CN111160961B - Information recommendation method and device - Google Patents
Information recommendation method and device Download PDFInfo
- Publication number
- CN111160961B CN111160961B CN201911329851.XA CN201911329851A CN111160961B CN 111160961 B CN111160961 B CN 111160961B CN 201911329851 A CN201911329851 A CN 201911329851A CN 111160961 B CN111160961 B CN 111160961B
- Authority
- CN
- China
- Prior art keywords
- terminal
- conversion
- click
- user
- data
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
Landscapes
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an information recommendation method and device, wherein the method comprises the following steps: the first terminal obtains conversion data containing user characteristics through the second terminal; the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model; and the first terminal determines a target user of the second terminal according to the click estimated model and the conversion estimated model, and pushes information to be recommended to the target user of the second terminal. When the method is applied to financial science and technology (Fintech), the users of information recommendation are screened, the users are concentrated in the range of target users and put in, the proportion of the users of information pushing to conversion users is improved, and therefore the conversion effect of information recommendation is improved.
Description
Technical Field
The invention relates to the field of financial science and technology (Fintech) and the field of computer software, in particular to an information recommendation method and device.
Background
With the development of computer technology, more and more technologies (big data, distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, and the traditional financial industry is gradually changing to financial technology (Fintech). At present, information recommendation in the field of financial science and technology is particularly important, and is a main channel for expanding users of financial products. The information recommendation is to put advertisement materials which are wanted to be put in by an information recommendation source into a user side through an information recommendation platform, display the advertisement materials to a user, and stimulate the user to click on the pushed information and consume the pushed information.
Along with the enrichment, diversification and refinement of information recommendation demands, the pursuit of information recommendation sources on the information recommendation post-conversion effect is stronger, and the number of users converted after the information recommendation has a remarkable relationship with how the information recommendation is. Reasonable targeted information recommendation can be aimed at accurate throwing of users. However, in the current information recommendation method, users are not distinguished, and unified delivery is performed blindly, so that the conversion effect of information recommendation is poor.
Disclosure of Invention
The embodiment of the application provides an information recommendation method and device, which solve the problem of poor information recommendation conversion effect in the prior art.
In a first aspect, an embodiment of the present application provides an information recommendation method: the first terminal obtains conversion data containing user characteristics through the second terminal; the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model; wherein, the click pre-estimation model and the conversion pre-estimation model are irrelevant to the behavior of any user; the first terminal determines a target user of the second terminal according to the click estimated model and the conversion estimated model, and pushes information to be recommended to the target user of the second terminal; the target user of the second terminal is a user with the conversion probability larger than a first preset probability threshold value among the users of the second terminal.
In the method, after the first terminal obtains the conversion data containing the user characteristics, the first terminal can combine the original exposure data, the original click data and the conversion data to perform machine learning training to obtain the click pre-estimation model and the conversion pre-estimation model, and as the pre-estimation value of any user by the click pre-estimation model and the conversion pre-estimation model is irrelevant to the behavior of the user, the training effect of the model is not affected by the change of the subsequent behavior of the user, the first terminal can determine the conversion probability of each user through the click pre-estimation model and the conversion pre-estimation model, further determine the second terminal meeting the preset condition, determine the target user of the second terminal according to the information to be pushed by the second terminal meeting the preset condition, and push the information to be recommended to the target user of the second terminal.
In an alternative embodiment, the first terminal obtains the conversion data including the user feature through the second terminal, including: the first terminal sends user identifiers of a plurality of click users in the original click data to the second terminal; the first terminal obtains tag values of the plurality of click users from the second terminal; the label values of the plurality of click users are determined by the second terminal according to the user identifications of the plurality of click users; and the first terminal uses the combined data obtained after the original click data are matched with the tag values of the click users as the conversion data according to the user identifications of the click users.
In the method, the conversion users accumulated by the first terminal are limited, but the first terminal can obtain the tag values determined by the second terminal according to the user identifications of the plurality of click users by sending the user identifications of the plurality of click users to the second terminal, so that more conversion users can be obtained according to the tag values of the click users, and the combined data obtained after the original click data are matched with the tag values of the plurality of click users is used as the conversion data, thereby providing a method for obtaining the conversion data according to the second terminal.
In an optional implementation manner, the tag value of any click user of the plurality of click users is an encrypted tag value obtained by encrypting the original tag by the second terminal according to a preset encryption algorithm.
In the method, the label value of the click user is an encrypted label value obtained by encrypting the original label by the second terminal according to the preset encryption algorithm, so that the label value of the click user is invisible to the first terminal, but the function of the label value can be reflected in the training process, and the privacy of the conversion user is protected on the basis of obtaining an effective conversion estimated model.
In an optional implementation manner, the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model, and includes: the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; the first terminal sends the encryption model parameters of the click pre-estimated model and the encryption model parameters of the conversion pre-estimated model to the second terminal; the first terminal obtains decryption model parameters of the click pre-estimation model and decryption model parameters of the conversion pre-estimation model from the second terminal, so that the click pre-estimation model and the conversion pre-estimation model are obtained; the decryption model parameters of the click pre-estimated model and the decryption model parameters of the conversion pre-estimated model are obtained by the second terminal according to a decryption algorithm corresponding to the preset encryption algorithm.
The method is a specific acquisition method of the conversion estimation model, machine learning training is performed according to the original exposure data, the original click data and the conversion data, encryption model parameters of the click estimation model and encryption model parameters of the conversion estimation model are obtained, in the process, a first terminal does not know what conversion users are, the encryption model parameters of the click estimation model and the encryption model parameters of the conversion estimation model are sent to a second terminal by the first terminal, decryption is performed by the second terminal, and the first terminal acquires decryption model parameters of the click estimation model and decryption model parameters of the conversion estimation model from the second terminal, so that the click estimation model and the conversion estimation model are obtained, the conversion estimation model capable of effectively predicting conversion estimation probability is obtained, safety in the conversion estimation model training process is further enhanced, and data privacy is protected.
In an optional implementation manner, the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; the first terminal determines a target user of the second terminal according to the click pre-estimation model and the conversion pre-estimation model, and the method comprises the following steps: inputting user characteristics of the user into the click pre-estimation model and the conversion pre-estimation model respectively for any user of the second terminal, and correspondingly obtaining encryption click pre-estimation probability and encryption conversion pre-estimation probability of the user respectively according to the encryption model parameters and the encryption model parameters; the first terminal sends the encryption click estimated probability and the encryption conversion estimated probability to the second terminal; the first terminal obtains the estimated probability of decryption click and the estimated probability of decryption conversion of the user from the second terminal; the decryption click estimated probability and the decryption conversion estimated probability correspond to each other, and the second terminal obtains the encryption click estimated probability and the encryption conversion estimated probability according to a decryption algorithm corresponding to the preset encryption algorithm; and the first terminal determines that the user is a target user of the second terminal according to the decryption click estimated probability and the decryption conversion estimated probability if the conversion probability of the user is determined to be larger than the first preset probability threshold.
In the above manner, machine learning training is performed according to the original exposure data, the original click data and the conversion data to obtain the encryption model parameters of the click prediction model and the encryption model parameters of the conversion prediction model, when a user is predicted, user features of the user are input into the click prediction model and the conversion prediction model respectively, according to the encryption model parameters and the encryption model parameters, the encryption click prediction probability and the encryption conversion prediction probability of the user are obtained correspondingly respectively, and are sent to the second terminal for decryption, and then the decryption click prediction probability and the decryption conversion prediction probability of the user from the second terminal are obtained.
In an optional implementation manner, for any user of the target users of the second terminal, if the user is a target user of multiple terminals at the same time, before the first terminal pushes the information to be recommended to the target user of the second terminal, the method further includes: the first terminal determines conversion prediction costs of the plurality of terminals for the user; the conversion prediction cost characterizes the cost required by the user to execute the preset behavior; the first terminal determines that a conversion predicted cost of the second terminal for the user is lowest among the plurality of terminals.
In the method, when the user is a target user of a plurality of terminals at the same time, the first terminal determines conversion prediction costs of the plurality of terminals for the user, and determines that the conversion prediction costs of the second terminal for the user are the lowest among the plurality of terminals, so that information recommendation is performed through the minimum conversion prediction costs.
After the first terminal determines the target user of the second terminal according to the click pre-estimation model and the conversion pre-estimation model, before the first terminal pushes the information to be recommended to the target user of the second terminal, the method further comprises: the first terminal adjusts the first preset probability threshold value to a second preset probability threshold value if the proportion of the target user of the second terminal in the user of the second terminal is larger than or equal to a preset proportion; and the first terminal re-determines a target user corresponding to the second terminal according to the second preset probability threshold value, wherein the second preset probability threshold value is larger than the first preset probability threshold value.
In the method, when the throwing proportion of the target users corresponding to the second terminal and the users of the second terminal is larger than or equal to the preset proportion, the first preset probability threshold is adjusted to be the second preset probability threshold, and the target users corresponding to the second terminal are determined again according to the second preset probability threshold, so that the number of the target users corresponding to the second terminal is reduced, advertisements are more intensively thrown, and the conversion effect of information recommendation is improved.
In a second aspect, the present application provides an information recommendation method apparatus, including: the acquisition module is used for acquiring conversion data containing user characteristics through the second terminal; the training module is used for performing machine learning training according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model; wherein, the click pre-estimation model and the conversion pre-estimation model are irrelevant to the behavior of any user; the processing module is used for determining a target user of the second terminal according to the click estimated model and the conversion estimated model and pushing information to be recommended to the target user of the second terminal; the target user of the second terminal is a user with the conversion probability larger than a first preset probability threshold value among the users of the second terminal.
In an alternative embodiment, the obtaining module is specifically configured to: sending user identifiers of a plurality of click users in the original click data to the second terminal; acquiring tag values of the plurality of click users from the second terminal; the label values of the plurality of click users are determined by the second terminal according to the user identifications of the plurality of click users; the label value is a click label value or a conversion label value; and according to the user identifications of the plurality of click users, the original click data and the conversion label values corresponding to the plurality of click users are matched to obtain combined data which is used as the conversion data.
In an optional implementation manner, the tag value of any click user of the plurality of click users is an encrypted tag value obtained by encrypting the original tag by the second terminal according to a preset encryption algorithm.
In an alternative embodiment, the training module is specifically configured to: performing machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; sending the encryption model parameters of the click pre-estimation model and the encryption model parameters of the conversion pre-estimation model to the second terminal; obtaining decryption model parameters of the click pre-estimation model and decryption model parameters of the conversion pre-estimation model from the second terminal, so as to obtain the click pre-estimation model and the conversion pre-estimation model; the decryption model parameters of the click pre-estimated model and the decryption model parameters of the conversion pre-estimated model are obtained by the second terminal according to a decryption algorithm corresponding to the preset encryption algorithm.
In an alternative embodiment, the training module is specifically configured to: performing machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; the processing module is specifically configured to: inputting user characteristics of the user into the click pre-estimation model and the conversion pre-estimation model respectively for any user of the second terminal, and correspondingly obtaining encryption click pre-estimation probability and encryption conversion pre-estimation probability of the user respectively according to the encryption model parameters and the encryption model parameters; transmitting the encryption click estimated probability and the encryption conversion estimated probability to the second terminal; obtaining a decryption click estimated probability and a decryption conversion estimated probability of the user from the second terminal; the decryption click estimated probability and the decryption conversion estimated probability correspond to each other, and the second terminal obtains the encryption click estimated probability and the encryption conversion estimated probability according to a decryption algorithm corresponding to the preset encryption algorithm; and according to the decryption click estimated probability and the decryption conversion estimated probability, if the conversion probability of the user is determined to be larger than the first preset probability threshold, determining that the user is a target user of the second terminal.
In an optional implementation manner, for any user of the target users of the second terminal, if the user is a target user of multiple terminals at the same time, the processing module is further configured to: determining conversion prediction costs of the plurality of terminals for the user; the conversion prediction cost characterizes the cost required by the user to execute the preset behavior; determining that a conversion predicted cost of the second terminal for the user is lowest among the plurality of terminals.
In an alternative embodiment, the processing module is further configured to: if the proportion of the target user of the second terminal in the user of the second terminal is larger than or equal to the preset proportion, the first preset probability threshold value is adjusted to be a second preset probability threshold value; and re-determining a target user corresponding to the second terminal according to the second preset probability threshold, wherein the second preset probability threshold is larger than the first preset probability threshold.
The advantages of the second aspect and the embodiments of the second aspect may be referred to the advantages of the first aspect and the embodiments of the first aspect, and will not be described here again.
In a third aspect, embodiments of the present application provide a computer device comprising a program or instructions which, when executed, are adapted to carry out the methods of the first aspect and the embodiments of the first aspect described above.
In a fourth aspect, embodiments of the present application provide a storage medium including a program or instructions, which when executed, are configured to perform the method of the first aspect and the respective embodiments of the first aspect.
Drawings
Fig. 1 is a schematic flow chart of steps of an information recommendation method according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of an information recommendation method device according to an embodiment of the present application.
Detailed Description
In order to better understand the above technical solutions, the following detailed description will be made with reference to the accompanying drawings and specific embodiments of the present application, and it should be understood that specific features in the embodiments and examples of the present application are detailed descriptions of the technical solutions of the present application, and not limiting the technical solutions of the present application, and the technical features in the embodiments and examples of the present application may be combined with each other without conflict.
The internet information recommendation is to deliver information materials which are wanted to be delivered by an information recommendation source to a user side through an information recommendation system, and a read information recommendation delivery mode is shown for the user, so that the information recommendation delivery mode benefits from development of internet big data, user portraits and high-efficiency computing capacity, and the effect after information recommendation is improved.
At present, the demand of an information recommendation source on the release effect data is stronger, and different user behaviors and information recommendation demands in different scenes have a remarkable relationship. Reasonable targeted information recommendation can be aimed at accurate throwing of users. However, in the current information recommendation method, users are not distinguished, and unified delivery is performed blindly, so that the conversion effect of information recommendation is poor.
For this reason, as shown in the step flowchart of fig. 1, the information recommendation method provided by the application is provided.
Step 101: the first terminal obtains conversion data containing user characteristics through the second terminal.
Step 102: and the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model.
Wherein the click prediction model and the conversion prediction model have no correlation with the behavior of any user.
Step 103: and the first terminal determines a target user of the second terminal according to the click estimated model and the conversion estimated model, and pushes information to be recommended to the target user of the second terminal.
The target user of the second terminal is a user with conversion probability larger than a first preset probability threshold value in the users of the second terminal, and the conversion probability of the user is determined according to the click estimated probability and the conversion estimated probability of the user. The click estimated probability is determined according to a click estimated model, and the conversion estimated probability is determined according to the conversion estimated probability.
It should be noted that, for information push, the behavior of the user can be divided into three phases: exposure, clicking and conversion. The exposure finger information is pushed to the user, such as recommendation information is inserted into a webpage of the user; clicking means that the user clicks the pushed recommendation information; the conversion user has performed a conversion action. Therefore, the exposure data is user characteristic data which pushes information but does not generate clicking action; the click data is user characteristic data which clicks push information but does not generate conversion behavior; the conversion data is user characteristic data of which push information is clicked and conversion actions occur. For example, the transformation behavior is a further preset behavior after clicking on registration, purchase, etc. The user must push the information before clicking and must click the pushed information before the conversion action occurs. The converted data clicks the characteristic data of the recommended information user; the original exposure data are characteristic data of users which are put in recommendation information but not clicked; the original click data is feature data of the click user to which recommendation information is put.
Step 101 may specifically be: the first terminal sends user identifiers of a plurality of click users in the original click data to the second terminal; the first terminal obtains tag values of the plurality of click users from the second terminal; the label values of the plurality of click users are determined by the second terminal according to the user identifications of the plurality of click users; the label value is a click label value or a conversion label value; and the first terminal uses the combined data obtained after the original click data are matched with the conversion label values corresponding to the click users as the conversion data according to the user identifications of the click users.
Since the security of the information recommendation source end for converting user data is the primary consideration goal of the scheme. The present application proposes a conversion protection technique to achieve this objective. The conversion protection encrypts click data generated by using certain recommended information after mixing the click data with the conversion data. The technology can make other parties including the information recommendation platform end unable to distinguish data, thereby successfully protecting the converted data. The conversion protection technology consists of two steps of data mixing and data encryption.
Data mixing: the information recommendation source terminal safely acquires the user identification of the clicking user of the recommendation information from the information recommendation platform terminalThen the information recommendation source terminal acquires click user identification +.>And then matching with the conversion user of the information recommendation source terminal. If user i appears as both click user and conversion user, the label corresponding to user representation i is set to 1, namelyIf user i is onlyFor clicking the user, the label corresponding to the user identifier i is 0, namely: />Obviously, a 0 or 1 corresponding to user identification i characterizes whether the user has transformed after clicking the data.
For example, the first terminal sends user identifiers of 10 click users to the second terminal, the second terminal obtains 4 click users as conversion users according to matching of the user identifiers of 10 click users with conversion users in the second terminal, the tag values of the 4 click users are set as conversion tag values, the tag values of the remaining 6 click users are set as click tag values, and the first terminal receives the click identifiers of 10 click users and the tag values, so that original click data can be matched with corresponding tag values according to the click identifiers to serve as conversion data. If the click user A uses the characteristics of the click user A, the characteristics and the label value corresponding to the click user A are used as conversion data of the click user A.
Data encryption: the information recommendation source end can utilize encryption algorithms such as homomorphic encryption algorithm and the like to strictly protect the safety of whether to convert the information. Specifically, the information recommendation source generates a pair of private key sk and public key pk by using homomorphic encryption method. Then, the information recommendation source encrypts the label which corresponds to each user i and is converted or not by using a public key, namely { i: [ [0,1 ]]] pk }. And finally, the information recommendation source terminal submits the encryption tag to the information recommendation platform terminal.
It should be noted that the tag value may be encrypted, and the encryption algorithm is not limited to the same as the encryption algorithm. Therefore, in one implementation, the tag value of any one click user of the plurality of click users is an encrypted tag value obtained by encrypting the original tag by the second terminal according to a preset encryption algorithm. For example, when the encrypted tag value of the user a is L (a), the first terminal obtains the tag value ciphertext, but does not know what the corresponding original text is, and only the second terminal can decrypt the encrypted tag value.
For example, a user is identified as X 1 The method comprises the steps of carrying out a first treatment on the surface of the In the above alternative embodiment, the information recommendation platform sends the end-to-end messageMessage recommendation source sends X 1 . Information recommendation source terminal receives X 1 After that, the local conversion data of the information recommendation source end can be searched whether X exists or not 1 . If present, X is then 1 The tag value of (2) is set to the conversion tag value, otherwise X is set to 1 Is set to the click data tag value. Then the information recommendation source terminal sends X 1 Is encrypted to obtain an encrypted tag value [ X ] 1 ] PK And (2) (3) is performed. In (3), the information recommendation platform side obtains X according to the obtained 1 、[X 1 ]pk, will [ X ] 1 ]pk as X in information recommendation platform end 1 Is a label value of (a).
In step 102, the information recommendation platform may obtain the transformation model according to the following embodiments, including the following first to third steps:
the first step: and the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model.
And a second step of: and the first terminal sends the encryption model parameters of the click pre-estimated model and the encryption model parameters of the conversion pre-estimated model to the second terminal.
And a third step of: the first terminal obtains decryption model parameters of the click pre-estimation model and decryption model parameters of the conversion pre-estimation model from the second terminal, so that the click pre-estimation model and the conversion pre-estimation model are obtained.
In the first step, the original exposure data, the original click data and the conversion data may be converted by a feature expression model and then subjected to machine learning training. Specifically:
the first terminal obtains encrypted feature expression data according to the original exposure data, the original click data and the conversion data and the feature expression model; the feature expression model is used to distinguish the exposure data from the encrypted mix data. And the first terminal obtains a conversion estimation model according to the encrypted characteristic expression data.
The characteristic expression model is trained based on a minimized preset loss function according to the exposure data and the encrypted mixed data; the preset loss function characterizes the degree of aggregation between exposure data and the degree of aggregation between click data and conversion data.
In the first to third steps, the decryption model parameters of the click pre-estimation model and the decryption model parameters of the conversion pre-estimation model are obtained by the second terminal according to a decryption algorithm corresponding to the preset encryption algorithm. It should be noted that, the training algorithm of the click estimation model and the conversion estimation model is not limited, and may be a gradient descent algorithm. In the implementation mode, the first terminal directly acquires the conversion pre-estimation model and the click pre-estimation model, and when the user needs to predict in the follow-up process, the first terminal calculates the conversion probability of the user by the decryption model parameters of the click pre-estimation model and the decryption model parameters of the conversion pre-estimation model.
An alternative embodiment of steps 102 to 103 is:
the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; inputting user characteristics of the user into the click pre-estimation model and the conversion pre-estimation model respectively for any user of the second terminal, and correspondingly obtaining encryption click pre-estimation probability and encryption conversion pre-estimation probability of the user respectively according to the encryption model parameters and the encryption model parameters; the first terminal sends the encryption click estimated probability and the encryption conversion estimated probability to the second terminal; the first terminal obtains the estimated probability of decryption click and the estimated probability of decryption conversion of the user from the second terminal; and the first terminal determines that the user is a target user of the second terminal according to the decryption click estimated probability and the decryption conversion estimated probability if the conversion probability of the user is determined to be larger than the first preset probability threshold.
In the foregoing embodiment, the decoding click estimated probability and the decoding conversion estimated probability correspond to each other, where the decoding click estimated probability and the decoding conversion estimated probability are obtained by the second terminal according to a decoding algorithm corresponding to the preset decoding algorithm. In the implementation mode, the first terminal does not directly acquire the conversion estimation model and the click estimation model, and then calculates the encryption click estimation probability and the encryption conversion estimation probability of the user according to the encryption model parameters of the click estimation model and the encryption model parameters of the conversion estimation model every time the user needs to be predicted, then decrypts the encrypted click estimation probability and the decryption conversion estimation probability into the decryption click estimation probability and the decryption conversion estimation probability by the second terminal, and further judges whether the user is a target user or not.
In an optional implementation manner of step 103, for any user to be evaluated of the second terminal, the first terminal may determine whether the user to be evaluated is a target user according to the click prediction model and the conversion prediction model. Specifically:
the first terminal acquires first user characteristics of a user to be evaluated; and the first terminal inputs the first user characteristics of the user to be evaluated into the conversion estimation model, and obtains the conversion estimation probability of the user to be evaluated based on the decrypted model parameters.
The conversion estimated probability of the user to be estimated is directly obtained according to the conversion estimated model obtained through training. The user to be evaluated is a new user which does not participate in conversion estimation model training, and the first user characteristics are characteristic information of the user to be evaluated. In particular, the characteristic information may include at least one of: the attribute characteristics of the user, the information characteristics of the information recommended by the user and the scene characteristics of the scene recommended by the user refer to basic attributes of the user, such as age, sex and the like, the information characteristics refer to basic attributes of the recommended information, such as format and layout of the recommended information, and the scene characteristics refer to the scene of recommending the recommended resources to the user, such as recommended places and user operations triggering the recommendation. In the following description, reference is also made to the second user feature and the third user feature, which are feature information containing at least one of the above items, but are different feature information from the first user feature.
In another embodiment of step 103, the trained transformation estimation model may include a first reference feature of the click data and a second reference feature of the transformation data. The average level of the features of the click data, which do not undergo conversion behavior, and the data of the consumer user, which are recommended by the click data, which are characterized by the second reference features, can be determined according to the following manner:
The first terminal acquires second user characteristics of a user to be evaluated; and the first terminal determines the conversion estimated probability of the user to be evaluated according to the relation between the second user characteristic of the user to be evaluated and the first reference characteristic and the second reference characteristic. In this way, the conversion estimated probability of the user to be evaluated can be determined by only performing simple operation, so that the efficiency of the conversion estimated probability of the user to be evaluated is improved.
Specifically, the method comprises the following steps:
and (2) the first terminal converts the second user characteristic into a third user characteristic through the characteristic expression model in the step (1).
Step (2) the first terminal determines a first euclidean distance of the third user feature from the first reference feature.
And (3) the first terminal is based on a second Euclidean distance between the third user characteristic and the second reference characteristic.
And (4) the first terminal determines the conversion estimated probability of the user to be evaluated according to the distance difference value of the first Euclidean distance and the second Euclidean distance.
It should be noted that, the first euclidean distance in the step (4) represents the similarity between the third user feature and the first reference feature, the second euclidean distance represents the similarity between the third user feature and the second reference feature, and the difference value can obtain whether the similarity between the third user feature and the conversion data is higher or the similarity between the third user feature and the click data is higher.
An alternative embodiment that may also be taken in step 103 is as follows:
the first terminal adjusts the first preset probability threshold value to a second preset probability threshold value if the proportion of the target user of the second terminal in the user of the second terminal is larger than or equal to a preset proportion; and the first terminal re-determines a target user corresponding to the second terminal according to the second preset probability threshold value, wherein the second preset probability threshold value is larger than the first preset probability threshold value.
The above-mentioned alternative embodiment is applicable to the situation that the number of target users whose conversion probability is larger than the first preset probability threshold is large. (namely, the throwing proportion of the target users corresponding to the first terminal in the plurality of target users is larger than or equal to a preset proportion). For example, the first preset probability threshold is 40%, for a total user of a certain first terminal, the user with a conversion probability of 40% accounts for 90% of the total user of the first terminal, but the user with a conversion probability of 80% accounts for only 30% of the total user of the first terminal. Obviously, users with the conversion probability of 80% are more accurate users for the first terminal, and the advertisement cost of each user is the same for the first terminal, so that the users with the conversion probability of 40% are more, 90% are obvious, the average exertion is not cost-effective for an information recommendation platform, and a strategy of 'grasp magnitude amplification' can be adopted to focus on the users with the conversion probability of 80%. Therefore, the preset probability threshold value can be correspondingly adjusted, for example, the second preset probability threshold value is 8%, the number of target users meeting the condition that the conversion probability is larger than the first preset probability threshold value is reduced, the number of target users is reduced, and better effect can be achieved with lower cost.
In the method from step 101 to step 103, if the competition relationship among multiple information recommendation sources of the same user is not involved, for each user, if multiple information recommendation sources want to deliver recommendation information of different product types to the same user (i.e. there is no competition relationship among the information recommendation sources), information can be pushed only by the information recommendation sources whose conversion probability reaches a preset probability threshold.
The following describes the construction of the transformation estimation model in step 103 in detail according to an embodiment:
the information recommendation platform end obtains encrypted conversion labels corresponding to all clicking users from the information recommendation source end, and then the triple data of the user labels, whether to convert and the user characteristics, namely the mixed encrypted data, can be obtained through data matching with the information recommendation platform. For example, for user i, the triplet data isIn addition, the information recommendation platform side has unencrypted exposure data +.>Click data->And the user characteristics corresponding to the exposure data and the click data respectively.
The information recommendation platform end establishes a conversion prediction model f (x|theta) by combining exposure data, click data and encrypted conversion data. Given the user characteristics x, f (x|θ) output the estimated conversion. Specifically, first, the conversion estimation model uses exposure data Click data->And feature expression data of transformation data learning data +.>And->It should be noted thatThe exposure data and the click data are directly acquired by the information recommendation platform end, and the information recommendation platform end can train the feature expression model by using the neural network. The neural network learns the feature expression function h (x) →e by minimizing the following preset loss function (1). The loss function attempts to provide the feature expression model with the ability to distinguish exposure data from click data and conversion data. Notably, the feature expression model can use massive exposure data, click data and conversion data to improve generalization capability.
Based on the method, a new triplet of clicking users and conversion users can be obtainedWherein [ (0, 1)]] pk Is an encryption tag indicating whether the user is converted or not, +.>Is the characteristic expression data corresponding to the conversion data,is the feature expression data corresponding to the click data. In addition, click data is distinguished from conversion data for better. The following function (2) can be established in the present application:
wherein the method comprises the steps ofCharacteristic expression data for evaluating a new user to be evaluated for characteristic expression e corresponding to transformation data +.>Similarity of (3)，/>Evaluating in an encryption space the characteristic expression data of the new user to be evaluated, characteristic expression e corresponding to click data +. >Is a similarity of (3). Specifically, & gt>A first reference feature can be used, by means of which the evaluation is made, then +.> Is a first Euclidean distance; />A second reference feature can be used, by means of which the evaluation is made, then +.>Is the second euclidean distance. Will pass-> To the parameter [ [ theta ]]] pk Training is carried out, and finally the conversion rate estimation function is obtained. Since the function structure is fixed when the conversion rate estimation function is constructed, only the parameter [ [ theta ] is]] pk Unknown, final conversion rate estimation function is estimated by using encrypted parameter [ [ theta ]]] pk In the form of a gel.
Then, the information recommendation source terminal firstly obtains encrypted conversion estimated model parameters [ [ theta ] from the information recommendation platform terminal]] pk . To enable real-time conversion prediction on the line, i.eThe information recommendation source terminal can obtain the estimated conversion rate in real time, and then the information recommendation source terminal decrypts the conversion estimated parameter by using the private key sk and has the decrypted theta. And finally, the information recommendation source terminal submits the decrypted conversion estimated parameters to the information recommendation platform terminal.
The information recommendation platform end obtains the conversion estimated model parameters theta decrypted by the information recommendation source end. When any user and the information recommendation source end request, the information recommendation platform end obtains the estimated conversion rate of the user based on the feature expression and the conversion estimated model parameter theta.
In the steps 101 to 103, a first terminal is an information recommendation platform, a second terminal is an information recommendation source, and the steps 101 to 103 are applicable to scenes in which a plurality of information recommendation source terminals entrust information recommendation platform terminals to recommend information to a plurality of users. The plurality of information recommendation sources all have respective users, and the same user or different users may exist between different information recommendation sources. The user of each information recommendation platform has conversion probability for the information recommendation platform. It should be noted that, for the same user, users who may belong to multiple information recommendation sources at the same time have corresponding conversion probabilities corresponding to the multiple information recommendation sources. Thus, one user may be the target user of information recommendation source one, but not the target user of information recommendation source two. From the point of view of the information recommendation source, if a target user exists in the information recommendation source, the delivery can be considered. If one target user is the same as the target user belonging to a plurality of information recommendation source ends, and the products corresponding to the information to be pushed by the information recommendation source ends are the same type of products, the information recommendation source ends are provided with the same type of products.
The information recommendation platform end can also determine the cost of information pushing in real time according to the estimated conversion rate. For example, after the feature x of the user i arrives, the information recommendation source request x is converted into e through the feature expression model x,i And then e x,i Inputting the information into a conversion prediction model to obtain conversion prediction probability P (e) corresponding to the information recommendation source terminal request x,i ). Can be obtained according to P (e) in the following manner x,i ) Determining whether to deliver and push informationThe cost of (2):
setting a probability threshold, such as 0.6, when P (e x,i ) When the probability threshold value is greater than or equal to 0.6, recommendation information is put in to the user i; otherwise, recommendation information is not put in to the user i. According to P (e) x,i ) Determining the cost of information push, the cost of information push for user i needs to be equal to P (e x,i ) And in positive correlation, namely, the user with higher estimated conversion probability is more worth throwing. In addition, when P (e x,i ) When the probability threshold value is smaller than or equal to 0.6, the cost of information push of the user i can be set to be 0.
When a competition relationship occurs among the plurality of information recommendation source ends, in step 103, the information recommendation platform end may perform delivery on the target user in the following manner:
that is, for any user of the target users of the second terminal, if the user is a target user of a plurality of terminals at the same time, before the first terminal pushes the information to be recommended to the target user of the second terminal, the following steps may be further performed:
The first terminal determines conversion prediction costs of the plurality of terminals for the user; the conversion prediction cost characterizes the cost required by the user to execute the preset behavior; the first terminal determines that a conversion predicted cost of the second terminal for the user is lowest among the plurality of terminals.
Specifically, for pushing information to be pushed of each information recommendation source end in the plurality of information recommendation source ends, the information recommendation platform end determines conversion prediction cost of each information recommendation source end for the target user; the conversion prediction cost is determined according to the competition cost of the information recommendation source terminal for the information to be pushed and the conversion probability of the target user for the information to be pushed; the conversion forecast cost characterizes a competing cost required by the target user to convert to a conversion user; the information recommendation platform pushes information to be recommended of a target information recommendation source end and puts the information to be recommended to the target user; the target information recommendation source terminal is the unique information recommendation source terminal with the lowest conversion prediction cost in the information recommendation source terminals. If a plurality of information recommendation source ends with the lowest conversion prediction cost exist, one information recommendation source end can be randomly selected from the information recommendation source ends to serve as target information recommendation source ends. And one information recommendation source terminal can be selected from the information recommendation source terminals as target information recommendation source terminals through other rules.
For example, each information recommendation source pays a fee for pushing information to the information recommendation platform, and the fee is used for the information recommendation platform to deliver users of the information recommendation sources. In general, the quotient obtained by dividing the information push fee paid by the information recommendation source by the total number of target users of the information recommendation source is the competition cost of information recommendation set by the information recommendation source for each user. For example, the information push fee of the information recommendation source is 10 ten thousand yuan, and the target users of the information recommendation source are 1 ten thousand, so that the competition cost of each target user is 10 yuan. The users may be classified according to attribute information of the users, and different competing costs may be set for different users, and the specific competing cost setting method is not limited herein. The conversion prediction cost K0 may be set as a quotient of a competition cost (hereinafter referred to as C0) of the information recommendation source for the to-be-pushed information and a conversion probability (P0) of the target user for the to-be-pushed information. I.e. k0=c0/P0. K0 means that an expected value of the information push fee input by the consumer is generated according to the conversion probability.
The above bidding process is illustrated below by way of example with respect to a user a.
For example, there are 5 information recommendation sources, i.e., an information recommendation source one, an information recommendation source two, an information recommendation source three, an information recommendation source four, and an information recommendation source five. The user A belongs to target users of the information recommendation source end I, the information recommendation source end II and the information recommendation source end III, but does not belong to target users of the information recommendation source end IV and the information recommendation source end five. Therefore, the information recommendation source terminal four and the information recommendation source terminal five are not competitive. The competition costs C1, C2 and C3 set for the user A by the information recommendation source end I, the information recommendation source end II and the information recommendation source end III are respectively 10 yuan, 8 yuan and 6 yuan correspondingly. The conversion probabilities P1, P2 and P3 of the user A corresponding to the information recommendation source end I, the information recommendation source end II and the information recommendation source end III are respectively 0.8, 0.5 and 0.6.
The conversion prediction cost K0 is exemplified by the quotient of C0 and P0. The conversion prediction costs K1, K2 and K3 of the information recommendation source end I, the information recommendation source end II and the information recommendation source end III are respectively 12.5 yuan, 16 yuan and 10 yuan. And recommending the information to be delivered of the information recommendation source terminal III to the user A.
As shown in fig. 2, the present application provides an information recommendation method apparatus, including: an obtaining module 201, configured to obtain, by using a second terminal, conversion data including a user feature; the training module 202 is configured to perform machine learning training according to the original exposure data, the original click data, and the transformation data, so as to obtain a click estimation model and a transformation estimation model; wherein, the click pre-estimation model and the conversion pre-estimation model are irrelevant to the behavior of any user; the processing module 203 is configured to determine a target user of the second terminal according to the click estimation model and the conversion estimation model, and push information to be recommended to the target user of the second terminal; the target user of the second terminal is a user with the conversion probability larger than a first preset probability threshold value among the users of the second terminal.
In an alternative embodiment, the obtaining module 201 is specifically configured to: sending user identifiers of a plurality of click users in the original click data to the second terminal; acquiring tag values of the plurality of click users from the second terminal; the label values of the plurality of click users are determined by the second terminal according to the user identifications of the plurality of click users; the label value is a click label value or a conversion label value; and according to the user identifications of the plurality of click users, the original click data and the conversion label values corresponding to the plurality of click users are matched to obtain combined data which is used as the conversion data.
In an optional implementation manner, the tag value of any click user of the plurality of click users is an encrypted tag value obtained by encrypting the original tag by the second terminal according to a preset encryption algorithm.
In an alternative embodiment, the training module 202 is specifically configured to: performing machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; sending the encryption model parameters of the click pre-estimation model and the encryption model parameters of the conversion pre-estimation model to the second terminal; obtaining decryption model parameters of the click pre-estimation model and decryption model parameters of the conversion pre-estimation model from the second terminal, so as to obtain the click pre-estimation model and the conversion pre-estimation model; the decryption model parameters of the click pre-estimated model and the decryption model parameters of the conversion pre-estimated model are obtained by the second terminal according to a decryption algorithm corresponding to the preset encryption algorithm.
In an alternative embodiment, the training module 202 is specifically configured to: performing machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model; the processing module 203 is specifically configured to: inputting user characteristics of the user into the click pre-estimation model and the conversion pre-estimation model respectively for any user of the second terminal, and correspondingly obtaining encryption click pre-estimation probability and encryption conversion pre-estimation probability of the user respectively according to the encryption model parameters and the encryption model parameters; transmitting the encryption click estimated probability and the encryption conversion estimated probability to the second terminal; obtaining a decryption click estimated probability and a decryption conversion estimated probability of the user from the second terminal; the decryption click estimated probability and the decryption conversion estimated probability correspond to each other, and the second terminal obtains the encryption click estimated probability and the encryption conversion estimated probability according to a decryption algorithm corresponding to the preset encryption algorithm; and according to the decryption click estimated probability and the decryption conversion estimated probability, if the conversion probability of the user is determined to be larger than the first preset probability threshold, determining that the user is a target user of the second terminal.
In an alternative embodiment, for any user of the target users of the second terminal, if the user is a target user of multiple terminals at the same time, the processing module 203 is further configured to: determining conversion prediction costs of the plurality of terminals for the user; the conversion prediction cost characterizes the cost required by the user to execute the preset behavior; determining that a conversion predicted cost of the second terminal for the user is lowest among the plurality of terminals.
In an alternative embodiment, the processing module 203 is further configured to: if the proportion of the target user of the second terminal in the user of the second terminal is larger than or equal to the preset proportion, the first preset probability threshold value is adjusted to be a second preset probability threshold value; and re-determining a target user corresponding to the second terminal according to the second preset probability threshold, wherein the second preset probability threshold is larger than the first preset probability threshold.
The embodiment of the application provides a computer device, which comprises a program or an instruction, and the program or the instruction is used for executing an information recommendation method and any optional method provided by the embodiment of the application when being executed.
The embodiment of the application provides a storage medium comprising a program or an instruction, which when executed, is used for executing an information recommendation method and any optional method provided by the embodiment of the application.
Finally, it should be noted that: it will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the scope of the application. Thus, it is intended that the present application also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (10)
1. An information recommendation method, comprising:
the first terminal obtains conversion data containing user characteristics through the second terminal;
the first terminal carries out machine learning training after converting the characteristic expression model according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model; wherein, the click pre-estimation model and the conversion pre-estimation model are irrelevant to the behavior of any user; the feature expression model is obtained by training based on a minimized preset loss function according to the original exposure data and the encrypted mixed data; the preset loss function characterizes the degree of aggregation between the original exposure data and the degree of aggregation between the original click data and the conversion data; the encrypted mixed data comprises a user tag, whether to convert or not, and a user characteristic;
The first terminal determines a target user of the second terminal according to the click estimated model and the conversion estimated model, and pushes information to be recommended to the target user of the second terminal; the target user of the second terminal is a user with the conversion probability larger than a first preset probability threshold value among the users of the second terminal.
2. The method of claim 1, wherein the first terminal obtains the conversion data including the user characteristics through the second terminal, comprising:
the first terminal sends user identifiers of a plurality of click users in the original click data to the second terminal;
the first terminal obtains tag values of the plurality of click users from the second terminal; the label values of the plurality of click users are determined by the second terminal according to the user identifications of the plurality of click users; the label value is a click label value or a conversion label value;
and the first terminal uses the combined data obtained after the original click data are matched with the conversion label values corresponding to the click users as the conversion data according to the user identifications of the click users.
3. The method of claim 2, wherein the tag value of any one of the plurality of click users is an encrypted tag value obtained by encrypting the original tag by the second terminal according to a preset encryption algorithm.
4. The method of claim 3, wherein the first terminal performs machine learning training according to the original exposure data, the original click data, and the conversion data to obtain a click prediction model and a conversion prediction model, comprising:
the first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model;
the first terminal sends the encryption model parameters of the click pre-estimated model and the encryption model parameters of the conversion pre-estimated model to the second terminal;
the first terminal obtains decryption model parameters of the click pre-estimation model and decryption model parameters of the conversion pre-estimation model from the second terminal, so that the click pre-estimation model and the conversion pre-estimation model are obtained; the decryption model parameters of the click pre-estimated model and the decryption model parameters of the conversion pre-estimated model are obtained by the second terminal according to a decryption algorithm corresponding to the preset encryption algorithm.
5. The method of claim 3, wherein the first terminal performs machine learning training according to the original exposure data, the original click data, and the conversion data to obtain a click prediction model and a conversion prediction model, comprising:
The first terminal performs machine learning training according to the original exposure data, the original click data and the conversion data to obtain encryption model parameters of the click prediction model and encryption model parameters of the conversion prediction model;
the first terminal determines a target user of the second terminal according to the click pre-estimation model and the conversion pre-estimation model, and the method comprises the following steps:
inputting user characteristics of the user into the click pre-estimation model and the conversion pre-estimation model respectively for any user of the second terminal, and correspondingly obtaining encryption click pre-estimation probability and encryption conversion pre-estimation probability of the user respectively according to the encryption model parameters and the encryption model parameters;
the first terminal sends the encryption click estimated probability and the encryption conversion estimated probability to the second terminal;
the first terminal obtains the estimated probability of decryption click and the estimated probability of decryption conversion of the user from the second terminal; the decryption click estimated probability and the decryption conversion estimated probability correspond to each other, and the second terminal obtains the encryption click estimated probability and the encryption conversion estimated probability according to a decryption algorithm corresponding to the preset encryption algorithm;
And the first terminal determines that the user is a target user of the second terminal according to the decryption click estimated probability and the decryption conversion estimated probability if the conversion probability of the user is determined to be larger than the first preset probability threshold.
6. The method according to any one of claims 1-5, wherein for any one of the target users of the second terminal, before the first terminal pushes the information to be recommended to the target user of the second terminal if the user is a target user of a plurality of terminals at the same time, the method further comprises:
the first terminal determines conversion prediction costs of the plurality of terminals for the user; the conversion prediction cost characterizes the cost required by the user to execute the preset behavior;
the first terminal determines that a conversion predicted cost of the second terminal for the user is lowest among the plurality of terminals.
7. The method of any of claims 1-5, wherein after the first terminal determines the target user of the second terminal according to the click prediction model and the conversion prediction model, before the first terminal pushes the information to be recommended to the target user of the second terminal, the method further comprises:
The first terminal adjusts the first preset probability threshold value to a second preset probability threshold value if the proportion of the target user of the second terminal in the user of the second terminal is larger than or equal to a preset proportion;
and the first terminal re-determines a target user corresponding to the second terminal according to the second preset probability threshold value, wherein the second preset probability threshold value is larger than the first preset probability threshold value.
8. An information recommendation device, characterized by comprising:
the acquisition module is used for acquiring conversion data containing user characteristics through the second terminal;
the training module is used for carrying out machine learning training after converting the characteristic expression model according to the original exposure data, the original click data and the conversion data to obtain a click estimation model and a conversion estimation model; wherein, the click pre-estimation model and the conversion pre-estimation model are irrelevant to the behavior of any user; the feature expression model is obtained by training based on a minimized preset loss function according to the original exposure data and the encrypted mixed data; the preset loss function characterizes the degree of aggregation between the original exposure data and the degree of aggregation between the original click data and the conversion data; the encrypted mixed data comprises a user tag, whether to convert or not, and a user characteristic;
The processing module is used for determining a target user of the second terminal according to the click estimated model and the conversion estimated model and pushing information to be recommended to the target user of the second terminal; the target user of the second terminal is a user with the conversion probability larger than a first preset probability threshold value among the users of the second terminal.
9. A computer device comprising a program or instructions which, when executed, performs the method of any of claims 1 to 7.
10. A storage medium comprising a program or instructions which, when executed, perform the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911329851.XA CN111160961B (en) | 2019-12-20 | 2019-12-20 | Information recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911329851.XA CN111160961B (en) | 2019-12-20 | 2019-12-20 | Information recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111160961A CN111160961A (en) | 2020-05-15 |
| CN111160961B true CN111160961B (en) | 2023-10-17 |
Family
ID=70557704
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911329851.XA Active CN111160961B (en) | 2019-12-20 | 2019-12-20 | Information recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111160961B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111931049B (en) * | 2020-08-02 | 2021-04-30 | 深圳前海大道金融服务有限公司 | Business processing method based on big data and artificial intelligence and block chain financial system |
| CN112579910A (en) * | 2020-12-28 | 2021-03-30 | 北京嘀嘀无限科技发展有限公司 | Information processing method, information processing apparatus, storage medium, and electronic device |
| CN113761348A (en) * | 2021-02-26 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Information recommendation method and device, electronic equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108259158A (en) * | 2018-01-11 | 2018-07-06 | 西安电子科技大学 | Efficient and secret protection individual layer perceptron learning method under a kind of cloud computing environment |
| CN109635462A (en) * | 2018-12-17 | 2019-04-16 | 深圳前海微众银行股份有限公司 | Model parameter training method, device, equipment and medium based on federation's study |
| WO2019076173A1 (en) * | 2017-10-20 | 2019-04-25 | 广州优视网络科技有限公司 | Content pushing method and apparatus, and computer device |
| CN110046952A (en) * | 2019-01-30 | 2019-07-23 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| CN110276210A (en) * | 2019-06-12 | 2019-09-24 | 深圳前海微众银行股份有限公司 | Based on the determination method and device of the model parameter of federation's study |
| CN110413877A (en) * | 2019-07-02 | 2019-11-05 | 阿里巴巴集团控股有限公司 | A kind of resource recommendation method, device and electronic equipment |
| CN110415032A (en) * | 2019-07-24 | 2019-11-05 | 深圳乐信软件技术有限公司 | A kind of exposure conversion ratio predictor method, device, equipment and storage medium |
| CN110570232A (en) * | 2019-08-05 | 2019-12-13 | 科大讯飞股份有限公司 | Internet advertisement putting method, device, server and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10037543B2 (en) * | 2012-08-13 | 2018-07-31 | Amobee, Inc. | Estimating conversion rate in display advertising from past performance data |
-
2019
- 2019-12-20 CN CN201911329851.XA patent/CN111160961B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019076173A1 (en) * | 2017-10-20 | 2019-04-25 | 广州优视网络科技有限公司 | Content pushing method and apparatus, and computer device |
| CN108259158A (en) * | 2018-01-11 | 2018-07-06 | 西安电子科技大学 | Efficient and secret protection individual layer perceptron learning method under a kind of cloud computing environment |
| CN109635462A (en) * | 2018-12-17 | 2019-04-16 | 深圳前海微众银行股份有限公司 | Model parameter training method, device, equipment and medium based on federation's study |
| CN110046952A (en) * | 2019-01-30 | 2019-07-23 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| CN110276210A (en) * | 2019-06-12 | 2019-09-24 | 深圳前海微众银行股份有限公司 | Based on the determination method and device of the model parameter of federation's study |
| CN110413877A (en) * | 2019-07-02 | 2019-11-05 | 阿里巴巴集团控股有限公司 | A kind of resource recommendation method, device and electronic equipment |
| CN110415032A (en) * | 2019-07-24 | 2019-11-05 | 深圳乐信软件技术有限公司 | A kind of exposure conversion ratio predictor method, device, equipment and storage medium |
| CN110570232A (en) * | 2019-08-05 | 2019-12-13 | 科大讯飞股份有限公司 | Internet advertisement putting method, device, server and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111160961A (en) | 2020-05-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xiong et al. | A task-oriented user selection incentive mechanism in edge-aided mobile crowdsensing | |
| CN111160961B (en) | Information recommendation method and device | |
| CN109033865A (en) | The method for allocating tasks of secret protection in a kind of space crowdsourcing | |
| CN113095510A (en) | Block chain-based federal learning method and device | |
| CN108932410A (en) | A kind of digital resource copyright method for authenticating, device, system and storage medium | |
| CN111027981B (en) | Method and device for multi-party joint training of risk assessment model for IoT (Internet of things) machine | |
| Zhai et al. | Towards secure and truthful task assignment in spatial crowdsourcing | |
| CN113449048B (en) | Data label distribution determining method and device, computer equipment and storage medium | |
| CN108667717A (en) | Block chain processing method, medium, device and computing device based on instant communication message record | |
| Sun et al. | Privacy-preserving verifiable incentive mechanism for online crowdsourcing markets | |
| CN112700281A (en) | Behavior prediction method, behavior prediction device, behavior prediction equipment and computer readable storage medium | |
| CN104765884B (en) | A kind of fingerprint identification method of HTTPS webpages | |
| Xu et al. | Knowledge as a service and knowledge breaching | |
| CN116383865A (en) | Federal learning prediction stage privacy protection method and system | |
| Kou et al. | Exgfair: A crowdsourcing data exchange approach to fair human face datasets augmentation | |
| CN113515778A (en) | Multi-party privacy data fusion method and device and electronic equipment | |
| WO2021120227A1 (en) | Method for information recommendation and device | |
| CN112948883A (en) | Multi-party combined modeling method, device and system for protecting private data | |
| CN113285798B (en) | Data processing method, bidding terminal equipment and computer readable medium | |
| CN111159761B (en) | Model training method and device | |
| US20190004999A1 (en) | Information processing device, information processing system, and information processing method, and program | |
| CN114610994A (en) | Push method and system based on joint prediction | |
| CN111988129B (en) | Influence maximization data set processing method, device and system | |
| CN115525907A (en) | Vector similarity judgment method under malicious adversary model and communication equipment | |
| CN111340453B (en) | Federal learning development method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |