shared two-party model of fragment division and multiplication triple calculation. FG The second secret fragment, the first participant based on FG The Secret Fragment One and FG The secret fragment two recovers the product of the third model score and the fourth model score. FG .

[0079] Step A2: Determine the expected value of the product model, the expected value of the third model, and the standard deviation of the third model.

[0080] In this embodiment, based on the obtained third model score and product model score, the first participant can perform statistical processing to obtain the expected value of the product model score, the expected value of the third model score, and the standard deviation of the third model score. The expected value can be calculated as a mean, by averaging the values ​​of each item in the product model score to obtain the expected value of the product model score. And calculate the mean of each item in the third model score to obtain the expected value of the third model score. The standard deviation is the arithmetic square root of the variance. It is obtained by calculating the arithmetic square root of the variances of each item in the third model. .

[0081] Step A3: Obtain the expected value and standard deviation of the fourth model score provided by the second participant.

[0082] In this embodiment of the disclosure, the first participant may also obtain the expected value of the fourth model score obtained by the second participant through statistical processing of the fourth model score. and the standard deviation of the fourth model The expected value and standard deviation can be referred to in the relevant description of step A2 above, and will not be repeated here to avoid repetition.

[0083] Step A4: Calculate the federal correlation coefficient using the expected value of the third model, the expected value of the fourth model, the standard deviation of the third model, the standard deviation of the fourth model, and the expected value of the product model.

[0084] Based on the aforementioned steps A1 to A3, the first participant can obtain the expected value of the third model. The fourth model has different expectations. Third model standard deviation The fourth model's standard deviation And the expected value of the product model By substituting into the aforementioned formula (5), the federal correlation coefficient in the form of the Pearson correlation coefficient can be calculated. , to represent the degree of linear correlation between the third model score and the fourth model score.

[0085] Based on this, the aforementioned federal correlation coefficient calculation protocol based on secret sharing is shown in Table 1 below:

[0086]

[0087] Step 203: When the federated correlation coefficient meets the first fusion condition, calculate the model discrimination of the fused third model score and fourth model score based on homomorphic encryption.

[0088] In this embodiment of the disclosure, in multi-party joint prediction, the lower the correlation between the prediction results of the first participant and the second participant, the richer the integrated information and the greater the improvement in analysis accuracy. However, excessively low correlation may also cause interference from invalid information, thereby affecting the accuracy of the analysis. During the evaluation phase, a first fusion condition corresponding to the federated correlation coefficient can be specifically set according to business needs. When the federated correlation coefficient meets the first fusion condition, it can be considered that the correlation between the first prediction model and the second prediction model meets the business needs, and subsequent evaluation can be performed; or, when the federated correlation coefficient does not meet the first fusion condition, it can be considered that the correlation between the first prediction model and the second prediction model does not meet the business needs, and the evaluation process can be stopped.

[0089] In this embodiment of the disclosure, based on the federated correlation coefficient meeting the first fusion condition, the model discrimination score after fusing the third and fourth model scores can be further calculated using homomorphic encryption. The third and fourth model scores can be fused based on encrypted operations, and the difference in the cumulative distribution of positive and negative samples in the predicted sample labels is used to determine the model discrimination score after fusing the third and fourth model scores. Generally, the larger this difference, the better the ability to distinguish between positive and negative samples, thus resulting in a higher model discrimination score. KS The (Kolmogorov-Smirnov) calculation can be performed as shown in formula (6):

[0090] (6)

[0091] The first evaluation sample is divided into: m Each model is divided into segments, based on the labeled sample labels. cumulative good i Indicates the first i The number of positive samples in each model partition cumulative bad i Indicates the first i The number of negative samples in each model partition, total good The total number of positive samples is represented by the symbol "total". bad This represents the total number of negative samples.

[0092] Based on the above formula (6), the largest difference is taken as the model discrimination. KS value.

[0093] In an optional embodiment of the method disclosed herein, ROC (Receiver Operating Characteristic) can also be used to evaluate the model's discriminative power. For example, the relationship between the True Positive Rate (TPR) and the False Positive Rate (FPR) can be used to determine the True Positive Rate, which represents the ratio of correctly identified positive samples, and the False Positive Rate, which represents the ratio of incorrectly identified negative samples as positive samples, by combining the predicted sample labels and the labeled sample labels. This allows for the evaluation of the model's discriminative power after the fusion of the third model score and the fourth model score.

[0094] In an optional method embodiment of this disclosure, the model discrimination is determined based on the above formula (6), and step 203 includes the following steps B1 to B6.

[0095] Step B1: Generate a homomorphic public key and a homomorphic private key.

[0096] The first participant can generate a homomorphic public key and a homomorphic private key pair. The homomorphic public key is used for encryption between the first participant and the second participant, while the homomorphic private key is used for decryption by the first participant.

[0097] Step B2: Encrypt the third model using the homomorphic public key to obtain the first encryption result.

[0098] The first participant can encrypt the third model based on the homomorphic public key to obtain the first encrypted result of the encrypted state.

[0099] Step B3: Provide the first encryption result and the homomorphic public key to the second participant.

[0100] Based on the first encryption result obtained by the first participant, the homomorphic public key and the first encryption result can be sent to the second participant to support the second participant in subsequent calculations.

[0101] Step B4: Obtain the first ciphertext sorting index provided by the second participant. The first ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the second fused ciphertext. The second fused ciphertext is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, and the fourth model.

[0102] After obtaining the first encryption result and the homomorphic public key, the second participant can encrypt the fourth model score based on the homomorphic public key to obtain the second encryption result. Then, based on the first encryption result and the second encryption result, a fusion operation of the homomorphic ciphertext is performed to obtain the fused model score ciphertext of the third model score and the fourth model score.

[0103] The second participant, having obtained the ciphertext of the fusion model, performs homomorphic encryption sorting to obtain the first ciphertext sorting index corresponding to the ciphertext of the fusion model, and provides this first ciphertext sorting index to the first participant. The homomorphic encryption sorting can employ homomorphic ciphertext sorting algorithms such as HE-Sort, HE Direct Sort, HE Greedy Sort, and HE Polynomial Rank Sort.

[0104] The first participant can obtain the first ciphertext sorting index provided by the second participant.

[0105] Step B5: Decrypt the first ciphertext sorting index using the homomorphic private key to obtain the first sample sorting index of the first evaluation sample.

[0106] The first participant can decrypt the first ciphertext sorting index with a homomorphic private key to obtain the sample sorting of the first evaluation sample indicated by the first sample sorting index.

[0107] Step B6: Calculate the model discrimination based on the sorting index of the first sample.

[0108] For the first participant, based on the first sample sorting index, calculate using the aforementioned formula (6). KS The value is used to obtain the model discrimination score obtained by fusing the third model score and the fourth model score.

[0109] In an optional method embodiment of this disclosure, step 203 includes the following steps C1 to C7.

[0110] Step C1: Generate a homomorphic public key and a homomorphic private key.

[0111] For the first participant, step C1 can refer to the relevant description of step B1 above.

[0112] Step C2: Weight the third model score based on the first business weight; the first business weight is determined by the first participant and the second participant through negotiation.

[0113] In the first participating party, the first prediction model and the second prediction model of the second participating party can be weighted and fused, using a first business weight negotiated and fused by the first and second participating parties. The first business weight can include the weight value sub-weighted by the first participating party for the third model, or it can include the weight value sub-weighted by the second participating party for the fourth model. The first and second participating parties hold their respective first business weights after negotiation. The first participating party performs a weighted calculation on the third model using the first business weights and then executes the subsequent encryption process.

[0114] Step C3: Encrypt the third model using the homomorphic public key to obtain the first encryption result.

[0115] In the first participating party, the third model score after being weighted by the first business weight is encrypted to obtain the first encryption result. Step C3 can refer to the relevant description of step B2 above.

[0116] Step C4: Provide the first encryption result and the homomorphic public key to the second participant.

[0117] For the first participant, step C4 can refer to the relevant description of step B3 above.

[0118] Step C5: Obtain the first ciphertext sorting index provided by the second participant. The first ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the second fused ciphertext. The second fused ciphertext is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the first business weight, and the fourth model score.

[0119] In the second participating party, step C5 can refer to the relevant description of step B4 above, wherein before encrypting the fourth model sub-score with a homomorphic public key, the fourth model sub-score can be weighted based on the negotiated first business weight.

[0120] For the first participant, step C5 can refer to the relevant description of step B4 above.

[0121] Step C6: Decrypt the first ciphertext sorting index using the homomorphic private key to obtain the first sample sorting index of the first evaluation sample.

[0122] For the first participant, step C6 can refer to the relevant description of step B5 above.

[0123] Step C7: Calculate the model discrimination based on the sorting index of the first sample.

[0124] For the first participant, step C7 can refer to the relevant description of step B6 above.

[0125] Based on this, the model's discriminative power is based on homomorphic encryption for encrypted states. KS The value is calculated based on the assumption that the first participant and the second participant have already negotiated the first business weight. w and (1- w The execution process of the HE-KS (Homomorphic Encryption-Kolmogorov-Smirnov) protocol is shown in Table 2 below:

[0126]

[0127] Step 204: When the model discrimination meets the second fusion condition, determine the target user's federated prediction with the second participant.

[0128] In this embodiment of the disclosure, if the aforementioned federated correlation coefficient meets the first fusion condition and the subsequent model discrimination also meets the second fusion condition, then the fusion effect of the first prediction model of the first participant and the second prediction model of the second participant meets the business requirements, and the second participant can be used as one of the optional objects for federated prediction of target users with the first participant; if the model discrimination does not meet the second fusion condition, then the second participant is rejected as one of the optional objects for federated prediction of target users with the first participant.

[0129] During the aforementioned evaluation phase, if the federated correlation coefficient meets the first fusion condition and the model discrimination meets the second fusion condition, the corresponding second participant and its second prediction model will be considered as one of the optional targets for joint prediction of target users. During the evaluation process, situations may arise where the federated correlation coefficient does not meet the first fusion condition or the model discrimination does not meet the second fusion condition. In such cases, the parameters and methods used can be adjusted according to actual business needs and evaluation conditions, or another second participant can be re-evaluated until at least one optional target for joint prediction of target users with the first participant is determined.

[0130] In an optional embodiment of the method disclosed herein, step D1 to step D3 are further included after step 202.

[0131] Step D1: When the federated correlation coefficient does not meet the first fusion condition, calculate the federated correlation coefficient between the third model score and the fifth model score provided by another second participant based on secret sharing.

[0132] Step D2: When the federated correlation coefficient meets the first fusion condition, calculate the model discrimination of the fused third model score and fifth model score based on homomorphic encryption.

[0133] Step D3: When the model discrimination meets the second fusion condition, determine the target user prediction with another second participant.

[0134] In this embodiment, when the federated correlation coefficient does not meet the first fusion condition, another second participant can be selected to re-evaluate. This second participant has a corresponding second prediction model and a second evaluation sample locally, and the second evaluation sample is aligned with the first evaluation sample. The second prediction model of this second participant obtains a fifth model score on the second evaluation sample and provides it to the first participant. The first participant can calculate the federated correlation coefficient between the third model score and the fifth model score based on secret sharing, as described in step 202 above; to avoid repetition, it will not be repeated here.

[0135] If the federated correlation coefficient calculated based on the third model score and the fifth model score meets the first fusion condition, the model discrimination degree after the fusion of the third model score and the fifth model score can be further calculated based on homomorphic encryption. Refer to the relevant description of step 203 above. To avoid repetition, it will not be repeated here.

[0136] If the model discrimination calculated based on the third model score and the fifth model score meets the second fusion condition, then the other second participant is considered as one of the optional targets for target user federated prediction with the first participant.

[0137] If the federated correlation coefficient calculated based on the third model score and the fifth model score does not meet the first fusion condition, other second participants can be re-selected for evaluation until at least one candidate is determined to conduct target user federated prediction with the first participant.

[0138] In an optional embodiment of the method disclosed herein, step 203 may be followed by steps E1 to E3.

[0139] Step E1: When the model discrimination does not meet the second fusion condition, calculate the federated correlation coefficient between the third model score and the sixth model score provided by another second participant based on secret sharing.

[0140] Step E2: When the federated correlation coefficient meets the first fusion condition, calculate the model discrimination of the fused third model score and sixth model score based on homomorphic encryption.

[0141] Step E3: When the model discrimination meets the second fusion condition, determine the prediction of the target user with another second participant.

[0142] In this embodiment, if the federated correlation coefficient meets the first fusion condition, the model discrimination may not meet the second fusion condition. Another second participant can be selected to re-evaluate. This second participant has a corresponding second prediction model and second evaluation samples locally, and the second evaluation samples are aligned with the first evaluation samples. The second prediction model of this second participant obtains a sixth model score on the second evaluation samples and provides it to the first participant. The first participant can calculate the federated correlation coefficient between the third model score and the sixth model score based on secret sharing, as described in step 202 above. To avoid repetition, it will not be repeated here.

[0143] If the federated correlation coefficient calculated based on the third model score and the sixth model score meets the first fusion condition, the model discrimination degree after the fusion of the third model score and the sixth model score can be further calculated based on homomorphic encryption. Refer to the relevant description of step 203 above. To avoid repetition, it will not be repeated here.

[0144] If the model discrimination calculated based on the third model score and the sixth model score meets the second fusion condition, then the other second participant can be considered as one of the options for conducting target user federated prediction with the first participant.

[0145] At this point, if the federated correlation coefficient calculated based on the third model score and the sixth model score does not meet the first fusion condition, or if the federated correlation coefficient meets the first fusion condition but the model discrimination does not meet the second fusion condition, other second participants can be re-selected for evaluation until at least one candidate is determined to conduct target user federated prediction with the first participant.

[0146] In an optional embodiment of the method disclosed herein, step 203 may be followed by steps F1 to F9.

[0147] Step F1: When the model's discrimination does not meet the second fusion condition, negotiate with the second participant to adjust the first business weight and obtain the second business weight.

[0148] Step F2: Weight the third model score based on the second business weight.

[0149] Step F3: Generate a homomorphic public key and a homomorphic private key.

[0150] Step F4: Encrypt the third model using the homomorphic public key to obtain the first encryption result.

[0151] Step F5: Provide the first encryption result and the homomorphic public key to the second participant.

[0152] Step F6: Obtain the second ciphertext sorting index provided by the second participant. The second ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the third fused ciphertext. The third fused ciphertext is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the second business weight, and the fourth model score.

[0153] Step F7: Decrypt the second ciphertext sorting index using the homomorphic private key to obtain the second sample sorting index of the first evaluation sample.

[0154] Step F8: Calculate the model discrimination based on the second sample sorting index.

[0155] Step F9: When the model discrimination meets the second fusion condition, determine the target user prediction with the second participant.

[0156] In this embodiment of the disclosure, if the federated correlation coefficient meets the first fusion condition, the model discrimination may not meet the second fusion condition. If the first participant and the second participant have negotiated the first business weight, the first business weight can be adjusted to obtain the second business weight, and the model score fusion can be adjusted by adjusting the weight allocation. Based on this, the model discrimination between the third model score and the fourth model score is calculated using the second business weight. Steps F3 to F9 can be referred to the relevant descriptions of steps C1 to C7 above, and will not be repeated here to avoid repetition.

[0157] Step 205: Obtain the first model score based on the first prediction model on the first selectable user group.

[0158] In this embodiment of the disclosure, step 205 can be referred to the relevant description of step 101 above. To avoid repetition, it will not be repeated here.

[0159] Step 206: Obtain the first fused ciphertext of the first model score and the second model score based on homomorphic encryption. The second model score is obtained by the second participant based on the second prediction model on the second optional user group; the first optional user group and the second optional user group are aligned.

[0160] In this embodiment of the disclosure, step 206 can be referred to the relevant description of step 102 above. To avoid repetition, it will not be repeated here.

[0161] Step 207: Decrypt the first fusion score ciphertext to obtain the first fusion model score.

[0162] In this embodiment of the disclosure, step 207 can be referred to the relevant description of step 103 above. To avoid repetition, it will not be repeated here.

[0163] Step 208: Determine the target user from the first selectable user group based on the first fusion model.

[0164] In this embodiment of the disclosure, step 208 can be referred to the relevant description of step 104 above. To avoid repetition, it will not be repeated here.

[0165] This disclosure provides a method for predicting target users. The scheme is initiated by a first participant, who obtains a first model score of a local first prediction model on a first selectable user group, and a second model score of a second prediction model from a second participant on the second selectable user group. The second model score is then obtained using homomorphic encryption and encrypted as a first fusion score. After decryption, the obtained first fusion model score is used to determine the target user within the first selectable user group. The first and second selectable user groups are aligned. The first and second prediction models are calculated by the first and second participants based on secret sharing, yielding a federated correlation coefficient that meets a first fusion condition. Furthermore, the model discrimination is calculated based on homomorphic encryption, meeting a second fusion condition. This scheme, based on privacy-preserving computation techniques such as secret sharing and homomorphic encryption, and after thorough evaluation of indicators such as federated correlation coefficient and model discrimination, can securely and effectively guarantee the model fusion effect. Furthermore, the first and second model scores can be fused using homomorphic encryption, and the target user can be determined using the first fusion model score. This achieves privacy protection in the model fusion computation process and improves the security, accuracy, and conversion rate of information delivery to target users.

[0166] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0167] This disclosure also provides a prediction device for a target user, configured in a first participant having a first prediction model. Specifically, refer to... Figure 3As shown, the target user prediction device 300 may include: a model score prediction module 301, used to obtain a first model score on a first selectable user group based on a first prediction model; a homomorphic encryption calculation module 302, used to obtain a first fusion score ciphertext of the first model score and the second model score based on homomorphic encryption, wherein the second model score is obtained by a second participant on a second selectable user group based on the second prediction model; the first selectable user group and the second selectable user group are aligned; the homomorphic encryption calculation module 302 is also used to decrypt the first fusion score ciphertext to obtain a first fusion model score; and a target user determination module 303, used to determine the target user in the first selectable user group based on the first fusion model score; wherein the first prediction model and the second prediction model are calculated by the first participant and the second participant during the evaluation process based on secret sharing to ensure that the federated correlation coefficient meets the first fusion condition, and based on homomorphic encryption to calculate the model discrimination to ensure that the model discrimination meets the second fusion condition; the first fusion condition includes that the federated correlation coefficient is less than or equal to a correlation threshold; and the second fusion condition includes that the model discrimination is greater than or equal to a discrimination threshold.

[0168] In an optional embodiment of this disclosure, the prediction device 300 for the target user includes a model score prediction module 301, which is further configured to obtain a third model score on a first evaluation sample based on a first prediction model; the prediction device 300 for the target user also includes a secret sharing calculation module, which is configured to calculate the federated correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing; the fourth model score is obtained by the second participant on a second evaluation sample based on a second prediction model; the first evaluation sample is aligned with the second evaluation sample; the homomorphic encryption calculation module 302 is further configured to calculate the model discrimination degree after fusing the third model score and the fourth model score based on homomorphic encryption when the federated correlation coefficient meets the first fusion condition; and the target user determination module 303 is further configured to determine the target user for federated prediction with the second participant when the model discrimination degree meets the second fusion condition.

[0169] In an optional embodiment of this disclosure, the secret sharing computation module is specifically used to perform secret sharing with the second participant based on the third model score and the fourth model score to obtain the product model score of the third model score and the fourth model score; determine the expected value of the product model score, the expected value of the third model score, and the standard deviation of the third model score; obtain the expected value of the fourth model score and the standard deviation of the fourth model score provided by the second participant; and calculate the federated correlation coefficient using the expected value of the third model score, the expected value of the fourth model score, the standard deviation of the third model score, the standard deviation of the fourth model score, and the expected value of the product model score.

[0170] In an optional embodiment of this disclosure, the homomorphic encryption calculation module 302 is specifically used to generate a homomorphic public key and a homomorphic private key; encrypt the third model score with the homomorphic public key to obtain a first encryption result; provide the first encryption result and the homomorphic public key to the second participant; obtain the first ciphertext sorting index provided by the second participant, the first ciphertext sorting index being obtained by the second participant performing homomorphic encryption sorting on the second fused score ciphertext, the second fused score ciphertext being obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, and the fourth model score; decrypt the first ciphertext sorting index with the homomorphic private key to obtain the first sample sorting index of the first evaluation sample; and calculate the model discriminability based on the first sample sorting index.

[0171] In an optional embodiment of this disclosure, the homomorphic encryption calculation module 302 is further configured to weight the third model score based on the first business weight; the first business weight is determined by negotiation between the first participant and the second participant; and the second fused score ciphertext is obtained by the second participant through homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the first business weight, and the fourth model score.

[0172] In an optional embodiment of this disclosure, the secret sharing computation module is further configured to calculate the federation correlation coefficient between the third model score and the fifth model score provided by another second participant based on secret sharing when the federation correlation coefficient does not meet the first fusion condition; the homomorphic encryption computation module 302 is further configured to calculate the model discrimination degree after fusing the third model score and the fifth model score based on homomorphic encryption when the federation correlation coefficient meets the first fusion condition; and the target user determination module 303 is further configured to determine the target user prediction with another second participant when the model discrimination degree meets the second fusion condition.

[0173] In an optional embodiment of this disclosure, the secret sharing computing module is further configured to calculate the federated correlation coefficient between the third model score and the sixth model score provided by another second participant based on secret sharing when the model discrimination does not meet the second fusion condition; the homomorphic encryption computing module 302 is further configured to calculate the model discrimination after fusing the third model score and the sixth model score based on homomorphic encryption when the federated correlation coefficient meets the first fusion condition; and the target user determination module 303 is further configured to determine the target user prediction with another second participant when the model discrimination meets the second fusion condition.

[0174] In an optional embodiment of this disclosure, the homomorphic encryption calculation module 302 is further configured to: negotiate with the second participant to adjust the first business weight and obtain the second business weight when the model discrimination does not meet the second fusion condition; weight the third model score based on the second business weight; generate a homomorphic public key and a homomorphic private key; encrypt the third model score with the homomorphic public key to obtain a first encryption result; provide the first encryption result and the homomorphic public key to the second participant; obtain the second ciphertext sorting index provided by the second participant, the second ciphertext sorting index being obtained by the second participant performing homomorphic encryption sorting on the third fusion score ciphertext, the third fusion score ciphertext being obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the second business weight, and the fourth model score; decrypt the second ciphertext sorting index with the homomorphic private key to obtain the second sample sorting index of the first evaluation sample; calculate the model discrimination based on the second sample sorting index; and determine to predict the target user with the second participant when the model discrimination meets the second fusion condition.

[0175] The specific details of each module in the aforementioned target user prediction device have been described in detail in the corresponding target user prediction method, so they will not be repeated here.

[0176] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0177] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0178] In exemplary embodiments of this disclosure, an electronic device capable of implementing the above-described methods is also provided. Those skilled in the art will understand that various aspects of this disclosure can be implemented as systems, methods, or program products. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: entirely hardware implementations, entirely software implementations (including firmware, microcode, etc.), or implementations combining hardware and software aspects, collectively referred to herein as circuits, modules, or systems.

[0179] The following reference Figure 4To describe an electronic device 400 according to such an embodiment of the present disclosure. Figure 4 The electronic device 400 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0180] like Figure 4 As shown, the electronic device 400 is manifested in the form of a general-purpose computing device. The components of the electronic device 400 may include, but are not limited to: at least one processing unit 410, at least one storage unit 420, a bus 430 connecting different system components (including storage unit 420 and processing unit 410), and a display unit 440.

[0181] The storage unit stores program code that can be executed by the processing unit 410, causing the processing unit 410 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 410 can perform actions such as... Figure 1 or Figure 2 The target user prediction method shown.

[0182] Storage unit 420 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 4201 and / or cache memory 4202, and may further include a read-only memory (ROM) 4203.

[0183] Storage unit 420 may also include a program / utility 4204 having a set (at least one) program module 4205, such program module 4205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0184] Bus 430 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0185] Electronic device 400 can also communicate with one or more external devices 500 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 400, and / or with any device that enables electronic device 400 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 450. Furthermore, electronic device 400 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 460. As shown, network adapter 460 communicates with other modules of electronic device 400 via bus 430. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 400, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0186] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (e.g., a personal computer, server, terminal device, or network device, etc.) to execute the user data prediction method according to the embodiments of this disclosure.

[0187] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code that, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. A program product for implementing the methods according to embodiments of this disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of this disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0188] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0189] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0190] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0191] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0192] Furthermore, the above figures are merely illustrative of the processes included in the method according to embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0193] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention described herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not invented by this disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A method for predicting target users, characterized in that, Applied to the first participant, the prediction method for the target user includes: The first model score is obtained based on the first prediction model on the first selectable user group; The first fused ciphertext of the first model score and the second model score is obtained based on homomorphic encryption. The second model score is obtained by the second participant on the second optional user group based on the second prediction model. The first optional user group is aligned with the second optional user group. Decrypt the first fusion ciphertext to obtain the first fusion model score; Based on the first fusion model, target users are identified from the first selectable user group; Before obtaining the first model score on the first selectable user group based on the first prediction model, the method further includes: The third model score is obtained based on the first prediction model on the first evaluation sample; The federated correlation coefficient between the third model score and the fourth model score provided by the second participant is calculated based on secret sharing; the fourth model score is obtained by the second participant on the second evaluation sample based on the second prediction model; the first evaluation sample is aligned with the second evaluation sample; the federated correlation coefficient characterizes the correlation between the prediction results of the first prediction model and the second prediction model on the same sample; When the federated correlation coefficient meets the first fusion condition, the model discrimination score after fusing the third model score and the fourth model score is calculated based on homomorphic encryption; the first fusion condition includes that the federated correlation coefficient is less than or equal to the correlation threshold; the model discrimination score characterizes the model's performance in distinguishing positive and negative samples. When the model discrimination meets the second fusion condition, it is determined to perform federated prediction of the target user with the second participant; the second fusion condition includes the model discrimination being greater than or equal to the discrimination threshold.

2. The target user prediction method according to claim 1, characterized in that, The calculation of the federal correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing includes: The second participant performs secret sharing based on the third model score and the fourth model score to obtain the product model score of the third model score and the fourth model score; Determine the expected value of the product model, the expected value of the third model, and the standard deviation of the third model; Obtain the expected value and standard deviation of the fourth model score provided by the second participant; The federal correlation coefficient is calculated using the expected value of the third model, the expected value of the fourth model, the standard deviation of the third model, the standard deviation of the fourth model, and the expected value of the product model.

3. The target user prediction method according to claim 1, characterized in that, The calculation of the model discriminability after fusing the third model score and the fourth model score based on homomorphic encryption includes: Generate homomorphic public and private keys; The third model is encrypted using the homomorphic public key to obtain a first encryption result; The first encryption result and the homomorphic public key are provided to the second participant; Obtain the first ciphertext sorting index provided by the second participant. The first ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the second fused ciphertext. The second fused ciphertext is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, and the fourth model score. The first ciphertext sorting index is decrypted using the homomorphic private key to obtain the first sample sorting index of the first evaluation sample; The model's discriminative power is calculated based on the first sample sorting index.

4. The target user prediction method according to claim 3, characterized in that, Before encrypting the third model using the homomorphic public key to obtain the first encryption result, the method further includes: The third model score is weighted based on the first business weight; the first business weight is determined by the negotiation between the first participant and the second participant; and the second fusion score ciphertext is obtained by the second participant through homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the first business weight, and the fourth model score.

5. The method for predicting target users according to claim 1, characterized in that, After calculating the federated correlation coefficient between the third model score and the fourth model score provided by the second participant based on secret sharing, the method further includes: When the federated correlation coefficient does not meet the first fusion condition, the federated correlation coefficient between the third model score and the fifth model score provided by another second participant is calculated based on secret sharing. When the federal correlation coefficient meets the first fusion condition, the model discriminability after fusing the third model score and the fifth model score is calculated based on homomorphic encryption. When the model's discriminative power meets the second fusion condition, it is determined to make a prediction of the target user with the other second participant.

6. The method for predicting target users according to claim 1, characterized in that, After calculating the model discriminability after fusing the third model score and the fourth model score based on homomorphic encryption, the method further includes: When the model discrimination does not meet the second fusion condition, the federated correlation coefficient between the third model score and the sixth model score provided by another second participant is calculated based on secret sharing; When the federal correlation coefficient meets the first fusion condition, the model discriminability after fusing the third model score and the sixth model score is calculated based on homomorphic encryption. When the model's discriminative power meets the second fusion condition, it is determined to make a prediction of the target user with the other second participant.

7. The target user prediction method according to claim 4, characterized in that, After calculating the model discriminability after fusing the third model score and the fourth model score based on homomorphic encryption, the method further includes: When the model's distinguishability does not meet the second fusion condition, the first business weight is adjusted in consultation with the second participant to obtain the second business weight. The third model score is weighted based on the second business weight; Generate homomorphic public and private keys; The third model is encrypted using the homomorphic public key to obtain a first encryption result; The first encryption result and the homomorphic public key are provided to the second participant; Obtain the second ciphertext sorting index provided by the second participant. The second ciphertext sorting index is obtained by the second participant performing homomorphic encryption sorting on the third fused ciphertext. The third fused ciphertext is obtained by the second participant performing homomorphic ciphertext calculation based on the first encryption result, the homomorphic public key, the second business weight, and the fourth model score. The second ciphertext sorting index is decrypted using the homomorphic private key to obtain the second sample sorting index of the first evaluation sample; The model's discriminative power is calculated based on the second sample sorting index; When the model's discriminative power meets the second fusion condition, it is determined to make a prediction of the target user with the second participant.

8. A device for predicting target users, characterized in that, The prediction device for the target user, applied to the first participant, includes: The model score prediction module is used to obtain the first model score on the first selectable user group based on the first prediction model. The homomorphic encryption calculation module is used to obtain the first fused ciphertext of the first model score and the second model score based on homomorphic encryption. The second model score is obtained by the second participant based on the second prediction model on the second optional user group. The first optional user group is aligned with the second optional user group. The homomorphic encryption calculation module is also used to decrypt the first fusion ciphertext to obtain the first fusion model score; The target user determination module is used to determine target users from the first selectable user group based on the first fusion model. In the prediction device for the target user, the model score prediction module is further configured to obtain a third model score based on the first prediction model on the first evaluation sample. The prediction device for the target user further includes a secret-sharing computing module, used to calculate the federated correlation coefficient between the third model score and the fourth model score provided by the second participant based on the secret sharing; the fourth model score is obtained by the second participant on the second evaluation sample based on the second prediction model; the first evaluation sample is aligned with the second evaluation sample; the federated correlation coefficient characterizes the correlation between the prediction results of the first prediction model and the second prediction model on the same sample; The homomorphic encryption calculation module is further configured to calculate the model discrimination degree after fusing the third model score and the fourth model score based on homomorphic encryption when the federated correlation coefficient meets the first fusion condition; the first fusion condition includes the federated correlation coefficient being less than or equal to a correlation threshold; the model discrimination degree characterizes the model's performance in distinguishing positive and negative samples; The target user determination module is further configured to determine, when the model discrimination meets the second fusion condition, to perform federated prediction of the target user with the second participant; the second fusion condition includes the model discrimination being greater than or equal to the discrimination threshold.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the target user prediction method according to any one of claims 1-7.

10. An electronic device, characterized in that, include: processor; and memory for storing the executable instructions of the processor; The processor is configured to execute the target user prediction method according to any one of claims 1-7 by executing the executable instructions.

Images