TWM621545U - User management devices capable of identifying anonymous users belonging to specific user groups - Google Patents

Abstract

The purpose of the present disclosure is how to use anonymous user data to group anonymous users to know which specific user group the anonymous user belongs to. In this way, the enterprise cannot only know which specific user group the real-name user belongs to, but also know which specific user group the anonymous user belongs to. Thus, marketers can use it to facilitate accurate marketing. Specifically, the present disclosure is to use real-name user data to find out the characteristics of real-name users that are important and meaningful to the specific user group, and then use the real-name user characteristics to map out the anonymous user characteristics corresponding to the specific user group. In this way, whether an anonymous user belongs to the specific user group can be determined based on the anonymous user data.

Description

A user management device that can determine whether an anonymous user belongs to a specific user group set

The present invention relates to a user management device, especially a user management device capable of judging whether an anonymous user belongs to a specific user group (eg, but not limited to high value user group, frequent purchase user group or high consumption user group).

For online shopping services, users may go through the process of viewing advertisements (for example, advertisements pushed by Facebook or Google), searching on the Internet, comparing plans and deciding to purchase, etc. during the purchase process. During the process of viewing advertisements, searching on the Internet, and comparing plans, users may not be logged in as members, but browse the web anonymously. On the other hand, many online shoppers currently allow users to directly purchase goods or services without registering and logging in as a member. Therefore, in the traditional user management system, it is impossible to know whether the anonymous user belongs to a specific user group, such as, but not limited to, a high-value user group, a frequent purchase user group, or a high-consumption user group. In addition, general companies will use demographic statistics to distinguish customer groups, and use membership data to predict valuable users, but the two often work independently, so it is still impossible to know what kind of anonymous users in behavioral characteristics will be potential high-value user groups.

In view of the problems of the prior art, if an anonymous user logs in to a member in advance, it is possible to predict which specific user group the anonymous user belongs to, and provide corresponding marketing strategies (such as, but not limited to, giving discounts, giving gifts or giving accumulations). discounts), it can effectively increase the willingness of anonymous users to purchase goods or subscribe to services, thereby increasing the transaction amount.

According to the purpose of the present invention, the present invention provides a user management device capable of judging whether an anonymous user belongs to a specific user group, which has a plurality of hardware circuits that are electrically connected to each other to perform the following steps: For the first specific user group, obtain multiple real-name user characteristics whose importance to the first specific user group is greater than a threshold value or whose importance is the top few as one of the first specific user groups. A plurality of first real-name user characteristics of an important real-name user characteristic set, wherein the feature engineering is to find the first important real-name user of the first specific user group based on a machine learning algorithm through the user management device Multiple first real-name user characteristics of the feature set; mapping one of the first specific user groups according to the multiple first real-name user characteristics of the first important real-name user characteristic set of the first specific user group a plurality of first anonymous user characteristics of a first set of important anonymous user characteristics, wherein at least a part of the plurality of first anonymous user characteristics is obtained from an anonymous profile of an anonymous user; At least a part of the plurality of first anonymous user characteristics obtained from the data determines whether the anonymous user belongs to the first specific user group.

Optionally, the user management device further performs the following step: using multiple real-name user profiles of the multiple real-name users to group the multiple real-name users, wherein at least a part of the multiple real-name users belong to the first real-name user. a specific group of users.

Optionally, use an RFM model (Recency Frequency Monetary Model) to group the multiple real-name users according to the multiple real-name user profiles of the multiple real-name users; The multiple static features and/or multiple dynamic features obtained from the multiple real-name user profiles are used to group the multiple real-name users; alternatively, a K-MEANS algorithm, a support vector machine algorithm and/or a machine learning algorithm are used. The multiple real-name user groups are grouped according to the multiple real-name user profiles.

Optionally, the user management device further performs the following steps: collecting the multiple real-name user data of the multiple real-name users, and acquiring the multiple real-name user data according to the multiple real-name user data. Multiple real-name user characteristics.

Optionally, find out from the plurality of first real-name user characteristics all of the plurality of first anonymous user characteristics that can be acquired through the anonymous data.

Optionally, find out the plurality of first anonymous user characteristics that can be acquired through the anonymous data as a part from the plurality of first real-name user characteristics, and use the plurality of first real-name user characteristics The multiple first real-name user characteristics and their numerical values of the anonymous data acquirer cannot be used as the other multiple first anonymous user characteristics and their numerical values.

Optionally, a grouping model of the anonymous users is trained through the plurality of first anonymous user characteristics, and the grouping model is used for at least a part of the plurality of the anonymous users obtained from the anonymous data of the anonymous user. A first anonymous user feature determines whether the anonymous user belongs to the first specific user group.

Optionally, a feature comparison is performed between the multiple first anonymous user characteristics of the anonymous user and the corresponding multiple first real-name user characteristics, so as to determine whether the anonymous user belongs to the first real-name user. a specific group of users.

Optionally, the feature comparison is to calculate a cosine similarity.

Compared with the prior art, which cannot perform precise marketing on anonymous users, the present invention can further perform precise marketing on anonymous users.

301: Device Information

302: Browsing behavior

303: Interests

304: Internet Protocol Address or Device Identifier

31: A collection of important anonymous user characteristics of a specific user group

311~31M: Anonymous user characteristics

321: Device Information

322: Browsing Behavior

323: Interests

324: Internet Protocol address or device identification

325: Demographics

326:Consumption record

327: Interactive Record

328: Account records

33: Collection of important real-name user characteristics of a specific user group

331~33N: Real-name user characteristics

4: Online real-name/anonymous user service system

41: Platform device

42: Anonymous user data capture device

43: Anonymous User Database

44: Real-name user data capture device

45: User management device

5: User management device

51: Real-name user grouping module

52: Real-name user feature engineering module

53: Anonymous user feature acquisition module

54: Anonymous/real-name user feature mapping module

55: Anonymous user grouping module

S61~S67: Process

FIG. 1 is a schematic diagram of the correspondence between real-name user features and anonymous user features according to the new embodiment of the present invention.

FIG. 2 is a block diagram of an online real-name/anonymous user service system using the user management device according to the new embodiment of the present invention.

3 is a block diagram of a user management device capable of judging whether an anonymous user belongs to a specific user group according to the new embodiment of the present invention.

FIG. 4 is a flowchart of a method for determining whether an anonymous user belongs to a specific user group according to an embodiment of the present invention.

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, and the protection scope of the present invention can be more clearly defined.

The user data collected by the company's platform devices usually include real-name user data (such as the user's member data registered on the company's shopping website or membership system) and anonymous user data (for example, masked or deleted data purchased from a third party). Identified information, the information that the user has not logged in as a registered user or the information of the user who is not registered in the company's shopping website or membership system). Further, real-name user data is named user data (for example, member data or verified user data, which It can be determined by login or third-party verification), real-name user information includes personal information with unique identification, email, phone number or device number that can be traced back to the user, but real-name user information also includes the corresponding User cookies, Internet Protocol address or device identification, browsing behavior, device information, interests, demographics, consumption records, interaction records and billing records. On the contrary, anonymous user information is anonymous user information, which only includes information such as small text files, Internet protocol addresses or device identification, browsing behavior, device information, interests and preferences, etc., and anonymous user information cannot be traced back to the user's personal information.

Generally, enterprises will use real-name user data to group real-name users to determine which specific user group the real-name users belong to. Although it may be possible to know where the anonymous user comes from from the Internet Protocol address of the anonymous user data, it is still not effective to know which specific user group the anonymous user data belongs to. For example, the Internet protocol address of the anonymous user data is the overseas Internet protocol address, but it is still impossible to know whether the anonymous user must not belong to the high-value user group, the frequent purchase user group and the high-consumption user group.

The purpose of this new model is how to use anonymous user data to group anonymous users, so as to know which group of specific user groups an anonymous user belongs to. In this way, in addition to knowing which specific user group a real-name user belongs to, enterprises can also know which group of anonymous users belong to. Which specific user group belongs to, so that marketers can use it to facilitate accurate marketing. In short, the present invention solves the technical problem that the prior art cannot perform accurate marketing on anonymous users because it is impossible to know which specific user group an anonymous user belongs to.

In this new model, the real-name user information of multiple real-name users is mainly collected first, and then, according to a grouping algorithm (for example, through a machine learning algorithm, but not limited to this) or a specific model (for example, an RFM model (Recency Frequency Monetary Model, that is, a classification model based on numerical grouping such as recent consumption, consumption frequency, consumption amount, etc.) to group multiple real-name users, and each real-name user will belong to one of the specific user groups. Taking high-value user groups as an example, If you want to know whether anonymous users also belong to high-value user groups, you can perform feature engineering on high-value user groups to extract important real-name user feature sets corresponding to high-value user groups (including many a real-name user feature). Next, anonymous user profiles of multiple anonymous users are collected, and multiple anonymous user features that may correspond to a part of multiple real-name user features of the important real-name user feature set are acquired as the important anonymous user feature set. Then, it is determined whether the anonymous user belongs to a high-value user group according to the multiple anonymous user characteristics of the anonymous user.

The aforementioned features refer to representative data, which can be understood as input parameters or variables, and meaningful input parameters or variables will affect the final user grouping result (that is, it will affect the prediction result of judging whether a user belongs to a specific user group). For example, features can be extracted from the user's browsing behavior. Further, after entering the webpage, the user often clicks on advertisement content, likes to watch videos, or stays on the webpage for less than 10 seconds, which may all be meaningful features.

In addition, the extracted common features can be divided into numerical type, category type and time type in terms of type. These three types of features can be converted to each other, for example, the features of the numerical type can be processed and converted into the features of the time type. The characteristic of the numerical type is to use the statistical description method to divide the overall data range into scores. The feature of the category type refers to what category it belongs to, for example, the device information may be a laptop, a mobile phone or a tablet, and which one of the device is a laptop, a mobile phone and a tablet is the feature of the category type. In addition, the features of the numerical type can also be converted into the features of the category type, or the features of the category type can also be weighted according to the weight of each category, so as to be converted into the features of the numerical type. The time-type feature may be the frequency, number of times, or the size of the numerical feature based on weekly/monthly/quarterly calculation methods, etc. There are many calculation methods, and the method described in the present invention is not limited. Simply put, any of the three types of features can be processed to generate features of the other.

First, please refer to FIG. 1 of the present invention. FIG. 1 is a schematic diagram of the correspondence between real-name user features and anonymous user features according to an embodiment of the present invention. In this diagram, anonymous user profiles may include including device information 301, browsing behavior 302, interests and preferences 303, Internet Protocol address or device identification 304, and through training algorithms or artificial rules of the analyst, it can be defined how to read the device information 301, browsing behavior 302, interest Preference 303, Internet Protocol address or device identification 304 captures a number of anonymous user characteristics. In addition, real-name user data may include device information 321, browsing behavior 322, interests and preferences 323, Internet Protocol address or device identification 324, demographic data 325, consumption records 326, interaction records 327, and account records 328, and through training Algorithms or artificial rules for analysts that can define how to identify 324 from device information 321, browsing behavior 322, interests 323, Internet Protocol address or device identification 324, demographic data 325, consumption records 326, interaction records 327 and Account record 328 captures multiple real-name user characteristics.

In a specific user group, multiple real-name users may have some real-name user characteristics that are important and meaningful to a specific user group. Through feature engineering, these real-name user characteristics can be extracted as a specific user group. The important real-name user feature set 33, wherein the important real-name user feature set 33 of a specific user group includes, for example, N real-name user features 331-33N. Generally speaking, the number of anonymous user features that can be obtained from anonymous user data is less than the number of real-name user features that can be obtained from real-name user data. N real-name user characteristics 331-33N are selected to find out the characteristics of corresponding anonymous users who can also obtain information. For example, the M anonymous user characteristics 311-31M of the important anonymous user characteristic set 31 of a specific user group can be obtained from The acquirer of anonymous user data can also be important and meaningful to a specific user group, where M is usually less than or equal to N, and M and N are positive integers. In this way, it is possible to judge whether an anonymous customer belongs to a specific user group according to the M anonymous user characteristics 311~31M of the anonymous user, and further formulate a marketing strategy to promote the success rate of transactions and subscriptions, so as to achieve accurate marketing to anonymous users. the goal of.

Please refer to FIG. 2 . FIG. 2 is a block diagram of an online real-name/anonymous user service system using the user management device according to the present invention. The online or offline (online example) real-name/anonymous user service system 4 in FIG. 2 can not only provide services to real-name users and anonymous users, but also provide services to anonymous users by checking whether the anonymous user belongs to which specific user group of real-name users. Users conduct precise marketing. The online real-name/anonymous user service system 4 includes a platform device 41, an anonymous user data capture device 42, an anonymous user database 43, a real-name user data capture device 44 and a user management device 45, wherein the anonymous user data capture device 42 signals The platform device 41, the anonymous user database 43 and the real-name user data capture device 44 are connected, and the user management device 45 is signally connected to the anonymous user database 43, the real-name user data capture device 44 and the platform device 41, and the real-name user data capture device 44 signal connection platform device 41 . In addition, the "signal connection" in the present invention refers to a wireless or wired connection through software or hardware, so that signals or information can be transmitted to each other in a plurality of components that are signal-connected.

The anonymous user connects to the platform device 41 via his user terminal device (such as, but not limited to, a mobile phone, a laptop or a tablet), and the platform device 11 can receive the data transmitted by the anonymous user. The platform device 41 may be a Software as a Service (SaaS) server, a web server, or a server that provides services based on applications installed on the user terminal device. The platform device 41 of the online real-name/anonymous user service system 4 can provide services to anonymous users and real-name users, such as online shopping, online transaction, online investment, online subscription or online consultation, etc., and the present invention is not limited by this. . Through the operation of the user terminal device, the user can choose to use the provided service after logging in as a member, or can choose not to log in as a member and use the provided service in an anonymous manner.

The anonymous user data capture device 42 will capture the data sent by the anonymous user to obtain anonymous user data, such as small text files, Internet Protocol addresses or device identification, browsing behavior, device information, interests and other data , and Anonymous User Data means no can be traced back to the user's personal information. The anonymous user database 43 may be an unstructured database (but the present invention is not limited to this, and may also be a structured database), and is used for storing anonymous user information.

The real-name user data capture device 44 is used to capture the data sent by the real-name user to obtain real-name user data. The anonymous user data of , is extracted as part of the real-name user data. The user management device 45 records and analyzes the real-name user data, and accordingly can group the real-name users according to the real-name user data, so as to determine which specific user group the real-name users belong to.

The user management device 45 can further process the anonymous user data of the anonymous user to find out a plurality of real-name user characteristics of a specific user group of the real-name user that can also be calculated from the anonymous user data, that is, to find out which can be calculated from the anonymous user data and Anonymous user characteristics corresponding to the real-name user characteristics of a specific user group of real-name users. Next, the user management device 45 determines whether the anonymous user belongs to a specific user group of real-name users according to the characteristics of these anonymous users, and labels these anonymous users (that is, marking that the anonymous user belongs to a specific user group), so as to determine whether the anonymous user belongs to a specific user group. Marketing strategies for precision marketing.

The user management device 45 is usually implemented by means of software and a computer device. For example, the computer device includes an operation unit, a storage unit and a communication unit, wherein the operation unit is electrically connected to the storage unit and the communication unit. The computing unit reads a plurality of code codes stored in the non-volatile storage medium, and after the plurality of code codes are executed, a method capable of judging whether the anonymous user belongs to a specific user group can be executed. In addition, the user management device 45 can also be implemented by a hardware circuit, for example, by designing Verilog or VHDL code to program a field programmable array (FPGA) chip to realize the user management device 45 of a pure hardware circuit, for example , by making a specific application A chip (ASIC) is used to realize the user management device 45 of pure hardware circuit. All in all, whether the user management device 45 is implemented using software or hardware is not intended to limit the present invention.

Please refer to FIG. 3 of the present invention. FIG. 3 is a block diagram of a user management device capable of determining whether an anonymous user belongs to a specific user group according to an embodiment of the present invention. Regardless of whether the user management device is implemented by software or hardware, it can be divided into multiple modules as shown in FIG. 3 . In FIG. 3, the user management device 5 includes a real-name user grouping module 51, a real-name user feature engineering module 52, an anonymous user feature acquisition module 53, an anonymous/real-name user feature mapping module 54 and an anonymous user grouping module 55, The real-name user grouping module 51 is signal-connected to the real-name user feature engineering module 52, and the anonymous/real-name user feature mapping module 54 is signal-connected to the real-name user feature engineering module 52, the anonymous user feature acquisition module 53, and the anonymous user grouping module. 55, and the anonymous user feature acquisition module 53 is signal-connected to the anonymous user grouping module 55. Of course, the module division in FIG. 3 is only one way to realize the present invention, and the method in FIG. 3 is not used to limit the present invention.

Please refer to FIG. 3 and FIG. 4 at the same time. FIG. 4 is a flowchart of a method for judging whether an anonymous user belongs to a specific user group according to an embodiment of the present invention. Of course, the flowchart of FIG. 4 is also one of the ways to implement the present invention. In fact, the execution sequence and details of the process of the entire method may be slightly changed according to actual requirements, that is, the approach in FIG. 4 is not intended to limit the present invention.

First, in the process S61, the real-name user grouping module 51 performs the step of collecting real-name user data. Specifically, the real-name user grouping module 51 will collect the real-name user data of multiple real-name users and process the real-name user data according to the system calculation result or the rules defined by humans, so as to obtain the real-name user characteristics, wherein the real-name user data is, for example, but not limited to It is the aforementioned real-name user information, and can also include information such as age, gender, occupation, registration time, RFM parameters, membership level or promotion status, account balance, and the pickup/delivery method of the used products. Real-name user characteristics such as weekly login frequency, average stay time, number of homepage stays per week or weekly Consumption amount, and this new model is not limited by this, under different circumstances, there will be different types of real-name user characteristics.

Next, in the process S62, the real-name user grouping module 51 performs the real-name user grouping step. Specifically, the real-name user grouping module 51 may use a defined model to group multiple real-name users. For example, using the RFM model for grouping, real-name users can be classified into high-value user groups and high-consumption user groups according to real-name user data. , frequent purchasers, recent high-active users, or other user groups. In addition to using the RFM model to group real-name users, other models can also be used to group users, such as grouping by static features (age, gender, occupation, registration time), or by dynamic features (RFM parameters, membership level or promotion) Status, account balance, pickup/delivery method of the goods used) are grouped. The way of grouping may be to use K-MEANS algorithm, support vector machine or machine learning (eg, neural network-like) and other algorithms to directly perform grouping according to the real-name user data of multiple real-name users.

The above-mentioned grouping may be dividing a plurality of specific user groups, or it may be only judging whether the real-name user belongs to a specific user group (for example, it only cares whether it is a high-value user group). The above-mentioned grouping can also be grouping for a specific activity or a single activity. For example, the grouping is to determine whether the real-name user belongs to a specific user group that has participated in various long-term promotion activities for a long time or whether it belongs to a specific user group that participates in a limited one-time idol product add-on purchase activity. . Simply put, the grouping method and grouping type of real-name users are not used to limit the present invention.

For example, in the case of the membership system commonly used in general e-commerce, each real-name user will get a user account after registration, and all future user behaviors in the system will be recorded in the system, such as when and how much the user consumes. Money, time, place and items of consumption, which pages were clicked on before consumption, how many items were put into the shopping cart, how many items were actually purchased, etc. After obtaining these real-name user data, through grouping methods (for example, RFM model, product hobby grouping or activity participation grouping, etc.), it is possible to define which real-name users spend the most and which real-name users spend the most. Name users spend the most frequently, and which real-name users have been active recently, so as to define a specific user group that currently has a high contribution to the enterprise.

Then, in the process 63 , the real-name user feature engineering module 52 performs a real-name user feature engineering step, so as to extract an important real-name user feature set associated with a specific user group of interest. Specifically, feature engineering may be implemented through machine learning or statistical algorithms, and the present invention is not limited to this, as long as at least one real-name user feature associated with a specific user group can be extracted.

Generally speaking, different real-name user characteristics have different importance to whether a real-name user belongs to a specific user group. Therefore, this step is mainly to pick out the real-name user characteristics that are important and meaningful to a specific user group as the important characteristics of a specific user group. Real-name user feature set. For example, it can be to select the real-name user features with the top few importance as the real-name user features in the important real-name user feature set of a specific user group, or it can select the real-name user features whose importance exceeds a certain threshold as the important real-name user features of a specific user group. The real-name user features in the real-name user feature set.

After knowing which real-name users belong to high-value user groups, and those real-name users are not high-value user groups, you can extract multiple characteristics of real-name users to understand which features are real-name user characteristics owned by high-value user groups. (That is, picking out real-name user characteristics that are important to high-value user groups). For example, after feature engineering, the characteristics of real-name users belonging to the high-value user group are: browsing the web page for more than 10 seconds per day, staying on the home page for more than 5 seconds every week, spending more than 1,000 yuan per week and high value The real-name users of the user group often log in through a mobile application rather than a web page through operating a computer. Therefore, these real-name user characteristics can be regarded as an important set of real-name user characteristics for high-value user groups.

For example, one of the real-name user data may be "ID: 0001; 2020-9-1 10: 00: 05; login homepage; device=iphone12", "ID: 0001; 2020-9-1 10:00:15; leave the homepage; device=iphone12", "ID: 0001; 2020-9-5 14:08:30; log in to the homepage; device=iphone12", "ID: 0001; 2020-9-5 14: 08:30; Consumption; Consumption amount=1,000; device=iphone12” and “ID: 0001; 2020-9-1 14:08:50; Leave homepage; device=iphone12”. Through the process S62, the real-name user grouping module 51 can obtain the characteristics of multiple real-name users of “ID: 0001” as “weekly login frequency=2”, “average stay time=15”, “weekly homepage stay times= 2" and "Weekly Consumption = 1,000".

After the real-name user data of each real-name user is converted into features, and each real-name user is grouped (after the process S62 is executed), in the process S63, a machine learning algorithm or a statistical algorithm can be used to perform feature engineering, extract and Identify which real-name user characteristics are important and meaningful to a specific user group (eg high-value user group). Usually, feature selection can be performed, and the first few real-name user features are selected according to the importance score as the important real-name user feature sets of a specific user group, and the important real-name user feature sets of these specific user groups are important and meaningful to the specific user group. Characteristics. Taking the above example as an example, through feature engineering, it can be found that the real-name user characteristics that are important and meaningful to the high-value user group are the weekly login frequency, average stay time, and weekly consumption amount. For real-name users, the weekly login frequency is greater than 2, the average stay time is greater than 10, and the weekly consumption amount is greater than 100. Real-name user features that are important and meaningful to high-value user groups can be used in the subsequent process S66.

In the process S64, the anonymous user feature acquisition module 53 executes the anonymous user data collection step to collect anonymous user data of each anonymous user. Then, in the process S65, the anonymous user feature acquisition module 53 processes the anonymous user data to obtain the anonymous user feature, wherein the anonymous user feature is obtained by processing the anonymous user data according to the system operation result or the rules defined by humans. The characteristics of anonymous users are, for example, the weekly login frequency, the average stay time or the number of times of staying on the home page per week, and this model is not limited by this. User characteristics. Some of the real-name user characteristics cannot be obtained through anonymous user data. For example, the aforementioned weekly consumption amount is a real-name user characteristic, but the anonymous user characteristic does not include the weekly consumption amount.

After obtaining the important real-name user feature set of the specific user group and the anonymous user feature of each anonymous user, in the process S66, the anonymous/real-name user feature mapping module 54 finds out the important real-name user feature set in the specific user group. What are the real-name user characteristics that can be obtained through anonymous user data, and take the anonymous user characteristics corresponding to these real-name user characteristics as the anonymous user characteristics of an important anonymous user characteristic set in a specific user group (for example, by comparing anonymous user characteristics) name and the name of the real-name user feature). For example, for real-name users in the high-value user group, the weekly login frequency is greater than 2, the average stay time is greater than 10, and the weekly consumption amount is greater than 100. However, only the weekly login frequency and average amount can be obtained through anonymous user data. The anonymous user characteristics of the stay time, therefore, the anonymous user's weekly login frequency and the average stay time of the anonymous user characteristics can be used to determine whether the anonymous user also belongs to the high-value user group.

Preferably, in the process S66, the anonymous user features of the important anonymous user feature set can be used for a model training, so as to establish a grouping model for judging whether the anonymous user belongs to a specific user group. To put it simply, the process S66 is mainly to find out the important anonymous user feature set that is important and meaningful for a specific user group and can be obtained through anonymous user data, and thereby trains a set of anonymous user features to be used according to the anonymous user data. The anonymous user characteristics obtained from the data are used to determine whether an anonymous user belongs to a specific user group.

Please note here that in the process S66, it is also possible to directly apply the grouping model of real-name users without additional training of the grouping model. The acquirer) can do feature filling. For example, set the weekly consumption amount to more than 100, so that the real-name user's points can be directly applied The group model groups anonymous users to determine whether anonymous users belong to high-value user groups. However, this new model is not limited by this. Another method is to directly compare the features of the anonymous user data with the approximate part of the real-name user data (for example, calculate the cosine similarity), and directly compare the approximate part according to the feature comparison result. The corresponding real-name user feature is given to anonymous users as anonymous user features, so as to further judge whether the anonymous user belongs to a high-value user group. Taking Figure 1 as an example, if the feature comparison result between the anonymous user data and the approximate part of the real-name user data is highly consistent, the approximate part of the real-name user data, for example, the corresponding real-name user feature 33N (which cannot be obtained from the anonymous user data) can be compared. user), directly set the value and provide it to the anonymous user as an anonymous user feature.

To put it simply, the method of how to perform anonymous/real-name user feature mapping in the process S66 is not intended to limit the present invention. The important anonymous user feature set of the specific user group obtained in the process S66 may be the same as or different from the dimension of the important real-name user feature set of the specific user group, and may be less than or equal to the important real-name user feature set. . Finally, in the process S67, the anonymous user grouping module 55 performs the anonymous user grouping step, which determines whether the anonymous user belongs to a specific user group according to the anonymous user characteristics and the grouping model of the aforementioned anonymous user. If it is determined that the anonymous user belongs to a specific user group, accurate marketing can be performed on the anonymous user, such as placing advertisements, discount coupons, discounts or promotions that are interested in the anonymous user. The method of grouping anonymous users can be achieved by using the grouping model established by machine learning algorithms as described above, or by calculating the similarity (for example, cosine similarity) between anonymous user characteristics and real-name user characteristics through a feature comparison algorithm. to determine whether an anonymous user belongs to a specific user group.

For example, in the process S66, the important real-name user feature set of the high-priced user group includes the weekly login frequency, the average stay time and the weekly consumption amount, and the weekly login frequency is greater than 2, the average stay time and the weekly consumption amount is greater than 15. Real-name users with a weekly consumption amount greater than 100 belong to a high-value user group, but because the weekly consumption amount cannot be obtained from anonymous user data, Therefore, the important set of anonymous user characteristics of the high-priced user group includes the weekly log-in frequency and the average dwell time.

Assume that the anonymous user characteristics of multiple anonymous users may be "ID: COOK0001; weekly login frequency=2; average stay time=15", "ID: COOK0002; weekly login frequency=0.5; average stay time=7", "ID: COOK0003; Weekly login frequency=1; Average stay time=50", "ID: COOK0004; Weekly login frequency=3; Average stay time=25" and "ID: COOK0005; Weekly login frequency=10; Average dwell time = 50". Then in the process S67, the weekly login frequency and average stay time of the anonymous user feature can be taken from the weekly login frequency and average stay time of the real-name user feature to calculate the cosine similarity. If in this embodiment, the cosine similarity The higher ones are the anonymous users of "ID: COOK0004" and "ID: COOK0005", so the anonymous users of "ID: COOK0004" and "ID: COOK0005" can be determined to belong to the high-value user group.

In addition, in this novel embodiment, each specific user group may further have two grouping models. Although the clustering model can be updated at regular intervals, if you do not want to update frequently, you can design one of the clustering models without considering any features related to the event, and the other clustering model considering the features related to the holding of the event. For example, if an animation festival is held, the user group may be mostly students. Therefore, through the Internet address in the anonymous user data, it can be known whether the anonymous user is from the school, and if so, the anonymous user is identified as The anonymous user characteristics of students whose occupation is a student are important and meaningful for this clustering model.

In some embodiments, the device of the present invention can execute the module or method of the present disclosure by a computer software program, wherein the computer software program is stored in a non-volatile storage medium. The non-volatile storage medium is used to determine the anonymous user data as belonging to the real-name user data set. The computer software program of the cluster, and after loading the computer software program through the computer, executes the following steps: obtaining and storing a first real-name user data set, marking the first real-name user data set as the first user cluster, and the first real-name user data set. The real-name user data set includes a plurality of real-name user original data; feature extraction is performed on the first real-name user data set to generate a first real-name user feature set from the first real-name user data set, wherein the first real-name user data set is The feature extraction includes performing feature importance analysis on the first real-name user data set with a machine learning algorithm to generate the first real-name user feature set; obtaining and storing a first anonymous user data set, the first anonymous user data The set contains a plurality of anonymous user original data, and based on the first real-name user feature set, feature extraction is performed on the first anonymous user data set to generate a first anonymous user feature set, and the first anonymous user feature set is used as the first anonymous user feature set. A subset of the real-name user feature set; feature selection is performed on the first anonymous user data set based on the first anonymous user feature set to generate a second anonymous user data set with the first anonymous user feature set; using the machine A learning algorithm fits the second anonymous user data set to generate an anonymous user grouping model; and uses the anonymous user grouping model to predict a third anonymous user data set, the third anonymous user data set includes at least one For anonymous user data, the anonymous user data predicted to be positive in the third anonymous user data set are marked as belonging to the first user cluster.

In some embodiments, the device of the present invention can obtain and store the first real-name user data set and the second real-name user data set by a computer program. The first real-name user data set and the second real-name user data set can reflect records of interactions between a plurality of real-name users and a software service, and the software service can be executed by a server. The first real-name user data set and the second real-name user data set may include a first periodic data set and a second periodic data set of a plurality of real-name users corresponding to the first real-name user data set and the second real-name user data set, respectively Sexual Data Set. The computer program can perform feature extraction on the first real-name user data set and the second real-name user data set to obtain the first periodic data set and the second periodic data set. the computer program can Feature engineering is performed on the real-name user data set and the second real-name user data set respectively, so as to obtain a first feature data set and a second feature data set corresponding to the first real-name user data set and the second real-name user data respectively. During the feature engineering stage, the computer program can use a first machine learning algorithm to fit the first real-name user data set and the second real-name user data, respectively, to generate the first feature data set and the Second feature dataset. For the first real-name user data set and the second real-name user data, the computer program can determine the importance of the feature data by calculating or finding the feature importance, and perform feature selection to generate the first feature data set and the second feature data set. The first feature data set may include a third periodic data set, and the third periodic data set may be a subset of the first periodic data set. The second feature data set may include a fourth periodic data set, and the fourth periodic data set may be a subset of the second periodic data set. The computer program can obtain and store an anonymous user data set, and the anonymous user data set can include the data of a plurality of anonymous users. The anonymous user data set can reflect the records of a plurality of anonymous users interacting with the software service. The computer program can perform feature comparison of the anonymous user data set with the first feature data set to generate a first anonymous feature data set. The first anonymous feature data set may include a plurality of anonymous features, and the anonymous features may be the same as some features of the first feature data set. The computer program can use the second machine learning algorithm to fit the first anonymous feature data set to generate a first anonymous user grouping model. The second machine learning algorithm may be the same as the first machine learning algorithm. The computer program can perform a feature comparison between the anonymous user data set and the second feature data set to generate a second anonymous feature data set. The second anonymous feature data set may include a plurality of anonymous features, and the anonymous features may be the same as some features of the second feature data set. The computer program can use the third machine learning algorithm to fit the second anonymous feature data set to generate a second anonymous user grouping model. The third machine learning algorithm may be the same as the first machine learning algorithm and/or the second machine learning algorithm. The computer program can use the first anonymous user grouping model and the second anonymous user grouping model to The anonymous user data set is predicted to obtain a first prediction result and a second prediction result respectively corresponding to the two anonymous user grouping models. The computer program can determine that the anonymous user data set is similar to the first anonymous user similar data set of the first real-name user data set according to the first prediction result, and can determine that the anonymous user data set is similar to the first real-name user data set according to the second prediction result. The second anonymous user similar data set of the second real-name user data set. In some examples, the computer program can compare the data quantity of the first anonymous user-like data set and the data quantity of the second anonymous user-like data set to determine which anonymous user-like data set to select for subsequent processing or application. For example, if the amount of data of the second anonymous user's similar data set is large, the computer program can select the second anonymous user's similar data set. If the aforementioned high-value user is taken as an example, the computer program can select the second anonymous user's similar data set. Similar data sets are judged to contain anonymous user data belonging to high-value user groups.

In some embodiments, the computer program can mark the real-name user data set as belonging to a particular user group. Continuing the above example of high-value users, the computer program can mark the first real-name user data set as a high-value user group data set, and the computer program can optionally mark the second real-name user data set as another Data sets for specific user groups, such as data sets for specific groups in RFM clusters. The computer program can mark the first real-name user data set and the second real-name user data set according to the existing grouping result, so that the first real-name user data set and the second real-name user data set belong to different user groups. In some examples, the computer program performs group prediction on the new anonymous user data using the first anonymous user grouping model and/or the second anonymous user grouping model, and according to the judgment of the above prediction results, the new anonymous user Profiles are marked as specific user groups corresponding to their similar identifiable user profiles. For example, the new anonymous user profile may be a single anonymous user profile. In another example, the computer program can predict new anonymous user data in real-time, near real-time and batches. For another example, the aforementioned batch prediction can be performed after accumulating data of a single anonymous user into several transactions.

In some embodiments, the first feature data set and the second feature data set may include a plurality of feature fields with different data attributes, each of the feature fields includes a plurality of feature values, and each of the feature fields has a plurality of The eigenvalues can constitute a eigenvalue range. The computer program may calculate or select a characteristic threshold value according to the characteristic value range of each of the characteristic fields. The data of the aforementioned feature data set may include feature threshold values. Each of the characteristic thresholds can be the maximum value, the minimum value, the mean value, the median, the mode, the percentile, the quartile, the standard deviation, and the like in the characteristic value range of each of the characteristic fields. According to the attributes of different characteristic values, the computer program can use different characteristic threshold value judgment rules to determine the characteristic threshold value. For example, the computer program can determine the correlation between the eigenvalue and the target variable or label fitted by the first machine learning algorithm, and if it is determined that the correlation is a positive correlation, it can calculate or select the minimum value in the range of eigenvalues as The characteristic threshold value, and if the correlation is judged to be negative correlation, the maximum value in the characteristic value range can be calculated or selected as the characteristic threshold value. Taking the feature selection of the aforementioned high-value users as an example, if the target variable is the consumption amount, the feature is the weekly login frequency, and the minimum value of the characteristic value range is 2, the computer program can determine that the consumption amount and the weekly login frequency have a positive correlation. Correlation, the minimum value can be selected as the characteristic threshold value. In some instances, if there is a weak correlation between the target variable and the feature, the computer program may calculate or select the arithmetic mean of the range of feature values as the feature threshold. In some instances, if the feature is a count attribute, the computer program may calculate or select the mode in the range of feature values as the feature threshold. In some examples, the computer program may set a threshold coefficient h=0, 0.5, 1, 1.8, 2, . For example, if h=0.5 is set, the characteristic threshold of the aforementioned weekly login frequency is 1. The computer program can set the threshold coefficient to adjust the similarity between anonymous user characteristics and real-name user characteristics, that is, to adjust the difference between the anonymous user data set and the first real-name user data set and/or the second real-name user data set Similarity, to determine the similar information of the first anonymous user and/or the similar information of the second anonymous user material. In some examples, the computer program can receive a setting command of a characteristic threshold, and set the characteristic threshold according to the setting command.

In some embodiments, when the computer program determines that the anonymous user data set is similar to the anonymous user similar data set of the real-name user data set according to the prediction result, the computer program can determine that the feature data set has at least The characteristic threshold value of a characteristic field is used as the basis for judging the similarity between the anonymous user data set and the real-name user data set. In some embodiments, when the computer program performs the feature selection, the feature threshold value of at least one feature field in the feature data set can be determined according to the aforementioned feature threshold value judgment rule, which is used as the basis for the anonymous user data set and the first feature field. 1. The basis for feature comparison of feature data sets. Taking the above-mentioned high-value users as an example, if the characteristic threshold of the weekly login frequency is 2, in the anonymous user characteristics of the data "ID: COOK0001" to "ID: COOK0005", the computer program can determine that the weekly login frequency is greater than The data of characteristic threshold values are "ID: COOK0004" and "ID: COOK0005". It should be noted that, under normal circumstances, the aforementioned anonymous feature data set will contain multiple features, so the anonymous feature data set may be an anonymous feature union. When generating the aforementioned anonymous user similar data set or the aforementioned feature comparison, the The data satisfying all the feature thresholds of the anonymous feature set is used as the condition for similarity judgment or feature comparison.

In some embodiments, the computer program can calculate the aforementioned feature importance using impurity models, entropy models, filtering, packing, embedding, other statistical methods, and the like. For example, the computer program can use the feature_importances_ property of the scikit-learn machine learning library to derive important features. In some embodiments, the aforementioned machine learning algorithms, machine learning models, or anonymous user grouping models are used in a classification manner recognized by those of ordinary skill in the art. For example, machine learning algorithms or models may include regression models and classification models.

In some embodiments, the aforementioned periodic data set may include data generated by iterative computations of the computer program for different periods. Different periods can include seconds, minutes, hours, A cycle in units of days, weeks, months, quarters, years, etc. Different periods can be obtained by time series analysis such as seasonal analysis. Periodic and/or time series analysis may be performed by a third party system or service, and the computer program may obtain different periodic, periodic data, or time series analysis results from the third party system or service.

Accordingly, the new model mainly uses the real-name user profiles of real-name users to obtain an important set of real-name user characteristics of a specific user group, and finds out the anonymous user profiles of anonymous users to find out whether there are corresponding anonymous user characteristics to determine whether the anonymous user belongs to Specific user groups, in order to further achieve the purpose of accurate marketing for anonymous users.

The above descriptions such as "the present new type..." or "the present new type is mainly...", the descriptions in the embodiments, the modifications and the contents disclosed in the drawings are only used to describe one or more of the new types described in the claims. The examples or implementation features only express the principle or spirit of the present invention, and do not limit the present invention, nor are they the only or necessary implementations of the present invention, and according to the spirit of the present invention, there may also be various types including different features or implementations. Therefore, the new type described in the claim is not limited by the content disclosed in the above-mentioned "this new type..." description, embodiments or drawings. The description in the first claim of the present application is merely an example, and the description in the claim can be appropriately changed according to the description in the specification, drawings, and the like.

5: User management device

51: Real-name user grouping module

52: Real-name user feature engineering module

53: Anonymous user feature acquisition module

54: Anonymous/real-name user feature mapping module

55: Anonymous user grouping module

Claims (10)

A user management device capable of judging whether an anonymous user belongs to a specific user group, comprising a plurality of hardware circuits that are electrically connected to each other, and configured with a real-name user feature engineering module, an anonymous user feature acquisition module, an anonymous/ A real-name user feature mapping module and an anonymous user grouping module, and the anonymous/real-name user feature mapping module is signal-connected to the real-name user feature engineering module, the anonymous user feature acquisition module, and the anonymous user grouping module. group, and the anonymous user feature acquisition module signal is connected to the anonymous user grouping module, wherein: the real-name user feature engineering module for a first specific user group of a plurality of real-name users, through a feature engineering acquisition For the first specific user group, multiple real-name user features whose importance is greater than a threshold or whose importance is the top few are taken as the majority of the first important real-name user feature set of one of the first specific user groups. a first real-name user feature, wherein the feature engineering is to find a plurality of first real-names of the first important real-name user feature set of the first specific user group based on a machine learning algorithm through the user management device User characteristics; the anonymous user characteristic acquisition module maps one of the first specific user groups according to the plurality of first real-name user characteristics in the first important real-name user characteristic set of the first specific user group A plurality of first anonymous user characteristics of the first important anonymous user characteristic set, wherein at least a part of the plurality of first anonymous user characteristics is obtained from an anonymous data of an anonymous user; and the anonymous user grouping module At least a part of the plurality of first anonymous user characteristics obtained from the anonymous data of the anonymous user is used to determine whether the anonymous user belongs to the first specific user group. The user management device capable of judging whether an anonymous user belongs to a specific user group as described in claim 1, wherein a plurality of hardware circuits electrically connected to each other are further configured with a real-name user grouping module, wherein: The real-name user grouping module uses multiple real-name user profiles of the multiple real-name users to group the multiple real-name users, wherein at least a part of the multiple real-name users belong to the first specific user group. The user management device capable of judging whether an anonymous user belongs to a specific user group according to claim 1, wherein an RFM model (Recency Frequency Monetary Model) is used to analyze the real-name user profiles of the real-name users for the real-name users. Grouping multiple real-name users; or, grouping the multiple real-name users according to multiple static characteristics and/or multiple dynamic characteristics obtained from the multiple real-name user profiles of the multiple real-name users; or, using A K-MEANS algorithm, a support vector machine algorithm and/or a machine learning algorithm group the multiple real-name user groups according to the multiple real-name user profiles. The user management device capable of determining whether an anonymous user belongs to a specific user group according to claim 1, wherein the user management device collects the plurality of real-name user data of a plurality of real-name users, and collects the real-name user data according to the plurality of real-name user data Acquire the multiple real-name user characteristics of the multiple real-name users. The user management device capable of judging whether an anonymous user belongs to a specific user group according to claim 1, wherein a person who can be acquired through the anonymous data is found from the characteristics of the plurality of first real-name users as all the plurality of users. The first anonymous user feature. The user management device capable of judging whether an anonymous user belongs to a specific user group according to claim 1, wherein the multiple first real-name user characteristics that can be acquired through the anonymous data as a part are found out. the first anonymous user characteristics, and the multiple first real-name user characteristics and their numerical values that cannot be obtained through the anonymous data acquirer among the multiple first real-name user characteristics as the other multiple first anonymous users features and their values. The user management device capable of judging whether an anonymous user belongs to a specific user group according to claim 1, wherein the anonymous user is trained through the plurality of first anonymous user characteristics and the grouping model is used to judge whether the anonymous user belongs to the first specific user according to at least a part of the characteristics of the plurality of first anonymous users obtained from the anonymous data of the anonymous user ethnic groups. The user management device capable of judging whether an anonymous user belongs to a specific user group according to claim 1, wherein the multiple first anonymous user characteristics of the anonymous user and the corresponding multiple first real-name user characteristics are performed. A feature comparison is used to determine whether the anonymous user belongs to the first specific user group. The user management device capable of judging whether an anonymous user belongs to a specific user group according to claim 8, wherein the feature comparison is to calculate a cosine similarity. A user management device capable of judging whether an anonymous user belongs to a specific user group, comprising a plurality of hardware circuits electrically connected to each other, the plurality of hardware circuits being used to read a computer program to configure a real-name user feature An engineering module, an anonymous user feature acquisition module, an anonymous/real-name user feature mapping module, and an anonymous user grouping module, the anonymous/real-name user feature mapping module signals are connected to the real-name user feature engineering module, The anonymous user feature acquisition module and the anonymous user grouping module, and the anonymous user feature acquisition module signal is connected to the anonymous user grouping module, the real-name user feature engineering module, the anonymous user feature The acquisition module, the anonymous/real-name user feature mapping module and the anonymous user grouping module are used to perform the following steps: acquiring and storing a first real-name user data set, and marking the first real-name user data set as the first User cluster, the first real-name user data set includes a plurality of real-name user original data; feature extraction is performed on the first real-name user data set to generate a first real-name user feature set from the first real-name user data set, wherein the The feature extraction of the first real-name user data set includes performing feature importance analysis on the first real-name user data set with a machine learning algorithm to generate the first real-name user feature set; Obtain and store a first anonymous user data set, the first anonymous user data set includes a plurality of anonymous user original data, and feature extraction is performed on the first anonymous user data set based on the first real-name user feature set to generate a first anonymous user data set User feature set, so that the first anonymous user feature set is a subset of the first real-name user feature set; based on the first anonymous user feature set, feature selection is performed on the first anonymous user data set to generate a feature set with the first anonymous user feature set. A second anonymous user data set of an anonymous user feature set; using the machine learning algorithm to fit the second anonymous user data set to generate an anonymous user grouping model; The user data set is predicted, the third anonymous user data set includes at least one anonymous user data, and the anonymous user data predicted as positive examples in the third anonymous user data set is marked as belonging to the first user cluster.

Images