JP7175006B2 - Information processing device and information processing method - Google Patents

Description

The present invention relates to an information processing apparatus and information processing method for identifying a terminal based on terminal information acquired from the terminal.

Conventionally, a method using cookies is widely known as a method of identifying terminals (so-called tracking) for terminals connected to a network, but there are cases where cookies cannot be used, cases where the use of cookies is not desired, etc. Based on the existence of , various technologies have been proposed to identify terminals without using cookies. For example, Patent Literature 1 describes a technique for identifying a terminal by using user's login information (for example, account given by SNS). In addition, technology that generates identification information from a combination of elements (parameters) that can be obtained from the terminal such as IP address, screen resolution, and number of CPU cores, and identifies the terminal using the generated identification information (so-called device finger) printing technology) is widely used.

JP 2017-16544 A

However, with the technique described in Patent Document 1, only users who have logged in for some service are identified (tracked), and there is a problem that identifiable users are limited. On the other hand, in conventional device fingerprinting technology, identification information is generally generated based on a group of elements including elements whose values change over time, such as IP addresses. As described above, there is a problem that reliability can only be guaranteed for a very short period of time.

The present invention was made to solve such problems, and it is possible to identify terminals with high reliability for as long as possible without using cookies and without limiting identifiable terminals. The purpose is to

In order to solve the above-described problems, in the present invention, first terminal information of a first terminal acquired at a first time point and second terminal information acquired at a second time point after the first time point are provided. When determining the possibility that the first terminal and the second terminal are the same based on the second terminal information of the terminal, the fixed parameter included in the first terminal information and the second terminal information In addition to the consistency with the included fixed parameters, the time change indicating the change from the first time point to the second time point, and the variable parameter included in the second terminal information from the variable parameter included in the first terminal information The possibility of identity is determined by reflecting a variable parameter change that indicates a change to.

The terminal information that can be obtained from the terminal includes a fixed parameter group consisting of fixed parameters whose values do not change with the passage of time, and a variable parameter group consisting of variable parameters whose values can change with the passage of time. There is a group. Since the values of fixed parameters do not change over time, if one fixed parameter group and another fixed parameter group have different values of a common type of fixed parameter, the terminal associated with the one fixed parameter group and the terminal associated with the other fixed parameter group are not the same. In addition, variable parameters are characterized in that there is a tendency in the mode of change of variable parameters with respect to changes in time for each type of variable parameter.

Based on this, according to the present invention configured as described above, when determining the possibility that the first terminal and the second terminal are the same, the fixed parameter group related to the first terminal and the second Consistency with the fixed parameter group related to the terminal is reflected. For this reason, by suitably utilizing the above-described characteristics of the fixed parameters, it is possible to reduce the possibility that two terminals that can be said to be definitely different from each other in view of the characteristics of the fixed parameters will be judged to be the same. It is possible to improve the reliability of judgment results.

Furthermore, according to the present invention, when determining the possibility that the first terminal and the second terminal are the same, the time change indicating the change from the first time point to the second time point and the first terminal information A change from the variable parameter included in the second terminal information to the variable parameter included in the second terminal information is reflected. Therefore, it is possible to determine the possibility of identity by taking into consideration the tendency of the aspect of change of the variable parameter with respect to time change. Even if the variable parameter changes over time, the possibility of identity can be determined with higher reliability than when such trends are not considered.

As described above, according to the present invention, terminals can be identified with high reliability for as long as possible without using cookies and without limiting identifiable terminals. Determining whether the terminal associated with the first terminal information and the terminal associated with the second terminal information are the same means that one terminal is identified from the viewpoint of whether it is the same as the other terminal. That's what it means.

1 is a block diagram showing a functional configuration example of an information processing system according to a first embodiment of the present invention; FIG. It is a list of examples of fixed parameters. It is a list of examples of variable parameters. FIG. 4 is a diagram showing the contents of records in a user management database; FIG. It is a figure which shows the relationship between the input of exclusive model M1, and an output. It is a figure which shows the calculation formula of a final evaluation value. FIG. 10 is a diagram showing a neural network of an individual variable para-evaluation value output model; FIG. 10 is a diagram showing an example of transition of the value of OSver over time; FIG. 9 is a diagram showing an example of transition of variable parameter change values with respect to time change values for OSver; FIG. 10 is a diagram showing a neural network of a weighted output model; FIG. 4 is a diagram showing the contents of teacher data; FIG. 10 is a diagram showing contents of terminal information at login and teacher data having a normal label; 7 is a flow chart showing the operation of the same possibility determination unit of the information processing apparatus according to the first embodiment of the present invention; It is a block diagram which shows the functional structural example of the information processing system which concerns on 2nd Embodiment of this invention. FIG. 10 is a diagram showing the relationship between the input and output of a dedicated model M1 according to modification 2, and the content of teacher data; FIG. 10 is a diagram showing a neural network of an individually variable para-evaluation value output model according to modified examples 3 and 4; FIG. 13 is a diagram showing a neural network of a weighted output model according to modification 6; FIG. 20 is a diagram showing a neural network of an individually variable para-evaluation value output model according to Modification 7; FIG. 14 is a diagram showing the relationship between the input and output of a dedicated model M1 according to Modification 7, and the content of teacher data;

<First embodiment>
A first embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a functional configuration example of an information processing system 1 according to this embodiment. As shown in FIG. 1, an information processing system 1 according to the present embodiment includes an information processing server 2 (corresponding to an "information processing apparatus" in claims). This information processing server 2 is connected to a network N including the Internet, a telephone network, and other communication networks, and a user terminal 3 and a site providing server 4 are connected to this network N.

The user terminal 3 is a terminal owned by any individual. The user terminal 3 may be a device in which a web browser is installed. As an example, the user terminal 3 may be a desktop computer, a notebook computer, a tablet computer, a mobile phone (a so-called smart phone or a mobile phone other than a smart phone). (including telephone), etc. FIG. 1 exemplarily illustrates one of the innumerable user terminals 3 .

The site providing server 4 is a web server on which a predetermined website accessible from a web browser is established. In this embodiment, the web page provided by the site providing server 4 in response to access from the user terminal 3 displays advertisement information (hereinafter referred to as "advertisement information") provided by the information processing server 2 by a method described later. contains a column that FIG. 1 exemplarily illustrates one of the innumerable site providing servers 4 .

As will be described in detail later, when the user terminal 3 accesses the site of the site providing server 4, the information processing server 2 identifies the user terminal 3 and presents an advertisement based on the preferences of the user associated with the user terminal 3. provide information. The advertisement information provided by the information processing server 2 is displayed in a predetermined column of the web page provided by the site providing server 4 . The information processing server 2 is operated by a predetermined business operator (hereinafter referred to as "service provider"). Although the information processing server 2 is represented by one block in FIG. 1, this does not mean that the information processing server 2 is composed of a single server device. The information processing server 2 may be composed of a plurality of devices, or may be part of a predetermined system.

As shown in FIG. 1, the information processing server 2 includes a server-side communication unit 6, an advertisement providing unit 7, and a similarity possibility determination unit 8 as functional configurations. In addition, the user terminal 3 has a user terminal side communication unit 9 and a web browser execution unit 10 as a functional configuration, and the web browser execution unit 10 has an integrated control unit 11 and a terminal information acquisition unit 12. . Each of the functional blocks 6 to 12 can be configured by hardware, DSP (Digital Signal Processor), or software. For example, when configured by software, each functional block 6 to 12 is actually configured with a computer CPU, RAM, ROM, etc., and a program stored in a recording medium such as RAM, ROM, hard disk, or semiconductor memory. is realized by the operation of This also applies to each functional block described in the second embodiment.

The information processing server 2 includes a server-side storage unit 13 as a storage medium. The server-side storage unit 13 stores a user management database 14, which will be described later. It should be noted that part or all of the data stored in the server-side storage unit 13 may be stored in an external device capable of communicating with the information processing server 2 .

The web browser execution unit 10 of the user terminal 3 executes various processes by having hardware such as a CPU read and execute web browsers and programs attached to web browsers (part of programs of OS, APIs, etc.). do. The user terminal side communication unit 9 of the user terminal 3 and the server side communication unit 6 of the information processing server 2 access the network N according to a predetermined communication standard and communicate with devices connected to the network N. FIG.

Processing of the user terminal 3, the site providing server 4, and the information processing server 2 when and after the web browser of the user terminal 3 accesses the site of the site providing server 4 will be described below.

The user who owns the user terminal 3 activates the web browser of the user terminal 3, instructs access to the URL corresponding to the site of the site providing server 4, and causes the web browser to access the URL. Hereinafter, for convenience of explanation, the user terminal 3 that has accessed the site providing server 4 will be referred to as a “processing target user terminal” and will be distinguished from other user terminals 3 . In response to this, the central control unit 11 of the web browser execution unit 10 of the processing target user terminal transmits an HTTP request to the site providing server 4 . The site providing server 4 responds with an HTTP response in response to the access to the URL (HTTP request). The HTML file included in the HTTP response incorporates a predetermined script written in a predetermined script language (hereinafter referred to as "dedicated script"). This dedicated script was developed and provided by a service provider.

The general control unit 11 of the web browser execution unit 10 of the processing target user terminal receives the HTTP response. The terminal information acquisition unit 12 of the web browser execution unit 10 executes the following processes by the function of the dedicated script (including accompanying programs) embedded in the HTTP response.

That is, the terminal information acquisition unit 12 obtains information regarding the hardware configuration/settings of the processing target user terminal, software configuration/settings, information regarding the network configuration/settings, and other information regarding the processing target user terminal, get what is specified. Hereinafter, a set of information acquired by the terminal information acquisition unit 12 will be referred to as "terminal information", and individual pieces of information classified by type will be referred to as "parameters". The parameters acquired by the terminal information acquisition unit 12 are classified into fixed parameters and variable parameters. Hereinafter, the set of all fixed parameters is called "fixed parameter group", and the set of all variable parameters is called "variable parameter group". That is, the terminal information is composed of a fixed parameter group and a variable parameter group, the fixed parameter group is composed of fixed parameters, and the variable parameter group is composed of variable parameters.

A fixed parameter is a parameter whose value does not change over time. FIG. 2 shows an example of a list of fixed parameters. When the value of a parameter does not change with the passage of time, it does not only mean that the value does not change at all no matter how much time passes. (e.g. the value changes only in very specific cases), meaning that in most cases the value does not change over time. For example, referring to the first row of the list in FIG. 2, the classification of devices such as mobile/desktop does not change over time, so the device type is a fixed parameter. Also, for example, referring to the third line of the list in FIG. 2, the time zone of the user system is basically changed only when the user moves from the original location to a different location across countries and regions. Therefore, the time zone as a parameter is a fixed parameter.

A variable parameter is a parameter whose value can change over time. FIG. 3 shows an example of variable parameters in the form of a list. For example, referring to the first line of the list in FIG. 3, the global IP address is currently not unique to each user terminal 3 in almost all cases, and changes each time. Address is a variable parameter. Also, for example, referring to the second line of the list in FIG. 3, the OSver as a parameter is a variable parameter because the version of the operating system changes from time to time according to updates that are performed each time.

After the terminal information acquisition unit 12 acquires the terminal information, the overall control unit 11 transmits an HTTP request requesting advertisement information to the information processing server 2 . At that time, the central control unit 11 acquires the terminal information acquired by the terminal information acquisition unit 12, the terminal information acquisition date and time information indicating the date and time when the terminal information was acquired by the terminal information acquisition unit 12, and the site accessed by the web browser. The URL of the providing server 4 (hereinafter referred to as "site URL") is included in the HTTP request. The advertisement provision unit 7 of the information processing server 2 receives the HTTP request, outputs the terminal information, the terminal information acquisition date and time information, and the site URL included in the HTTP request to the identity possibility determination unit 8, and determines the unique ID (described later). Request a response.

The same possibility determination unit 8 receives terminal information, terminal information acquisition date and time information, and site URL from the advertisement providing unit 7, and executes the following processing in response to a unique ID response request. That is, the identity possibility determination unit 8 acquires a fixed parameter group included in the terminal information. Next, the same possibility determination unit 8 generates a pre-hash fixed parameter-derived ID by arranging each of the fixed parameters according to a rule. Any method for generating a pre-hash fixed parameter-derived ID can be any method as long as a pre-hash fixed parameter-derived ID with a common value is always generated when the fixed parameter group has a common value. There may be. Next, the same possibility determination unit 8 uses a hash function prepared in advance to generate a hash value of the pre-hash fixed parameter derived ID, and uses this as a fixed parameter derived ID (hereinafter abbreviated as “fixed parameter ID”). . Next, the possibility-of-sameness determination unit 8 refers to the user management database 14 stored in the server-side storage unit 13 .

FIG. 4 is a diagram schematically showing the contents of one record in the user management database 14. As shown in FIG. As will become clear later, when a certain user terminal 3 accesses the site providing server 4 (more precisely, any of the plurality of site providing servers 4), the user management database 14 stores information about the user terminal 3. When the possibility-of-identity determination unit 8 determines that the corresponding record is not registered, one record (the record corresponding to the user terminal 3) is newly registered in the user management database 14. FIG. On the other hand, if the possibility of identity determination unit 8 determines that a record corresponding to the user terminal 3 has already been registered in the user management database 14, the record corresponding to the user terminal 3 in the user management database 14 is updated. be done. Therefore, ideally, one record is registered in the user management database 14 for each user terminal 3 that has accessed the site providing server 4 in the past.

As shown in FIG. 4, one record in the user management database 14 has fields in which a unique ID, a fixed parameter ID, terminal information acquisition date and time information, a fixed parameter group, a variable parameter group, and history information are stored. . A unique ID is a unique value for each record and functions as identification information for identifying the user terminal 3 . The history information is information in which URLs of sites accessed by the user terminal 3 in the past are accumulated.

The possibility-of-identification determination unit 8 refers to the user management database 14 stored in the server-side storage unit 13, and selects a fixed parameter having the same value as the generated fixed parameter ID (=fixed parameter ID of the user terminal to be processed). It is determined whether (1) there is no record having the ID, (2) there is one record, and (3) there are two or more records. Hereinafter, a record having a fixed parameter ID having the same value as the value of the generated fixed parameter ID (=fixed parameter ID of the user terminal to be processed) will be referred to as a "candidate record" as appropriate.

If there are no candidate records, the following conditions apply. That is, as described above, fixed parameters do not change in value over time. Therefore, if there is at least one fixed parameter with a different value between one fixed parameter group and another fixed parameter group, the user terminal 3 associated with the one fixed parameter group and the other fixed parameter group It can be determined that it is not the same as the user terminal 3 . Then, if there is at least one fixed parameter with a different value between the one fixed parameter group and the other fixed parameter group, the fixed parameter ID generated by hashing the one fixed parameter group and , is different from the variable parameter ID generated by hashing the other fixed parameter group. Based on this, if there is no record having a fixed parameter ID with the same value as the value of the fixed parameter ID of the user terminal to be processed, the record corresponding to the user terminal to be processed is not registered in the user management database 14. is.

Based on the above, when it is determined that there is no record having a fixed parameter ID that is the same as the value of the fixed parameter ID of the user terminal to be processed, the possibility of identity determination unit 8 determines whether the user terminal to be processed and each record It is determined that the terminal is different from the corresponding user terminal 3 and the record corresponding to the processing target user terminal is not registered in the user management database 14 . By judging in this way, it is possible to eliminate the possibility that the user terminal 3, which is certainly different from the user terminal to be processed when considering the characteristics of the fixed parameters, is judged to be the same, and the reliability of the judgment result can be eliminated. can improve sexuality. Then, the possibility-of-sameness determining unit 8 newly registers one record corresponding to the user terminal to be processed in the user management database 14 by the following method.

That is, first, the possibility-of-sameness determination unit 8 generates a unique ID. For example, the sameness possibility determination unit 8 generates data by arranging all fixed parameters included in the fixed parameter group and all variable parameters included in the variable parameter group according to rules, and uses a hash function to generate data. A hash value of data is generated and used as a unique ID. In this embodiment, the terminal information contains a sufficiently large number of fixed parameters and variable parameters, and in this way a unique ID can be generated for each record. However, as long as the uniqueness of the value can be ensured, any method may be used to generate the unique ID.

Next, the same possibility judgment unit 8 determines the generated unique ID, the generated fixed parameter ID, the input terminal information date and time information, the fixed parameter group included in the input terminal information, the variable parameter group included in the input terminal information, Then, one record stored in the field corresponding to each of the input site URLs is registered in the user management database 14 . Next, the same possibility determination unit 8 responds to the advertisement providing unit 7 with the generated unique ID.

On the other hand, when there is only one candidate record, the state is as follows. That is, at least the values of the fixed parameter group are the same for the user terminal to be processed and the user terminal 3 corresponding to the candidate record, and it is possible that the user terminal to be processed and the user terminal 3 corresponding to the candidate record are the same. It is a state of being sexual. Based on this, if there is only one candidate record, the possibility-of-sameness determination unit 8 executes the following process.

That is, the same possibility judgment unit 8 executes the following processing using the machine-learned dedicated model M1 (corresponding to the "model" in the scope of claims). The dedicated model M1 is based on the inputs described later, one terminal for which terminal information was obtained at one point in time, and another terminal for which terminal information was obtained at another point in time after the one point in time. It is a model that outputs a final evaluation value that indicates the possibility that are the same. However, the value of the fixed parameter group included in the terminal information related to the one terminal matches the value of the fixed parameter group included in the terminal information related to the other terminal. Note that matching the values of the fixed parameter group means that the values of all types of fixed parameters match. In the following description of the dedicated model M1, the two terminals to be determined as being the same or not are referred to as "one terminal" and "the other terminal", respectively, and the time point at which the terminal information relating to one terminal is acquired is , is older (=before in terms of time) than the time when the terminal information related to the other terminal was acquired. Further, the terminal information related to one terminal is called "one terminal information", and the terminal information related to the other terminal is called "other terminal information". As described above, the one terminal information and the other terminal information have the same fixed parameter group values.

FIG. 5 is a diagram conceptually showing the relationship between the input and output of the dedicated model M1. As shown in FIG. 5, the dedicated model M1 includes:1. 2. variable parameter group related to one terminal (hereinafter referred to as "one side variable parameter group"); 3. variable parameter group related to the other terminal (hereinafter referred to as "other variable parameter group"); 4. Acquisition date/time information indicating the point in time (date/time) when terminal information was acquired (hereinafter referred to as "one-side terminal information acquisition date/time information"); 4. Acquisition date and time information indicating the time (date and time) when the other terminal information was acquired (hereinafter referred to as "other terminal information acquisition date and time information"); A fixed parameter group common to one terminal information and the other terminal information (hereinafter referred to as "common fixed parameter group") is input. Also, the dedicated model M1 outputs a final evaluation value (an evaluation value indicating the possibility that one terminal and the other terminal are the same). In this embodiment, to simplify the explanation, the final evaluation value takes a value between "0" and "100", and the larger the value, the higher the possibility of being the same.

Now, when there is only one candidate record, the identity possibility determination unit 8 determines the variable parameter group (corresponding to the variable parameter group on the one hand) included in the candidate record, the variable parameter group related to the user terminal to be processed (on the other hand, the variable parameter group group), terminal information acquisition date and time information included in the candidate record (corresponding to terminal information acquisition date and time information on the one hand), terminal information acquisition date and time information input from the advertisement providing unit 7 (corresponding to terminal information acquisition date and time information on the other hand), and , the fixed parameter group (corresponding to the common fixed parameter group) related to the user terminal (or candidate record) to be processed is input to the dedicated model M1, and the final evaluation value output by the dedicated model M1 is obtained. The process of obtaining the final evaluation value using the model M1 corresponds to the process of determining the possibility of identity in claims.

Next, the possibility-of-sameness determination unit 8 determines whether or not the final evaluation value is equal to or greater than a predetermined threshold value T1. The threshold T1 is a reference value for determining whether or not one terminal and the other terminal can be regarded as identical, and is determined in advance based on tests and simulations. If the final evaluation value is less than the threshold value T1, the identity possibility determination unit 8 determines that the user terminal to be processed and the user terminal 3 corresponding to the candidate record are not the same, and correspond to the user terminal to be processed by the method described above. The record is registered in the user management database 14 and the unique ID is notified to the advertisement providing unit 7 .

On the other hand, if the final evaluation value is equal to or greater than the threshold value T1, the identity possibility determination unit 8 determines that the processing target user terminal and the user terminal 3 corresponding to the candidate record are the same (=identifying the processing target user terminal). ), perform the following processing: That is, the possibility-of-sameness determination unit 8 updates the terminal information acquisition date/time information, the fixed parameter group, and the variable parameter group of the candidate record, respectively, with the information input from the advertisement providing unit 7 . Furthermore, the same possibility judgment unit 8 additionally stores the site URL input from the advertisement providing unit 7 in the history information field. As described above, the candidate record is updated based on the latest information received from the processing target user terminal. Next, the possibility-of-sameness determination unit 8 notifies the unique ID of the candidate record to the advertisement providing unit 7 .

In this embodiment, even if there is only one candidate record, if the final evaluation value is less than the threshold value T1, the possibility-of-sameness determination unit 8 performs processing with the user terminal 3 corresponding to the candidate record. It is determined that the target user terminal is not the same. This is for the following reasons. That is, the fixed parameter group consists of only fixed parameters and does not include variable parameters, so the number of parameters included in the fixed parameter group is reduced accordingly. For example, since general fingerprint information in conventional device fingerprinting technology includes both fixed parameters and variable parameters, the number of parameters included is large compared to the fixed parameter group of this embodiment. Therefore, the probability of matching fixed parameter groups by chance for different user terminals 3 is higher than the probability of matching groups containing both fixed and variable parameters. Based on this, in the present embodiment, the above processing is performed for different terminals so as to reduce the possibility of being determined to be the same when fixed parameter groups coincidentally match. However, if the number of fixed parameters constituting the fixed parameter group is sufficiently large and it is rare for the values of the fixed parameter group to match for different user terminals 3, when there is only one candidate record, the It may be determined that the user terminal 3 corresponding to one candidate record and the user terminal to be processed are the same.

Also, when there are two or more candidate records, the state is as follows. That is, there is a possibility that each of the user terminals 3 corresponding to multiple candidate records is the same as the user terminal to be processed. Based on this, when there are two or more candidate records, the possibility-of-sameness determining unit 8 executes the following process.

That is, the identity possibility determination unit 8 uses the dedicated model M1 to acquire the final evaluation value indicating the identity possibility with the target user terminal for each of the candidate records. If all of the obtained final evaluation values are below the threshold T1, the identity possibility determination unit 8 determines that the user terminals 3 corresponding to all candidate records are not the same as the target user terminal. In this case, the possibility-of-sameness determining unit 8 registers a record corresponding to the user terminal to be processed in the user management database 14 by the method described above, and notifies the advertisement providing unit 7 of the unique ID.

If only one final evaluation value among the obtained final evaluation values is equal to or higher than the threshold T1 and the other final evaluation values are lower than the threshold T1, the possibility-of-sameness determining unit 8 selects candidates for final evaluation values equal to or higher than the threshold T1. It is determined that the user terminal 3 corresponding to the record and the user terminal to be processed are the same. In this case, the possibility-of-sameness determination unit 8 updates the candidate record with the latest information received from the processing target user terminal by the method described above, and notifies the advertisement providing unit 7 of the unique ID of the candidate record.

When two or more final evaluation values among the obtained final evaluation values are equal to or greater than the threshold T1, the possibility-of-sameness determination unit 8 selects the candidate record associated with the final evaluation value having the largest final evaluation value among the final evaluation values equal to or greater than the threshold T1. and the user terminal to be processed are the same. In this case, the possibility-of-sameness determination unit 8 updates the candidate record with the latest information received from the processing target user terminal by the method described above, and notifies the advertisement providing unit 7 of the unique ID of the candidate record. As will be described later, the dedicated model M1 is optimized by machine learning so that the final evaluation value is as large as possible when one terminal and the other terminal are the same, and the final evaluation value is small when they are not the same. be done. Therefore, ideally, when one record corresponding to the same user terminal 3 as the user terminal to be processed is registered in the user management database 14, the final evaluation value of only that record is equal to or greater than the threshold value T1. becomes.

Upon receiving notification of the unique ID from the possibility-of-sameness determination unit 8, the advertisement providing unit 7 refers to the user management database 14 stored in the server-side storage unit 13, and specifies a record corresponding to the notified unique ID. . Next, the advertisement providing unit 7 recognizes history information (one or more site URLs) of the specified record. The advertisement providing unit 7 estimates the user's preference based on one or more recognized site URLs, and generates drawing data for displaying advertisement information based on the user's preference. The process of estimating the user's preference is appropriately performed based on existing technology, and data necessary for generating drawing data is prepared in advance. The advertisement providing unit 7 responds drawing data to the processing target user terminal by HTTP response.

The general control unit 11 of the user terminal to be processed receives the drawing data, and displays the website to which the advertisement information is added based on the drawing data and the HTML file received from the site providing server 4 on the display device.

Next, the dedicated model M1 will be described. FIG. 6 is a diagram showing a formula for calculating the final evaluation value. The calculation formula in FIG. 6 is mainly intended for understanding the concept of the final evaluation value, and the contents are simplified in a manner suitable for explanation. As shown in FIG. 6, the final evaluation value is the sum of individual variable parameter evaluation values (hereinafter abbreviated as "individual variable parameter evaluation values") multiplied by individual weights. In the simplified example of FIG. 6, the sum of all weights is "1" and each individual variable para-evaluation value ranges from "0" to "100". Therefore, the final evaluation value in this example can also be said to be the weighted average value of the variable parameter individual evaluation values.

An individual variable parameter evaluation value exists for each type of variable parameter. The individual variable parameter evaluation value of one type of variable parameter is an evaluation value (evaluation) that indicates the possibility that one terminal and the other terminal are the same when focusing on the one type of variable parameter. The individual weight is a weight for adjusting the degree of influence of the evaluation of each type of variable parameter (individual variable parameter evaluation value) on the final determination of the possibility of identity (final evaluation value). Individual variable para-evaluation values and individual weights will be described in detail later. As shown in FIG. 6, the dedicated model M1 includes an individual variable parameter evaluation value output model that outputs an individual variable parameter evaluation value for each type of variable parameter. The dedicated model M1 also includes a weight output model that outputs each of the individual weights.

First, the individual variable parameter evaluation value output model will be described. In the following, the individually variable para-evaluation value output model will be described assuming that the individual variable para-evaluation value output model is a neural network. FIG. 7 is a diagram showing the neural network of the individually variable para-evaluation value output model simplified in a manner suitable for explanation. As shown in FIG. 7, the inputs of the 0th layer (input layer) of the individually variable para-evaluation value output model include time-varying values. The time change value is a value that indicates a change between the date and time indicated by the date and time of acquisition of one terminal information (hereinafter referred to as "one date and time") and the date and time indicated by the date and time of acquisition of the other terminal information (hereinafter referred to as "other date and time"). , are expressed in different ways depending on the type of variable parameter. The time change value is calculated by inputting the one terminal information acquisition date and time information and the other terminal information acquisition date and time information to the time change value output algorithm AR1.

For example, if the type of variable parameter is OSver, the time change value is a value that represents the elapsed time from one date and time to the other date and time in units of “days” (this is just an example, and it is said that it is appropriate to express it in this way. not). In this case, the time change value output algorithm AR1 related to OSver calculates the elapsed time from one date and time to the other date and time based on the inputted one terminal information acquisition date and time information and the other terminal information acquisition date and time information, and The time change value indicating the calculated elapsed time in seconds is output to the neuron of the 0th layer (input layer) of the neural network. In addition, the time change value is, for example, a value representing the elapsed time from one date and time to the other date and time in seconds, or whether the elapsed time from one date and time to the other date and time exceeds a predetermined time. It is conceivable to use a value indicating whether or not

Further, as shown in FIG. 7, the input of the individual variable parameter evaluation value output model includes variable parameter change values (hereinafter abbreviated as "variable parameter change values"). A variable parameter change value related to one type of variable parameter is a value that indicates a change between the one type of variable parameter of one terminal information and the one type of variable parameter related to the other terminal information. A variable parameter change value is expressed in different ways depending on the type of variable parameter. The variable parameter change value is a variable parameter related to one terminal information (hereinafter referred to as "one variable parameter") and a variable parameter related to the other terminal information (hereinafter referred to as "other variable parameter") for the variable parameter change value output algorithm AR2. is calculated by inputting

For example, when the variable parameter type is OSver, the variable parameter change value output algorithm AR2 related to OSver executes the following processing. For example, assume that OSver is described as "○.x.□" (where ○, ×, and □ are arbitrary integers) in the order of major item, medium item, and minor item. In this case, the variable parameter change value output algorithm AR2 appropriately considers the match/mismatch of major items and the match/mismatch of major items and medium items according to a predetermined rule, and changes the version of the OS to be compared as a whole. A variable para-variation value indicative of the quantity is calculated. If the OS version has changed to the negative side, the variable parameter change value is a value that indicates that the OS version has changed to the negative side (this is just an example, and can be expressed in this way. not appropriate).

Also, for example, when the type of variable parameter is a global IP address (IPv4), the variable parameter change value output algorithm AR2 related to the global IP address recognizes the class of the IP address, and if the class is different, it is variable according to a predetermined rule. If the class is the same, the value of the variable parameter change value is increased, and conversely, if the classes are the same, the value of the variable parameter change value is decreased according to a predetermined rule. do). In addition, the variable parameter change value output algorithm AR2 distinguishes between the network part and the host part when the classes are the same, and if the values of the network part match, the variable parameter change value is reduced according to a predetermined rule, and the value of the network part is reduced. If the values differ, the variable parameter change value is increased according to a predetermined rule, and if the values of the network part match, the magnitude of the difference in the value of the host part is reflected in the variable parameter change value according to a predetermined rule (only This is just an example, and is not meant to be expressed in this way.)

In addition, as shown in FIG. 7, the input of the individual variable para-evaluation value output model includes each of the fixed parameters included in the fixed parameter group. Although the details are omitted, each of the fixed parameters is appropriately subjected to necessary data conversion, data shaping, normalization, vectorization, etc. so that it has a value suitable as an input to the neural network. For example, regarding the device type, the value indicating mobile is set to "1", the value to indicate desktop is set to "-1", the value to indicate mobile is set to "100", and the value to indicate desktop. is set to "0", and the value of the device type is represented by a two-dimensional vector. (These examples are just examples, and it is not appropriate to do this).

The individual variable para-evaluation value output model is expected to output individual variable para-evaluation values in the following manner through machine learning.

Here, the inventors have tended to change aspects of variable parameters with respect to changes in time for each type of variable parameter, and the aspect of change of variable parameters is affected by the value of a specific fixed parameter. I found out.

FIG. 8A shows the value of OSver <variable parameter> (=OS version number) with respect to the passage of time when the value of OS type <fixed parameter> is a value indicating a certain type (specified as type SX). value) is shown by a graph in which the horizontal axis is time and the vertical axis is OS version coordinates. From the example in FIG. 8A, it can be seen that for OSver <variable parameter>, when the OS type <fixed parameter> is type SX, the value tends to increase little by little in proportion to the passage of time (change in time It can be read that there is a tendency of the aspect of change of the variable parameter with respect to Note that FIG. 9 shows a graph obtained by converting the graph of FIG. 8A into a graph showing the relationship between the time change value and the variable parameter change value. In FIG. 9, following the above trend, the variable para change value gradually increases in proportion to the increase in the time change value.

FIG. 8B shows an example of transition of the OS version value when the value of the OS type <fixed parameter> is a value indicating the type SY different from the type SX on the same coordinates as in FIG. 8A. 1 is a diagram illustrated by a graph of FIG. From the example of FIG. 8B, for OSver <variable parameter>, when the OS type <fixed parameter> is type SY, the value gradually increases over time, while the value increases at substantially constant intervals. It can be read that there is a tendency (the tendency of the aspect of change of variable parameters with change in time) that the version of the OS increases significantly after . Assuming that there is such a tendency, the variable parameter change value related to OSver <variable parameter> when the OS type is type SY follows the trend (influences the trend) with respect to changes in the time change value. It is thought that it will change in response to

In addition, by comparing each diagram in FIG. 8, it can be read that the tendency of the mode of change of OSver <variable parameter> with respect to time varies depending on the value of OS type <fixed parameter>.

In addition, for example, regarding the global IP address <variable parameter>, if the device type is mobile, the value (address value) does not change (hardly changes) for a very short period of several tens of seconds from a certain reference time. , over that short period of time, one can assume a tendency for the values (address values) to change chaotically from those at the reference time. Regarding the global IP address <variable parameter>, if the device type is a desktop, it can be assumed that the value (address value) tends to change within a certain range over time. Based on these assumptions, it is assumed that the global IP address <variable parameter> tends to change over time, and that this change is affected by the value of the device type <fixed parameter>. can.

Now, referring to FIG. 7, in the individual variable parameter evaluation value output model, each of the time change value, the variable parameter change value, and the fixed parameters included in the fixed parameter group is input to the neural network. For this reason, for one type of individual variable parametric evaluation value output model, the following individual variable parametric evaluation values are output by adjusting the weights of the links connecting neurons in the neural network by machine learning. There is expected. That is, the more the relationship between the time-varying value and the variable parameter-varying value conforms to the trend of the change in the value of the one type of variable parameter with respect to the change in time reflecting the value of the predetermined fixed parameter, the better. , the output individual variable parameter evaluation value increases, and the more the relationship deviates from the trend, the smaller the output individual variable parameter evaluation value is expected.

For example, referring to FIG. 9, when the OS type is type SX, as shown in the graph of FIG. do. In addition, the value of the time change value input to the individual variable parameter evaluation value output model related to OSver <variable parameter> is Q1 (see FIG. 9), and the value of OS type <fixed parameter> input to the model is Assume that the value indicates the type SX. In this case, when the value of the variable parameter change value is P1 (see FIG. 9), the relationship between the time change value and the variable parameter change value is the change in OSver <variable parameter> when the OS type is type SX. It can be said that it conforms to the trend of the mode. Therefore, in this case, it is expected that the variable parametric evaluation value output by the individual variable parametric evaluation value output model after machine learning will increase. On the other hand, when the value of the variable parameter change value is P2 or P3, it can be said that the relationship does not conform to the trend. Therefore, in such a case, it is expected that the variable parametric evaluation value output by the individual variable parametric evaluation value output model after machine learning will be small.

In this embodiment, all the fixed parameters included in the fixed parameter group are input to the individual variable parameter evaluation value output model related to the variable parameters. For fixed parameters of the type that affect the variable parameters, the degree of influence on the variable parameter evaluation value increases, and for fixed parameters that do not affect the change, the degree of influence decreases (ideally, the degree of influence disappears). , it is assumed that the weight of each link of the neural network converges.

In addition, in this embodiment, the type of fixed parameters to be input to the individual variable parameter evaluation value output model is not selected by humans, and all fixed parameters included in the fixed parameter group are used as inputs. For example, regarding the individual variable parameter evaluation value output model related to OSver <variable parameter>, only the OS type <fixed parameter> is assumed to affect the change in the value of OSver <variable parameter>. It is not considered as an input. As a result, it is possible to influence the individual variable parameter evaluation value even for fixed parameters that are difficult for humans to notice although they originally affect changes in the variable parameters.

As described above, in the present embodiment, focusing on the fact that there is a tendency in the aspect of change in the value of the variable parameter with respect to time change, the neural network of the individual variable parameter evaluation value output model There is a feature that has not existed in the past in that it inputs Then, due to this feature, the dedicated model M1 is machine-learned, so that for the individual variable parameter evaluation value output model related to one type of variable parameter, the mode of change of the one type of variable parameter with respect to time change is learned, and a highly reliable evaluation value that reflects this tendency is output.

Next, the weighted output model will be described. The weighted output model will be described below assuming that the weighted output model is a neural network. FIG. 10 is an illustratively simplified representation of a weighted output model neural network. As shown in FIG. 10, the input of the weighted output model includes time-varying values and fixed parameters. Therefore, it is expected that the weight output model will output individual weights with the following contents by performing machine learning. That is, the individual weight corresponding to the individual variable parameter evaluation value that can be determined to be reliable considering the time change value and the value of each fixed parameter increases, while the individual variable that can be determined to be unreliable A smaller value is expected for the individual weights corresponding to the para-evaluation values.

For example, regarding the individual variable parameter evaluation value related to the global IP address <variable parameter>, when the device type is mobile, if the time change value is smaller than a certain value (for example, a value indicating a period of about several tens of seconds) , is reliable to some extent, but becomes almost unreliable when the time-varying value exceeds a certain value. In this case, if the value of the device type <fixed parameter> input to the weight output model is a value that indicates mobile, and the time change value exceeds a certain value, the "global Individual weight corresponding to IP address <variable parameter> is expected to be very small.

Further, for example, regarding the individual variable parameter evaluation value related to OSver <variable parameter>, if the OS type is a predetermined type, the time change value exceeds a certain value (for example, a value indicating a period of about one year). , OSver <variable parameter> is relatively more reliable than the individual variable parameter evaluation values of other variable parameters. In this case, the value of the OS type <fixed parameter> input to the weighted output model is a value indicating the predetermined type, and if the time change value exceeds a certain value, the weighted output model after machine learning is output. , the individual weight corresponding to OSver <variable parameter> is expected to be relatively large compared to other individual weights.

As described above, in the present embodiment, focusing on the fact that the reliability of the individual variable parametric evaluation value changes with time, the time-varying value is input to the neural network of the weighted output model. has features that are not found in With this feature, the dedicated model M1 is machine-learned to learn changes in reliability with respect to changes in time, and highly reliable individual weights that reflect this tendency are output. .

As described above, each individual variable para-evaluation value is calculated, and each individual weight is calculated. For this reason, the final evaluation value derived by the calculation formula of FIG. The value takes into consideration the characteristic that changes in , and the reliability can be dramatically improved compared to the case where such characteristic is not taken into account.

In the present embodiment, the dedicated model M1 is composed of an individual variable parameter evaluation value output model and a weighted output model independent of this, but this is for convenience of explanation. , and a single neural network may be configured to output individual variable parametric evaluation values and individual weights. Further, for the individual variable para-evaluation value output model and the weight output model, the values to be input to the neurons of the 0th layer (input layer) may be input to the neurons of the hidden layer (intermediate layer). Also, the dedicated model M1 does not need to be a neural network, and may be a model capable of machine learning and capable of outputting a final evaluation value according to an input.

Next, learning of the dedicated model M1, especially preparation of teacher data will be described. FIG. 11 is a diagram showing the contents of teacher data. As shown in FIG. 11, learning of the dedicated model M1 includes a group of variable parameters on the one hand, a group of variable parameters on the other hand, terminal information acquisition date and time information, terminal information acquisition date and time information, a common fixed parameter group, and Teacher data including a label indicating whether the terminal and the other terminal are the same is used. In order to improve the accuracy of the dedicated model M1, it is important to prepare a large amount of teacher data of labels (hereinafter referred to as "correct labels") indicating that one terminal and the other terminal are the same.

Teacher data is prepared, for example, as follows. That is, every time there is a login for online services that require login (services provided by service providers or services provided by other providers), the user terminal 3 Terminal information acquisition date and time information, a fixed parameter group, and a variable parameter group are collected from the , and stored in association with the login ID. At that time, the fixed parameter group is grouped for each login ID. Note that the user may log in with a different user terminal 3, but the values of the fixed parameter group collected when the user logs in to the service are the fixed parameters collected when the user logged in at a certain timing in the past. If they match the values of the parameter group, it can be considered that the user terminals 3 used for login are the same.

As described above, groups are formed in which combinations of login IDs, terminal information acquisition date and time information, fixed parameter groups, and variable parameter groups (hereinafter referred to as "login terminal information") are accumulated. In each group, the value of the login ID of each piece of terminal information at login and the value of the fixed parameter group are the same. FIG. 12A shows the contents of a plurality of log-in terminal information belonging to a certain group.

Then, teacher data having a positive label is generated based on a group in which a sufficient amount of login terminal information has been accumulated. More specifically, all login terminal information belonging to a group is based on the same user terminal 3 . Therefore, as shown in FIG. 12(B), based on any two log-in terminal information belonging to one group, one case having a positive label (represented as "1" in FIG. 12(B)) of training data can be generated. Therefore, it is possible to generate a plurality of positive-label teacher data by variously combining the login-time terminal information belonging to one group. Teacher data is prepared by generating teacher data for each of the plurality of groups in the above procedure. Note that the method of generating teacher data is not limited to the methods illustrated above. For example, for a plurality of terminals prepared for testing, a variable parameter group is collected at any time, and associated with terminal information acquisition date and time information and a fixed parameter group. , and based on this, teacher data may be generated.

As described above, the dedicated model M1 is machine-learned based on the teacher data to obtain the tendency of the change of the one type of variable parameter with respect to time and the individual variable parameter according to time. The weight of the link connecting each neuron is adjusted so that the final evaluation value reflecting changes in the reliability of the evaluation value is output, making it possible to output a highly reliable final evaluation value.

Next, the operation of the information processing server 2, especially the operation of the same possibility determination unit 8 will be described using a flowchart. FIG. 13 is a flow chart showing the operation of the same possibility determination unit 8. As shown in FIG.

As shown in FIG. 13, the same possibility determination unit 8 receives terminal information, terminal information acquisition date and time information, and a site URL from the advertisement providing unit 7 (step SA1). Next, the same possibility judgment unit 8 acquires a fixed parameter group included in the terminal information (step SA2). Next, the same possibility judgment unit 8 generates a pre-hash fixed parameter-derived ID (step SA3). Next, the possibility-of-sameness judgment unit 8 generates a fixed para ID (step SA4). It is determined whether or not a record having a fixed parameter ID with the same value as the value exists (step SA5).

If it does not exist (step SA5: NO), the possibility-of-sameness determining unit 8 generates a unique ID (step SA6). Next, the possibility-of-sameness determination unit 8 determines whether the unique ID, fixed parameter ID, terminal information date and time information, fixed parameter group, variable parameter group, and site URL generated in step SA6 are stored in corresponding fields. record is registered (step SA7). Next, the possibility-of-sameness determination unit 8 responds with the generated unique ID to the advertisement providing unit 7 (step SA8). After the processing of step SA8, the flowchart of FIG. 13 ends.

On the other hand, if candidate records exist (step SA5: YES), the possibility-of-sameness determining unit 8 determines whether or not there is one candidate record (step SA9). If there is one (Step SA9: YES), the identity possibility determination unit 8 determines the variable parameter group included in the candidate record, the variable parameter group related to the user terminal to be processed, the terminal information acquisition date and time information included in the candidate record, the advertisement The terminal information acquisition date and time information input from the provision unit 7 and the fixed parameter group related to the user terminal (or candidate record) to be processed are input to the dedicated model M1, and the final evaluation value output by the dedicated model M1 is obtained (step SA10 ). Next, the possibility-of-sameness determination unit 8 determines whether or not the final evaluation value is equal to or greater than the threshold value T1 (step SA11). If it is not equal to or greater than the threshold T1 (step SA11: NO), the process proceeds to step SA6.

If it is equal to or greater than the threshold T1 (step SA11: YES), the identity possibility determination unit 8 updates the candidate record based on the latest information received from the process target user terminal (step SA12). Next, the possibility-of-sameness determining unit 8 notifies the unique ID of the candidate record to the advertisement providing unit 7 (step SA13). After the processing of step SA13, the flowchart of FIG. 13 ends.

If the number of candidate records is not one (=two or more) in step SA9 (step SA9: NO), the possibility-of-sameness determination unit 8 uses the dedicated model M1 to determine the final evaluation value for each of the candidate records. Acquire (step SA14). Next, the possibility-of-sameness determination unit 8 determines whether or not all of the obtained final evaluation values are below the threshold value T1 (step SA15). If less than (step SA15: YES), the process proceeds to step SA6. If not less than (step SA15: NO), the possibility-of-sameness determination unit 8 determines whether or not only one final evaluation value is equal to or greater than the threshold value T1 (step SA16). If there is only one (step SA16: YES), the possibility-of-sameness determination unit 8 updates the candidate record for the final evaluation value equal to or greater than the threshold T1 (step SA17), and notifies the unique ID (step SA18). After the processing of step SA18, the flowchart of FIG. 13 ends. On the other hand, if there is more than one (=two or more) (step SA16: NO), the possibility-of-sameness determination unit 8 determines the final evaluation value having the largest value among the final evaluation values equal to or greater than the threshold T1. The candidate record is updated (step SA19), and the unique ID is notified (step SA20). After the processing of step SA20, the flowchart of FIG. 13 ends.

As described above, according to the present embodiment, when determining the possibility that two different terminals (a first terminal and a second terminal) are the same, the fixed parameter group for the first terminal and the fixed parameter group related to the second terminal are reflected. Therefore, it is possible to reduce the possibility that two terminals, which can be said to be definitely different from each other in view of the characteristics of the fixed parameters, will be judged to be the same, thereby improving the reliability of the judgment result. Furthermore, according to the present embodiment, when determining the possibility that the first terminal and the second terminal are the same, the time change indicating the change from the first time point to the second time point and the first terminal A change from the variable parameter included in the information to the variable parameter included in the second terminal information is reflected. Therefore, it is possible to determine the possibility of identity by taking into consideration the tendency of the aspect of change of the variable parameter with respect to time change. Even if the variable parameter changes over time, the possibility of identity can be determined with higher reliability than when such trends are not considered. As described above, according to the present embodiment, terminals can be identified with high reliability for as long as possible without using cookies and without limiting identifiable terminals.

<Second embodiment>
Next, a second embodiment will be described. In the following description of the second embodiment, elements that are the same as those of the first embodiment are denoted by the same reference numerals, and description thereof will be omitted. FIG. 14 is a block diagram showing a functional configuration example of an information processing system 1A according to the second embodiment. As shown in FIG. 14, an information processing system 1A according to this embodiment includes a user terminal 3A that can access a network N, to which a site providing server 4A and an information processing server 2A are connected. The user terminal 3A functions as an "information processing device" in the scope of claims. The information processing system 1A according to the present embodiment and the information processing system 1 according to the first embodiment are largely different in that the entity that executes the dedicated model M1 is the user terminal 3A.

As shown in FIG. 14, the user terminal 3A includes a web browser execution unit 10A instead of the web browser execution unit 10 according to the first embodiment. It has an acquisition unit 12A and an identity possibility determination unit 8A. Further, the information processing server 2A includes an advertisement providing section 7A instead of the advertisement providing section 7, and a response section 20 instead of the same possibility determining section 8. FIG.

Processing of the processing target user terminal, the site providing server 4A, and the information processing server 2A after the web browser of one user terminal 3A (processing target user terminal) accesses the site of the site providing server 4A will be described below.

In response to an instruction from the user, the central control unit 11A of the web browser execution unit 10A of the processing target user terminal transmits an HTTP request to the site providing server 4A and obtains an HTTP response as a response. The HTTP response contains an HTML file for displaying a web page and a dedicated script. In the dedicated script, the dedicated model M1 is implemented as a script in addition to the script having the function of acquiring the terminal information.

The terminal information acquisition unit 12A of the web browser execution unit 10A acquires terminal information (fixed parameter group and variable parameter group). The same possibility determination unit 8A generates a fixed para ID based on the terminal information acquired by the terminal information acquisition unit 12A, and selects a record (candidate record) is registered in the user management database 14 to the response unit 20 of the information processing server 2A.

The response unit 20 of the information processing server 2A refers to the user management database 14 in response to the inquiry from the identity possibility determination unit 8A, and if there is no candidate record, notifies that effect. Also, when there is one or more candidate records, the response unit 20 responds to the same possibility determination unit 8A with all the candidate records.

When the possibility-of-sameness determination unit 8A receives a notification that no candidate record exists,
A unique ID is generated based on the terminal information, and the generated unique ID, terminal information, terminal information acquisition date and time information, and site URL are transmitted to the response unit 20 to instruct registration of a new record. The response unit 20 registers a new record (record corresponding to the user terminal to be processed) in the user management database 14 according to the instruction. Furthermore, the same possibility determination unit 8A notifies the unique ID to the advertisement providing unit 7A.

Further, when there is a response of one or more candidate records, the sameness possibility determination unit 8A uses the dedicated model M1 to determine the same user terminal 3A as the user terminal to be processed in the same manner as in the first embodiment. , or determine that no such candidate record exists. When one candidate record corresponding to the same user terminal 3A as the user terminal to be processed is specified, the possibility-of-identification determination unit 8A transmits the terminal information, the terminal information acquisition date and time information, and the site URL to the response unit 20, Instructs to update that one candidate record. The response unit 20 updates the content of the candidate record according to the instruction. Furthermore, the same possibility determination unit 8A notifies the unique ID to the advertisement providing unit 7A.

If it is determined that there is no candidate record corresponding to the user terminal 3A that is the same as the user terminal to be processed, the possibility-of-identification determining unit 8A generates a unique ID based on the terminal information, and adds the generated unique ID, the terminal information, It transmits the terminal information acquisition date and time information and the site URL to the response unit 20, and instructs registration of a new record (=record corresponding to the user terminal to be processed). The response unit 20 registers a new record in the user management database 14 according to the instruction. Furthermore, the same possibility determination unit 8A notifies the unique ID to the advertisement providing unit 7A.

When the advertisement provision unit 7A receives notification of the unique ID from the same possibility determination unit 8A of the user terminal to be processed, the advertisement provision unit 7A refers to the user management database 14 and generates drawing data of the advertisement information in the same manner as in the first embodiment. and transmitted to the integrated control unit 11A of the web browser execution unit 10A of the user terminal to be processed. The general control unit 11A receives the drawing data, and displays the website to which the advertisement information is added on the display device based on the drawing data and the HTML file received from the site providing server 4A.

As described above, the user terminal 3A may use the dedicated model M1 to determine the possibility of identity between itself and the user terminal 3A corresponding to the record registered in the user management database 14. FIG. Note that the second embodiment may have the following configuration. That is, the identity possibility determination unit 8A requests the response unit 20 of the information processing server 2A to transmit the user management database 14. FIG. The identity possibility determination unit 8A receives the user management database 14 from the response unit 20 and stores it in a predetermined storage area. The same possibility determination unit 8A executes processing using the user management database 14 stored in the predetermined storage area.

As described above, the first and second embodiments of the present invention have been described. scope should not be construed as limited. Thus, the invention may be embodied in various forms without departing from its spirit or essential characteristics.

<Modification 1>
For example, in each of the above-described embodiments, all fixed parameters included in the fixed parameter group are input to the individual variable parameter evaluation value output model and weight output model of the dedicated model M1. However, for an individual variable parameter evaluation value output model related to one variable parameter, a configuration may be adopted in which fixed parameters that do not affect the individual variable parameter evaluation value related to the one variable parameter are not input. Also, fixed parameters that do not affect individual weights may not be input to the weight output model.

<Modification 2>
Further, a configuration may be adopted in which no fixed parameters are input to the dedicated model M1 (and thus each of the individual variable parametric evaluation value output models and the weight output model). This is because regardless of the value of the fixed parameter, variable parameters tend to change over time for each type of variable parameter. For example, OSver <variable parameter> tends to increase in value over time regardless of the OS type <fixed parameter>. Also, this is because the reliability of the individual variable parameter evaluation value changes according to the time-varying value, regardless of the value of the fixed parameter.

Note that it is also assumed that there are variable parameters whose change over time is strongly influenced by the value of a specific fixed parameter. For example, the mode of change in the value of the global IP address <variable parameter> is considered to vary greatly depending on the value (mobile/desktop) of the device type <fixed parameter>. In this way, for variable parameters that are affected by the values of the fixed parameters, it is assumed that the reliability of the individual variable parameter evaluation values will decrease if the fixed parameters are not input to the input of the individual variable parameter evaluation value output model. However, by appropriately performing machine learning, the individual weight for such individual variable para-evaluation values is reduced, and it is expected that the influence on the final evaluation value will be kept low.

FIG. 15A is a diagram showing the relationship between the input and the output of the dedicated model M1 when the dedicated model M1 is configured so as not to input the common fixed parameter group. FIG. 15B is a diagram showing the contents of teacher data used for learning the dedicated model M1 in this case. As shown in FIG. 15B, in this case, the teacher data includes a variable parameter group, a variable parameter group, terminal information acquisition date and time information, terminal information acquisition date and time information, and a label. Configured.

<Modification 3>
In addition, for the individually variable para-evaluation value output model (in this example, a neural network; however, it may not be a neural network as described above), instead of the time change value, one terminal information acquisition date and time information and the other terminal It may be configured to input the information acquisition date and time information. Further, it is also possible to input one terminal information acquisition date and time information and the other terminal information acquisition date and time information together with the time change value to the individually variable para-evaluation value output model. Even in these cases, the machine-learned exclusive model M1 can output the final evaluation value reflecting the change from the time when the one terminal information is acquired to the time when the other terminal information is acquired. In these cases, in the preprocessing, the terminal information acquisition date/time information and the terminal information acquisition date/time information are appropriately converted into values that can be processed by the neural network through data shaping and normalization.

<Modification 4>
Alternatively, instead of the variable parameter change value, one variable parameter and the other variable parameter may be input to the individual variable parameter evaluation value output model. Further, a configuration may be adopted in which one variable parameter and the other variable parameter are input together with the variable parameter change value to the individual variable parameter evaluation value output model. Even in these cases, the machine-learned exclusive model M1 can reflect the change from one variable parameter to the other variable parameter to output the final evaluation value. In these cases, the preprocessing appropriately converts values that can be processed by the neural network through data shaping and normalization for one variable parameter and the other variable parameter. FIG. 16 shows, with respect to Modifications 3 and 4, one terminal information acquisition date and time information and the other terminal information acquisition date and time information are input instead of time change values, and one variable parameter and the other variable parameter are input instead of variable parameter change values. 2 is a diagram showing a simplified neural network of an individual variable para-evaluation value output model when configured to input .

<Modification 5>
Alternatively, instead of the time change value, one terminal acquisition date and time information and the other terminal information acquisition date and time information may be input to the weight output model. Further, the weight output model may be configured to input the one terminal acquisition date and time information and the other terminal information acquisition date and time information together with the time change value. Even in these cases, the learned weight output model can output individual weights that reflect changes over time.

<Modification 6>
Also, one or more individual variable parameter evaluation values may be further input to the weighted output model. FIG. 17 is a diagram showing a simplified neural network of the weighted output model according to Modification 6. As shown in FIG. According to the configuration of Modified Example 6, the individual weight corresponding to the individual variable parameter evaluation value that can be determined to be reliable by considering the time change value, the value of each fixed parameter, and one or more individual variable parameter evaluation values. is expected to increase, while the individual weight corresponding to the individual variable para-evaluation value that can be determined to be unreliable is expected to decrease.

For example, it is assumed that the variable parameter change value related to OSver <variable parameter> is a value indicating that the OS version has changed to the minus side. As described above, the version of the OS has a feature that it gradually increases with the passage of time. and the other terminal are very unlikely to be the same). In this case, the individual variable parameter evaluation value related to OSver <variable parameter> is highly reliable. Considering the trend of changes in OS versions over time, it is unlikely that the same terminal will have a lower version of the OS (the possibility of it becoming lower is extremely small), and "each terminal may be the same" This is because there is a very high probability that the assessment that the Therefore, in such a case, the individual weight related to OSver <variable parameter> should be relatively large so that the individual variable parameter evaluation value related to OSver <variable parameter> affects the final evaluation value as much as possible. There is expected.

<Modification 7>
Also, as shown in FIG. 18(A), for one terminal, each of a plurality of one-way variable parameters (variable parameter types are common) acquired at a plurality of different points in time, together with corresponding terminal information acquisition date and time information, , may be input to an individual variable para-evaluation value output model. Alternatively, as shown in FIG. 18B, the following configuration may be used. That is, each combination of the terminal information acquisition date and time information and one variable parameter is input to the variable parameter distribution value output algorithm AR3. The variable parameter distribution value output algorithm AR3 outputs variable parameter distribution values based on the input. A variable parameter distribution value is a value that indicates a change in a plurality of variable parameters with respect to a change in time. is information representing a waveform representing how the value of the variable parameter changes over time. This variable parameter distribution value may be input to the individual variable parameter evaluation value output model. On the other hand, each combination of the variable parameter and the terminal information acquisition date and time information and the variable parameter distribution value may be input to the individual variable parameter evaluation value output model. In addition, each combination of the terminal information acquisition date and time information and one variable parameter and the combination of the other variable parameter and the other terminal information acquisition date and time information are input to the variable parameter distribution value output algorithm AR3, and the variable parameter distribution The variable parameter distribution value may be output from the value output algorithm.

FIG. 19A is a diagram showing the relationship between the input and output of the dedicated model M1 according to Modification 7. FIG. FIG. 19B is a diagram showing contents of teacher data according to Modification 7. As shown in FIG. As shown in FIG. 19A, the dedicated model M1 according to this modified example has the point that a plurality of "combinations of one-side variable parameter groups and one-side terminal information acquisition date and time information" are input as information related to one-side terminals. , is different from the dedicated model M1 according to the first embodiment. Further, as shown in FIG. 19B, in this modification, the teacher data has a plurality of combinations of one variable parameter group and one terminal information acquisition date and time, and furthermore, the other variable parameter group and the other terminal information acquisition It has a combination with date and time, a set of common fixed parameters, and a label.

In the case of this modification, the identity possibility determination unit 8 according to the first embodiment executes the following processing. That is, when the possibility-of-identification determining unit 8 determines that the user terminal to be processed and the user terminal 3 corresponding to one candidate record in the user management database 14 are the same, the latest Instead of updating the candidate record based on the information, the combination of the terminal information acquisition date/time information and the variable parameter group acquired this time is added to the combination of the field of the terminal information acquisition date/time information and the field of the variable parameter group. As a result, every time a certain user terminal 3 accesses the site providing server 4 and the user terminal 3 is recognized, the variable parameter group is entered in the variable parameter group field of the record of the user terminal 3 in the user management database 14. being added. Therefore, in the user management database 14, one unique ID may be associated with a plurality of different "combinations of terminal information acquisition date and time information and variable parameter groups".

On the other hand, when judging whether or not the user terminal to be processed and the user terminal 3 corresponding to a certain candidate record are the same, the possibility-of-sameness determination unit 8 determines whether the terminal information acquisition date and time information and the variable parameter The combination with the group is input to the dedicated model M1 as information relating to the other terminal. Furthermore, the sameness possibility determination unit 8 determines each of the "combinations of the terminal information acquisition date and time information and the variable parameter group" included in the candidate record (thus, if there are multiple such combinations, each of the multiple combinations) is input to the dedicated model M1 as information related to the terminal. In addition, the identity possibility determination unit 8 also inputs the common fixed parameter group to the dedicated model M1. Then, the identity possibility determination unit 8 acquires the final evaluation value output by the dedicated model M1, and uses this to determine the identity possibility between the user terminal to be processed and the user terminal 3 corresponding to a certain candidate record. . The above processing may be performed for the second embodiment.

As described above, the possibility-of-identity determination unit 8 according to the present modification executes the following process when determining the possibility of identity with two different terminals (a first terminal and a second terminal). . That is, the possibility-of-identity determination unit 8 determines the consistency between the fixed parameters related to the first terminal and the fixed parameters related to the second terminal, as well as a plurality of points in time when the terminal information related to the first terminal is acquired and A time change based on a second point in time when the terminal information on the second terminal is acquired, each variable parameter on the first terminal at a plurality of points of time, and a variable parameter on the second terminal at the second point of time The likelihood of identity is determined by reflecting variable parameter changes based on the parameters.

<Other Modifications>
Although a plurality of modified examples have been described above, it is a matter of course that one or a plurality of modified examples may be combined and applied to the first embodiment or the second embodiment.

Moreover, the method by which the terminal information acquisition unit 12 acquires the fixed parameter group and the variable parameter group is not limited to the method illustrated. For example, when using the elements described in the header part of the HTTP request as fixed parameters or variable parameters, the terminal information acquisition unit 12 may analyze the header part and acquire fixed parameters or variable parameters. good.

Further, the method of using the determination result (final evaluation value) of the possibility of identity is not limited to the one exemplified in each of the above embodiments. For example, the first embodiment may have the following configuration. That is, a server device different from the information processing server 2 has a function of providing advertisement information, and the server device stores a unique ID and history information (as described above, the history information is not limited to the site URL). to manage. Then, when identifying a unique ID in response to access to the site providing server 4 by the user terminal 3, the identity possibility determination unit 8 of the information processing server 2 transmits the unique ID to the server device. The server device identifies history information corresponding to the unique ID, and provides the user terminal 3 with an advertisement image reflecting the history information. The second embodiment may also have a similar configuration.

In addition, the result of determining the possibility of identity (final evaluation value) can also be used for digital forensics and risk-based authentication. In risk-based authentication, for example, every time a user logs into an online service, the dedicated model M1 is used to determine whether the terminal used for previous login is the same as the terminal used for current login. If not, stronger authentication is required.

2 Information processing server (information processing device)
3A user terminal (information processing device)
8, 8A Identity possibility judgment unit

Claims (13)

Information about terminals that can be connected to a network. A group of fixed parameters consisting of fixed parameters whose values do not change with the passage of time, and a variable group consisting of variable parameters whose values can change with the passage of time. An information processing device that processes terminal information including a parameter group,
Based on the first terminal information of the first terminal acquired at a first time point and the second terminal information of the second terminal acquired at a second time point after the first time point an identity possibility determination unit that determines an identity possibility, which is a possibility that the first terminal and the second terminal are the same,
The identity possibility determination unit
A fixed parameter related to the first terminal, which is a fixed parameter included in the first terminal information of the first terminal acquired at the first time point, and a second terminal acquired at the second time point In addition to consistency with the fixed parameter related to the second terminal, which is a fixed parameter included in the second terminal information, a time change indicating a change from the first time point to the second time point; the second terminal obtained at the second time point from the variable parameter relating to the first terminal, which is a variable parameter included in the first terminal information of the first terminal obtained at the first time point; and a variable parameter change indicating a change to a variable parameter related to the second terminal, which is a variable parameter included in the second terminal information, and a variable parameter change indicating a change to the variable parameter related to the second terminal. Device. The identity possibility determination unit
Each of the first terminal information of the first terminal acquired at a first plurality of time points indicating a plurality of time points, and the second terminal information of the second terminal acquired at the second time point determining the possibility of identity based on the terminal information,
When determining the possibility of identity, in addition to matching between the fixed parameter for the first terminal and the fixed parameter for the second terminal, the first plurality of time points and the second time point are also used. reflecting the time change and the variable parameter change based on each of the variable parameters of the first terminal at the first plurality of time points and the variable parameter of the second terminal at the second time point; 2. The information processing apparatus according to claim 1, wherein possibility of identity is determined. The identity possibility determination unit
When the fixed parameter group related to the first terminal and the fixed parameter group related to the second terminal do not match, regardless of the time change and the variable parameter change, the first terminal and the second 3. The information processing apparatus according to claim 1 or 2, wherein it is determined that the terminal of the terminal is not the same as the terminal of the terminal. The identity possibility determination unit
When the fixed parameter group related to the first terminal and the fixed parameter group related to the second terminal match,
Evaluating the possibility of identity by reflecting the time change and the change in the variable parameter for each type of variable parameter, and determining the final possibility of identity by reflecting the evaluation for each type of variable parameter. 4. The information processing apparatus according to any one of claims 1 to 3, characterized by: The identity possibility determination unit
When evaluating the possibility of identity for one type of variable parameter, the probability of identity is evaluated by reflecting the value of one or more fixed parameters in addition to the time change and the variable parameter change. 5. The information processing apparatus according to claim 4. The identity possibility determination unit
When determining the final possibility of identity by reflecting the evaluation for each type of variable parameter, the evaluation of each type of variable parameter is reflected in the final determination of the possibility of identity by reflecting the change over time 6. The information processing apparatus according to claim 4, wherein the degree of influence is adjusted. The identity possibility determination unit
When determining the final possibility of identity by reflecting the evaluation for each type of variable parameter, the evaluation for each type of variable parameter is performed by reflecting the time change and the value of one or more fixed parameters. 7. The information processing apparatus according to claim 6, wherein the degree of influence on the final determination of the possibility of identity is adjusted. The identity possibility determination unit
When the fixed parameter group related to the first terminal and the fixed parameter group related to the second terminal match,
Data based on one terminal information and other terminal information whose fixed parameter groups match each other, wherein the variable parameter group related to the one terminal, the variable parameter group related to the other terminal information, and the one Information indicating one point in time when the terminal information was acquired, information indicating another point in time when the other terminal information was acquired, and a terminal related to the one terminal information and a terminal related to the other terminal information Using a model trained using data including a label indicating whether or not it is the same as teacher data,
From the variable parameter group related to the first terminal, the variable parameter group related to the second terminal information, the information indicating the first time point, and the information indicating the second time point, the possibility of identity is determined 2. The information processing apparatus according to claim 1, wherein said evaluation value is output. The model is
The variable parameter group related to the one terminal information and the variable parameter group related to the other terminal information are changed from the variable parameter related to the one terminal information to the variable parameter related to the other terminal information for each type of variable parameter converted to a variable parameter change value indicating a change to , this is used as an input, and
Information indicating the one point in time when the terminal information was obtained and information indicating the other point in time when the other terminal information was obtained, and the change from the one point in time to the other point 9. The information processing apparatus according to claim 8, wherein the time-varying value is converted into an indicated time-varying value and used as an input. The identity possibility determination unit
When the fixed parameter group related to the first terminal information and the fixed parameter group related to the second terminal information match,
Data based on one piece of terminal information and other terminal information whose fixed parameter groups match each other, wherein a variable parameter group related to the one terminal information, a variable parameter group related to the other terminal information, and the one information indicating one point in time when the terminal information was acquired, information indicating another point in time when the other terminal information was acquired, one or more fixed parameter values, and the terminal related to the one terminal information and a label indicating whether or not the terminal related to the other terminal information is the same using a model learned using data as training data,
The first variable parameter group related to the terminal information, the second variable parameter group related to the terminal information, information indicating the first time point, information indicating the second time point, and one or more 2. The information processing apparatus according to claim 1, wherein an evaluation value indicating the possibility of identity is output from values of fixed parameters. The model is
The variable parameter group related to the one terminal information and the variable parameter group related to the other terminal information are changed from the variable parameter related to the one terminal information to the variable parameter related to the other terminal information for each type of variable parameter converted to a variable parameter change value indicating a change to , this is used as an input, and
Information indicating the one point in time when the terminal information was obtained and information indicating the other point in time when the other terminal information was obtained, and the change from the one point in time to the other point 11. The information processing apparatus according to claim 10, wherein the time-varying value is converted into an indicated time-varying value and used as an input. The identity possibility determination unit
When the fixed parameter group related to the first terminal and the fixed parameter group related to the second terminal match,
Data based on one piece of terminal information and other piece of terminal information whose fixed parameter groups match each other, wherein a plurality of variable parameter pieces related to the piece of piece of terminal information acquired at a plurality of points in time and the piece of other piece of terminal information , information indicating each of a plurality of points in time when the one terminal information was acquired, information indicating other points in time when the other terminal information was acquired, and one or more fixed parameters Using a model trained using data including a value and a label indicating whether the terminal associated with the one terminal information and the terminal associated with the other terminal information are the same as training data,
Each of a plurality of variable parameter groups related to the first terminal acquired at the first plurality of time points, a variable parameter group related to the second terminal, information indicating the first time point, and the second 3. The information processing apparatus according to claim 2, wherein the evaluation value indicating the possibility of identity is output from the information indicating the point in time. Information about terminals that can be connected to a network. A group of fixed parameters consisting of fixed parameters whose values do not change with the passage of time, and a variable group consisting of variable parameters whose values can change with the passage of time. An information processing method by an information processing device that processes terminal information including a parameter group,
The identity determination unit of the information processing device acquires a fixed parameter included in the first terminal information of the first terminal acquired at a first time point and a second time point after the first time point. a first step of determining consistency with a fixed parameter included in the second terminal information of the second terminal;
a time change indicating a change from the first time point to the second time point when the identity determination unit of the information processing apparatus determines that there is a match in the first step; The first terminal and the second terminal are the same, reflecting a variable parameter change indicating a change from a variable parameter included in the terminal information to a variable parameter included in the second terminal information. a second step of determining the likelihood;
An information processing method comprising:

Images