WO2018130201A1

WO2018130201A1 - Method for determining associated account, server and storage medium

Info

Publication number: WO2018130201A1
Application number: PCT/CN2018/072381
Authority: WO
Inventors: 戴智君; 谢毅
Original assignee: 腾讯科技（深圳）有限公司
Priority date: 2017-01-16
Filing date: 2018-01-12
Publication date: 2018-07-19
Also published as: CN108322317A; CN108322317B

Abstract

Disclosed in the embodiments of the present application is a method for determining an associated account, the method being applied to a server. The method comprises: acquiring, from a terminal device, usage data concerning one or more users using the terminal device by means of their respective first user accounts, the usage data of each first user account comprising: the identifier of the terminal device, the first user account, and the usage records of the first user account; determining, according to the usage records corresponding to each first user account, the usage parameters of at least two dimensions of each first user account; calculate, using the usage parameters of at least two dimensions of each first user account and a preset association recommendation model, the first score corresponding to each first user account; and determining, according to the first score corresponding to each first user account, the first account associated with the terminal device. Further provided in the embodiments of the present application are a corresponding server and storage medium.

Description

Method, server and storage medium for determining associated accounts

This application claims the priority of the Chinese Patent Application filed on Jan. 16, 2017, the application number of which is hereby incorporated by reference. in.

Technical field

The present application relates to Internet technologies in the field of communications, and in particular, to a method, server and storage medium for determining associated accounts.

Background technique

With the rapid development of communication technology, the functions and intelligence of the client are becoming more and more abundant. When a user performs a business application through a client, such as a video application played by a client, a service request such as a request to play a video or a request to download a video, the server collects these service requests and determines the user according to the service request. Confidence, intelligently recommend relevant media information (such as a variety video or TV drama video) or user information to the user through confidence, etc., thereby facilitating the user to select more relevant media information of interest or association with the same hobby The user's information, for example, when the user watches the video, the video client can play the video according to the type of the video selected by the user, and the server intelligently recommends the video associated with the video to the user according to the type of the video played by the video client, which is convenient for the user to select. Therefore, how to dig out a friend or item with the same interest from a large variety of item big data is a problem that needs attention.

Technical content

The application example provides a method for determining an associated account, which is applied to a server, and the method includes:

Obtaining, by the terminal device, the usage data when the user uses the terminal device by using the corresponding first user account, the usage data of each first user account includes: the identifier of the terminal device, the first a user account and a usage record of the first user account;

Determining, according to the usage record corresponding to each first user account, usage parameters of at least two dimensions of each first user account;

Calculating, by using the usage parameter of the at least two dimensions of each first user account and the preset association recommendation model, a first score corresponding to each first user account;

Determining, according to the first score corresponding to each first user account, a first account associated with the terminal device.

The application examples also provide a server comprising one or more processors and one or more memories, the one or more memories comprising computer readable instructions configured to be executed by the one or more processors achieve:

Obtaining usage data when the one or more users use the terminal device by using the corresponding first user account, the usage data of each first user account includes: an identifier of the terminal device, the first a user account and a usage record of the first user account;

The present application examples also provide a non-transitory computer readable storage medium storing computer readable instructions that cause at least one processor to perform the method as described above.

DRAWINGS

1 is a schematic diagram of various hardware entities in a data processing system in an example of the present application;

2A is a schematic flowchart of an account identification association method provided by an example of the present application;

2B is a schematic diagram 1 of a framework for an account identification association method provided by an example of the present application;

3 is a schematic diagram 2 of a framework for an account identification association method provided by an example of the present application;

4 is a schematic diagram 3 of a framework for an account identification association method provided by an example of the present application;

FIG. 5 is a diagram 1 showing an exemplary relationship between an exemplary terminal and a user account provided by an example of the present application; FIG.

6 is a diagram 2 showing an exemplary relationship between an exemplary terminal and a user account provided by an example of the present application;

FIG. 7 is a schematic diagram 4 of a framework for an account identification association method provided by an example of the present application; FIG.

FIG. 8 is a schematic structural diagram 1 of a server provided by an example of the present application; FIG.

FIG. 9 is a schematic structural diagram 2 of a server provided by an example of the present application;

FIG. 10 is a schematic structural diagram 3 of a server provided by an example of the present application.

detailed description

The technical solutions in the examples of the present application will be clearly and completely described below in conjunction with the drawings in the examples of the present application.

Terminal device: refers to mobile electronic devices, also known as mobile devices (mobile devices), mobile devices, handheld devices, wearable devices, etc., is an embedded chip-based computing device, usually has a Small display, touch input, or a small keyboard.

Machine learning: relying on theories of probability, statistics, and neural communication to enable computers to simulate human learning behaviors to acquire new knowledge or skills, and to reorganize existing knowledge structures to continuously improve their performance.

Model training: The manually selected samples are input to the machine learning system, and the accuracy of the sample identification is optimized by continuously adjusting the model parameters.

The International Mobile Equipment Identity (IMEI) is the unique identification number of the mobile phone.

RFM model: In RFM mode, R (Recency) indicates how far the customer last used, F (Frequency) indicates the number of times the customer has used in the most recent period of time, and M (Monetary) indicates that the customer has used it in the most recent period of time. The amount. R used in this scheme indicates the reporting time, F indicates the reporting frequency, and M indicates the reporting source.

IMEI-User Account: A relationship data between a terminal (IMEI) and a user account.

FIG. 1 is a schematic diagram of various hardware entities in an architecture of a data processing system in the example of the present application. FIG. 1 includes: one or more servers 1, terminal devices 21-25, and network 3, in network 3. Network entities including routers, gateways, etc., are not shown in Figure 1. The terminal device 21-25 performs service product information interaction with the server 1 through a wired network or a wireless network, so as to acquire time-related data generated by the user's use of the terminal device from the terminal 21-25, and transmit the acquired data to the server. 1. The data is usage data generated by the user using the terminal device. For example, the usage data may include an identifier of the terminal device, a user account, and a historical access record (the user browses news, articles, watches videos, and accesses recorded data of the social networking site). . The type of the terminal device is as shown in FIG. 1, and includes a mobile phone (terminal 23), a tablet or PDA (terminal 25), a desktop (terminal 22), a PC (terminal 24), an all-in-one (terminal 21), and the like. Among them, the terminal device is equipped with various application function modules required by various users, such as an application having entertainment functions (such as a video application, an audio playback application, a game application, a reading software), and an application having a service function (such as a map navigation). Applications, group purchase applications, shooting applications, etc., and then set system functions such as applications.

Based on the hardware entity shown in FIG. 1 , the user generates corresponding usage data of the use terminal by using the application on the terminal or the terminal, where the usage data includes the identifier of the terminal device, the user account, and the user generated by using the terminal device by using the terminal device. Use records. The user account may be a communication account corresponding to the terminal (for example, a mobile phone number) or a login account of an application on the terminal device. The terminal device transmits the usage data to the server 1. The identifier of the terminal device may be an IMEI (International Mobile Equipment Identity). When different users use the application on the terminal device or the terminal device, different users correspond to different user accounts, that is, use different user accounts to log in to the application, and the server stores different user accounts corresponding to the terminal device. data. The server performs a score calculation on each user account by using the usage data of each user account corresponding to the terminal device acquired by the terminal device, and obtains a score of each user account. And in combination with the correspondence between the preset score and the confidence, the confidence of each user account is determined. Confidence is used to characterize the accuracy of determining a user account as a user account associated with the terminal device. According to the preset selection rule, the confidence level is selected, and the associated data may be pushed by the terminal device according to the usage data corresponding to the user account corresponding to the selected confidence level. The server 1 can be a push platform, such as an advertisement push platform, an article push platform, and the like. The account identification association method provided by the example of the present application is applied to the server, and based on the usage data of each user account, the rating result of each user account is obtained by preset the association recommendation model, and then the user account associated with the terminal is determined.

In some examples, the server recommends relevant data for the second user or item associated with the user's attention or request for the first user intelligence by determining the similarity or relevance. However, in reality, people often change the communication account (user account) corresponding to the terminal device or the terminal device. Therefore, in the full amount of relationship data, there may be a case where one terminal corresponds to multiple user accounts, or a user account corresponds to more The situation of the terminal. In such a case, the server uses the most frequent user account as the user account associated with the terminal.

However, due to changes in the user account used by the terminal, etc., there may be a user account used on the current terminal that is not the most frequently used user account currently counted on the terminal. Therefore, the server is the one that will be used the most before. The account as the user account associated with the terminal may make the determined user account inaccurate. The example of FIG. 1 is only an example of a system architecture that implements an example of the present application. The example of the present application is not limited to the system structure described in FIG. 1 above, and various examples of the present application are proposed based on the system architecture.

To solve the above technical problem, the example of the present application provides a method for determining an associated account. As shown in FIG. 2A, the method may include:

S110. Acquire a first historical record of the user, where the first historical record of the user includes relationship data of at least one first user account corresponding to the user at the terminal, and at least two dimensions of each first user account in the relationship data are used. recording.

When the step S101 is performed, the usage data (corresponding to the relationship data in the foregoing) when the one or more users use the terminal device by using the corresponding first user account is performed, and the first user accounts are used. The usage data includes: an identifier of the terminal device, the first user account, and a usage record of the first user account; and determining, according to the usage record corresponding to each first user account, at least two of the first user accounts. Usage parameters of the dimensions (corresponding to at least two dimensions used above).

It should be noted that the account identification association method provided by the example of the present application is that the server uses the usage data of each user account of the terminal, and obtains a score result of multiple user accounts corresponding to the terminal by using a preset association recommendation model, thereby determining the terminal association. The process of user accounts.

When the user uses the terminal device or the application on the terminal, the user reports the usage data to the server 1. The format of the usage data may be: {identity of the terminal device, user account identifier, usage record}, wherein the usage record includes the reporting time of the usage data and the source of use. The reporting time of the usage data may be the time of logging in to the first user account, or the time of exiting the first user account. The source may be used by the user to identify the identifier of the application on the terminal device used by the corresponding user account by using the user account, for example, WeChat, Weibo, news, video, etc., or browse the website for the user by using the identifier of the user account. Domain name identification, for example, NetEase mailbox, Sibo forum, etc. The source of use also includes the active source and the passive source. The active source refers to the source corresponding to the usage data actively reported by the terminal served by the server in the application, and the passive source refers to the source corresponding to the usage data obtained from other platforms. The plurality of pieces of the usage data of a terminal constitute a first history record of the terminal. When multiple users use the same terminal, the first history record includes multiple pieces of usage data of the foregoing format. For example, including usage data {terminal 1, account 1, usage record 1}, {terminal 1, account 2, usage record 2}, {terminal 1, account 1, usage record 1}, {terminal 1, account 2, usage record 2}. The usage data of the terminal acquired by the server may be the usage data reported by the terminal in the latest period of time, for example, the usage data reported by the terminal in the most recent month, so that the determined user account associated with the terminal is more accurate.

The terminal in the example of the present application is an electronic device in which various applications are installed.

In the example of the present application, since the server performs data interaction with the terminal, when the user performs the use or operation of the application on the terminal, the server may obtain the usage data reported by the terminal, where the usage data includes the identifier of the terminal device. The identifier of the user account, and the usage record of the terminal corresponding to the user corresponding to the user account. For a terminal, the terminal can correspond to multiple user accounts (multiple users use the terminal). Among them, when the user uses the terminal through each user account, a usage data is generated respectively. The terminal can report the usage data to the server, and the server adds each usage data to the first history record corresponding to the terminal. The server can determine the usage data corresponding to each user account according to the first history record of the terminal. In summary, the server may obtain a first history record corresponding to one terminal, and determine usage data corresponding to each user account according to the first history record.

When determining the usage parameters of the at least two dimensions according to the usage data corresponding to each user account, first determining one or more usage data corresponding to each user account, and determining at least one account corresponding to each account according to one or more usage data corresponding to each account. Two dimensions use parameters. In the example of the present application, the at least two dimension usage parameters of each first user account include at least two of the usage time of each user account, the number of uses of each user account, and the source of use of each user account. The at least two dimension usage parameters shown may include at least two of usage time, number of uses, and source of use. According to a plurality of usage data corresponding to one user account, the current latest reporting time in each usage data is used as the usage time of the user account, and the number of usage data corresponding to the user account is used as the usage number of the user account, and each usage is used. The source of the data is used as the source of the account.

However, the example of the present application may not limit the number of parameters and the data type of the at least two dimensions, wherein the usage parameter of one data type corresponds to one dimension, for example, the usage parameter may include usage time, usage times, and usage source. Dimensions, the number of usage parameters of each dimension may be one or more, for example, the usage source may be multiple.

Preferably, in the example of the present application, the server may use the RFM model to process at least two dimension usage parameters of each user account. For example, the server may acquire the usage time of each user account corresponding to the terminal, and the usage times of each user account. Use parameters such as the source of each user account. The RFM model determines a first score of each user account according to at least two dimension usage parameters of each user account; and determines a first account associated with the terminal device according to the first score corresponding to each first user account.

In the example of the present application, the user account may be a communication account corresponding to the terminal (for example, a mobile phone number) or a login account of the application on the terminal, that is, the at least one user account includes: at least one communication account corresponding to the terminal or an application on the terminal. At least one login account, the application example is not limited. The application is a function application that needs to be logged in or registered by the user, and the specific application type is not limited.

It should be noted that the identifier of the terminal device can be represented by the IMEI. When the terminal reports the usage data based on the user behavior on the terminal to the server, the terminal simultaneously reports its own identifier and the user account, for example, using The data includes the IMEI of the terminal device and the user account.

S112. Call the preset association recommendation model to process at least two dimension usage records of each first user account, and output a first score corresponding to each first user account.

When the step S112 is performed, the first score corresponding to each first user account is calculated by using the usage parameter of the at least two dimensions of each first user account and the preset association recommendation model.

After the server obtains the usage data of the terminal based on the historical behavior of the user on the terminal, the server determines at least two dimension usage parameters of each user account corresponding to the terminal according to the usage data. And the preset association recommendation model is established in the server, and the preset association recommendation model is used for performing comprehensive scoring of multiple dimensions on each user account according to at least two dimension usage parameters of each user account. Therefore, the server processes at least two dimension usage parameters of each user account by calling a preset association recommendation model, and outputs a first score corresponding to each first user account. Then, the server can obtain at least one first score corresponding to the at least one first user account corresponding to the terminal.

It should be noted that, in the example of the present application, the preset association recommendation model may be composed of two parts, and a part is a preset first model for using an importance score (second score) for each dimension, and a part is for Each dimension outputted after the first model is preset uses a second score corresponding to the parameter to perform a preset second model of the integrated weighted score.

In the example of the present application, the first score is a comprehensive score result of the multi-dimensional evaluation of each first user account of the at least one first user corresponding to the terminal, wherein the comprehensive score result is used to represent the first user account corresponding to the terminal. The accuracy of the correspondence of the terminal devices. That is, the higher the comprehensive rating result of the first user account, the higher the accuracy rate corresponding to the first user account, so the server determines that the first user account is a common user account on the terminal, and The first user account is associated with the identity of the terminal device.

Exemplarily, it is assumed that at least two dimension usage parameters of each first user account include: the usage time of each first user account, the number of uses of each first user account, and the source of use of each first user account, Then, the server may calculate the importance score (second score) of each first user account usage time and the importance score of each first user account usage count by using the preset association recommendation model (second score) And the importance score of the source of use of each first user account (second score), and the importance score according to the usage time of each first user account, and the importance of the number of uses of each first user account The score value and the importance score of each first user account source and the preset association recommendation model are obtained, and the comprehensive score result of each first user account is obtained.

According to the account identification association method provided by the application, the server determines at least two dimension usage parameters of each first user account in each first user account corresponding to the terminal according to the usage data of the terminal, according to at least two of the first user accounts. The dimension usage parameter and the preset association recommendation model determine a comprehensive score (first score) of each first user account, and determine a first user account associated with the terminal device according to the comprehensive score of each first user account. The first user account is evaluated according to the usage parameters of the multiple dimensions of the first user account, so that the determined first user account associated with the terminal device is more accurate.

In some examples, the method for determining an associated account provided by the present application further includes the following steps:

S11: Push the associated data to the terminal device according to the usage record corresponding to the first user account associated with the terminal device.

The determined first user account associated with the terminal is a user account commonly used by the terminal, and the associated data is recommended to the terminal according to the usage data corresponding to the most commonly used account of the terminal, so that the recommended data is more accurate. For example, the user's interest characteristics corresponding to the account associated with the terminal are determined according to the usage record in the usage data of the account, and the associated data is recommended to the terminal according to the interest feature, for example, pushing advertisements, news, articles, and the like.

In some examples, the account identification association method provided by the present application further includes the following steps:

S21. Determine a confidence level corresponding to each first user account according to the first score corresponding to each first user account and the correspondence between the preset score and the confidence level.

The server processes the at least two dimension usage parameters of each first user account by calling the preset association recommendation model, and obtains a first score corresponding to each first user account. After obtaining at least one first score, the server obtains the At least one first score of the correspondence between the terminal and the at least one user account, the server stores a correspondence between the preset score and the confidence, and the correspondence corresponding to the first score is obtained according to the correspondence between the preset score and the confidence degree. That is, the server obtains the confidence of the correspondence between each first user account and the terminal.

Specifically, the server matches the at least one first score with the correspondence between the preset score and the confidence, for example, when the first score matches the fourth score in the correspondence between the preset score and the confidence, the server The confidence level corresponding to the fourth score in the correspondence between the preset score and the confidence is determined as the confidence corresponding to the first score.

S22. Acquire a first confidence level that satisfies a preset selection rule from the at least one confidence level, and use the at least two dimension usage records of the first user account corresponding to the first confidence level to push the associated data to the terminal, where the preset selection is performed. The rules are determined by the type of actual push associated data.

When the step S22 is performed, the confidence level that satisfies the preset selection rule is obtained from the confidence levels corresponding to the first user accounts, and the first user account corresponding to the confidence level that satisfies the preset selection rule is corresponding. The usage record pushes the associated data for the terminal device.

After the server obtains at least one confidence level corresponding to the at least one first score according to the correspondence between the at least one first score and the preset score and the confidence level, in the example of the present application, the server may perform the terminal and the first according to the confidence level. The determination of the degree of association of an account. Therefore, the server may determine a confidence level from at least one confidence level, and then perform recommendation of the associated data according to the usage data of the first user account corresponding to the confidence level.

It should be noted that, since the type of associated data recommended by the server to the terminal may be various, for example, advertisement, video, audio, and the like, the server may adopt different types of recommended associated data. Different rules, that is, the preset selection rule selects a confidence level from the at least one confidence level, so as to push the terminal according to the usage data of the first user account corresponding to the confidence level (the selected confidence may correspond to one or more relationship data) Associated data.

Specifically, in the example of the present application, the server may obtain the confidence with the highest degree of confidence from at least one confidence level, and push the associated data for the terminal according to the usage data of the first user account corresponding to the confidence level. Specifically, the relevant data is pushed for the terminal based on the usage record in the usage data. For example, when the server is to recommend the associated video to the terminal, the selected confidence level corresponds to the usage record in the usage data corresponding to the one or more first user accounts to determine the interest feature corresponding to the terminal, and the server may select A related video having a high degree of relevance to the interest characteristics of the terminal is used as the recommended video. In the example of the present application, the higher the confidence, the more accurate the association is. Therefore, the server can select the user's usage record corresponding to the user account corresponding to the highest confidence level as a reference for the user's preference or hobby, and perform related data. Recommended.

In another example, the server may further obtain, from the at least one confidence level, a confidence level corresponding to the first user account, and associate the terminal with the usage record in the usage data of the first user account corresponding to the confidence level. data. For example, when the server is to recommend the associated advertisement to the terminal, the server may select a confidence level corresponding to at least one first user account, for example, the acquired at least one confidence includes the first confidence and the second confidence. The first confidence level is greater than the second confidence level, and the first confidence level corresponds to a first user account (ie, corresponding to one user, the user is a common user of the terminal), and the second confidence level corresponds to three second user accounts (corresponding to three The user is not the user of the terminal. In this case, the second confidence is selected, and the related advertisement is recommended to the terminal according to the usage record in the usage data of the first user account corresponding to the second confidence. That is, the user characteristics are determined according to the usage records of the above three users, and then the advertisement is pushed to the terminal according to the user characteristics. The pushed advertisement is matched with more users who use the terminal. In this case, the recommended ads will make as many users as possible interested in buying. Therefore, in the example of the present application, the server selects the user usage record in the usage data corresponding to the first confidence level corresponding to the at least one first user account as a reference of the user's preference or hobby, and performs recommendation of the associated data. 2B is a detailed flow chart for pushing associated data to a terminal based on confidence.

It can be understood that, because the server can calculate the first score corresponding to each first user account by the terminal by using the preset association recommendation model. Furthermore, the confidence level corresponding to each first user account is such that the server can select different confidence levels according to preset selection rules. The user corresponding to the user account corresponding to the selected confidence level is used as the user to be recommended. The associated data is pushed to the terminal according to the usage record in the usage data of the user account. In the case that the terminal corresponds to multiple user accounts, the user account associated with the terminal is adaptively determined, so that the associated data is recommended for the user corresponding to the user account associated with the terminal, and the accuracy of the association recommendation is improved.

Based on the implementation process of the foregoing example, as shown in FIG. 3, the application example provides an account identification association method, where the server invokes a preset association recommendation model, where the association recommendation model is used to process at least two of each first user account. The dimension uses the parameter to output a first score corresponding to each of the first user accounts. The preset association recommendation model includes a first model and a second model, and the process of obtaining at least one first rating may include:

S201: Calling the preset first model to process at least two dimension usage records of each first user account, and outputting at least two second scores corresponding to at least two dimension usage records of each first user account, where The preset association recommendation model includes: a preset first model, wherein the preset first model is used to respectively score the importance degree of the usage record of at least two dimensions of each first user account.

When the foregoing S201 is executed, the use of the first model to calculate the usage parameters of the at least two dimensions of each of the first user accounts is performed, and the second score of the usage parameters of each dimension of each first user account is output.

In the example of the present application, the preset association recommendation model may be composed of two parts, one part is a preset first model for importance degree scoring using parameters for each dimension, and one part is for each output after the preset first model is output. The dimension uses the first score corresponding to the parameter to perform a preset second model of the integrated weighted score.

Exemplarily, it is assumed that at least two dimension usage parameters of each first user account include: the usage time of each first user account, the number of uses of each first user account, and the source of use of each first user account, Then, the server may calculate the importance score (second score) of each first user account usage time and the importance score of each first user account usage count by using the first model in the preset association recommendation model. The value (second score) and the importance score (second score) of the source of use of each first user account.

In the example of the present application, the preset first model may include at least two of formula (1), formula (2), and formula (3), wherein formula (1), formula (2), and formula (3) are :

Lg(1/(1+(n/30)*n))/N (2)

Lg(j/k)/J (3)

Among them, the importance score of the use source is determined by the formula (1), the importance score of the use time is determined by the formula (2), and the importance score of the use count is determined by the formula (3). Where m is the total number of sources of use of each first user account, h _i is a preset score corresponding to the i th use source of each first user account, H is the total number of preset use sources, and M is a pre- Set the first normalization parameter and m is less than H. n is the number of days when the usage time of each first user account is from the current time, N is a preset second normalization parameter; j is the number of times of use of each first user account, k is a preset time, and J is a preset The third normalization parameter.

In the example of the present application, M, N and J are normalization parameters and are positive integers. Specifically, the value of M can be

The value of the value falls within the absolute value of the upper limit of two adjacent integer intervals; the value of N can be lg (1/(1+(n/30)*n)) and the value falls between two adjacent integers The absolute value of the upper limit value in the interval; the value of J may be the absolute value of the upper limit value when the value of lg(j/k) falls within two adjacent integer intervals.

It should be noted that, in the example of the present application, formula (1) is a model for calculating a second score of the source of use of each first user account; formula (2) is used to calculate each first user account. A model of the second score using time; formula (3) is a model for calculating a second score for the number of uses of each first user account.

Specifically, when the server scores the source of use of each first user account, since the source of use of each first user account can be classified into an active source and a passive source, the h _i corresponding to different usage sources is That is, the preset score corresponding to the ith source of use of each first user account is also different. The value of h _i corresponding to the active source is higher than the value of h _i corresponding to the passive source, and the value of h _i corresponding to the passive source may be assigned different values according to the calculated accuracy rate of the active source, and the same as the active source. Then the value of h _i is higher. One or more correspondences of IMEI-user accounts included in the usage data obtained from passive sources such as other platforms and one or more correspondences of IMEI-user accounts included in the usage data acquired by the active source are included The more the corresponding correspondences, the closer the passive source is to the active source. For example, the value of h _i corresponding to the active source may be 2, and the value range of h _i corresponding to the passive source may be between 1 and 1.9. The more the source of use of each first user account, the higher the score corresponding to the source of use of each first user account obtained by formula (1). When the server scores the usage time of each first user account, formula (2) indicates that the longer the n is from the current time, the lower the score obtained by the usage time of each first user account, and in order to optimize the time decay, The first user account reported in a 30-day period gives a higher score and slower decay; and the data reported over 30 days is attenuated faster, so the denominator of the logarithm in equation (2) is added (n /30) to optimize. In short, the closer the usage time of each first user account is to the current time, the higher the score. When the server scores the number of uses of each first user account, the higher the number of uses of each first user account, the higher the score.

Further, the example of the present application does not limit the form of the preset first model, as long as the preset first model can be used to represent the scores of the plurality of dimension usage records, which is not limited by the examples in the present application. For example, formula (1), formula (2), and formula (3) can also be correspondingly transformed into: formula (4), formula (5), and formula (6), as follows:

1-lg(1/(1+(n/30)*n))/N (5)

1-lg(j/k)/J (6)

Finally, the logarithmic operation in equations (1)-(6) is used to normalize the score.

S202: The preset second model is processed to process the at least two second scores, and the first score corresponding to each first user account is output until the at least one first score is obtained, where the preset association recommendation model further includes: A second model is provided, the preset second model is used to weight the at least two second scores to obtain a total score.

When performing the above step S202, executing the second model to process the second score of each dimension usage record of each first user account, and outputting the first score of each first user account, where the first The second model is configured to weight the second score of each dimension usage record of each first user account to obtain the first score of each first user account.

The server processes the at least two dimension usage parameters of each first user account by calling the preset first model, and outputs at least two second scores corresponding to the at least two dimension usage parameters of each first user account. Thereafter, since the preset second model is used to weight the at least two second scores to obtain a total score, the server may call the preset second model to process the at least two second scores, and output the first The first score corresponding to the user account, and in the same manner, the server may obtain the first score corresponding to each first user account in the at least one first user account.

In the example of the present application, in the preset second model, at least two second scores corresponding to each first user account correspond to different weight values, and the server may perform at least two second scores corresponding to each first user account. Weighting gives a comprehensive score, the first score. In the three dimensions of using source, usage time and usage times, usage time is the most important factor, so the usage time corresponds to a higher weight and the other two weights are lower. Based on this principle, the preset second model in the example of the present application can be obtained according to the training model, and the specific implementation process will be described in the following examples.

The construction or generation method of the preset second model in the example of the present application can be performed by a common classification method of machine learning, for example, support vector machine, logistic regression, decision tree, GBDT or neural network. In the example of the present application, by using at least two second scores corresponding to each first user account as target variables, the constructed samples are called for training, and the weight parameters are adjusted to obtain an optimal model capable of comprehensive scoring based on multiple dimensions. .

It can be understood that, because the server can calculate the first score corresponding to each first user account by using the preset association recommendation model, the confidence level corresponding to each first user account is obtained, so that the server can select according to the preset. The rule implementation achieves different confidence levels in different situations, and the user corresponding to the user account corresponding to the selected confidence level is used as the user to be recommended. In the case that the terminal corresponds to multiple user accounts, the user account associated with the terminal is adaptively determined, so that the associated data is recommended for the usage data corresponding to the user account associated with the terminal, and the accuracy of the association recommendation is improved.

Based on the description of the implementation process of the above example, the example of the present application provides a preset second model based on the introduction of machine learning technology, and all feature dimensions are considered when the first score is obtained, and then the judgment is comprehensively performed. In the initial stage of forming the preset second model, it is still necessary to manually select as many features as possible (ie, the characteristics of the sample) for the machine learning model training, and determine which features are selected according to the degree of discrimination of the first training result. There is basically no problem of manual intervention selection parameters here, and machine learning can learn the appropriate parameters by itself. Since the meaning of the feature is more intuitive than the meaningless parameter, the distribution of the feature is easier to understand. Firstly, based on the real-time comprehensive evaluation of the machine learning model, the comprehensive evaluation involves the comprehensive consideration of the parameters used in multiple dimensions. Improve the accuracy of the comprehensive score. In addition, the model itself has the function of evolutionary learning. Even if the allowable range is updated or deleted, by simply re-training the model (sometimes requiring fine-tuning of the feature), the adjustment of the preset second model can be performed to maximize the accuracy of the comprehensive scoring result.

The application of machine learning technology in multiple dimensions of comprehensive scoring can be shared and disseminated freely, because the machine learning comprehensive score is comprehensive and self-evolving, not specific to a specific user account, and therefore can be disclosed even to different user accounts of the same terminal. A comprehensive scoring approach based on machine learning models. Based on the foregoing example, the application example provides a method for forming a preset second model. As shown in FIG. 4, the method includes:

S301. Acquire a positive sample and a negative sample according to a preset configuration ratio, where the positive sample and the negative sample include a correspondence between each terminal of the at least two terminals and at least one second user account, and At least two third scores of each second user account of each terminal obtained by the model.

When performing the foregoing step S301, the second user account and the third user account corresponding to each terminal device in the at least one terminal device are obtained, and at least two dimensions of the second user account and the third user account corresponding to each terminal device are obtained. Using the parameter; determining, according to the first model, at least two third scores of the second user account corresponding to each terminal device and at least two third scores of the third account.

The training sample selects multiple terminals, each terminal corresponds to a positive sample and one or more negative samples, and the positive sample is a user account that the terminal is using. For example, the positive sample is the second user account, and the negative sample is the third user account. Determining, according to the usage data corresponding to the second user account, the usage parameters of the at least two dimensions of the second user account, and determining at least two third scores of the second user account, where each third score corresponds to the usage parameter of each dimension . In the same manner, at least two third scores of each third user account are determined. Among them, the negative sample is an account that has been used but has not been used. Here, in the actual operation process, there will be a certain ratio between the positive sample and the negative sample, and this ratio is the configuration ratio. When the preset second model is formed, the configuration of the training data by the server (the sample of the existing user behavior and the corresponding comprehensive scoring result) also needs to be set according to the configuration ratio. The positive sample and the negative sample are user accounts corresponding to the first terminal.

S302. Call the set training model to process the positive sample or the negative sample to obtain the first training result.

When performing the above step S302, the following steps are performed:

Presetting parameters of the second model;

Performing the following processing for each terminal device in the at least one terminal device:

Determining, according to at least two third scores of the second user account corresponding to the terminal device, a first score of the second user account, and determining, according to at least two third scores of the third user account corresponding to the terminal device, Describe the first rating of the third user account;

Determining a user account associated with the terminal device according to the first rating of the second user account and the first rating of the third user account.

It should be noted that the server in the example of the present application obtains at least two third scores obtained by presetting the first model for each second user account corresponding to the positive sample and the negative sample; and obtaining at least each third user account pair. The process of the two third ratings is the same as the principle of obtaining at least one second score corresponding to each first user account.

It can be understood that the more complete the allowable range involved in the positive and negative samples in the example of the present application, the more accurate the subsequent comprehensive score evaluation is.

When the second model is determined by machine learning, the parameters of the second model are preset. The second model is used to perform weighted summation of the third scores of the respective dimensions to obtain a first score, and the parameters of the second model include weights corresponding to the second scores of the respective dimensions. Firstly, the weight corresponding to the second score of each dimension is preset, and for any sample of the positive sample (second user account) and the negative sample (third user account) of each terminal, one parameter is determined according to the preset parameter of the second model. The first score of the sample determines the user account associated with the terminal according to each user account corresponding to one terminal. For example, the user account with the highest highest score is determined as the user account associated with the terminal. When the user account associated with the terminal is the second user account (positive sample), the user associated with the terminal determined by the second model is correct. For each terminal device in the at least one terminal device, the account corresponding to the terminal device may be determined, and when the account is a positive sample, the evaluation is correct, otherwise, the evaluation error

S303. Continuously detecting the training model until the first training result satisfies the preset condition, and using the training model that meets the preset condition as the preset second training model, the preset condition is used to represent the pre-condition The data output result obtained by the second model is used to determine the most common user account of the terminal, and the real common user account closest to the terminal.

When performing the above step S303, the following steps are performed:

Determining an accuracy rate of the second model according to a user account associated with each terminal device in the at least one terminal device;

The accuracy of adjusting the parameters of the second model to the second model satisfies a preset condition.

The accuracy of the second model is determined by determining the correct terminal device according to the user account corresponding to each terminal device. For example, the ratio of the terminal device that is correctly evaluated to the total terminal device may be used as the accuracy of the second model. The parameters of the second model are adjusted, that is, the weights of the second scores are adjusted, and step S302 is repeatedly performed until the accuracy of the acquired second model reaches a maximum, and the parameters of the second model at this time are optimal.

In the example of the present application, regardless of the training model, when the training is started, the entry of the training model includes the features of the at least two dimensions described above, and if the feature does not have a favorable influence on the first training result or is wrong At the same time, the feature of the dimension or the weight of the data is lowered. If the feature has a favorable influence on the first training result, the weight of the feature or the data is increased. If the weight of one parameter is reduced to 0, then in the training model, Features will not have any effect. Through the final experiment of the examples of the present application, the characteristics of the above different dimensions that ultimately have a positive impact on the first training result are long-term features (ie, at least two third scores in the examples of the present application). It is assumed below that the features of different dimensions include only at least two third scores corresponding to at least two dimensions using parameters (ie, other non-conforming features have been eliminated), then the formation process of the preset second model substantially includes And inputting at least two dimensions of the positive sample or the negative sample into the training model using the at least two third scores corresponding to the parameters (ie, invoking the training model), and obtaining the first training result from the training model. The training model constructed therein has at least two third scores, and each of the third scores has a corresponding weight (preset priority). The first training result is continuously monitored until the preset condition is met (when the accuracy of the second model reaches a maximum), then the training model is taken as the preset second model. The first training result is a determined user account associated with each terminal.

The preset condition in the example of the present application may be that the accuracy of the comprehensive result reaches a preset threshold, and the preset threshold may be 99%, and the specific preset threshold may be set. The example in the application is not limited, but the preset threshold is The higher the setting, the more accurate the preset second model of the comprehensive score that reaches the preset threshold or preset condition.

Exemplarily, as shown in FIG. 5, the accuracy relationship between the terminal and the user account is shown in FIG. 1. In the process of obtaining the preset second model, the server expresses the accuracy of the comprehensive result by the confidence accuracy rate, with a preset threshold. 99%. The RFM model is used to obtain three dimension usage parameters, that is, the usage time (R) of each first user account, the usage frequency (F) of each first user account, and the usage source (S) of each first user account are For example, the usage parameters of at least two dimensions are as shown in FIG. 5. When the usage time of each first user, that is, the weight value of R is 0.7, the confidence accuracy rate satisfies the preset condition, and therefore, the server trains out. The usage time of each first user of the preset second model corresponds to a weight of 0.7, and the sum of the usage times of each first user account and the weight of each first user account is 0.3, specifically, The usage number of each first user account may be 0.2 and the weight value corresponding to the source of use of each first user account may be 0.1.

Moreover, for example, from the relationship between the terminal and the user account shown in FIG. 6, the accuracy rate chart 2 shows that if the server adopts a single dimension (R or F or M), the confidence accuracy rate (76.5%) does not adopt the RFM model. The three dimensions obtained are higher by the confidence accuracy (88.20%) achieved by using the weighted total score corresponding to the data. As can be seen from Figure 5, if the server uses a single dimension (R, F, and M) or directly uses the weight value to directly score the second dimension (RFM total score) of the multi-dimensional, the confidence accuracy is not obtained by the RFM model. The three two dimensions use the weight-weighted total score corresponding to the record to achieve a high confidence rate, and the highest value when the weight of R is 0.7, and the confidence accuracy rate is 99%.

As can be seen from the above process, 1) the example of the present application adopts a comprehensive scoring method based on the preset second model, when constructing a first user account and at least two of the relationship data between the terminal and the at least one first user account The second scoring performs a comprehensive scoring based on multiple dimensions, and fully utilizes multiple dimension usage parameters corresponding to each first user account on the terminal to obtain a preset second model, which can effectively obtain each first user on the terminal. The use of the account's trustworthiness indicator enables evaluation of each first user account on the relevant terminal. 2) The example of the present application introduces the use of different usage parameters in the data to train the training model, and determines the weight of the second score of each dimension in the second model according to the first training result, and then determines according to the second model. The comprehensive score of each first user account thus improves the accuracy of the comprehensive score of the user account. 3) A remarkable feature of the preset second model adopted in the application example is that the model can self-evolve, automatically adjust the weight according to the transformation of the record using at least two dimensions, and avoid the rule-based manual frequent intervention adjustment parameter.

It can be understood that, in the example of the present application, the application example uses the usage data of the terminal, and determines multiple dimensions of each first user account corresponding to the terminal according to the usage data. Use parameters as the primary data source. The scoring process and the model construction process are simple and easy, and do not need to use various complicated coding, clustering, and filtering methods to perform complex construction and processing on the features, which greatly reduces the workload of data processing and makes the preset second model simple. Available.

Further, as shown in FIG. 7 , after S303, in a method for determining an associated account provided by the example of the present application, the server may obtain a correspondence between a preset score and a confidence level, which may include:

S304. Call the preset second model to process the positive sample and the negative sample, and obtain the second training result.

After the server obtains the specific preset second model, that is, after determining the weight value of each second score, the server may input the positive sample and the negative sample into the preset second model (ie, call the preset second model), The second training result is obtained. The second training result is a first score corresponding to each first user account,

It should be noted that the second training result in the example of the present application is a comprehensive score for each sample on the basis of the highest confidence accuracy.

S305: Call a second training result and a correspondence between the preset sample and the confidence accuracy, and obtain a confidence accuracy rate corresponding to the second training result.

The server inputs the positive sample and the negative sample into the preset second model, and after obtaining the second training result, the server can know the correspondence between the second training result and each sample (each sample in the positive sample and the negative sample). . In the example of the present application, the correspondence between the preset sample and the confidence accuracy rate is further set in the server, wherein the correspondence between the preset sample and the confidence accuracy rate is based on the second score of the usage time in the sample. High, the higher the confidence level of the corresponding confidence is set. That is, the closer the usage time of the user account in the relationship data between the terminal and the user account is to the current time, the higher the confidence rate of the comprehensive score of the server is represented, and the formula (2) or formula (5) is based on the principle. Calculate the score of the time used. Therefore, the server matches the correspondence between each sample and the preset sample and the confidence accuracy rate, and obtains the confidence accuracy rate corresponding to each sample, and according to the correspondence between the second training result and each sample, The first confidence accuracy corresponding to the second training result can be obtained.

S306. The correspondence between the second training result and the first confidence accuracy is used as a correspondence between the preset score and the confidence.

After the server obtains the first confidence accuracy rate corresponding to the second training result, the higher the first score obtained by the server through the preset second model is used to represent the relationship between the terminal corresponding to the first score and the at least one first user account. The relational data corresponding to the data has the highest relationship accuracy rate, that is, the first user account corresponding to the first rating is the most commonly used user account in the relationship data between the terminal and the at least one first user account. Therefore, the correspondence between the second training result and the first confidence accuracy rate may be used to represent the relationship accuracy between the terminal corresponding to the first rating and the relationship data of the at least one first user account. Therefore, the server may use the second training result and the first training result. The correspondence between the confidence rate and the confidence is used as the correspondence between the preset score and the confidence. Then, the server can determine the user corresponding to the confidence that the preset selection rule wants to select by the correspondence between the preset score and the confidence. The user corresponding to the account.

Exemplarily, taking the three dimensions of RFM as an example, the second training result is obtained by formula (1), formula (2) and formula (3), and preset second model. As shown in Table 1, the second training result (weighted total score), the confidence level, and the correspondence table of the preset samples are summarized.

Table 1

It can be seen from Table 1 that the higher the weighted total score, the higher the confidence, and the confidence level indicates the relationship accuracy between the terminal and the at least one first user account. Therefore, the server can select the adaptive selection by the preset selection rule. The confidence of the most commonly used users on the terminal enables the push of associated data, or the most trusted confidence of the user to use to push the associated data.

It can be understood that, because the server can calculate the first score corresponding to each first user account by using the preset association recommendation model, the confidence level corresponding to each first user account is obtained, so that the server can select according to the preset. The rule selects the confidence level under different conditions, and pushes the associated data to the terminal according to the usage data of the user account corresponding to the selected confidence level. In the case that the user account used on the terminal is changed, the account associated with the terminal is adaptively determined, so that the associated data is recommended for the user corresponding to the associated account on the terminal, and the accuracy of the association recommendation is improved.

As shown in FIG. 8, the example of the present application provides a server 1, which may include:

The obtaining unit 10 is configured to acquire a first history record of the user, where the first history record of the user includes relationship data of at least one first user account corresponding to the user, and each first user account in the relationship data. Use records for at least two dimensions.

The calling unit 11 is configured to invoke a preset association recommendation model, where the preset association recommendation model is configured to process at least two dimension usage records of each of the first user accounts, and output corresponding to each of the first user accounts. The first score is obtained by at least one first score, wherein the at least one first user account corresponds to the at least one first score respectively.

The obtaining unit 10 is further configured to invoke the correspondence between the at least one first score and the preset score and the confidence, acquire at least one confidence level corresponding to the at least one first score, and from the at least one Confidence acquisition obtains a first confidence level that satisfies the preset selection rule, and uses at least two dimension usage records of the first user account corresponding to the first confidence level to push the associated data for the terminal, the preset selection rule It is determined by the type of actual push associated data.

The at least two dimension usage records of each of the first user accounts acquired by the obtaining unit 10 include: a usage time of each of the first user accounts, a usage count of each of the first user accounts, and the At least two of the sources of use of each first user account.

The calling unit 11 is configured to: call the preset first model to process the at least two dimension usage records of each of the first user accounts, and output the at least two dimension usage records corresponding to each of the first user accounts. At least two second scores, wherein the preset association recommendation model includes: the preset first model, the preset first model being used for at least two of each of the first user accounts respectively The dimension is scored using the importance of the record; and the preset second model is invoked to process the at least two second scores, and the first score corresponding to each of the first user accounts is output until the location is obtained Determining at least one first rating, wherein the preset association recommendation model further comprises: the preset second model, wherein the preset second model is configured to weight the at least two second scores to obtain a total score .

Based on FIG. 8, as shown in FIG. 9, the server 1 further includes a detecting unit 12.

The obtaining unit 10 is further configured to obtain a positive sample and a negative sample according to a preset configuration ratio, where the positive sample and the negative sample are correspondences between the first terminal and the at least one second user account, and each of the first At least two third scores obtained by the two user accounts through the preset first model.

The calling unit 11 is further configured to invoke the set training model to process the positive sample or the negative sample to obtain a first training result.

The detecting unit 12 is configured to continuously detect the training model until the first training result satisfies a preset condition.

The acquiring unit 10 is further configured to use, as the preset second model, the training model that meets the preset condition that the first training result meets the preset condition, where the preset condition is used to represent according to the preset The data output result obtained by the second model is used to determine the most common user account of the terminal, and is closest to the real user account of the terminal.

The calling unit 11 is further configured to: after the training model that meets the preset condition that the first training result meets the preset condition, use the preset second model to process the The positive sample and the negative sample, the second training result is obtained.

The acquiring unit 10 is further configured to acquire, according to the correspondence between the second training result and the preset sample and the confidence accuracy, the first confidence accuracy rate corresponding to the second training result; Corresponding relationship between the second training result and the first confidence accuracy rate is used as a correspondence between the preset score and the confidence.

The preset first model includes:

At least two of lg(1/(1+(n/30)*n))/N and lg(j/k)/J;

Where m is the total number of sources of use of each of the first user accounts, h _i is a preset score corresponding to the ith source of use of each of the first user accounts, and H is the total number of preset usage sources , M is a preset first normalization parameter, and m is less than H; n is the number of days from the current time of each first user account, and N is a preset second normalization parameter; j is The number of times each first user account is used, k is a preset time, and J is a preset third normalization parameter.

The obtaining unit 10 is specifically configured to obtain the first confidence that is the most reliable from the at least one confidence level.

The obtaining unit 10 is specifically configured to obtain, from the at least one confidence level, the first confidence level that corresponds to the at least one first user account.

The at least one first user account acquired by the acquiring unit 10 includes: at least one communication account corresponding to the terminal or at least one login account of the first application on the terminal.

As shown in FIG. 10, the present application also provides a server, including a processor 14 and a storage medium 15, which is linked to the processor 14 via a system bus 16. The processor 14 is embodied by a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA). The storage medium 15 is for storing executable program code, and the program code includes computer operation instructions. The storage medium 15 may include a high speed RAM memory, and may also include a nonvolatile memory, for example, at least one disk storage. The program code stored in the memory is configured to be executed by the processor to implement the method of determining an associated account in the present application described above and to implement the functions of the various modules in the server in the present application.

Those skilled in the art will appreciate that examples of the present application can be provided as a method, system, or computer program product. Accordingly, the examples of the present application may take the form of a hardware instance, a software example, or an example of combining software and hardware aspects. Moreover, the examples of the present application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) including computer usable program code.

The present application examples are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to examples of the present application. It will be understood that each flow and/or block of the flowchart illustrations and/or FIG. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine for the execution of instructions for execution by a processor of a computer or other programmable data processing device. Means for implementing the functions specified in one or more of the flow or in a block or blocks of the flow chart.

The computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device. The apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.

These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device. The instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.

The above descriptions are only preferred examples of the examples of the present application, and are not intended to limit the scope of protection of the examples of the present application.

Claims

A method for determining an associated account is applied to a server, and the method includes:

Obtaining usage data when the one or more users use the terminal device by using the corresponding first user account, the usage data of each first user account includes: an identifier of the terminal device, the first a user account and a usage record of the first user account;

Determining, according to the usage record corresponding to each first user account, usage parameters of at least two dimensions of each first user account;

Calculating, by using the usage parameter of the at least two dimensions of each first user account and the preset association recommendation model, a first score corresponding to each first user account;

Determining, according to the first score corresponding to each first user account, a first account associated with the terminal device.
The method of claim 1 wherein the method further comprises:

And uploading the associated data to the terminal device according to the usage record corresponding to the first user account associated with the terminal device.
The method of claim 1 wherein the method further comprises:

Determining a confidence level corresponding to each first user account according to a correspondence between a preset score and a confidence level, and a first score corresponding to each of the first user accounts;

And obtaining a confidence level that meets the preset selection rule, and the usage record corresponding to the first user account corresponding to the confidence level of the preset selection rule is pushed by the terminal device according to the confidence level corresponding to the first user account. Associated data.
The method of claim 1 wherein

The usage parameters of the at least two dimensions of each first user account include data of at least two types of data of the following three types: usage time of each first user account, usage times of each first user account, and each The source of the first user account.
The method according to claim 1, wherein the preset association recommendation model comprises a first model and a second model, wherein the usage parameters and the preset association recommendation of at least two dimensions of each of the first user accounts are utilized. The first score corresponding to each model of the first user account is calculated by the model:

Invoking the first model to calculate usage parameters of at least two dimensions of each first user account, and outputting a second score of usage parameters of each dimension of each first user account;

And calling the second model to weight the second score of the usage parameter of each dimension of each first user account to obtain the first score of each first user account.
The method according to claim 5, wherein the at least two dimension usage parameters of each first user account comprise: a usage time of each first user account, a usage number of each first user account, and each first user account. Source of use;

The invoking the first model to calculate the usage parameters of the at least two dimensions of the first user accounts, and outputting the second scores of the usage parameters of each dimension of each first user account includes:

Determining a second rating of the source of use of each first user account by the following formula (1);

Determining a second rating of the usage time of each first user account by the following formula (2);

Lg(1/(1+(n/30)*n))/N (2)

Determining a second score of the number of uses of each first user account by the following formula (3);

Lg(j/k)/J (3)

Where m is the total number of sources used by each first user account, hi is the preset score corresponding to the ith source of each first user account, H is the total number of preset usage sources, and M is a preset The first normalization parameter, and m is less than H; n is the number of days from the current time of each first user account, N is a preset second normalization parameter; j is the use of each first user account The number of times, k is the preset time, and J is the preset third normalization parameter.
The method of claim 5 wherein the method further comprises:

Obtaining a second user account and a third user account corresponding to each terminal device in the at least one terminal device, and acquiring usage parameters of at least two dimensions of the second user account and the third user account corresponding to each terminal device;

Determining, according to the first model, at least two third scores of the second user account corresponding to each terminal device and at least two third scores of the third account;

Presetting parameters of the second model;

Performing the following processing for each terminal device in the at least one terminal device:

Determining, according to at least two third scores of the second user account corresponding to the terminal device, a first score of the second user account, and determining, according to at least two third scores of the third user account corresponding to the terminal device, Describe the first rating of the third user account;

Determining, according to the first score of the second user account and the first score of the third user account, the second user account or the third user account as a user account associated with the terminal device;

The method further includes:

Determining an accuracy rate of the second model according to a user account associated with each terminal device in the at least one terminal device;

The accuracy of adjusting the parameters of the second model to the second model satisfies a preset condition.
The method according to claim 3, wherein the obtaining a confidence level that satisfies a preset selection rule from the confidence levels corresponding to the first user accounts comprises:

The confidence level with the highest confidence is obtained from the confidence levels corresponding to the first user accounts.
The method according to claim 3, wherein the obtaining a confidence level that satisfies a preset selection rule from the confidence levels corresponding to the first user accounts comprises:

Obtaining a maximum confidence level corresponding to the first user account from the confidence levels corresponding to the first user accounts.
A server comprising one or more processors and one or more memories, the one or more memories comprising computer readable instructions configured to be executed by the one or more processors to:

Obtaining usage data when the one or more users use the terminal device by using the corresponding first user account, the usage data of each first user account includes: an identifier of the terminal device, the first a user account and a usage record of the first user account;

Determining, according to the usage record corresponding to each first user account, usage parameters of at least two dimensions of each first user account;

Calculating, by using the usage parameter of the at least two dimensions of each first user account and the preset association recommendation model, a first score corresponding to each first user account;

Determining, according to the first score corresponding to each first user account, a first account associated with the terminal device.
The server of claim 10 wherein said one or more processors execute said computer readable instructions to:

And uploading the associated data to the terminal device according to the usage record corresponding to the first user account associated with the terminal device.
The server of claim 10 wherein said one or more processors execute said computer readable instructions to:

Determining a confidence level corresponding to each first user account according to a correspondence between a preset score and a confidence level, and a first score corresponding to each of the first user accounts;

And obtaining a confidence level that meets the preset selection rule, and the usage record corresponding to the first user account corresponding to the confidence level of the preset selection rule is pushed by the terminal device according to the confidence level corresponding to the first user account. Associated data.
The server of claim 10 wherein said one or more processors execute said computer readable instructions to:

The usage record of the at least two dimensions of each first user account includes data of at least two types of the following three types of data: the usage time of each first user account, the usage times of each first user account, and each The source of the first user account.
The server of claim 10, wherein the preset associated recommendation model comprises a first model and a second model, the one or more processors executing the computer readable instructions to:

Invoking the first model to calculate usage parameters of at least two dimensions of each first user account, and outputting a second score of usage parameters of each dimension of each first user account;

And calling the second model to weight the second score of the usage parameter of each dimension of each first user account to obtain the first score of each first user account.
The server according to claim 14, wherein the at least two dimension usage parameters of each first user account include: a usage time of each first user account, a usage number of each first user account, and each first user account. Source of use;

Wherein the one or more processors execute the computer readable instructions to:

Determining a second rating of the source of use of each first user account by the following formula (1);

Determining a second rating of the usage time of each first user account by the following formula (2);

Lg(1/(1+(n/30)*n))/N (2)

Determining a second score of the number of uses of each first user account by the following formula (3);

Lg(j/k)/J (3)

Where m is the total number of sources used by each first user account, hi is the preset score corresponding to the ith source of each first user account, H is the total number of preset usage sources, and M is a preset The first normalization parameter, and m is less than H; n is the number of days from the current time of each first user account, N is a preset second normalization parameter; j is the use of each first user account The number of times, k is the preset time, and J is the preset third normalization parameter.
The server of claim 14 wherein said one or more processors execute said computer readable instructions to:

Obtaining a second user account and a third user account corresponding to each terminal device in the at least one terminal device, and acquiring usage parameters of at least two dimensions of the second user account and the third user account corresponding to each terminal device;

Determining, according to the first model, at least two third scores of the second user account corresponding to each terminal device and at least two third scores of the third account;

Presetting parameters of the second model;

Performing the following processing for each terminal device in the at least one terminal device:

Determining, according to at least two third scores of the second user account corresponding to the terminal device, a first score of the second user account, and determining, according to at least two third scores of the third user account corresponding to the terminal device, Describe the first rating of the third user account;

Determining a user account associated with the terminal device according to the first rating of the second user account and the first rating of the third user account;

Wherein the one or more processors execute the computer readable instructions to:

Determining an accuracy rate of the second model according to a user account associated with each terminal device in the at least one terminal device;

The accuracy of adjusting the parameters of the second model to the second model satisfies a preset condition.
The server of claim 12 wherein said one or more processors execute said computer readable instructions to:

The confidence level with the highest confidence is obtained from the confidence levels corresponding to the first user accounts.
The server of claim 12 wherein said one or more processors execute said computer readable instructions to:

Obtaining a maximum confidence level corresponding to the first user account from the confidence levels corresponding to the first user accounts.
A non-transitory computer readable storage medium storing computer readable instructions for causing at least one processor to perform the method of any of claims 1-9.