Target user classification method, target user classification system, and target user classification program

Abstract

To provide a technique for efficient targeting with 1st Party Data without using 3rd Party Cookie.SOLUTION: A target user classification method includes the steps of: holding a trained model for calculating a score representing purchase trend of a user, the trained model being generated through machine learning by using, as training data, accumulated user information of a plurality of users, the user information being information on users including at least action information on websites; calculating the score of a target user by inputting user information of the target user to the held trained model; setting segment conditions which are the number of a plurality of segments and thresholds set for classifying the target user according to the score; and classifying the target user into one of the segments on the basis of the set segment conditions and the calculated score of the target user.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a target user classification system that classifies target users into one of a plurality of segments according to their purchasing tendencies. [Background technology]

[0002] The key to increasing the effectiveness of Internet advertising is to consider "who to deliver to." If an advertisement is delivered to someone who has no interest in the advertised product, the likelihood that that person will purchase it is low. Therefore, the effectiveness of advertising is increased by narrowing down the users who are likely to make a purchase and delivering the advertisement to those users.

[0003] User attribute information is often used to narrow down users. For example, information about the user, such as gender, age, occupation, place of residence, income, hobbies, and marital status. When narrowing down targets based on these user attributes, the user's nature and tendencies are generalized and defined according to the attributes. For example, imagine an image of "users with the attribute of being single men in their 30s with hobbies like ___, who are highly interested in traveling," and deliver advertisements for travel plans or products such as luggage to users with such attributes.

[0004] In other words, travel agencies and luggage sellers can expect to improve the effectiveness of their advertisements by narrowing down the target of advertisement delivery to users who have the above attributes. Such advertisement delivery is carried out by collecting user attribute information, etc. by tracking user behavior such as web browsing history and search keywords using 3rd Party Cookies (Patent Document 1). [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2019-057104 A Summary of the Invention [Problem to be solved by the invention]

[0006] As mentioned above, the use of third-party cookies is beneficial for those who operate online sales and advertising businesses. However, from the perspective of protecting individual privacy, the use of third-party cookies has become problematic in recent years, and there have been growing moves to regulate their use.

[0007] In light of these circumstances, there is a demand for technology that enables effective targeting by utilizing first-party data without using third-party cookies. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems, the present invention provides the following target user classification method, etc. That is, the target user classification method is executed by a computer and classifies a target user into one of a plurality of segments according to the purchasing tendency, and is characterized by including: a model holding step of holding a trained model that calculates a score representing a user's purchasing tendency, the model being generated by machine learning using accumulated data for a plurality of users, the user information being information about the user including at least behavioral information on a website, as training data; a score calculation step of inputting the user information of the target user into the trained model that has been held and calculating the score of the target user; a condition setting step of setting segment conditions, which are the number of a plurality of segments and a threshold value that are set in order to classify the target user according to the score; and a classification step of classifying the target user into one of the segments based on the set segment conditions and the calculated score of the target user.

[0009] The present invention also provides a target user classification method, further comprising a delivery step of delivering advertisements to the target users according to the segments classified in addition to the characteristics.

[0010] In addition, the present invention provides a target user classification method that further includes, in addition to the above-mentioned characteristics, an acquisition step of acquiring purchasing history of the classified target user, and a resetting step of resetting segment conditions based on the acquired purchasing history, the score of the target user, and the classified segment.

[0011] In addition to the above features, the present invention provides a method for classifying target users, in which the model retention step includes a modification substep of modifying and retaining the retained trained model using the acquired purchasing history of the target user, user information of the target user, and the score.

[0012] In addition to the features, the trained model stored in the model storage step provides a method for classifying target users, which is generated for each product category that is subject to purchase.

[0013] The present invention also provides a target user classification program, which is a program for causing a computer to execute a target user classification method for classifying a target user into one of a plurality of segments according to purchasing tendencies, the target user classification program including: a model retaining step for retaining a trained model for calculating a score representing a user's purchasing tendencies, the trained data being generated by machine learning of accumulated data for a plurality of users, the user information being information about the user including at least behavioral information on a website; a score calculation step for inputting the user information of the target user into the trained model retained to calculate the score of the target user; a condition setting step for setting segment conditions, which are the number of a plurality of segments and threshold values ​​set in order to classify the target user according to the score; and a classification step for classifying the target user into one of the segments based on the set segment conditions and the calculated score of the target user.

[0014] The present invention also provides a target user classification system, which is a target user classification device that classifies a target user into one of a plurality of segments according to their purchasing tendencies, and which includes: a model holding unit that holds a trained model that calculates a score representing a user's purchasing tendencies, generated by machine learning using accumulated data for a plurality of users as training data, the user information being information about the user including at least behavioral information on websites; a score calculation unit that inputs the user information of the target user into the held trained model and calculates the score of the target user; a condition setting unit that sets segment conditions, which are the number of a plurality of segments and threshold values ​​that are set to classify the target user according to the score; and a classification unit that classifies the target user into one of the segments based on the set segment conditions and the calculated score of the target user. Effect of the Invention

[0015] According to the target user classification method of the present invention, by classifying users into segments according to their purchasing tendencies, it is possible to deliver advertisements according to the segment, thereby contributing to improving the efficiency of advertising. [Brief description of the drawings]

[0016] [Figure 1] FIG. 1 is a conceptual diagram showing an example of the overall configuration of a target user classification system according to an embodiment of the present invention. [Diagram 2] FIG. 1 is a flow diagram showing an example of a process flow of a target user classification method. [Diagram 3] A diagram showing an example of user information. [Figure 4] A diagram showing an example of segments and thresholds set according to scores [Diagram 5] A diagram showing a case where conditions are set taking into account the proportion of each segment [Figure 6] A block diagram showing one embodiment of a functional configuration of a target user classification system. [Figure 7] A conceptual diagram showing an example of the hardware configuration of a target user classification system. [Figure 8]A conceptual diagram showing an example of the hardware configuration of a target user classification system. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Note that the present invention should not be limited to these embodiments, and can be embodied in various forms without departing from the spirit of the present invention.

[0018] <Example 1> <Summary> The target user classification method of this embodiment classifies target users into segments according to their purchasing tendencies. FIG. 1 is a conceptual diagram showing an example of the overall configuration of a target user classification system according to this embodiment. As shown in the figure, a target user classification server 101, which is the main processing entity in the target user classification system, a web server 102 that provides advertiser websites embedded with tags that acquire user information, an advertisement delivery server 103, and user terminals 104 to 106 of multiple users are configured to be able to exchange information with each other via a network 107. Note that the number of user terminals is not limited to that shown in the figure, and there may be many more.

[0019] In such a configuration, the target user classification server 101 acquires user behavior information of the user terminals 104-106 on the advertiser's website from the web server 102, and calculates the user's tendency to purchase products, etc. on the website as a score using a model generated based on the accumulated user behavior information. Then, the user is classified into segments according to the calculated score. Furthermore, the advertisement delivery server 103 delivers advertisements to the target users according to the segments into which they have been classified.

[0020] The following describes the functions and processing flow of the target user classification system and the target user classification method, as well as the contents of the hardware. The functional blocks of the system described below can be realized as a combination of hardware and software. Specifically, if a computer is used, the hardware components include a CPU (Central Processing Unit), main memory, bus, or secondary storage device (hard disk drive, non-volatile memory, storage media such as CDs and DVDs, and read drives for these media), input devices used for inputting information, printers, display devices, and other external peripheral devices, as well as interfaces for the external peripheral devices, communication interfaces, driver programs and other application programs for controlling these hardware devices, and user interface applications. Then, by the arithmetic processing of the CPU according to the program deployed on the main memory, data input from input devices and other interfaces and stored in memory and hard disks is processed and stored, and commands for controlling the above hardware and software are generated. Alternatively, the functional blocks of the system may be realized by dedicated hardware.

[0021] This invention can be realized not only as a system, but also as a method. A part of such an invention can be configured as software. Furthermore, a program used to cause a computer to execute such software, and a recording medium on which the program is fixed, are naturally included in the technical scope of this invention (the same applies throughout this specification).

[0022] <Processing flow> 2 is a flow diagram showing an example of a process flow of the target user classification method according to the present invention. As shown in FIG. 2, the method includes step S201 (model storage step), step S202 (score calculation step), step S203 (condition setting step), and step S204 (classification step).

[0023] Furthermore, following step S204, step S205 (distribution step), step S206 (acquisition step), and step S207 (resetting step) can be configured.

[0024] Note that the execution entity of each step described below, for example, step S201 (model retention step), step S202 (score calculation step), step S203 (condition setting step), step S204 (classification step), step S206 (acquisition step), and step S207 (resetting step), is mainly executed by the target user classification server 101 shown in FIG. 1, and step S205 (distribution step) is mainly executed by the advertisement distribution server 103.

[0025] The target user classification method according to the present invention is a method for classifying target users into one of a plurality of segments according to their purchasing tendencies. The target users are users who are the subject of classification in the present invention, and more specifically, refer to those who browse websites using an information processing device equipped with a computer such as a personal computer or a smartphone.

[0026] Step S201 (model holding step) is a step for holding a trained model that calculates a score representing a user's purchasing tendency, generated by machine learning using data accumulated for multiple users as training data, the user information being information about the user including at least behavioral information on the website. Here, the purchasing tendency refers to the degree to which a user is expected to purchase a product or service offered by a specific website from which the user information is obtained. A strong purchasing tendency means that the user is highly likely to make a purchase.

[0027] "Website behavior information" is information about behaviors taken by a user on a website that the user accessed. The "website" referred to here is a website that displays advertisements for the company's products and services to users who accessed the website and sells the products and the like that it provides. It is assumed that the advertiser advertises to users classified using the target user classification method according to the present invention.

[0028] Examples of "user information" include the website URL, device information of users who visited the website, OS information of the device, time spent on the site, maximum scroll value, referrer (source of flow to the site, source of reference), products and services offered, regional information, time of access, etc. User information including the above-mentioned behavioral information can be obtained by embedding tags (measurement tags) on the website, and is not obtained by using third-party cookies.

[0029] 3 is a diagram showing an example of user information. As shown in the figure, the user information for user "aa0256" includes "www.sakataya.com" in the URL field, "google" in the referrer field, "20s" in the duration field, "30%" in the maximum scroll value field, "smart phone" in the user device field, "iOS" in the device OS field, "18:25" in the access time field, and "liquor" in the product information field.

[0030] Such user information is accumulated for multiple users who accessed the website, and learning data for machine learning is generated based on this accumulated user information. For example, as shown in Figure 3, user information for multiple users who accessed the website of a liquor store called "Sakataya" is accumulated for one month, and features are extracted from the accumulated data to generate learning data.

[0031] Then, using the generated training data, machine learning is performed using an algorithm such as a decision tree or random forest to generate a model for calculating a score of purchasing tendency, and the model is stored as a trained model. In addition, deep learning may be used instead of or in addition to the above-mentioned statistical approach to generate a model for score calculation. Since the two have different characteristics, one of them may be selected or both may be used in combination for model generation and score calculation depending on various conditions such as the amount of training data and the structure of the website.

[0032] The trained model to be retained may be configured to be generated for each product category that is the subject of purchase. Product categories include the category of services. Depending on the type of product, etc. that is the subject of purchase, the relationship between the user's behavior on the website and the target product, etc. (purchase, request for information, etc.) may differ. For this reason, it is preferable to generate and retain a trained model for calculating purchasing tendency as a score for each product category.

[0033] Step S202 (score calculation step) is a step of inputting user information of a target user into the above-mentioned stored trained model to calculate the score of the target user. The score representing the purchasing tendency can be defined as, for example, "1" to "10." In this case, the higher the score, the higher the purchasing tendency, and a score of "10" is assigned to a user who actually makes a purchase. The score may be defined as "1" to "100," or "1" to "1000."

[0034] Step S203 (condition setting step) is a step for setting segment conditions, which are the number of multiple segments and thresholds set to classify target users according to the scores. FIG. 4 is a diagram showing an example of segments and thresholds set according to scores. In this example, the scores are defined as "1" to "10" and four segments are set. The relationship between each segment and the score is set as follows: segment "high" has a score of "7 to 9", segment "mid" has a score of "4 to 6", and segment "low" has a score of "1 to 3". The score for dividing each segment is the threshold. In this example, the score does not have a decimal point, and the upper and lower score limits for dividing each segment ("1" and "3" for the low segment) are the thresholds.

[0035] Segment conditions may also be set based on the proportion of users classified into each segment relative to the overall population. FIG. 5 shows the segment conditions shown in FIG. 4 set with the proportion of each segment taken into consideration. As shown in the figure, the proportion of the "high" segment is set to "20%", the proportion of the "mid" segment is set to "30%, and the proportion of the "low" segment is set to "50%". Note that the proportion of the "buyer" segment is not set in order to classify according to purchasing trends. Therefore, it can be said that the number of segments set is three.

[0036] It is preferable to set the ratio of each segment so that the classification of users is not biased. For example, if there are an extremely large number of users classified as the "high" segment, it will not be possible to narrow down the users with a high propensity to purchase from the total population, and it will not be possible to improve the effectiveness of advertisement distribution using this classification.

[0037] Also, in the above example, the number of segments to be classified is set to three, but this can be increased or decreased. The number of segments to be classified can be appropriately selected depending on how the classified results are to be used. For example, if the target users are not narrowed down too much and advertisements are to be distributed widely, it is sufficient to set fewer segments (divide into "high" and "low" and distribute to "high" users). On the other hand, if there are multiple desired levels of results, such as high-level results such as product purchases or requests for information, and medium-level results such as awareness of products without aiming for such results, it is preferable to set a relatively large number of segments according to the number of those levels of results.

[0038] Then, based on the segment conditions set as described above and the target user's score calculated in the score calculation step, the target user is classified into one of the segments (step S204: classification step).

[0039] Then, in step S205 (distribution step), advertisements are distributed to the target user according to the segment into which the target user has been classified. The target user is a user who has accessed the website from which user information is acquired, who is classified into a segment, and who is the target of advertisement distribution according to the segment. When this user accesses the website, user information is acquired immediately, and as soon as the user information required for calculation is acquired, the above-mentioned score is calculated, the user is classified into one of the segments, and advertisements are displayed according to the segment.

[0040] In some cases, advertisements are delivered according to the classified segments, such as by delivering advertisements only to target users classified in a specific segment and not delivering advertisements to target users classified in other segments. In other cases, different advertisements are delivered according to the classified segments. In other cases, the presence or absence of delivery according to segments and the delivery of multiple types of advertisements according to segments may be combined.

[0041] By classifying users according to their purchasing tendencies and delivering advertisements according to the classified segments, it is possible to suppress the display of unnecessary advertisements to user groups for which advertising effects are unlikely to be expected, while effectively delivering advertisements to user groups for which advertising effects are likely to be expected, thereby improving advertising effectiveness. In addition, by matching the desired performance level with a segment and delivering advertisements according to each segment, advertising effectiveness can be improved.

[0042] Furthermore, instead of classifying users based on fixed segment settings, the number of segments and thresholds can be set arbitrarily, making it possible to optimize user segment classification according to various factors such as the type of product, the level of awareness of the product, and the desired level of results.

[0043] Furthermore, the method can be configured to include an acquisition step S206 and a reset step S207. The acquisition step S206 acquires the purchasing history of the classified target users. In this step, advertisements are delivered to the target users according to the segments classified as described above (including cases where advertisements are not delivered), and the purchasing history of the target users is acquired accordingly. The purchasing history is various information related to the purchases of the target users, such as the products or services purchased, the quantity purchased, the contract period, and the period from advertisement delivery to purchase. This information is acquired from the website from which the user information was acquired.

[0044] Then, in a resetting step S207, the segment conditions are reset based on the acquired purchase history and the calculated score and classified segment for the target user. By feeding back the purchase history to the setting of the segment conditions, it is possible to optimize advertisement delivery.

[0045] The above-mentioned model holding step S201 can also be configured to further include a correction sub-step (not shown). In the correction sub-step, the retained trained model is corrected and held using the acquired purchase history of the target user, the user information of the target user, and the calculated score. Correcting the trained model means further performing machine learning as described in the model holding step, using each piece of listed information as learning data. By reflecting such information related to the purchase history in the trained model, it is possible to contribute to improving the accuracy of calculation of scores according to purchasing trends and improving the effectiveness of advertisement delivery using the scores.

[0046] <Target user classification system> As described above, each processing step has been described as a target user classification method, but the present invention can also be realized as a target user classification system. Fig. 6 is a block diagram showing one embodiment of the functional configuration of the target user classification system.

[0047] 6, the target user classification system 600 includes a model storage unit 601, a score calculation unit 602, a condition setting unit 603, a classification unit 604, a distribution unit 605, an acquisition unit 606, and a resetting unit 607. The target user storage unit 601 also includes a correction unit 608.

[0048] The model holding unit 601 has a function of holding a trained model that calculates a score representing a user's purchasing tendency, generated by machine learning using data that is accumulated as training data for a plurality of users and is user information that is information about the user including at least behavioral information on the website. The process by the model holding unit 601 is as described in the model holding step S201.

[0049] The score calculation unit 602 has a function of inputting user information of a target user into the stored trained model and calculating the score of the target user. The process by the score calculation unit 602 is as described in the score calculation step S202.

[0050] The condition setting unit 603 has a function of setting segment conditions, which are the number of multiple segments and thresholds set to classify target users according to the scores. The process by the condition setting unit 603 is as described in the condition setting step S203.

[0051] The classification unit 604 has a function of classifying the target user into any one of the segments based on the set segment conditions and the calculated score of the target user. The process by the classification unit 604 is as described in the classification step S204.

[0052] The distribution unit 605 has a function of distributing advertisements according to the segments into which the target users are classified. The process performed by the distribution unit 605 is as described in the distribution step S205.

[0053] The acquiring unit 606 has a function of acquiring the purchase history of the classified target users. The process by the acquiring unit 606 is as described in the acquiring step S206.

[0054] The resetting unit 607 has a function of resetting the segment conditions based on the acquired purchase history and the score of the target user and the segment classified. The process by the resetting unit 607 is as described in the resetting step S207.

[0055] The correction means 608 has a function of correcting and storing the stored trained model using the acquired purchase history of the target user, the user information of the target user, and the score. The process by the correction means 608 is as described in the correction sub-step (not shown).

[0056] <Hardware configuration> 7 and 8 are conceptual diagrams showing an example of a hardware configuration of the target user classification system. First, as shown in FIG. 7, the target user classification server 700 constituting the target user classification system has a CPU 701 that performs various arithmetic processing, a RAM 702 that is a volatile recording medium, a storage 703 such as a flash memory or HDD that is a non-volatile storage medium, an input / output interface 704, and a communication interface 705. The RAM 702 reads out a program that performs various arithmetic processing to have the CPU 701 execute it, and provides a work area (work area) for the program. In addition, a plurality of addresses are assigned to the RAM 702, and the program executed by the CPU 701 can exchange data with each other and perform processing by identifying and accessing the addresses. Furthermore, various programs are executed, and data is exchanged with the advertisement delivery server 710, the web server 800 shown in FIG. 8, and the user terminal 810 via the communication interface 705 and the network 730 (the same applies throughout this specification).

[0057] Also, as shown in FIG. 7, the advertisement delivery server 710 constituting the target user classification system has a CPU 711 that performs various arithmetic processing, a RAM 712 which is a volatile recording medium, a storage 713 such as a flash memory or HDD which is a non-volatile storage medium, an input / output interface 714, and a communication interface 715.

[0058] Also, as shown in FIG. 8, the web server 800 constituting the target user classification system has a CPU 801 that performs various arithmetic processing, a RAM 802 which is a volatile recording medium, storage 803 such as a flash memory or HDD which is a non-volatile storage medium, a communication interface 804, and an input / output interface 805.

[0059] Also, as shown in FIG. 8, the user terminal 810 constituting the target user classification system has a CPU 811 that performs various arithmetic processing, a RAM 812 which is a volatile recording medium, storage 813 such as a flash memory or HDD which is a non-volatile storage medium, a communication interface 814, and an input / output interface 815.

[0060] Here, the functions of the model holding unit 601 and the correction means 608 of the target user classification system 600 in Figure 6 are mainly realized by the CPU 701, RAM 702 and storage 703 in the target user classification server 700 in Figure 7, the functions of the score calculation unit 602 and the acquisition unit 606 in Figure 6 are mainly realized by the CPU 701, RAM 702 and communication interface 705 in Figure 7, and the functions of the condition setting unit 603 and the classification unit 604 in Figure 6 are mainly realized by the CPU 701 and RAM 702 in Figure 7.

[0061] 7. The functions of the distribution unit 605 of the target user classification system 600 shown in FIG. 6 are mainly realized by the CPU 711, the RAM 712, and the communication interface 715 in the advertisement distribution server 710 in FIG.

[0062] <Effects> According to the target user classification system of this embodiment, it is possible to classify users into segments according to their purchasing tendencies and deliver advertisements according to the classified segments, thereby improving the efficiency of advertisements.

[0063] 101: Target user classification server 102: Advertisement distribution server 103: Web server 104~106: User terminal 601: Model holding section 602: Score calculation unit 603: Condition setting section 604: Classification department 605: Distribution Department 606: Acquisition Department 607: Resetting section 608: Correction means 700: Target user classification server 701:CPU 702:RAM 703: Storage 704: Input / Output Interface 705: Communication interface

Claims

1. The computer executes A method for classifying target users into one of a plurality of segments according to purchasing tendencies, comprising: a model holding step for holding a trained model that calculates a score representing a user's purchasing tendency, the trained model being generated by machine learning using accumulated data of user information, which is information about the user including at least behavioral information on a website, for a plurality of users as training data; A score calculation step of inputting user information of a target user into the stored trained model to calculate the score of the target user; A condition setting step of setting segment conditions, which are the number of multiple segments and a threshold value set to classify target users according to the score; a classification step of classifying the target user into any one of the segments based on the set segment conditions and the calculated score of the target user; A method for classifying a target user, comprising:

2. A delivery step of delivering an advertisement to the target user according to the classified segments, The method of claim 1 , further comprising:

3. An acquisition step of acquiring purchase records of the classified target users; A resetting step of resetting segment conditions based on the acquired purchase history and the score of the target user and the segment classified into the segment; The method of claim 2, further comprising:

4. The model retaining step includes: The method for classifying target users according to claim 3, further comprising a correction sub-step of correcting and retaining a retained trained model using the acquired purchasing history of the target user, the user information of the target user, and the score.

5. The method for classifying target users according to claim 1 or 2, characterized in that the trained model stored in the model storage step is generated for each product category that is subject to purchase.

6. On the computer, A program for executing a target user classification method for classifying target users into one of a plurality of segments according to purchasing tendencies, a model holding step for holding a trained model that calculates a score representing a user's purchasing tendency, the trained model being generated by machine learning using accumulated data of user information, which is information about the user including at least behavioral information on a website, for a plurality of users as training data; A score calculation step of inputting user information of a target user into the stored trained model to calculate the score of the target user; A condition setting step of setting segment conditions, which are the number of multiple segments and a threshold value set to classify target users according to the score; a classification step of classifying the target user into any one of the segments based on the set segment conditions and the calculated score of the target user; A target user classification program comprising:

7. A target user classification system that classifies target users into one of a plurality of segments according to purchasing tendencies, a model storage unit that stores a trained model that calculates a score representing a user's purchasing tendency, the trained model being generated by machine learning using accumulated data of user information, which is information about the user including at least behavioral information on a website, for a plurality of users as training data; a score calculation unit that inputs user information of a target user into the stored trained model and calculates the score of the target user; a condition setting unit that sets a segment condition, which is a number of a plurality of segments and a threshold value that are set in order to classify target users according to the score; A classification unit that classifies the target user into any one of the segments based on the set segment conditions and the calculated score of the target user; A target user classification system comprising:

Images