WO2021120229A1 - 一种数据处理方法、装置及系统 - Google Patents

一种数据处理方法、装置及系统 Download PDF

Info

Publication number
WO2021120229A1
WO2021120229A1 PCT/CN2019/127230 CN2019127230W WO2021120229A1 WO 2021120229 A1 WO2021120229 A1 WO 2021120229A1 CN 2019127230 W CN2019127230 W CN 2019127230W WO 2021120229 A1 WO2021120229 A1 WO 2021120229A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
client
data processing
encrypted
sent
Prior art date
Application number
PCT/CN2019/127230
Other languages
English (en)
French (fr)
Inventor
曹义成
刘博�
王达
郑文琛
杨强
Original Assignee
深圳前海微众银行股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳前海微众银行股份有限公司 filed Critical 深圳前海微众银行股份有限公司
Priority to PCT/CN2019/127230 priority Critical patent/WO2021120229A1/zh
Publication of WO2021120229A1 publication Critical patent/WO2021120229A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data

Definitions

  • the present invention relates to the technical field of financial technology (Fintech), and in particular to a data processing method, device and system.
  • the client entrusts the server to predict the user’s behavior. Accordingly, after receiving the client’s entrustment, the server can use the conversion model to predict the users to be promoted who may perform preset behaviors, so that the users who are to be promoted can perform Promotion to obtain better conversion effect. It can be seen that the transformation effect of the transformation model directly affects the promotion effect. If you want to have a better promotion effect, you need to improve the transformation effect of the transformation model.
  • the client can also expose the conversion data to the server so that the server can use more comprehensive data to train the conversion model , So as to improve the conversion effect of the conversion model.
  • the conversion data includes the user's private data, such as the user's residential address, mobile phone number, ID number, etc. If the conversion data is exposed to the server, the user's private data may not be protected; and the server itself There is a risk of leaking the user's private data. If the server sells the user's private data to other clients of the same type, the experience of the client will be poor and the client will suffer greater losses.
  • the embodiments of the present invention provide a data processing method, device, and system to solve the technical problems of insecure user privacy data and poor client experience caused by exposing and converting data to the server in the prior art.
  • an embodiment of the present invention provides a data processing method, which is applied to a data processing system, and the data processing system connects a client and a server, and the method includes:
  • the first encrypted data is obtained by encrypting the converted data by the client using the preset public key; further, an encrypted conversion model is obtained by training according to the first encrypted data and the second encrypted data, and the The encryption conversion model is sent to the client, and the decryption conversion model sent by the client is received.
  • the decryption conversion model is that the client uses a preset private key corresponding to the preset public key to convert the encryption The model is decrypted.
  • a data processing request sent by the client is also received, and the data processing request includes a data range
  • the data processing request is sent to the server, so that the server obtains the click data that meets the range of the data; accordingly, the click data sent by the server is received, and the click The data is sent to the client, so that the client can obtain the conversion data corresponding to the click data.
  • the data processing request also includes the user name and key of the client; in specific implementation, before sending the data processing request to the server, it is also based on the The corresponding relationship between the user name and the key of each client registered by the data processing system is determined, and the user name and the key corresponding to the client in the data processing request are determined, and it is determined that the client has data processing authority.
  • an embodiment of the present invention provides a data processing system.
  • the data processing system includes an interactive platform and a modeling platform.
  • the interactive platform and the modeling platform are connected by a data interactive link, and the interactive platform also connect the client and server;
  • the interactive platform is configured to receive the first encrypted data and the preset public key sent by the client, and obtain the click data sent by the server, and use the preset public key to encrypt the click data to obtain Second encrypted data; the first encrypted data is obtained by encrypting the converted data by the client using the preset public key;
  • the interaction platform is further configured to send the first encrypted data and the second encrypted data to the modeling platform through the data interaction link;
  • the modeling platform is configured to train to obtain an encryption conversion model according to the first encrypted data and the second encrypted data, and send the encryption conversion model to the interactive platform through the data interaction link;
  • the interactive platform is further configured to send the encryption conversion model to the client, and receive the decryption conversion model sent by the client; the decryption conversion model is that the client uses the preset public key The corresponding preset private key is obtained by decrypting the encryption conversion model.
  • the data exchange link includes a data output link and a data input link arranged between the first isolation end and the second isolation end, and the data output link defines the data to pass through all the links in sequence.
  • the interaction platform, the first isolation terminal, and the second isolation terminal are transmitted to the modeling platform, and the data input link defines data sequentially via the modeling platform, the second isolation terminal, and the The first isolation terminal is transmitted to the interactive platform; the interactive platform is also used to: transmit the first encrypted data and the second encrypted data through the data output link; accordingly, the modeling platform also Used for: transmitting the encryption conversion model through the data input link.
  • the interactive platform is provided with a plurality of interactive nodes deployed in a cluster, and the plurality of interactive nodes are connected to the client through a load balancing node; wherein, the load balancing node uses After receiving the data processing request sent by the client, and according to the resource usage of multiple interactive nodes in the interactive platform, selecting a target interactive node from the multiple interactive nodes, and sending the data processing request to all The target interaction node.
  • the data processing request includes the user name and key of the client; in a specific implementation, after receiving the data processing request sent by the client, the load balancing node also performs The corresponding relationship between the user name and the key of each client registered by the data processing system is determined, the user name of the client in the data processing request corresponds to the key, and it is determined that the client has data processing authority.
  • the data processing request also includes a data range; in specific implementation, the target interaction node is used to: send the data processing request to the server, so that the server obtains the satisfaction The click data within the data range; and, receiving the click data sent by the server, and sending the click data to the load balancing node; correspondingly, the load balancing node is also used for: The click data is sent to the client, so that the client obtains the conversion data corresponding to the click data.
  • the data processing system further includes a model application platform, which is connected to the interaction platform; in specific implementation, the interaction platform is also used to: send the decryption transformation model to The model application platform; correspondingly, the model application platform is also used to predict users to be promoted using the decryption conversion model, perform information promotion on the users to be promoted, and update the decryption according to the results of the promotion Conversion model.
  • the interaction platform and the model application platform perform one-way data interaction.
  • an embodiment of the present invention also provides a data processing device, the device including:
  • the transceiver module is configured to receive the first encrypted data and the preset public key sent by the client; the first encrypted data is obtained by the client using the preset public key to encrypt the converted data;
  • a processing module configured to obtain click data sent by the server, and encrypt the click data using the preset public key to obtain second encrypted data
  • a training module configured to train to obtain an encryption conversion model according to the first encrypted data and the second encrypted data, and send the encryption conversion model to the client;
  • the transceiver module is further configured to receive a decryption transformation model sent by the client, where the decryption transformation model is that the client uses a preset private key corresponding to the preset public key to decrypt the encryption transformation model owned.
  • the transceiver module before the transceiver module receives the first encrypted data and the preset public key sent by the client, it is also used to: receive a data processing request sent by the client, where the data processing request includes Data range; send the data processing request to the server, so that the server obtains the click data that meets the data range; receive the click data sent by the server, and send the click data To the client, so that the client obtains the conversion data corresponding to the click data.
  • the data processing request also includes the user name and key of the client; before the transceiver module sends the data processing request to the server, it is also used to: The corresponding relationship between the user name and the key of each client registered by the data processing system is determined, the user name of the client in the data processing request corresponds to the key, and it is determined that the client has data processing authority.
  • the training module is further used to predict the users to be promoted using the decryption conversion model, and respond to the Promoted users conduct information promotion.
  • the training module after the training module performs information promotion on the user to be promoted, it is further used to update the decryption conversion model according to the promotion result of the user to be promoted.
  • the preset public key is determined based on a differential privacy encryption algorithm.
  • an embodiment of the present invention also provides a computing device, including a program or instruction, and when the program or instruction is executed, the following steps are implemented:
  • the client receives a decryption conversion model sent by the client, where the decryption conversion model is obtained by the client using a preset private key corresponding to the preset public key to decrypt the encryption conversion model.
  • an embodiment of the present invention also provides a storage medium, including a program or instruction, and when the program or instruction is executed, the following steps are implemented:
  • the client receives a decryption conversion model sent by the client, where the decryption conversion model is obtained by the client using a preset private key corresponding to the preset public key to decrypt the encryption conversion model.
  • the entire conversion data is transmitted in an encrypted manner, and the model is also trained based on the encrypted conversion data and the encrypted click data, and the decryption key is always kept by the client, so that other devices (such as servers and data The processing system) cannot decrypt the user’s private data in the converted data.
  • the user’s private data is safer and the client’s experience is better; moreover, by using the data processing system to isolate the client and the server, the server and the client cannot directly communicate with each other. Data interaction, thereby ensuring the security of private data.
  • FIG. 1 is a schematic diagram of a possible system architecture provided by an embodiment of the present invention
  • FIG. 2 is a schematic diagram of the interaction flow of a data processing method provided by an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of another possible system architecture provided by an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of the interaction flow of a data processing method provided by an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of the interaction flow of another data processing method provided by an embodiment of the present invention.
  • FIG. 1 is a schematic diagram of a possible system architecture provided by an embodiment of the present invention.
  • the system architecture includes a data processing system 100, a server 110, and at least one client, such as a client 121, a client 122, and Client 123; wherein, the data processing system 100 connects the server 110 and at least one client, and the connection mode may be a wired mode or a wireless mode, which is not specifically limited.
  • the data processing system 100 may also be connected to at least one client terminal, such as the client terminal 131, the client terminal 132, and the client terminal 133 via a network.
  • the user terminal can be any terminal device, such as a mobile phone, a tablet computer, an Ipad, a notebook, etc., without limitation.
  • the data processing system can train the conversion model by combining the click data in the server and the conversion data in the client while not leaking the conversion data in the client.
  • the click data is used to identify whether the user is Behavior interest
  • conversion data is used to identify whether a user has performed a preset behavior, so that the conversion model can be used to predict the possibility of any user performing a preset behavior.
  • Fig. 2 is a schematic flow diagram corresponding to a data processing method provided by an embodiment of the present invention, and the method includes:
  • Step 201 The client sends the first encrypted data and the preset public key to the data processing system.
  • the first encrypted data is obtained by the client encrypting the converted data using a preset public key, and the private key corresponding to the preset public key is kept by the client.
  • the conversion data includes data about whether the user purchases the advertising product, such as user name, mobile phone number, address, purchase quantity, purchase amount, etc.
  • the identification of the conversion data is set based on the user's identification.
  • the conversion data will contain more privacy data of the user.
  • the prior art usually requires the client to send the conversion data in plain text to the server so that the server can train a conversion model with better effect.
  • this method will leak the privacy of the user.
  • Step 202 The data processing system obtains the click data sent by the server.
  • the client after the client determines to promote an advertisement product, it can send a data processing request to the data processing system.
  • the data processing request can include the user name and key of the client.
  • the data processing request is used for
  • the data processing system authenticates the client.
  • the data processing system after the data processing system receives the data processing request sent by the client, it can first parse the data processing request to obtain the user name and key of the client, and then according to the user name and password of each client registered in the data processing system in advance. Correspondence of the key to determine whether the user name of the client corresponds to the key.
  • the data processing system can send a response message that the authentication fails to the client; accordingly, if the user name of the client exists, and the user name of the client corresponds to the key, it means that the client is already in The data processing system is registered, and the identity of the client has not been forged. In this way, the data processing system can then query to determine whether the client has data processing permissions based on the corresponding relationship between the client and the permissions. If not, you can contact the client Send a response message for permission verification failure. If it has, you can start the data processing process according to the processing conditions set by the client.
  • the client can carry the user name and key of the client every time it sends data to the data processing system, so that the data processing system can verify the client's identity and authority every time. Therefore, in addition to identifying the illegal request of the illegal client when the data processing request is received for the first time, the tampered client can also be identified during the data processing process, so that the data processing process is always in a safe state.
  • the data processing system can identify the illegal request of the illegal client. In this way, the data processing response to the illegal request of the illegal client can be avoided. Respond to the data processing of the client that is safe and has not been tampered with, so as to ensure the security of data processing and improve the credibility of the processing environment in the data processing system.
  • Step 203 The data processing system uses the preset public key to encrypt the click data to obtain second encrypted data.
  • the click data is used to identify whether the user is interested in the preset behavior before performing the preset behavior, and the identification of the click data is set based on the user's identification.
  • the click data can include the user's operation information on the advertising interface of the advertising product, such as user name, whether to click on the advertising interface of the advertising product, browsing time, etc., and can also include the user's exposure information of the advertising product, such as Whether the advertising interface of the advertising product is displayed to the user, etc.
  • step 201 to step 203 is only an example of the execution flow, and does not constitute a limitation on the order of execution of step 201 to step 203.
  • the following describes three possible ways of generating the first encrypted data and the second encrypted data.
  • the client decides to promote a certain advertising product, it can first customize the data range, then generate a data processing request based on the data range and the identifier of the advertising product, and send the data processing request to the data processing system; accordingly; After receiving the data processing request, the data processing system may send the data processing request to the server, so that the server obtains the click data that meets the data range from the click data corresponding to the advertisement product and sends it back to the data processing system.
  • the data range can be set by those skilled in the art according to actual business needs.
  • the data range corresponding to different clients can be different.
  • the data range can include any of the time range, the amount range, the user location area range, and the user identification range.
  • One or any number of items, or other data ranges may also be included, without limitation.
  • the data processing system can send the click data corresponding to the advertising product to the client.
  • the client can identify the click data of each user in the click data.
  • the client can use the preset public key to encrypt the conversion data of each user to obtain a piece of encrypted conversion data, and can set an identifier for the piece of encrypted conversion data based on the user's identifier, thereby obtaining the first encrypted data , And can send the first encrypted data and the preset public key to the data processing system.
  • the data processing system After the data processing system receives the first encrypted data and the preset public key sent by the client, it can also use the preset public key to encrypt each user's click data sent by the server to obtain an encrypted click data, and Set an identifier for the encrypted click data based on the user’s identifier, thereby obtaining the second encrypted data; wherein the method of setting the identifier of the encrypted click data and the method of setting the encrypted conversion data can be consistent, so that the encrypted data of the same user
  • the identification of the click data and the identification of the encrypted conversion data may be consistent, so that the encrypted click data of the same user corresponds to the encrypted conversion data.
  • the preset public key can be determined based on a differential privacy encryption algorithm. Since the differential privacy encryption algorithm can encrypt different feature values of the same feature tag into irregular encrypted feature values, the differential privacy encryption algorithm is used to transform users After the data is encrypted, even if the data processing system obtains the encrypted conversion data, the data processing system cannot determine whether the user has successfully purchased the advertising product based on the encrypted conversion data, thereby ensuring the security of the user's private data. For example, user 1 is a user who has purchased advertising product A in the last 2 months, and users 2 to 5 are users who have clicked on the advertisement of advertising product A in the last 2 months but have not purchased advertising product A.
  • the conversion data of users 1 to 5 is encrypted using the preset public key determined by the differential privacy encryption algorithm, the conversion result of user 1 corresponds to the encrypted feature value of 11100001, and the conversion result of user 2 corresponds to
  • the encrypted feature value of user 3 is 11110001, the encrypted feature value of user 3’s conversion result is 10110001, the encrypted feature value of user 4’s conversion result is 10110101, and the encrypted feature value of user 5’s conversion result is 10110011; obviously, even
  • the data processing system knows that one of the users 1 to 5 has purchased the advertising product A in the last 2 months, and the data processing system cannot determine which user purchased the advertising product A based on the encrypted feature values of the conversion results of users 1 to 5 ⁇ advertised product A.
  • the data processing request is initiated by the client, and the data processing system and the server respond, which realizes the initiative of the client to control the conversion process; and, the embodiment of the present invention supports the client to process the data according to actual needs.
  • the data range is set in the request, and the conversion model is obtained by training the conversion data and click data that meet the data range in the client and the server, so that the effect of the model is better, and the result of data processing is more in line with the needs of the actual scene.
  • the client decides to promote a certain advertising product, it can first customize the data range, and then generate a data processing request based on the data range, preset public key, and the identifier of the advertising product, and send the data processing request to Data processing system; accordingly, after receiving the data processing request, the data processing system can send the data processing request to the server so that the server can obtain click data that meets the data range from the click data corresponding to the advertising product.
  • the data processing system may encrypt the click data corresponding to the advertising product to obtain the second encrypted data, and send the second encrypted data to the client
  • the client can obtain the conversion data of each user according to the identification of each user's click data in the second encrypted data, so as to obtain the conversion data corresponding to the click data.
  • the client can use the preset public key to encrypt the converted data to obtain the first encrypted data, and can send the first encrypted data to the data processing system.
  • the data processing system may encrypt the click data corresponding to the advertisement product to obtain the second encrypted data, and the data processing system may convert the click data Send to the client, so that the client obtains the conversion data of each user according to the identifier of each user's click data in the click data, and obtains the conversion data corresponding to the click data.
  • the client can use the preset public key to encrypt the converted data to obtain the first encrypted data, and can send the first encrypted data to the data processing system.
  • the client if it decides to promote a certain advertising product, it can obtain the conversion data corresponding to the advertising product within the data range.
  • the conversion data corresponding to the advertising product within the data range includes the landing page of the advertising product within the data range.
  • the client can use a preset public key to encrypt the conversion data to obtain an encrypted conversion data, and can set an identifier for the encrypted conversion data based on the user's identifier; In this way, the client terminal can obtain the first encrypted data containing multiple pieces of encrypted conversion data, and each piece of encrypted conversion data is used to identify whether each corresponding user has purchased an advertisement product.
  • the client after the client obtains the first encrypted data, it can also generate a data processing request according to the first encrypted data, the preset public key, the data range, and the identification of the advertising product, and then send the data processing request to the data processing
  • the data processing system can send the data range and the identification of the advertising product to the server, so that the server obtains the click data that meets the data range and sends it to the data processing system.
  • the data processing system can first use the first encrypted data
  • the click data is filtered to obtain the click data matching the first encrypted data, and then the matching click data is encrypted to obtain the second encrypted data.
  • the methods for generating the first encrypted data and the second encrypted data are not limited to the above three.
  • the data processing system may also send the data range and the preset public key to the server at the same time, so that The server sends the second encrypted data obtained by encrypting the click data satisfying the data range by using the preset public key to the data processing system, etc., which is not specifically limited.
  • Step 204 The data processing system trains to obtain an encryption conversion model based on the first encrypted data and the second encrypted data.
  • the type of encryption conversion model can be set by those skilled in the art based on experience.
  • it can be a classification model, or can also be a neural network model, or can also be a regression model, which is not limited in the embodiment of the present invention.
  • the encrypted conversion model may be any one or more of the encrypted purchase conversion model, the encrypted click conversion model, and the encrypted comprehensive conversion model. If the encrypted click data and the encrypted conversion data are used for training If the encrypted purchase conversion model is obtained, the purchase conversion model can predict the probability of users who click on the advertisement page of the advertisement product to purchase the advertisement product; if the encrypted click conversion model is trained using encrypted exposure data and encrypted click data, then the click conversion model It can predict the probability that users who watch the advertisement page of the advertisement product click on the advertisement page; if the encrypted exposure data, encrypted click data and encrypted conversion data are used to train to obtain an encrypted comprehensive conversion model, the comprehensive conversion model can predict the viewing of the advertisement The probability that users on the product's advertising page will purchase the advertised product.
  • Step 205 The data processing system sends the encrypted conversion model to the client.
  • Step 206 The client uses the preset private key corresponding to the preset public key to decrypt the encryption conversion model to obtain the decryption conversion model.
  • the client may store an advertisement product, a table of correspondences between the preset public key and the preset private key, and the client may use different preset public keys to encrypt the conversion data corresponding to multiple advertisement products.
  • the preset public keys of the conversion data corresponding to any two advertising products are different; in this way, after receiving the encrypted conversion model corresponding to any advertising product sent by the data processing system, the client can query the correspondence table to determine the corresponding advertising product
  • the preset private key of the advertisement product is used to decrypt the encryption conversion model corresponding to the advertisement product by using the preset private key corresponding to the advertisement product to obtain the decryption conversion model.
  • the preset public key and the preset private key may be a private key and a public key determined using a differential privacy algorithm, or may also be a private key and a public key determined using a homomorphic encryption algorithm, which are not specifically limited.
  • Step 207 The client sends the decryption transformation model to the data processing system.
  • the client can directly send the decryption conversion model to the data processing system in plaintext, or in order to ensure the security of the decryption conversion model in the data transmission process, first use the set encryption algorithm to convert the decryption The model is encrypted, and then the encrypted conversion model is sent to the data processing system; the set encryption algorithm is negotiated in advance by the client and the data processing system, and the data processing system stores the decryption algorithm corresponding to the set encryption algorithm.
  • the encryption algorithm can be set by those skilled in the art based on experience. For example, it can be a preset hash algorithm, a symmetric encryption algorithm, or an asymmetric encryption algorithm, which is not limited.
  • the data processing system can also use the decryption conversion model to predict the predicted probability of each user buying an advertising product. If the predicted probability of a user buying an advertising product is greater than or Equal to the preset probability, the user can be regarded as a user to be promoted to promote the advertisement product to the user, such as displaying the advertisement page of the advertisement product on the user terminal of the user; accordingly, if the user purchases the advertisement product, the predicted probability If it is less than the preset probability, the user will not be regarded as the user to be promoted.
  • the data processing system can The user corresponding to the user end 131 and the user corresponding to the user end 133 are regarded as two users to be promoted. Therefore, the web browser of the user end 131 and the web browser of the user end 133 can be used to display advertising products to the two users to be promoted. advertising.
  • the internal structure of the data processing system may have multiple possibilities.
  • the data processing system may include only one interactive node, use the interactive node to interact with the client and the server, and use the interactive node to train to obtain encryption Conversion model; or the data processing system may include a load balancing node and multiple interactive nodes, and the load balancing node is used to determine the target interactive node for performing data processing from the multiple interactive nodes, so that the target interactive node is used to perform data with the client and server Interaction, and use the target interaction node to train to obtain the encryption transformation model, and so on.
  • FIG. 3 is a schematic diagram of a possible architecture of a data processing system provided by an embodiment of the present invention.
  • the data processing system 100 may be provided with an interactive platform 101 and a modeling platform 102 connected to each other, and the interactive platform 101 may The client and the server 110 are respectively connected, and the connection method may be a wired method or a wireless method, which is not limited.
  • the data processing system may further include a model application platform 103, the model application platform 103 may also be connected to the interactive platform 101, and the model application platform 103 is connected to at least one client terminal, such as client terminal 131, client terminal 132, and user terminal through a network. ⁇ 133.
  • the computer rooms of the interaction platform 101, the modeling platform 102, and the model application platform 103 can be arranged independently, and the data are isolated from each other to ensure the security of their respective data.
  • Fig. 4 is an interactive flowchart of a data processing method provided by an embodiment of the present invention.
  • the data processing method involves various platforms in the data processing system, specifically including: an interactive platform and a model training platform And model application platform, etc.
  • the method includes:
  • Step 401 The client sends a data processing request to the interactive platform, and the data processing request includes a data range.
  • Step 402 The interactive platform authenticates the client according to the data processing request, and determines whether the client has passed the authentication, if the authentication fails, then executes step 403, and if the authentication passes, then executes step 404.
  • the interactive platform is provided with a load balancing node 1011 and multiple interactive nodes deployed in a cluster, such as interactive node 1012, interactive node 1013, and interactive node 1014; multiple interactive nodes pass through the load balancing node 1011 connect to the client.
  • the data processing request sent by the client can be first obtained by the load balancing node 1011. Accordingly, the load balancing node 1011 can make statistics on the resource usage of each interactive node in the interactive platform at the current moment, and according to the resource usage from more The most idle interactive node among the interactive nodes is selected. Since the interactive node has the strongest processing capability at the current moment, the load balancing node 1011 can use the interactive node as the target interactive node and send a data processing request to the target interactive node.
  • multiple interactive nodes can process multiple data processing requests at the same time, and new data processing requests can be based on each interactive node
  • the resource usage of is allocated to the interactive node with the most idle or best processing capability. In this way, the concurrency of data processing can be guaranteed and the efficiency of data processing can be improved.
  • the load balancing node 1011 may also authenticate the client before determining the target interaction node.
  • the data processing request may also include the client's identification.
  • the load balancing node 1011 receives the data processing request sent by the client, it can be based on the client's identification Query the corresponding relationship between the client and the authority. If the client has the authority to process data, it is determined that the authentication result of the client is authentication passed. If the client does not have the authority to process data, the authentication of the client is determined The result is that the authentication fails.
  • the data processing request can also include the user name and key of the client.
  • the load balancing node 1011 receives the data processing request sent by the client, it can According to query the correspondence between the user name and key of each client registered in the data processing system in advance, if the user name of the client does not correspond to the key, it is determined that the identity authentication of the client is not passed, and step 403 is executed accordingly; If the user name and key of the client are legal, it is determined that the identity authentication of the client is passed.
  • the load balancing node 1011 can query the correspondence between the client and the authority to determine whether the client has data processing capabilities. If the authority does not have, it is determined that the authority authentication of the client is not passed, and step 403 is executed. If it is, it is determined that the authority authentication of the client is passed, so that the authentication result of the client is authentication passed.
  • Step 403 The interactive platform sends a response message of authentication failure to the client.
  • the load balancing node 1011 may send a response message that the identity authentication fails to the client; if the authority authentication of the client fails, the load balancing node 1011 may send the client The terminal sends a response message that the permission authentication fails.
  • Step 404 The interactive platform sends the data range to the server.
  • the load balancing server 1011 can determine the target interactive node from each interactive node according to the resource usage of each interactive node, and send the data processing request to the target interactive node. Node so that the target interactive node sends the data range in the data processing request to the server.
  • the interactive platform in the following embodiments of the present invention all refers to a target interactive node.
  • one-way data interaction is performed between the interactive platform and the server, that is, the server can transmit data to the interactive platform, but cannot obtain the data in the interactive platform.
  • the server by setting the interactive platform to perform one-way data interaction with the server, the server cannot obtain the encrypted conversion data in the interactive platform, so that the security of the conversion data can be better guaranteed.
  • one-way data interaction only limits the transmission flow of data, that is, the output party of one-way data interaction can send data to the input party of one-way data interaction, and the input party of one-way data interaction cannot communicate with one-way data.
  • the output party sends data.
  • one-way data interaction does not limit the transmission flow of the request message.
  • the interactive platform can send data processing requests to the server and receive data response messages sent by the server. Click the one-way transmission process of data, and the server cannot fetch the data from the interactive platform, so as to prevent the server from obtaining the encrypted conversion data.
  • Step 405 After receiving the data range sent by the interactive platform, the server obtains click data that meets the data range, and sends the click data to the interactive platform.
  • the click data is used to identify whether the user is interested in the preset behavior before performing the preset behavior, for example, whether the user clicks on the advertisement page of the advertisement product.
  • Step 406 The interactive platform sends the click data to the client.
  • Step 407 After receiving the click data sent by the interactive platform, the client obtains the conversion data corresponding to the click data, and encrypts the conversion data using the preset public key to obtain the first encrypted data.
  • the conversion data is used to identify whether the user has performed a preset behavior, such as whether a user who meets the data range has purchased an advertising product.
  • Step 408 The client sends the first encrypted data and the preset public key to the interactive platform.
  • Step 409 The interactive platform uses the preset public key to encrypt the click data to obtain the second encrypted data.
  • Step 410 The interactive platform sends the first encrypted data and the second encrypted data to the modeling platform.
  • the interactive platform can be connected to the modeling platform through a fixed data interactive link, and the data interactive link can include a first isolation end and a second isolation end.
  • the first isolation terminal is set on the side of the interactive platform
  • the second isolation terminal is set on the side of the modeling platform.
  • the first isolation terminal and the second isolation terminal follow the private communication Protocol to ensure that the data output link limited data is transmitted to the modeling platform via the interactive platform, the first isolation terminal and the second isolation terminal in turn, and that the data input link limited data is transmitted via the modeling platform, the second isolation terminal and the second isolation terminal in turn An isolated end is transmitted to the interactive platform.
  • first isolation terminal and the second isolation terminal can be any components with input and output isolation functions, such as third-party equipment, signal isolators, signal isolation barriers, etc., which are not specifically limited.
  • the interactive platform can transmit the first encrypted data and the second encrypted data to the modeling platform through the data output link. Accordingly, the modeling platform uses the first encrypted data and the second encrypted data to train to obtain the encrypted conversion model. , You can also transmit the encrypted conversion model to the interactive platform through the data input link.
  • Step 411 The modeling platform uses the first encrypted data and the second encrypted data to train the model to obtain the encrypted conversion model.
  • Step 412 The modeling platform sends the encrypted conversion model to the interactive platform.
  • Step 413 The interactive platform sends the encrypted conversion model to the client.
  • Step 414 The client uses the preset private key corresponding to the preset public key to decrypt the encryption conversion model to obtain the decryption conversion model.
  • step 415 the client sends the decryption transformation model to the interactive platform.
  • Step 416 The interactive platform sends the decryption transformation model to the model application platform.
  • the data processing process of the data processing system can be functionally divided into a data interaction process, a model training process, and a model application process.
  • a data interaction process By setting the data, it is managed by the interactive platform, and the model is trained by the modeling platform and applied by the model.
  • the actual launch of the platform on the market can further functionally isolate the data interaction process, model training process, and model application process to ensure data security, model training security, and model application security.
  • one-way data interaction is performed between the interactive platform and the model application platform, that is, the model application platform can only receive the model sent by the interactive platform, but cannot obtain the data in the interactive platform.
  • the model application platform cannot obtain the encrypted conversion data in the interaction platform, so that the security of the conversion data can be better guaranteed.
  • Step 417 The model application platform uses the decryption transformation model to promote information, and updates the decryption transformation model according to the promotion result.
  • the model application platform can use the decryption conversion model to predict each user and determine the probability of each user performing a preset behavior. If the probability is greater than the preset probability, then The user is regarded as a user to be promoted, and information related to the preset behavior is promoted to the user, for example, an advertisement page of an advertisement product is displayed to the user.
  • the model application platform can also optimize the decryption conversion model based on the promotion result of the user to be promoted. For example, if the promotion result of the user to be promoted is that the user is not interested in the preset behavior or does not perform the preset behavior, the model application platform can Correspondingly reduce the weight of the features of the user to be promoted in the conversion model. If the promotion result of the user to be promoted is to be interested in or perform a preset behavior, the model application platform can correspondingly increase the weight of the user to be promoted in the conversion model. Promote the weight of the characteristics of the user.
  • Figure 5 is an interactive flowchart of another data processing method provided by an embodiment of the present invention.
  • the data processing method involves various platforms in the data processing system, including: interactive platforms, model training Platform and model application platform, etc.
  • the method includes:
  • Step 501 The client sends a data processing request to the interactive platform.
  • the data processing request includes a data range and a preset public key.
  • step 502 the interactive platform authenticates the client according to the data processing request, and determines whether the client has passed the authentication. If the authentication fails, execute step 503; if the authentication passes, execute step 504.
  • Step 503 The interactive platform sends a response message of authentication failure to the client.
  • Step 504 The interactive platform sends the data range to the server.
  • Step 505 After receiving the data range sent by the interactive platform, the server obtains click data that meets the data range, and sends the click data to the interactive platform.
  • the click data is used to identify whether the user is interested in the preset behavior before performing the preset behavior.
  • Step 506 The interactive platform uses the preset public key to encrypt the click data to obtain the second encrypted data.
  • Step 507 The interactive platform sends the click data to the client.
  • Step 508 After receiving the click data sent by the interactive platform, the client obtains the conversion data corresponding to the click data, and encrypts the conversion data using the preset public key to obtain the first encrypted data.
  • the conversion data is used to identify whether the user has performed a preset behavior.
  • Step 509 The client sends the first encrypted data to the interactive platform.
  • Step 510 The interactive platform transmits the first encrypted data and the second encrypted data to the modeling platform through the data output link.
  • Step 511 The modeling platform uses the first encrypted data and the second encrypted data to train the model to obtain the encrypted conversion model.
  • Step 512 The modeling platform transmits the encrypted conversion model to the interactive platform through the data input link.
  • Step 513 The interactive platform sends the encrypted conversion model to the client.
  • Step 514 The client uses the preset private key corresponding to the preset public key to decrypt the encryption conversion model to obtain the decryption conversion model.
  • step 515 the client sends the decryption transformation model to the interactive platform.
  • Step 516 The interactive platform sends the decryption transformation model to the model application platform.
  • step 517 the model application platform uses the decryption transformation model to promote information, and updates the decryption transformation model according to the promotion result.
  • the data processing system may also be provided with a node for managing the life cycle (not shown in FIG. 3), and the node for managing the life cycle may monitor each stage of the data processing process and manage the execution sequence of each stage.
  • the life cycle table is used to record the execution status of the data interaction phase, the modeling phase and the model application phase (such as the unexecuted status, the executing status or the completed status), and it can also store the data interaction phase, the modeling phase and the model application The dependency relationship of the stages, so that each node in the data processing system determines the next stage to be executed based on the dependency relationship.
  • the model application phase depends on the modeling phase
  • the modeling phase depends on the data interaction phase.
  • the data processing system receives the first encrypted data and the preset public key sent by the client, and obtains the click data sent by the server, and uses the preset The public key encrypts the click data to obtain second encrypted data; the first encrypted data is obtained by encrypting the converted data by the client using the preset public key; further, the data processing system An encryption conversion model is obtained by training according to the first encrypted data and the second encrypted data, the encryption conversion model is sent to the client, and the decryption conversion model sent by the client is received, the decryption conversion model It is obtained by decrypting the encryption conversion model by the client using the preset private key corresponding to the preset public key.
  • the entire conversion data is transmitted in an encrypted manner, and the model is also trained based on the encrypted conversion data and the encrypted click data, and the decryption key is always kept by the client, so that other devices (such as servers and data The processing system) cannot decrypt the user’s private data in the converted data.
  • the user’s private data is safer and the client’s experience is better; moreover, by using the data processing system to isolate the client and the server, the server and the client cannot directly communicate with each other. Data interaction, thereby ensuring the security of private data.
  • an embodiment of the present invention also provides a data processing device, which includes:
  • the transceiver module is configured to receive the first encrypted data and the preset public key sent by the client; the first encrypted data is obtained by the client using the preset public key to encrypt the converted data;
  • a processing module configured to obtain click data sent by the server, and encrypt the click data using the preset public key to obtain second encrypted data
  • a training module configured to train to obtain an encryption conversion model according to the first encrypted data and the second encrypted data, and send the encryption conversion model to the client;
  • the transceiver module is further configured to receive a decryption transformation model sent by the client, where the decryption transformation model is that the client uses a preset private key corresponding to the preset public key to decrypt the encryption transformation model owned.
  • the transceiver module before the transceiver module receives the first encrypted data and the preset public key sent by the client, it is further configured to: receive a data processing request sent by the client, where the data processing request includes a data range; The data processing request is sent to the server, so that the server obtains the click data that meets the data range; receives the click data sent by the server, and sends the click data to the client Terminal, so that the client terminal obtains the conversion data corresponding to the click data.
  • the data processing request further includes the user name and key of the client; before the transceiver module sends the data processing request to the server, it is also used to: The corresponding relationship between the user name and the key of each registered client is determined, and the user name of the client in the data processing request is determined to correspond to the key, and it is determined that the client has the data processing authority.
  • the training module is further configured to: use the decryption conversion model to predict users to be promoted, and perform operations on the users to be promoted. Information promotion.
  • the training module performs information promotion on the user to be promoted, it is further used to update the decryption conversion model according to the promotion result of the user to be promoted.
  • the preset public key is determined based on a differential privacy encryption algorithm.
  • an embodiment of the present invention also provides a computing device, including a program or instruction.
  • the program or instruction When the program or instruction is executed, the following steps are implemented: receiving the first encrypted data and the preset public key sent by the client
  • the first encrypted data is obtained by encrypting the conversion data by the client using the preset public key; obtaining the click data sent by the server, and encrypting the click data using the preset public key, Obtain second encrypted data; train to obtain an encryption conversion model according to the first encrypted data and the second encrypted data, and send the encryption conversion model to the client, and receive the decryption conversion sent by the client Model, the decryption conversion model is obtained by decrypting the encryption conversion model by the client using a preset private key corresponding to the preset public key.
  • an embodiment of the present invention also provides a storage medium, including a program or instruction.
  • the program or instruction When the program or instruction is executed, the following steps are implemented: receiving the first encrypted data and the preset public key sent by the client
  • the first encrypted data is obtained by encrypting the conversion data by the client using the preset public key; obtaining the click data sent by the server, and encrypting the click data using the preset public key,
  • Obtain second encrypted data train to obtain an encryption conversion model based on the first encrypted data and the second encrypted data, send the encryption conversion model to the client, and receive the decryption conversion sent by the client Model, the decryption conversion model is obtained by decrypting the encryption conversion model by the client using a preset private key corresponding to the preset public key.
  • the embodiments of the present invention can be provided as a method or a computer program product. Therefore, the present invention may adopt the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
  • a computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.

Abstract

一种数据处理方法、装置及系统，涉及金融科技(Finetch)技术领域。所述数据处理系统(100)接收客户端(121,122,123)发送的第一加密数据和预设公钥后，使用预设公钥加密点击数据得到第二加密数据，根据第一加密数据和第二加密数据训练得到加密转化模型，并将加密转化模型发送给客户端(121,122,123)，接收客户端(121,122,123)使用预设私钥解密加密转化模型得到的解密转化模型。通过客户端(121,122,123)存储解密密钥，并设置转化数据全程加密，使得数据处理系统(100)无法解密得到转化数据中的隐私数据，从而隐私数据较为安全；且，通过使用数据处理系统(100)隔离客户端(121,122,123)与服务器(110)，使得服务器(110)与客户端(121,122,123)无法进行直接的数据交互，从而更加保证了隐私数据的安全性。

Description

一种数据处理方法、装置及系统 技术领域
本发明涉及金融科技(Fintech)技术领域，尤其涉及一种数据处理方法、装置及系统。
背景技术
随着计算机技术的发展，越来越多的技术应用在金融领域，传统金融业正在逐步向金融科技(Fintech)转变，然而，由于金融行业的安全性、实时性要求，金融领域也对技术提出了更高的要求。在用户转化技术领域，客户端委托服务器对用户的行为进行预测，相应地，服务器接收到客户端的委托后，可以使用转化模型预测得到可能会执行预设行为的待推广用户，从而对待推广用户进行推广，以获取较好的转化效果。由此可知，转化模型的转化效果直接影响着推广效果，若想要具有较好的推广效果，则需要提高转化模型的转化效果。
对于客户端来说，转化效果较好的转化模型通常意味着需要较高的转化成本，为了降低转化成本，客户端也可以将转化数据曝光给服务器，以使服务器使用较为全面的数据训练转化模型，从而提高转化模型的转化效果。然而，转化数据中包括用户的隐私数据，比如用户的居住地址、手机号码、身份证号码等，若将转化数据曝光给服务器，则可能会使得用户的隐私数据得不到保护；且，服务器本身存在着泄漏用户的隐私数据的风险，若服务器向其它相同类型的客户端售卖用户的隐私数据，则会使得客户端的体验较差，对客户端造成较大的损失。
综上，目前亟需一种数据处理方法，用以解决现有技术曝光转化数据给服务器所导致的用户的隐私数据不安全、客户端的体验较差的技术问题。
发明内容
本发明实施例提供一种数据处理方法、装置及系统，用以解决现有技术曝光转化数据给服务器所导致的用户的隐私数据不安全、客户端的体验较差的技术问题。
第一方面，本发明实施例提供的一种数据处理方法，所述方法应用于数据处理系统，所述数据处理系统连接客户端和服务器，所述方法包括：
接收所述客户端发送的第一加密数据和预设公钥，并获取所述服务器发送的点击数据，使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；进一步地，根据所述第一加密数据和所述第二加密数据训练得到加密转化模型，将所述加密转化模型发送给所述客户端，并接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
在一种可能的设计中，所述接收所述客户端发送的第一加密数据和预设公钥之前，还接收所述客户端发送的数据处理请求，所述数据处理请求中包括数据范围，将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围之内的所述点击数据；相应地，接收所述服务器发送的所述点击数据，并将所述点击数据发送给所述客户端，以 使所述客户端获取所述点击数据对应的所述转化数据。
在一种可能的设计中，所述数据处理请求中还包括所述客户端的用户名和密钥；具体实施中，所述将所述数据处理请求发送给所述服务器之前，还根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
第二方面，本发明实施例提供的一种数据处理系统，所述数据处理系统包括交互平台和建模平台，所述交互平台与所述建模平台通过数据交互链路连接，所述交互平台还连接客户端和服务器；
所述交互平台，用于接收所述客户端发送的第一加密数据和预设公钥，并获取所述服务器发送的点击数据，使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；所述第一加密数据是所述客户端使用所述预设公钥对所述转化数据进行加密得到的；
所述交互平台，还用于通过所述数据交互链路将所述第一加密数据和所述第二加密数据发送给所述建模平台；
所述建模平台，用于根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并通过所述数据交互链路将所述加密转化模型发送给所述交互平台；
所述交互平台，还用于将所述加密转化模型发送给所述客户端，以及接收所述客户端发送的解密转化模型；所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
在一种可能的设计中，所述数据交互链路包括设置在第一隔离端与第二隔离端之间的数据输出链路和数据输入链路，所述数据输出链路限定数据依次经由所述交互平台、所述第一隔离端和所述第二隔离端传输至所述建模平台，所述数据输入链路限定数据依次经由所述建模平台、所述第二隔离端和所述第一隔离端传输至所述交互平台；所述交互平台还用于：通过所述数据输出链路传输所述第一加密数据和所述第二加密数据；相应地，所述建模平台还用于：通过所述数据输入链路传输所述加密转化模型。
在一种可能的设计中，所述交互平台中设置有按照集群式部署的多个交互节点，所述多个交互节点通过负载均衡节点连接所述客户端；其中，所述负载均衡节点，用于接收所述客户端发送的数据处理请求，并根据所述交互平台中多个交互节点的资源使用情况，从所述多个交互节点中选取目标交互节点，将所述数据处理请求发送给所述目标交互节点。
在一种可能的设计中，所述数据处理请求中包括所述客户端的用户名和密钥；具体实施中，所述负载均衡节点在接收所述客户端发送的数据处理请求之后，还根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
在一种可能的设计中，所述数据处理请求中还包括数据范围；具体实施中，所述目标交互节点用于：将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围之内的所述点击数据；以及，接收所述服务器发送的所述点击数据，将所述点击数据发送给所述负载均衡节点；相应地，所述负载均衡节点还用于：将所述点击数据发送给所述客户端，以使所述客户端获取所述点击数据对应的所述转化数据。
在一种可能的设计中，所述数据处理系统还包括模型应用平台，所述模型应用平台连接所述交互平台；具体实施中，所述交互平台还用于：将所述解密转化模型发送给所述模型应用平台；相应地，所述模型应用平台还用于：使用所述解密转化模型预测出待推广的 用户，并对所述待推广的用户进行信息推广，根据推广结果更新所述解密转化模型。
在一种可能的设计中，所述交互平台与所述模型应用平台进行单向数据交互。
第三方面，本发明实施例还提供了一种数据处理装置，所述装置包括：
收发模块，用于接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
处理模块，用于获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；
训练模块，用于根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端；
所述收发模块，还用于接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
在一种可能的设计中，所述收发模块接收客户端发送的第一加密数据和预设公钥之前，还用于：接收所述客户端发送的数据处理请求，所述数据处理请求中包括数据范围；将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围的所述点击数据；接收所述服务器发送的所述点击数据，并将所述点击数据发送给所述客户端，以使所述客户端获取所述点击数据对应的所述转化数据。
在一种可能的设计中，所述数据处理请求中还包括所述客户端的用户名和密钥；所述收发模块将所述数据处理请求发送给所述服务器之前，还用于：根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
在一种可能的设计中，所述收发模块接收所述客户端发送的解密转化模型之后，所述训练模块还用于：使用所述解密转化模型预测出待推广的用户，并对所述待推广的用户进行信息推广。
在一种可能的设计中，所述训练模块对所述待推广的用户进行信息推广之后，还用于：根据所述待推广的用户的推广结果更新所述解密转化模型。
在一种可能的设计中，所述预设公钥基于差分隐私加密算法确定。
第四方面，本发明实施例还提供了一种计算设备，包括程序或指令，当所述程序或指令被执行时，实现如下步骤：
接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据，根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端，接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
第五方面，本发明实施例还提供了一种存储介质，包括程序或指令，当所述程序或指令被执行时，实现如下步骤：
接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据，根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端，接收所述客户 端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
本发明实施例中，转化数据全程采用加密的方式进行传输，模型也是基于加密的转化数据和加密的点击数据进行训练，而解密密钥始终由客户端进行保管，从而其它设备(比如服务器和数据处理系统)无法解密得到转化数据中的用户的隐私数据，用户的隐私数据较为安全，客户端的体验较好；且，通过使用数据处理系统隔离客户端与服务器，使得服务器与客户端无法进行直接的数据交互，从而更加保证了隐私数据的安全性。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简要介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域的普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例提供的一种可能的系统架构示意图；
图2为本发明实施例提供的一种数据处理方法的交互流程示意图；
图3为本发明实施例提供的又一种可能的系统架构示意图；
图4为本发明实施例提供的一种数据处理方法的交互流程示意图；
图5为本发明实施例提供的又一种数据处理方法的交互流程示意图。
具体实施方式
为了使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步地详细描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
图1为本发明实施例提供的一种可能的系统架构示意图，如图1所示，该系统架构中包括数据处理系统100、服务器110和至少一个客户端，比如客户端121、客户端122和客户端123；其中，数据处理系统100连接服务器110和至少一个客户端，连接的方式可以为有线方式，也可以为无线方式，具体不作限定。
如图1所示，数据处理系统100还可以通过网络连接至少一个用户端，比如用户端131、用户端132和用户端133。其中，用户端可以为任意终端设备，比如手机、平板电脑、Ipad、笔记本等，不作限定。
本发明实施例中，数据处理系统能够在不泄露客户端中的转化数据的同时，联合服务器中的点击数据和客户端中的转化数据训练得到转化模型，点击数据用于标识用户是否对预设行为感兴趣，转化数据用于标识用户是否执行了预设行为，从而转化模型能够用于预测任意用户执行预设行为的可能性。
为了便于理解，本发明的下列实施例以广告推广领域为例描述本发明实施例中的数据处理方法。
基于图1所示意的系统架构，图2为本发明实施例提供的一种数据处理方法对应的流程示意图，该方法包括：
步骤201，客户端向数据处理系统发送第一加密数据和预设公钥。
其中，第一加密数据由客户端使用预设公钥对转化数据进行加密得到，而预设公钥对应的私钥由客户端来保管。在广告产品转化场景中，转化数据包含用户是否购买广告产品的数据，比如用户名、手机号码、住址、购买数量、购买金额等，转化数据的标识基于用户的标识进行设置。
显然地，转化数据中会包含用户较多的隐私数据，现有技术通常需要客户端发送明文的转化数据给服务器，服务器才能训练得到效果较好的转化模型，然而该种方式会泄露用户的隐私数据，导致用户的隐私数据不安全；而本发明实施例由客户端发送加密的转化数据给数据处理系统，且解密密钥由客户端保管，从而数据处理系统无法对加密的转化数据进行解密，保护转化数据的安全。
步骤202，数据处理系统获取服务器发送的点击数据。
在一种可能的实现方式中，客户端在确定对某一广告产品进行推广后，可以向数据处理系统发送数据处理请求，数据处理请求中可以包括客户端的用户名和密钥，数据处理请求用于数据处理系统对客户端进行鉴权。具体实施中，数据处理系统接收到客户端发送的数据处理请求后，可以先解析数据处理请求得到该客户端的用户名和密钥，然后根据预先在数据处理系统中注册的各个客户端的用户名与密钥的对应关系，确定该客户端的用户名和密钥是否对应，若不存在该客户端的用户名，或者该客户端的用户名和密钥不对应，则说明该客户端未在数据处理系统注册或该客户端的身份被伪造，因此，数据处理系统可以向客户端发送身份验证失败的响应消息；相应地，若存在该客户端的用户名，且该客户端的用户名和密钥对应，则说明该客户端已在数据处理系统注册，且该客户端的身份未被伪造，如此，数据处理系统可以再根据客户端与权限的对应关系，查询确定客户端是否具有数据处理的权限，若不具有，则可以向客户端发送权限验证失败的响应消息，若具有，则可以按照客户端设置的处理条件启动数据处理过程。
在另一种可能的实现方式中，客户端每次向数据处理系统发送数据时，均可以携带客户端的用户名和密钥，以使数据处理系统每次都对客户端进行身份验证和权限验证，从而除了可以在初次接收到数据处理请求时识别出非法客户端的非法请求，还可以在数据处理的过程中识别出被篡改的客户端，以使数据处理过程始终处于安全状态。
本发明实施例中，通过在执行数据处理中对客户端的身份和权限进行验证，数据处理系统可以识别出非法客户端的非法请求，如此，可以避免对非法客户端的非法请求进行数据处理响应，而仅对安全且未被篡改的客户端进行数据处理响应，从而可以保证数据处理的安全性，提高数据处理系统中处理环境的可信度。
步骤203，数据处理系统使用预设公钥对点击数据进行加密，得到第二加密数据。
其中，点击数据用于标识用户在执行预设行为之前是否对预设行为感兴趣，点击数据的标识基于用户的标识进行设置。在广告产品转化场景中，点击数据可以包含用户对广告产品的广告界面的操作信息，比如用户名、是否点击广告产品的广告界面、浏览时长等，还可以包括用户对广告产品的曝光信息，比如广告产品的广告界面是否显示给用户等。
需要说明的是，步骤201至步骤203的编号仅为执行流程的一种示例，并不构成对步骤201至步骤203的执行先后顺序的限定。
下面描述第一加密数据和第二加密数据的三种可能的生成方式。
生成方式一
具体实施中，若客户端确定对某一广告产品进行推广，则可以先自定义数据范围，然后根据数据范围和广告产品的标识生成数据处理请求，并将数据处理请求发送给数据处理系统；相应地，数据处理系统接收到数据处理请求后，可以将数据处理请求发送给服务器，以使服务器从该广告产品对应的点击数据中获取满足数据范围的点击数据并回传给数据处理系统。其中，数据范围可以由本领域技术人员根据实际业务需要进行设置，不同客户端对应的数据范围可以不同，比如，数据范围可以包括时间范围、金额范围、用户位置区域范围、用户的标识范围中的任意一项或任意多项，或者也可以包括其它数据范围，不作限定。
相应地，当接收到服务器发送的广告产品对应的点击数据后，数据处理系统可以将广告产品对应的点击数据发送给客户端，如此，客户端可以根据点击数据中每个用户的点击数据的标识获取每个用户的转化数据，从而得到广告产品对应的点击数据对应的转化数据。相应地，客户端可以使用预设公钥对每个用户的转化数据进行加密，得到一条加密的转化数据，并可以基于用户的标识为该条加密的转化数据设置标识，从而得到第一加密数据，并可以将第一加密数据和预设公钥发送给数据处理系统。
当数据处理系统接收到客户端发送的第一加密数据和预设公钥后，还可以使用预设公钥对服务器发送的每个用户的点击数据进行加密，得到一条加密的点击数据，并可以基于用户的标识为该条加密的点击数据设置标识，从而得到第二加密数据；其中，设置加密的点击数据的标识的方式与设置加密的转化数据的方式可以一致，从而使得同一用户的加密的点击数据的标识与加密的转化数据的标识可以保持一致，以使同一用户的加密的点击数据与加密的转化数据对应。
在一个示例中，预设公钥可以基于差分隐私加密算法确定，由于差分隐私加密算法可以将同一特征标签的不同特征值加密为无规律的加密特征值，因此采用差分隐私加密算法对用户的转化数据进行加密后，即使数据处理系统获取到了加密的转化数据，数据处理系统也无法根据加密的转化数据确定用户是否成功购买了广告产品，从而可以保证用户的隐私数据的安全性。举例来说，用户1为最近2个月内购买了广告产品A的用户，用户2～用户5为最近2个月内点击了广告产品A的广告但未购买广告产品A的用户，则针对于转化结果这一特征标签，使用差分隐私加密算法确定的预设公钥对用户1～用户5的转化数据进行加密后，用户1的转化结果对应的加密特征值为11100001，用户2的转化结果对应的加密特征值为11110001，用户3的转化结果对应的加密特征值为10110001，用户4的转化结果对应的加密特征值为10110101，用户5的转化结果对应的加密特征值为10110011；显然地，即使数据处理系统知晓用户1～用户5中存在一个用户在最近的2个月内购买了广告产品A，数据处理系统也无法根据用户1～用户5的转化结果的加密特征值确定是哪一个用户购买了广告产品A。
本发明实施例中，数据处理请求由客户端发起，并由数据处理系统和服务器进行响应，实现了客户端控制转化过程的主动性；且，本发明实施例支持客户端根据实际需要在数据处理请求中设置数据范围，以联合客户端和服务器中满足数据范围的转化数据和点击数据训练得到转化模型，从而模型的效果更好，数据处理的结果更加符合实际场景的需要。
生成方式二
具体实施中，若客户端确定对某一广告产品进行推广，则可以先自定义数据范围，然后根据数据范围、预设公钥和广告产品的标识生成数据处理请求，并将数据处理请求发送 给数据处理系统；相应地，数据处理系统接收到数据处理请求后，可以将数据处理请求发送给服务器，以使服务器从该广告产品对应的点击数据中获取满足数据范围的点击数据。
在一个示例中，当接收到服务器发送的该广告产品对应的点击数据后，数据处理系统可以对该广告产品对应的点击数据进行加密，得到第二加密数据，并将第二加密数据发送给客户端，如此，客户端可以根据第二加密数据中每个用户的点击数据的标识获取每个用户的转化数据，从而得到点击数据对应的转化数据。如此，客户端可以使用预设公钥对转化数据进行加密，得到第一加密数据，并可以将第一加密数据发送给数据处理系统。
在另一个示例中，当接收到服务器发送的该广告产品对应的点击数据后，数据处理系统可以对该广告产品对应的点击数据进行加密，得到第二加密数据，且数据处理系统可以将点击数据发送给客户端，以使客户端根据点击数据中每个用户的点击数据的标识获取每个用户的转化数据，得到点击数据对应的转化数据。如此，客户端可以使用预设公钥对转化数据进行加密，得到第一加密数据，并可以将第一加密数据发送给数据处理系统。
生成方式三
具体实施中，若客户端确定对某一广告产品进行推广，则可以获取数据范围内该广告产品对应的转化数据，数据范围内该广告产品对应的转化数据包括数据范围内在该广告产品的落地页上购买了该广告产品的所有用户的转化数据。进一步地，针对于每个用户的转化数据，客户端可以使用预设公钥对该转化数据进行加密，得到一条加密的转化数据，并可以基于用户的标识为该条加密的转化数据设置标识；如此，客户端即可得到包含多条加密的转化数据的第一加密数据，每条加密的转化数据用于标识对应的每个用户是否购买了广告产品。
在该种生成方式中，客户端得到第一加密数据后，还可以根据第一加密数据、预设公钥、数据范围和广告产品的标识生成数据处理请求，然后将数据处理请求发送给数据处理系统；相应地，数据处理系统可以将数据范围和广告产品的标识发送给服务器，以使服务器获取满足数据范围的点击数据并发送给数据处理系统，如此，数据处理系统可以先使用第一加密数据对点击数据进行过滤，得到与第一加密数据匹配的点击数据，然后再对匹配的点击数据进行加密，得到第二加密数据。
需要说明的是，本发明实施例中，第一加密数据和第二加密数据的生成方式不限于上述三种，比如数据处理系统还可以将数据范围和预设公钥同时发送给服务器，以使服务器将使用预设公钥加密满足数据范围的点击数据得到的第二加密数据发送给数据处理系统，等等，具体不作限定。
步骤204，数据处理系统根据第一加密数据和第二加密数据训练得到加密转化模型。
需要说明的是，加密转化模型的类型可以由本领域技术人员根据经验进行设置，比如可以为分类模型，或者也可以为神经网络模型，或者还可以为回归模型，本发明实施例对此不作限定。
本发明实施例中，加密的转化模型可以为加密的购买转化模型、加密的点击转化模型和加密的综合转化模型中的任意一个或任意多个，若使用加密的点击数据和加密的转化数据训练得到加密的购买转化模型，则购买转化模型可以预测点击广告产品的广告页面的用户购买广告产品的概率；若使用加密的曝光数据和加密的点击数据训练得到加密的点击转化模型，则点击转化模型可以预测观看到广告产品的广告页面的用户点击广告页面的概率；若使用加密的曝光数据、加密的点击数据和加密的转化数据训练得到加密的综合转化模型， 则综合转化模型可以预测观看到广告产品的广告页面的用户购买广告产品的概率。
步骤205，数据处理系统将加密转化模型发送给客户端。
步骤206，客户端使用预设公钥对应的预设私钥对加密转化模型进行解密，得到解密转化模型。
在一个示例中，客户端中可以存储有广告产品、预设公钥与预设私钥的对应关系表，客户端可以分别使用不同的预设公钥对多个广告产品对应的转化数据进行加密，任意两个广告产品对应的转化数据的预设公钥不同；如此，当接收到数据处理系统发送的任一广告产品对应的加密转化模型后，客户端可以查询对应关系表确定该广告产品对应的预设私钥，从而使用该广告产品对应的预设私钥对该广告产品对应的加密转化模型进行解密，得到解密转化模型。
其中，预设公钥与预设私钥可以为使用差分隐私算法确定的私钥和公钥，或者也可以为使用同态加密算法确定的私钥和公钥，具体不作限定。
步骤207，客户端将解密转化模型发送给数据处理系统。
本发明实施例中，客户端可以直接以明文的方式将解密转化模型发送给数据处理系统，或者也可以为了保证解密转化模型在数据传输过程中的安全性，先使用设定加密算法对解密转化模型进行加密，然后将加密后的转化模型发送给数据处理系统；其中，设定加密算法由客户端与数据处理系统预先协商，数据处理系统中存储有设定加密算法对应的解密算法，设定加密算法可以由本领域技术人员根据经验进行设置，比如可以为预设哈希算法、对称加密算法或非对称加密算法，不作限定。
在一个示例中，数据处理系统接收到客户端发送的解密转化模型后，还可以使用解密转化模型预测得到每一个用户购买广告产品的预测概率，若某一用户用户购买广告产品的预测概率大于或等于预设概率，则可以将该用户作为一个待推广用户，从而向该用户推广广告产品，比如在该用户的用户端上显示广告产品的广告页面；相应地，若用户购买广告产品的预测概率小于预设概率，则不将该用户作为待推广用户。举例来说，在预设概率为80％时，若解密转化模型确定用户端131～用户端133对应的用户购买广告产品的概率分别为90％、60％和85％，则数据处理系统可以将用户端131对应的用户和用户端133对应的用户作为两个待推广用户，因此可以分别通过用户端131的web浏览器和用户端133的web浏览器向这两个待推广用户显示广告产品的广告。
本发明实施例中，数据处理系统的内部结构可以具有多种可能，比如数据处理系统可以仅包括一个交互节点，使用该交互节点与客户端和服务器进行数据交互，并使用该交互节点训练得到加密转化模型；或者数据处理系统可以包括负载均衡节点和多个交互节点，使用负载均衡节点从多个交互节点中确定出执行数据处理的目标交互节点，从而使用目标交互节点与客户端和服务器进行数据交互，并使用目标交互节点训练得到加密转化模型，等等。
下面描述一种可能的数据处理系统的结构示意图。
图3为本发明实施例提供的一种数据处理系统的可能的架构示意图，如图3所示，数据处理系统100中可以设置有相互连接的交互平台101和建模平台102，交互平台101可以分别连接客户端和服务器110，连接的方式可以为有线方式，也可以为无线方式，不作限定。
在一个示例中，数据处理系统还可以包括模型应用平台103，模型应用平台103也可 以连接交互平台101，且模型应用平台103通过网络连接至少一个用户端，比如用户端131、用户端132和用户端133。
本发明实施例中，交互平台101、建模平台102和模型应用平台103的机房可以独立布置，数据相互隔离，以保证各自数据的安全性。
基于图3所示意的系统架构，图4为本发明实施例提供的一种数据处理方法的交互流程图，该数据处理方法涉及数据处理系统中的各个平台，具体包括：交互平台、模型训练平台和模型应用平台等。
如图4所示，该方法包括：
步骤401，客户端向交互平台发送数据处理请求，数据处理请求包括数据范围。
步骤402，交互平台根据数据处理请求对客户端进行鉴权，并判断客户端是否鉴权通过，若鉴权不通过，则执行步骤403，若鉴权通过，则执行步骤404。
在一种可能的实现方式中，交互平台中设置有负载均衡节点1011和按照集群式部署的多个交互节点，比如交互节点1012、交互节点1013和交互节点1014；多个交互节点通过负载均衡节点1011连接客户端。
具体实施中，客户端发送的数据处理请求可以先被负载均衡节点1011获取，相应地，负载均衡节点1011可以统计当前时刻交互平台中每个交互节点的资源使用情况，并根据资源使用情况从多个交互节点中选择最为空闲的交互节点，由于当前时刻该交互节点的处理能力最强，因此负载均衡节点1011可以将该交互节点作为目标交互节点，并将数据处理请求发送给目标交互节点。
本发明实施例中，通过在交互平台中设置负载均衡节点和集群式布置的多个交互节点，使得多个交互节点可以同时处理多个数据处理请求，且新的数据处理请求能够基于各个交互节点的资源使用情况被分给最为空闲或处理能力最好的交互节点，如此，可以保证数据处理的并发性，并可以提高数据处理的效率。
具体实施中，负载均衡节点1011在接收到客户端发送的数据处理请求后，确定目标交互节点之前，还可以对客户端进行鉴权。鉴权的方式可以有多种，下面描述两种可能的鉴权方式：
在一种可能的鉴权方式中，数据处理请求中除了包括数据范围外，还可以包括客户端的标识，如此，负载均衡节点1011在接收到客户端发送的数据处理请求后，可以根据客户端的标识查询客户端与权限的对应关系，若该客户端具有数据处理的权限，则确定该客户端的鉴权结果为鉴权通过，若该客户端不具有数据处理的权限，则确定该客户端的鉴权结果为鉴权不通过。
在另一种可能的鉴权方式中，数据处理请求中除了包括数据范围外，还可以包括客户端的用户名和密钥，如此，负载均衡节点1011在接收到客户端发送的数据处理请求后，可以根据查询预先在数据处理系统注册的各个客户端的用户名与密钥的对应关系，若该客户端的用户名与密钥不对应，则确定对该客户端的身份鉴权不通过，执行步骤403；相应地，若该客户端的用户名与密钥合法，则确定对该客户端的身份鉴权通过，如此，负载均衡节点1011可以再查询客户端与权限的对应关系，确定该客户端是否具有数据处理的权限，若不具有，则确定对该客户端的权限鉴权不通过，执行步骤403，若具有，则确定对该客户端的权限鉴权通过，从而该客户端的鉴权结果为鉴权通过。
步骤403，交互平台向客户端发送鉴权失败的响应消息。
其中，若对该客户端的身份鉴权不通过，则负载均衡节点1011可以向客户端发送身份鉴权失败的响应消息，若对该客户端的权限鉴权不通过，则负载均衡节点1011可以向客户端发送权限鉴权失败的响应消息。
步骤404，交互平台将个数据范围发送给服务器。
具体实施中，若负载均衡服务器1011对客户端的鉴权通过，则负载均衡服务器1011可以根据各个交互节点的资源使用情况从各个交互节点中确定出目标交互节点，并将数据处理请求发送给目标交互节点，以使目标交互节点将数据处理请求中的数据范围发送给服务器。
为了统一描述，本发明的下列实施例中交互平台均是指目标交互节点。
在一个示例中，交互平台与服务器之间进行单向数据交互，即服务器可以向交互平台传输数据，而无法获取交互平台中的数据。在该示例中，通过设置交互平台与服务器进行单向数据交互，使得服务器无法获取到交互平台中加密的转化数据，从而可以较好地保证转化数据的安全性。
需要说明的是，单向数据交互仅限定数据的传输流向，即单向数据交互的输出方可以向单向数据交互的输入方发送数据，而单向数据交互的输入方无法向单向数据交互的输出方发送数据，然而，单向数据交互并不限定请求消息的传输流向，比如在单向交互中，交互平台可以向服务器发送数据处理请求，并可以接收服务器发送的数据响应消息，从而实现点击数据的单向传输过程，而服务器无法从交互平台中取出数据，从而避免服务器获取到加密的转化数据。
步骤405，服务器接收交互平台发送的数据范围后，获取满足数据范围的点击数据，并将点击数据发送给交互平台。
其中，点击数据用于标识用户在执行预设行为之前是否对预设行为感兴趣，比如用户是否点击了广告产品的广告页面。
步骤406，交互平台将点击数据发送给客户端。
步骤407，客户端接收交互平台发送的点击数据后，获取点击数据对应的转化数据，并使用预设公钥对转化数据进行加密，得到第一加密数据。
其中，转化数据用于标识用户是否执行了预设行为，比如满足数据范围的用户是否购买了广告产品。
步骤408，客户端将第一加密数据和预设公钥发送给交互平台。
步骤409，交互平台使用预设公钥对点击数据进行加密，得到第二加密数据。
步骤410，交互平台将第一加密数据和第二加密数据发送给建模平台。
在一种可能的实现方式中，如图3所示，交互平台可以通过固定的数据交互链路与建模平台连接，数据交互链路可以包括设置在第一隔离端与第二隔离端之间的数据输出链路和数据输入链路，第一隔离端设置在交互平台一侧，第二隔离端设置在建模平台一侧；其中，第一隔离端和第二隔离端之间遵循私有通信协议，从而保证数据输出链路限定数据依次经由交互平台、第一隔离端和第二隔离端传输给建模平台，且保证数据输入链路限定数据依次经由建模平台、第二隔离端和第一隔离端传输给交互平台。
需要说明的是，第一隔离端和第二隔离端可以为具有输入输出隔离功能的任意组件，比如第三方设备、信号隔离器、信号隔离栅等，具体不作限定。
具体实施中，交互平台可以通过数据输出链路传输第一加密数据和第二加密数据给建 模平台，相应地，建模平台在使用第一加密数据和第二加密数据训练得到加密转化模型后，也可以通过数据输入链路传输加密转化模型给交互平台。
在上述实现方式中，通过在交互平台和建模平台之间设置固定的输出链路和固定的输入链路，使得数据能够按照设置好的链路在交互平台和建模平台之间进行双向传输，如此，可以避免数据在交互平台和建模平台之间传输时被其它的非法机构获取，提高数据的安全性。
步骤411，建模平台使用第一加密数据和第二加密数据训练模型，得到加密转化模型。
步骤412，建模平台将加密转化模型发送给交互平台。
步骤413，交互平台将加密转化模型发送给客户端。
步骤414，客户端使用预设公钥对应的预设私钥对加密转化模型进行解密，得到解密转化模型。
步骤415，客户端将解密转化模型发送给交互平台。
步骤416，交互平台将解密转化模型发送给模型应用平台。
本发明实施例中，数据处理系统的数据处理过程在功能上可以划分为数据交互过程、模型训练过程和模型应用过程，通过设置数据由交互平台管理，模型由建模平台训练，且由模型应用平台实际投入市场，可以进一步地在功能上隔离数据交互过程、模型训练过程和模型应用过程，保证数据的安全性、模型训练的安全性和模型应用的安全性。
在一个示例中，交互平台与模型应用平台之间进行单向数据交互，即模型应用平台仅能接收交互平台发送的模型，而无法获取交互平台中的数据。在该示例中，通过设置模型应用平台与交互平台进行单向数据交互，使得模型应用平台无法获取到交互平台中加密的转化数据，从而可以较好地保证转化数据的安全性。
步骤417，模型应用平台使用解密转化模型进行信息推广，并根据推广结果更新解密转化模型。
具体实施中，模型应用平台接收到交互平台发送的解密转化模型后，可以使用解密转化模型对每一个用户进行预测，确定每一个用户执行预设行为的概率，若概率大于预设概率，则可以将该用户作为一个待推广的用户，并向该用户推广与预设行为相关的信息，比如向用户显示广告产品的广告页面。
在一个示例中，模型应用平台还可以根据待推广用户的推广结果优化解密转化模型，比如，若待推广用户的推广结果为对预设行为不感兴趣或不执行预设行为，则模型应用平台可以相应地降低转化模型中该待推广用户所具有的特征的权重，若待推广用户的推广结果为对预设行为感兴趣或执行预设行为，则模型应用平台可以相应地提升转化模型中该待推广用户所具有的特征的权重。
基于图3所示意的系统架构，图5为本发明实施例提供的又一种数据处理方法的交互流程图，该数据处理方法涉及数据处理系统中的各个平台，具体包括：交互平台、模型训练平台和模型应用平台等。
如图5所示，该方法包括：
步骤501，客户端向交互平台发送数据处理请求，数据处理请求包括数据范围和预设公钥。
步骤502，交互平台根据数据处理请求对客户端进行鉴权，并判断客户端是否鉴权通过，若鉴权不通过，则执行步骤503，若鉴权通过，则执行步骤504。
步骤503，交互平台向客户端发送鉴权失败的响应消息。
步骤504，交互平台将数据范围发送给服务器。
步骤505，服务器接收交互平台发送的数据范围后，获取满足数据范围的点击数据，并将点击数据发送给交互平台。
其中，点击数据用于标识用户在执行预设行为之前是否对预设行为感兴趣。
步骤506，交互平台使用预设公钥对点击数据进行加密，得到第二加密数据。
步骤507，交互平台将点击数据发送给客户端。
步骤508，客户端接收交互平台发送的点击数据后，获取点击数据对应的转化数据，并使用预设公钥对转化数据进行加密，得到第一加密数据。
其中，转化数据用于标识用户是否执行了预设行为。
步骤509，客户端将第一加密数据发送给交互平台。
步骤510，交互平台通过数据输出链路传输第一加密数据和第二加密数据给建模平台。
步骤511，建模平台使用第一加密数据和第二加密数据训练模型，得到加密转化模型。
步骤512，建模平台通过数据输入链路传输加密转化模型给交互平台。
步骤513，交互平台将加密转化模型发送给客户端。
步骤514，客户端使用预设公钥对应的预设私钥对加密转化模型进行解密，得到解密转化模型。
步骤515，客户端将解密转化模型发送给交互平台。
步骤516，交互平台将解密转化模型发送给模型应用平台。
步骤517，模型应用平台使用解密转化模型进行信息推广，并根据推广结果更新解密转化模型。
本发明实施例中，数据处理系统中还可以设置有管理生命周期的节点(图3未示意)，管理生命周期的节点可以对数据处理过程的各个阶段进行监控，并管理各个阶段的执行顺序。其中，生命周期表用于记录数据交互阶段、建模阶段和模型应用阶段的执行状态(比如未执行状态、正在执行状态或已完成状态)，还可以存储数据交互阶段、建模阶段和模型应用阶段的依赖关系，以使数据处理系统中的各个节点基于依赖关系确定下一步的待执行阶段。举例来说，模型应用阶段依赖建模阶段，建模阶段依赖数据交互阶段，如此，若建模阶段的执行状态为未执行状态或正在执行状态，则数据处理系统无法启动模型应用阶段，相应地，若数据交互阶段的执行状态为未执行状态或正在执行状态，则数据处理系统也无法启动建模阶段。
从上述内容可以看出：本发明的上述实施例中，数据处理系统接收所述客户端发送的第一加密数据和预设公钥，并获取所述服务器发送的点击数据，使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；进一步地，所述数据处理系统根据所述第一加密数据和所述第二加密数据训练得到加密转化模型，将所述加密转化模型发送给所述客户端，并接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。本发明实施例中，转化数据全程采用加密的方式进行传输，模型也是基于加密的转化数据和加密的点击数据进行训练，而解密密钥始终由客户端进行保管，从而其它设备(比如服务器和数据处理系统)无法解密得到转化数据中的用户的隐私数据，用户的隐私数据较为安全，客户端的体验较好；且，通过使 用数据处理系统隔离客户端与服务器，使得服务器与客户端无法进行直接的数据交互，从而更加保证了隐私数据的安全性。
基于同一发明构思，本发明实施例还提供了一种数据处理装置，所述装置包括：
收发模块，用于接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
处理模块，用于获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；
训练模块，用于根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端；
所述收发模块，还用于接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
可选地，所述收发模块接收客户端发送的第一加密数据和预设公钥之前，还用于：接收所述客户端发送的数据处理请求，所述数据处理请求中包括数据范围；将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围的所述点击数据；接收所述服务器发送的所述点击数据，并将所述点击数据发送给所述客户端，以使所述客户端获取所述点击数据对应的所述转化数据。
可选地，所述数据处理请求中还包括所述客户端的用户名和密钥；所述收发模块将所述数据处理请求发送给所述服务器之前，还用于：根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
可选地，所述收发模块接收所述客户端发送的解密转化模型之后，所述训练模块还用于：使用所述解密转化模型预测出待推广的用户，并对所述待推广的用户进行信息推广。
可选地，所述训练模块对所述待推广的用户进行信息推广之后，还用于：根据所述待推广的用户的推广结果更新所述解密转化模型。
可选地，所述预设公钥基于差分隐私加密算法确定。
基于同一发明构思，本发明实施例还提供了一种计算设备，包括程序或指令，当所述程序或指令被执行时，实现如下步骤：接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端，接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
基于同一发明构思，本发明实施例还提供了一种存储介质，包括程序或指令，当所述程序或指令被执行时，实现如下步骤：接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据，根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端，接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
本领域内的技术人员应明白，本发明的实施例可提供为方法、或计算机程序产品。因 此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
显然，本领域的技术人员可以对本发明进行各种改动和变型而不脱离本发明的精神和范围。这样，倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内，则本发明也意图包含这些改动和变型在内。

Claims (20)

  1. 一种数据处理方法，其特征在于，所述方法包括：
    接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
    获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；
    根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端；
    接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
  2. 根据权利要求1所述的方法，其特征在于，所述接收所述客户端发送的第一加密数据和预设公钥之前，还包括：
    接收所述客户端发送的数据处理请求，所述数据处理请求中包括数据范围；
    将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围的所述点击数据；
    接收所述服务器发送的所述点击数据；
    将所述点击数据发送给所述客户端，以使所述客户端获取所述点击数据对应的所述转化数据。
  3. 根据权利要求2所述的方法，其特征在于，所述数据处理请求中还包括所述客户端的用户名和密钥；
    所述将所述数据处理请求发送给所述服务器之前，还包括：
    根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
  4. 根据权利要求1所述的方法，其特征在于，所述接收所述客户端发送的解密转化模型之后，还包括：
    使用所述解密转化模型预测出待推广的用户，并对所述待推广的用户进行信息推广。
  5. 根据权利要求4所述的方法，其特征在于，所述对所述待推广的用户进行信息推广之后，还包括：
    根据所述待推广的用户的推广结果更新所述解密转化模型。
  6. 根据权利要求1至5中任一项所述的方法，其特征在于，所述预设公钥基于差分隐私加密算法确定。
  7. 一种数据处理系统，其特征在于，所述数据处理系统包括交互平台和建模平台，所述交互平台与所述建模平台通过数据交互链路连接，所述交互平台还连接客户端和服务器；
    所述交互平台，用于接收所述客户端发送的第一加密数据和预设公钥，并获取所述服务器发送的点击数据，使用所述预设公钥对点击数据进行加密，得到第二加密数据；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
    所述交互平台，还用于通过所述数据交互链路将所述第一加密数据和所述第二加密数 据发送给所述建模平台；
    所述建模平台，用于根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并通过所述数据交互链路将所述加密转化模型发送给所述交互平台；
    所述交互平台，还用于将所述加密转化模型发送给所述客户端，以及接收所述客户端发送的解密转化模型；所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
  8. 根据权利要求7所述的数据处理系统，其特征在于，所述数据交互链路包括设置在第一隔离端与第二隔离端之间的数据输出链路和数据输入链路，所述数据输出链路限定数据依次经由所述交互平台、所述第一隔离端和所述第二隔离端传输至所述建模平台，所述数据输入链路限定数据依次经由所述建模平台、所述第二隔离端和所述第一隔离端传输至所述交互平台；
    所述交互平台，还用于通过所述数据输出链路传输所述第一加密数据和所述第二加密数据；
    所述建模平台，还用于通过所述数据输入链路传输所述加密转化模型。
  9. 根据权利要求7所述的数据处理系统，其特征在于，所述交互平台中设置有按照集群式部署的多个交互节点，所述多个交互节点通过负载均衡节点连接所述客户端；
    所述负载均衡节点，用于接收所述客户端发送的数据处理请求，并根据所述交互平台中多个交互节点的资源使用情况，从所述多个交互节点中选取目标交互节点，将所述数据处理请求发送给所述目标交互节点。
  10. 根据权利要求9所述的数据处理系统，其特征在于，所述数据处理请求中包括所述客户端的用户名和密钥；
    所述负载均衡节点在接收所述客户端发送的数据处理请求之后，还用于：
    根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
  11. 根据权利要求9或10所述的数据处理系统，其特征在于，所述数据处理请求中还包括数据范围；
    所述目标交互节点，用于将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围的所述点击数据；以及，接收所述服务器发送的所述点击数据，将所述点击数据发送给所述负载均衡节点；
    所述负载均衡节点，还用于将所述点击数据发送给所述客户端，以使所述客户端获取所述点击数据对应的所述转化数据。
  12. 根据权利要求11所述的数据处理系统，其特征在于，所述数据处理系统还包括模型应用平台，所述模型应用平台连接所述交互平台；
    所述交互平台，还用于将所述解密转化模型发送给所述模型应用平台；
    所述模型应用平台，还用于使用所述解密转化模型预测出待推广的用户，并对所述待推广的用户进行信息推广，根据推广结果更新所述解密转化模型。
  13. 根据权利要求12所述的数据处理系统，其特征在于，所述交互平台与所述模型应用平台进行单向数据交互。
  14. 一种数据处理装置，其特征在于，所述装置包括：
    收发模块，用于接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
    处理模块，用于获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；
    训练模块，用于根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端；
    所述收发模块，还用于接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
  15. 根据权利要求14所述的装置，其特征在于，所述收发模块接收客户端发送的第一加密数据和预设公钥之前，还用于：
    接收所述客户端发送的数据处理请求，所述数据处理请求中包括数据范围；将所述数据处理请求发送给所述服务器，以使所述服务器获取满足所述数据范围的所述点击数据；接收所述服务器发送的所述点击数据，并将所述点击数据发送给所述客户端，以使所述客户端获取所述点击数据对应的所述转化数据。
  16. 根据权利要求15所述的装置，其特征在于，所述数据处理请求中还包括所述客户端的用户名和密钥；
    所述收发模块将所述数据处理请求发送给所述服务器之前，还用于：
    根据预先在所述数据处理系统注册的各个客户端的用户名与密钥的对应关系，确定所述数据处理请求中所述客户端的用户名和密钥对应，并确定所述客户端具有数据处理的权限。
  17. 根据权利要求16所述的装置，其特征在于，所述收发模块接收所述客户端发送的解密转化模型之后，所述训练模块还用于：
    使用所述解密转化模型预测出待推广的用户，并对所述待推广的用户进行信息推广。
  18. 根据权利要求17所述的装置，其特征在于，所述训练模块对所述待推广的用户进行信息推广之后，还用于：
    根据所述待推广的用户的推广结果更新所述解密转化模型。
  19. 一种计算设备，其特征在于，包括程序或指令，当所述程序或指令被执行时，实现如下步骤：
    接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
    获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；
    根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端；
    接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
  20. 一种存储介质，其特征在于，包括程序或指令，当所述程序或指令被执行时，实现如下步骤：
    接收客户端发送的第一加密数据和预设公钥；所述第一加密数据是所述客户端使用所述预设公钥对转化数据进行加密得到的；
    获取服务器发送的点击数据，并使用所述预设公钥对所述点击数据进行加密，得到第二加密数据；
    根据所述第一加密数据和所述第二加密数据，训练得到加密转化模型，并将所述加密转化模型发送给所述客户端；
    接收所述客户端发送的解密转化模型，所述解密转化模型为所述客户端使用所述预设公钥对应的预设私钥对所述加密转化模型进行解密得到的。
PCT/CN2019/127230 2019-12-20 2019-12-20 一种数据处理方法、装置及系统 WO2021120229A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/127230 WO2021120229A1 (zh) 2019-12-20 2019-12-20 一种数据处理方法、装置及系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/127230 WO2021120229A1 (zh) 2019-12-20 2019-12-20 一种数据处理方法、装置及系统

Publications (1)

Publication Number Publication Date
WO2021120229A1 true WO2021120229A1 (zh) 2021-06-24

Family

ID=76477053

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/127230 WO2021120229A1 (zh) 2019-12-20 2019-12-20 一种数据处理方法、装置及系统

Country Status (1)

Country Link
WO (1) WO2021120229A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108205575A (zh) * 2016-12-20 2018-06-26 北京国双科技有限公司 数据处理方法和装置
CN110059501A (zh) * 2019-04-16 2019-07-26 广州大学 一种基于差分隐私的安全外包机器学习方法
US20190294805A1 (en) * 2018-03-22 2019-09-26 Via Science, Inc. Neural-network training using secure data processing
CN110399742A (zh) * 2019-07-29 2019-11-01 深圳前海微众银行股份有限公司 一种联邦迁移学习模型的训练、预测方法及装置
CN110516464A (zh) * 2019-09-02 2019-11-29 深圳市网心科技有限公司 基于神经网络计算的数据保护方法及相关设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108205575A (zh) * 2016-12-20 2018-06-26 北京国双科技有限公司 数据处理方法和装置
US20190294805A1 (en) * 2018-03-22 2019-09-26 Via Science, Inc. Neural-network training using secure data processing
CN110059501A (zh) * 2019-04-16 2019-07-26 广州大学 一种基于差分隐私的安全外包机器学习方法
CN110399742A (zh) * 2019-07-29 2019-11-01 深圳前海微众银行股份有限公司 一种联邦迁移学习模型的训练、预测方法及装置
CN110516464A (zh) * 2019-09-02 2019-11-29 深圳市网心科技有限公司 基于神经网络计算的数据保护方法及相关设备

Similar Documents

Publication Publication Date Title
CN109167695B (zh) 基于联邦学习的联盟网络构建方法、设备及可读存储介质
WO2022042301A1 (zh) 一种数据处理方法、装置、智能设备及存储介质
TWI672648B (zh) 業務處理方法、裝置、資料共享系統及儲存介質
US20220278962A1 (en) Generating and linking private transaction identifiers to distributed data repositories
US20220318907A1 (en) Systems and methods for generating secure, encrypted communications across distributed computer networks for authorizing use of cryptography-based digital repositories in order to perform blockchain operations in decentralized applications
WO2022247576A1 (zh) 一种数据处理方法、装置、设备及计算机可读存储介质
JP2019503533A5 (zh)
US11949671B2 (en) Systems and methods for managing secure sharing of online data
CN112804354B (zh) 跨链进行数据传输的方法、装置、计算机设备和存储介质
WO2023005838A1 (zh) 数据共享方法和电子设备
CN105407097A (zh) 基于第三方的数据供应方法及装置
US10909264B2 (en) Personal data providing system, personal data providing method, and information processing apparatus
CN107302524A (zh) 一种云计算环境下的密文数据共享系统
US11388147B2 (en) System and method for redirecting data access to local trust managers via an indirection logic service
CN110807209B (zh) 一种数据处理方法、设备及存储介质
US10970741B2 (en) Method and system for integrating audience data with a social media site
CN111125734B (zh) 一种数据处理方法及系统
CN113821811B (zh) 基于区块链的数据获取方法及系统、电子设备及存储介质
CN110034922A (zh) 请求处理方法、处理装置以及请求验证方法、验证装置
WO2021120229A1 (zh) 一种数据处理方法、装置及系统
CN111131227B (zh) 一种数据处理方法及装置
WO2017159067A1 (ja) 情報処理装置およびエージェントシステム
CN115242486B (zh) 数据处理方法、装置及计算机可读存储介质
WO2024078108A1 (zh) 一种基于区块链的数据处理方法、装置、电子设备、计算机可读存储介质及计算机程序产品
JP2014215884A (ja) 複数のネットワーク・プラットフォームの秘密鍵管理方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19956339

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 12/10/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19956339

Country of ref document: EP

Kind code of ref document: A1