WO2021181900A1 - Target user feature extraction method, target user feature extraction system, and target user feature extraction server - Google Patents

Abstract

In the present invention, a computer having a processor and a memory: acquires, as user data, session data storing historical information of a user terminal which has accessed content on a web server, and user attribute data storing user attribute information; acquires, as contributor data, page attribute data storing attributes of the content, and contributor attribute data storing attributes of a contributor that provided the content; acquires, as a target type, a contributor to be extracted and a user feature that the contributor sets as a capture target; calculates an item and a value range for data to be extracted from the target type; calculates the data to be extracted corresponding to the item from the user data and the contributor data; and calculates an access feature amount on the basis of the item value range from the contributor data and the data to be extracted.

Description

Target user feature extraction method, target user feature extraction system and target user feature extraction server Capture by reference

This application claims the priority of Japanese Patent Application No. 2020-039825, which is a Japanese application filed on March 9, 2020, and incorporates it into this application by referring to its contents.

The present invention relates to a technique for extracting a specific user feature from the history information of a user who browses a website.

On websites on the Internet, there is known a technique for determining the content of an advertisement to be displayed on the content based on the behavior history (browsing, viewing or searching history) of the user who accesses the content.

In addition, on websites that provide products and services, there is known a technology that estimates user preferences based on the behavior history of users who access products and reviews, and determines recommended products and services.

For example, Patent Documents 1 and 2 are known as a web analysis technique for analyzing the preference of a user who accesses a website.

Patent Document 1 discloses a technique for estimating a recommendation candidate item for each user by referring to the user information storage unit and the user history information storage unit. Further, in Patent Document 2, the user's preference distribution is analyzed from the selection history of the item selected by the user, a recommendation index close to the center of the favorable distribution and away from the preference distribution shape is calculated, and the calculated recommendation is obtained. A technique for displaying recommended items based on an index is disclosed.

Japanese Unexamined Patent Publication No. 2015-148975 Japanese Unexamined Patent Publication No. 2011-96025

A poster who provides information such as content to a website may intend to create a new business or dig up an existing user depending on the user who accesses the provided content. When extracting the characteristics of the user targeted by the poster (hereinafter referred to as user characteristics), the items to be emphasized depend on the preference and target (acquisition target) of the poster.

For example, some contributors emphasize approaching existing customers in order to expand their current business, while others emphasize exploring potential customers in order to create new businesses. There is. In order to extract the user characteristics targeted by the poster, it is necessary to accurately extract the intention of the poster in addition to the web analysis of the user.

In Patent Document 1 of the above-mentioned conventional example, after determining a user's preference, a plan for presenting information to the user is calculated from the user's behavior history. The preference determination unit disclosed in Patent Document 1 only uses the user's attribute information and history information to determine the degree of similarity, and the intention (or preference) regarding the target on the side of providing the item (content). Was not taken into account.

Further, in Patent Document 2 of the conventional example, a user's preference is analyzed, a recommendation index away from the preference distribution shape is calculated, and an unexpected item is provided. However, there is a problem that this Patent Document 2 does not consider the target intended by the provider of the item.

Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to extract user characteristics that a poster who provides information to a website wants to acquire from the history of users who have accessed the website.

The present invention is a target user feature extraction method in which a computer having a processor and a memory extracts user features targeted by a poster from history information of accessing the contents of a web server, wherein the computer is the web server. The user data acquisition step of acquiring the session data storing the history information of the user terminal that has accessed the contents of the above and the user attribute data storing the attribute information of the user who uses the user terminal as user data, and the computer , The poster data acquisition step of acquiring the page attribute data storing the attributes of the content and the poster attribute data storing the attributes of the poster who provided the content as poster data, and the computer The preference acquisition step of accepting the poster to be extracted and acquiring the information of the user targeted by the poster as the target type, and the item of the data to be extracted from the target type of the poster by the computer. A target calculation step for calculating a range of values of the item, a session feature calculation step for the computer to calculate extraction target data corresponding to the item from the user data and the poster data, and the extraction by the computer. It includes an access feature extraction step of calculating an access feature amount based on a range of values of the item from the target data and the poster data.

Therefore, the present invention makes it possible to extract user characteristics according to the preference of the poster who provides the content from the history information of the user who has accessed the website. As a result, in addition to extracting the users expected by the poster of the information, it is possible to extract new user characteristics that are different from the intention of the poster, and it is possible to create a new business.

Details of at least one implementation of the subject matter disclosed herein are described in the accompanying drawings and in the description below. Other features, aspects, and effects of the disclosed subject matter are manifested in the disclosures, drawings, and claims below.

It is a block diagram which shows the Example of this invention and shows an example of the structure of the target user feature extraction system. It is a block diagram which shows the Example of this invention and shows an example of the structure of the target user feature extraction server. It is a figure which shows the Example of this invention and shows the outline of the process performed in the target user feature extraction server. It is a figure which shows the Example of this invention and shows an example of a session data. It is a figure which shows the Example of this invention and shows an example of the user attribute data. It is a figure which shows the Example of this invention and shows an example of the data to be extracted. It is a figure which shows the Example of this invention and shows an example of the range conversion information. It is a flowchart which shows the Example of this invention and shows an example of the processing performed in the session feature calculation part of the target user feature extraction server. It is a flowchart which shows the Example of this invention and shows an example of the processing performed in the target calculation part. It is a flowchart which shows the Example of this invention and shows an example of the processing performed in the range conversion part. It is a figure which shows the Example of this invention and shows an example of the process performed in the range conversion part. It is a flowchart which shows the Example of this invention and shows an example of the processing performed in the target determination item processing part. It is a figure which shows the Example of this invention and shows an example of the selection data for learning. It is a graph which shows the Example of this invention and shows an example of selection data. It is a figure which shows the Example of this invention and shows an example of the category table. It is a figure which shows the Example of this invention and shows an example of the condition table. It is a graph which shows the Example of this invention and shows an example of selection data. It is a figure which shows the Example of this invention and shows an example of selection data. It is a figure which shows the Example of this invention and shows an example of the industry similarity degree map. It is a figure which shows the Example of this invention and shows an example of the browsing number data. It is a figure which shows the Example of this invention and shows an example of the industry similarity degree map. It is a figure which shows the Example of this invention and shows an example of statistical data. It is a figure which shows the Example of this invention and shows an example of the industry similarity degree map. It is a figure which shows the Example of this invention and shows an example of the extraction result screen. It is a figure which shows the Example of this invention and shows an example of the extraction target.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention and showing an example of the configuration of a target user feature extraction system.

The target user feature extraction system supplies information to the web server 200 that manages the website including the content 210 and the advertisement 220, the user terminals 100-1 to 100-3 that access the website information, and the web server 200. The users (target types) that the posters who provide information from the posting terminals 300-1 to 300-3 and the posting terminals 300-1 to 300-3 want to acquire are extracted from the access history (log 230) of the web server 200. The target user feature extraction server 1 is included.

As the code of the user terminals 100-1 to 100-3, the code "100" is used, omitting the "-" and subsequent parts when not individually specified. Similar codes are used for the codes of other components.

Posting terminals 300-1 to 300-3 are operated by contributors A, B, and C in different industries, and each contributor A to C also serves as an advertiser to provide content 210 and advertisement 220.

In this embodiment, the poster who operates the posting terminal 300 serves as both the provider of the content 210 and the advertiser, but the present invention is not limited to this, and the poster and the advertiser of the content 210 are different. May be good. Further, the user terminals 100-1 to 100-3 are operated by users in different industries a, b, and c, and browse the contents 210 and the advertisement 220 of the web server 200.

The web server 200 is composed of a computer, and transmits the access history (history information) of the user terminal 100, the information of the poster who uses the posting terminal 300, and the attribute data of the content 210 to the target user feature extraction server 1. The web server 200 may be connected to a database server, an application server, or the like to build a website.

The target user feature extraction server 1 is a history of users who have accessed the web server 200 from the users (users of the user terminal 100) that the posters A to C who provide information to the website provided by the web server 200 want to acquire. Extract user characteristics from session data). Further, the target user feature extraction server 1 analyzes the content (page) 210 provided by the posting terminal 300 and extracts it as a page feature.

The target user feature extraction server 1 collects the access history of the user terminal 100 at a predetermined cycle (for example, one month), and extracts the access feature amount including the user feature and the page feature for the poster to be extracted. Notify the posting terminal 300.

The posting terminal 300 notifies the target user feature extraction server 1 in advance of the user information that the poster wants to acquire as the target type. Alternatively, the poster may notify the web server 200 of the target type from the posting terminal 300, and the target user feature extraction server 1 may acquire the target type from the web server 200.

FIG. 2 is a block diagram showing an example of the configuration of the target user feature extraction server 1. The target user feature extraction server 1 is a computer including a processor 11, a memory 12, a storage device 13, an input device 14, an output device 15, and a communication device 16.

The communication device 16 is connected to the network 400 and communicates with the web server 200 and the posting terminal 300. The output device 15 is composed of a display or the like. The input device 14 is composed of a keyboard, a mouse, or a touch panel.

The memory 12 includes a processing target selection unit 21, a session feature calculation unit 22, a target calculation unit 23, an access feature extraction unit 27, a target determination item processing unit 28, a data processing unit 30, and a learning unit 31. It is loaded as a program and executed by the processor 11.

The processor 11 operates as a functional unit that provides a predetermined function by executing processing according to the program of each functional unit. For example, the processor 11 functions as the session feature calculation unit 22 by executing the process according to the session feature extraction program. The same applies to other programs. Further, the processor 11 also operates as a functional unit that provides each function of a plurality of processes executed by each program. A computer and a computer system are devices and systems including these functional parts.

In the storage device 13, session data 41, user attribute data 42, page attribute data 43, poster attribute data 44, poster target data 45, and range conversion information 46 are used as data used by each of the above programs. Is stored.

The session data 41 shows the access history of the user terminal 100 that has accessed the content 210 (or the advertisement 220) among the logs 230 collected by the web server 200. The user attribute data 42 indicates the attributes of the user who uses the user terminal 100. The page attribute data 43 indicates the attributes of the content 210. The poster attribute data 44 indicates the attribute of the poster. In the poster target data 45, the user group (target type) that the posters A to C want to acquire is set as qualitative information. The poster target data 45 can also set a range (or threshold value) of items and values. In the range conversion information 46, an item of analysis target data that specifies a user group for each target type and a range (or threshold value) of the value of the item are set. The details of each data will be described later.

Next, the outline of each program running on the target user feature extraction server 1 will be described.

The processing target selection unit 21 receives the period of the session data 41 used for analysis and the poster to be analyzed from the input device 14 or the like in the access history (session data 41) of the user terminal 100 acquired from the web server 200. It should be noted that the analysis may be performed on the targets of all the contributors by designating the period of the session data 441 without designating the contributors.

The target calculation unit 23 accepts the poster to be analyzed, acquires the range of user characteristics that the poster wants to acquire from the poster target data 45 as target information, and analyzes the session data based on the target information. Determine the range of values.

As will be described later, the items and ranges for analyzing session data are set according to the target information (acquisition target) and preference for each poster A to C. For example, as an item for determining the target of a poster, a web server. Using the number of views and staying time at 200, the range can be specified by the range of these numerical values, the threshold value, and the like.

The item and range for analyzing the session data 41 may be determined by the range calculation unit 26 with reference to the range conversion information 46, or the target determination model 25 may calculate the item and range.

When the range conversion information 46 corresponding to the target information of the poster target data 45 exists, the range conversion unit 24 causes the range calculation unit 26 to refer to the range conversion information 46 to determine the item and the range. Further, when the range conversion information 46 corresponding to the target information does not exist, the range conversion unit 24 has the session data 41, the user attribute data 42, the page attribute data 43, the poster attribute data 44, and the target for the specified period. Information is input to the target determination model 25 to generate items and ranges to be analyzed.

The session feature calculation unit 22 acquires the session data 41 of the period accepted by the processing target selection unit 21, the user attribute data 42 of the user, the page attribute data 43, and the poster attribute data 44 included in the session data 41, and targets the target. The data of the items determined by the calculation unit 23 is generated as the extraction target data indicating the characteristics of the session. If the determined item exists in the session data 41, the session data 41 for the specified period is set as the extraction target data 50.

The session feature calculation unit 22 uses the target determination item processing unit 28 and the data processing unit 30 to generate the extraction target data 50 corresponding to the determined item, as will be described later. The target determination item processing unit 28 uses the similarity calculation unit 29 according to the target item.

Further, in the session feature calculation unit 22, the user terminal 100 sets the web server 200 for each page of the content 210 accessed by the user terminal 100, for each tag of the page attribute data 43, or for each poster who provides the content 210. The visit history can be calculated as data showing the characteristics of the session.

The access feature extraction unit 27 receives the extraction target data from the session feature calculation unit 22 and the items and ranges from the target calculation unit 23, and extracts the user features and the features (page features) of the accessed content 210.

First, the access feature extraction unit 27 calculates the user feature amount as the user feature based on the extraction target data from the session feature calculation unit 22, the item from the target calculation unit 23, and the range of the value of the item. Further, the access feature extraction unit 27 receives the attribute data of the poster (poster attribute data 44) and the attribute data of the content 210 provided by the poster to the web server 200 (page attribute data 43), and is accessed by the user. The feature amount of the content 210 related to the above is extracted as a page feature.

The user features and page features extracted by the access feature extraction unit 27 are notified to the posting terminal 300. In addition, the access feature extraction unit 27 can display the extracted user features and page features on the output device 15.

The user features extracted by the access feature extraction unit 27 may include, for example, the ratio of the type of industry of the user who accessed the content 210 of the poster to be extracted, the characteristics of the session (the number of repeats), and the like as the feature amount. can.

Further, the page feature extracted by the access feature extraction unit 27 can include, for example, the ratio of tags of the accessed content 210, the average staying time of each page, and the like as the feature amount.

The learning unit 31 inputs session data 41, user attribute data 42, poster attribute data 44, page attribute data 43, and poster target data 45, performs machine learning, and generates a target determination model 25. The target determination model 25 is generated in advance before the user feature 51 and the page feature 52 are extracted.

<Data>
Next, the data used by each program will be described. FIG. 4 is a diagram showing an example of session data 41. The session data 41 is historical information collected by the target user feature extraction server 1 from the web server 200 at a predetermined cycle or the like.

The session data 41 is a table that includes the ID 411, the access time 412, the visit page 413, the number of repeats 414, and the departure time 415 in one record.

ID411 stores the identifier of the user terminal 100. ID411 is a value given by the web server 200, and may be a unique value in the target user feature extraction system.

The access time 412 stores the date and time when the user terminal 100 started accessing the page. The visit page 413 stores the URL of the content 210 accessed by the user terminal 100.

The repeat count 414 stores the cumulative number of times the page has been accessed. The withdrawal time 415 stores the time when the user terminal 100 finishes browsing the page.

FIG. 5 is a diagram showing an example of user attribute data 42. The user attribute data 42 is a table set by the target user feature extraction server 1. The user attribute data 42 is a table that includes ID 421, IP 422, industry 423, and sales 424 in one record.

ID 421 stores the identifier of the user terminal 100. ID 421 is the same value as ID 411 of the session data 41. IP422 stores the IP address of the user terminal 100.

The industry 423 stores the industry of the user's company (or group) that uses the user terminal 100. Since the industry 423 can identify the company to which the user belongs from the IP address of the user terminal 100, the industry may be determined from the information of the company. Sales 424 stores the sales of the company to which the user belongs.

The type of business and sales of the user who uses the user terminal 100 may be set by the administrator of the target user feature extraction server 1 or the like, or may be set from a preset database or the like.

FIG. 6 is a diagram showing an example of the extraction target data 50. The extraction target data 50 is intermediate data calculated by the session feature calculation unit 22. In the illustrated example, the number of views and the average staying time are output from the target calculation unit 23 as the items of the extraction target data for specifying the user group.

In the case of the illustrated extraction target data 50, the session feature calculation unit 22 aggregates the pages viewed by each user for each contributor from the session data 41 within the period accepted by the processing target selection unit 21, and the user attribute data 42. Here is an example combined with.

The extraction target data 50 is a table that includes ID 501, contributor 502, number of views 503, average stay time 504, and industry 505 in one record.

The ID 501 stores the identifier of the user terminal 100. ID501 is the same value as ID411 of the session data 41. The contributor 502 stores the identifier of the contributor of the content 210 viewed by the user of the ID 501. The identifier of the poster of the content 210 is information preset for each page constituting the content 210, and is acquired from the page attribute data 43 transmitted from the web server 200.

The number of views 503 stores the total number of pages provided by the poster 502 viewed by the user of the ID 501. The average stay time 504 stores the average time that the user of the ID 501 stays (views) on the page provided by the poster 502. The industry 505 stores the industry 423 of the user attribute data 42.

FIG. 7 is a diagram showing an example of the range conversion information 46. The range conversion information 46 is a table for converting the qualitative information of the poster target data 45 into a range of items and values to be extracted.

The range conversion information 46 is information in which the items of the extraction target data 50 calculated from the session data 41, the user attribute data 42, and the like and the data range 462 are preset for each target type 461 that classifies the user group that the poster wants to acquire. Is. The target type 461 is a value of the target information of the poster target data 45.

As an example of the target type 461, an example in which "new", "existing", "people who subscribe over time", "repeater", "good customer", and "people who are interested in cutting" are set is shown. There is.

The "new" target type 461 indicates that the poster provides information on the content 210 and the advertisement 220 to the website of the web server 200 for the purpose of acquiring new users. In this embodiment, a range 462 is set in advance in which a user who has viewed the content 210 of the corresponding poster 50 times or less is regarded as a "new" user.

The "existing" target type 461 indicates that the poster provides information to the web server 200 for the purpose of digging up existing users. In this embodiment, a range 462 is set in advance in which a user whose content 210 is viewed by the corresponding contributor exceeds 50 as an "existing" user.

The target type 461 of the "person who subscribes over time" indicates that the content 210 is provided to the web server 200 for the purpose of acquiring users who browse the content 210 of the poster over time. In this embodiment, a range 462 for determining a user whose average staying time 504 of the content 210 of the corresponding poster is 500 seconds or more per page as the corresponding user is set in advance.

The target type 461 of the "repeater" indicates that information is provided to the web server 200 for the purpose of acquiring users who repeatedly browse the content 210 of the poster. In this embodiment, a range 462 for determining a user whose content 210 of the corresponding poster has a repeat number of 414 of 2 or more and a visit interval of 1 week or less as the corresponding user is set in advance.

The target type 461 of the "excellent customer" is preset with a range 462 for determining a user who accesses the content 210 of the poster and whose sales 424 of the company to which the user belongs is 1 billion yen or more as the corresponding user. NS.

The target type 461 of the "person who is interested in cutting" is preset with a range 462 for determining the user who has accessed the page including the "cutting" tag in the content 210 of the poster as the corresponding user.

When the target type 461 corresponding to the target information of the poster target data 45 does not exist in the range conversion information 46, the range conversion unit 24 adds the session data 41 and the user attribute data 42 to the target determination model 25 as described later. And page attribute data 43 and poster attribute data 44 are input to generate items and ranges.

Although not shown, the page attribute data 43 is a table that includes a URL, a tag indicating the type of the content 210, and an identifier of the poster who provides the content 210 for each page of the content 210. The page attribute data 43 may include static information such as words used in the content 210, or may include features of sentences and articles calculated by word2vec or the like.

Although not shown, the poster target data 45 is set with the poster identifier and the target information selected in advance by the poster. The target information of the poster target data 45 corresponds to the value of the target type 461 of the range conversion information 46 described above, but a value not included in the target type 461 of the range conversion information 46 can be set. Further, the poster target data 45 can be set with information including an item and a range of values in addition to qualitative information. Although not shown, the poster attribute data 44 stores the identifier of the poster, the type of business of the poster, and the department to which the poster belongs.

<Extraction process>
Hereinafter, an example of the processing performed by the target user feature extraction server 1 will be described. FIG. 3 is a diagram showing an outline of processing performed by the target user feature extraction server 1. This process is started based on the command of the user of the target user feature extraction server 1.

The processing target selection unit 21 accepts the extraction target period and the poster. As described above, when the contributor is not input, all the contributors of the web server 200 are extracted.

First, the target calculation unit 23 receives posters from the processing target selection unit 21, acquires the target type for each poster from the poster target data 45, and corresponds to the target information from the range conversion information 46 or the target determination model 25. Determine the item and value range to be used.

The target calculation unit 23 determines the item and range of the extraction target data 50 for each contributor using the range conversion unit 24, outputs the item to the session feature calculation unit 22, and outputs the range to the access feature extraction unit 27. do.

As described above, when the target type 461 corresponding to the target information does not exist in the range conversion information 46, the range conversion unit 24 transfers the session data 41, the user attribute data 42, the page attribute data 43, and the target determination model 25 to the target determination model 25. The poster attribute data 44 is input to determine the item and range to be extracted.

When the target type 461 corresponding to the target information does not exist in the range conversion information 46, the range conversion unit 24 generates an item and a range to be extracted by the target determination model 25, thereby generating an access feature extraction unit. 27 can extract user features that match the target information.

The target determination model 25 is a model generated in advance by machine learning. The learning unit 31 of the target user feature extraction server 1 generates the target determination model 25 by machine learning the poster attribute data 44 and the page attribute data 43 in the session data 41 and the user attribute data 42 of the user terminal 100.

The session feature calculation unit 22 acquires the session data 41 within the period received from the processing target selection unit 21, and acquires the user attribute data 42 corresponding to the ID 411 of the session data 41.

The session feature calculation unit 22 receives items from the target calculation unit 23 and generates extraction target data 50 including the items specified from the session data 41 and the user attribute data 42 within the specified period.

The item of the extraction target data 50 is determined according to the content of the range 462 corresponding to the target type 461 of the range conversion information 46 or the output of the target determination model 25. The generated extraction target data 50 is output to the access feature extraction unit 27. The session feature calculation unit 22 may generate the extraction target data 50 for each poster to be extracted, or may generate the extraction target data 50 including all the items of the poster to be extracted.

The access feature extraction unit 27 receives a range of values to be extracted from the target calculation unit 23, and receives the extraction target data 50 from the session feature calculation unit 22. The access feature extraction unit 27 applies a well-known or known analysis technique to extract user features corresponding to the range 462 specified from the extraction target data 50 for each contributor, and sets the user feature 51 as the feature amount of the session. Output.

For example, when the feature extraction model generated by machine learning is used, the access feature extraction unit 27 uses the target type of the poster as the explanatory variable and the range of the number of views as the objective variable, and estimates the user features included in the target information. do.

Further, the access feature extraction unit 27 acquires the poster attribute data 44 and the page attribute data 43, extracts the page accessed by the user included in the extraction target data 50, and indicates the page feature 52 indicating the feature amount of the session. Is output as. The access feature extraction unit 27 can also estimate the extraction of the page feature 52 by machine learning in the same manner as described above. The access feature extraction unit 27 is not limited to the machine learning model, and may apply statistical values such as an average value and a median value.

FIG. 23 is a diagram showing an example of the user feature 51 extracted by the access feature extraction unit 27 and the extraction result screen 600 of the page feature 52. Further, FIG. 24 is a diagram showing an example of session data 41 analyzed by the access feature extraction unit 27.

In FIG. 24, users 1 to 3 using the user terminal 100 access pages A1 and A2 of poster A and page B1 of poster B, and page features of page D1 of poster D are also pages A1 and A2. An example similar to B1 is shown.

FIG. 23 shows an example in which the user feature 51 of the user corresponding to the target type 461 of the poster A and the extraction result of the page feature 52 are displayed as extraction targets.

The target type 461 of the contributor A shows an example in which users 1 to 3 shown in FIG. 24 correspond. As the user characteristic 51, it is shown that the metal industry accounts for 67% and the material manufacturer accounts for 33% in the industries of users 1 to 3, and the access of users 1 to 3 is extracted as a feature that the number of repeats is 414.

Further, the page feature 52 accessed by the users 1 to 3 includes metal and processing as a tag of the page attribute data 43, and it is displayed that the average stay time 504 is long as a feature of the session data 41.

By the above processing, the target user feature extraction server 1 is obtained from the ID 411 for each session, the visit page 413, the time information (412, 415), the industry of the user attribute data 42, the tag of the page attribute data 43, and the poster. It is possible to extract user features that match the target information (poster's preference) from the extraction target data 50.

The target information of the poster is, for example, a qualitative value of "targeting a new customer", and the item and range obtained by quantitatively converting this target information are "the number of views to the article of the poster is 30". An industry that is less than 50 and has a distance (similarity) of 10 or more from the attributes of the poster.

One of the session features (user features) to be extracted by the access feature extraction unit 27 is the data of the number of visits (views) of the user to the content 210 of the poster for each industry. The other is a feature indicating the distance between the attributes of the user's industry, and for this, the result of calculating the similarity from the number of visits to the tag of the user's industry and the page attribute data 43 can be used.

Therefore, the data processing unit 30 calculates the total number of page visits for each user for each poster's content 210 and for each attribute (industry 423) associated with the user's ID 411. Further, regarding the distance, the data processing unit 30 calculates the distance related to the feature amount by using, for example, a method of calculating the similarity such as a multidimensional scaling method, and constitutes the extraction target data 50 with these data.

With respect to the extraction target data 50, the access feature extraction unit 27 sets the user's industry and the number of visits as the user's characteristics as the characteristics of the session that matches the poster's preference of "targeting new customers". It can be presented as 51. Further, the access feature extraction unit 27 extracts the features of the page visited by the user's industry when the session data is narrowed down by the link destination of the user's industry included in the session features and the content 210 of the poster. It can be output as page feature 52.

As the range 462, in addition to the above, the distance of the feature amount (similarity) of the industry among a plurality of users who have accessed the visit page 413 is calculated by using the industry 423 of the user attribute data 42. The access feature extraction unit 27 can present the content 210 as the user feature 51 as a group of users according to the distance for each poster.

FIG. 8 is a flowchart showing an example of processing performed by the session feature calculation unit 22 shown in FIG. When the session feature calculation unit 22 receives a period from the processing target selection unit 21 and receives an item from the target calculation unit 23, the session feature calculation unit 22 performs the following processing.

The session feature calculation unit 22 acquires the data within the received period from the session data 41 (S1). Next, the session feature calculation unit 22 acquires the user attribute data 42 of the user (user terminal 100) included in the session data 41 within the designated period (S2).

The session feature calculation unit 22 combines the session data 41 acquired in step S1 with the user attribute data 42 in which the user IDs 411 and 421 match to generate the combined data (S3).

The session feature calculation unit 22 determines whether or not the item received from the target calculation unit 23 is included in the combined data generated in step S3 (S4). When the combined data includes all the items to be extracted, the session feature calculation unit 22 outputs the combined data as the extraction target data 50 as it is. On the other hand, when the session feature calculation unit 22 does not include all the items to be extracted in the combined data, the session feature calculation unit 22 proceeds to step S5 and generates data of the received items from the combined data by the data processing unit 30.

The data processing unit 30 generates data of the items to be extracted determined by the target calculation unit 23 from the combined data for each user.

For example, when the item is the average stay time, the data processing unit 30 calculates the difference between the departure time 415 and the access time 412 for the record in which the ID 411 of the session data 41 and the visit page 413 match, and averages the same visit page 413. Calculate the value as the average staying time. Further, the data processing unit 30 may specify the contributor (identifier) of each visit page 413 with reference to the page attribute data 43, and calculate the average staying time for each contributor.

Next, the session feature calculation unit 22 outputs the data generated for each of the above items in step S6 to the access feature extraction unit 27 as the extraction target data 50.

Through the above processing, the session feature calculation unit 22 calculates the data of the items used for determining the target information from the session data 41 and the user attribute data 42 within the designated period, and outputs the data as the extraction target data 50.

FIG. 9 is a flowchart showing an example of processing performed by the target calculation unit 23 shown in FIG. The target calculation unit 23 receives a poster from the processing target selection unit 21 and starts the following processing.

The target calculation unit 23 acquires target information from the poster target data 45 for the accepted poster (S11). The target calculation unit 23 determines whether or not the acquired target information is information including an item and a range (or a threshold value) of a value (S12). If the item and range are included, the process proceeds to step S14, and if not, the process proceeds to step S13.

In step S13, the target information is qualitative information. In this case, the target calculation unit 23 uses the range conversion unit 24 to convert the qualitative information into items and ranges. Then, in step S14, the converted item and the range of values are output to the session feature calculation unit 22 and the access feature extraction unit 27.

FIG. 10 is a flowchart showing an example of processing performed by the range conversion unit 24 of the target calculation unit 23. The target calculation unit 23 determines whether or not the range conversion information 46 corresponding to the target information exists (S21). If the range conversion information 46 exists, the process proceeds to step S22, and if the range conversion information 46 does not exist, the process proceeds to step S23.

In step S22, the range conversion unit 24 refers to the range conversion information 46, acquires the range 462 from the target type 461 corresponding to the target information, and determines the range of items and values set in the range 462.

In step S23, the range conversion unit 24 inputs the session data 41, the user attribute data 42, the page attribute data 43, and the poster attribute data 44 into the target determination model 25 in the target determination model 25, and the item and range to be extracted. To decide.

By the above processing, when the target information of the poster target data 45 is qualitative information, the range conversion information 46 or the target determination model 25 determines the items to be extracted and the range of values.

FIG. 11 is a diagram showing an example of the target determination item processing unit 28 performed by the range conversion unit 24 of the target calculation unit 23. When the target determination model 25 is used, the range conversion unit 24 processes the user data 510 of the session data 41 and the user attribute data 42 by the target determination item processing unit 28 for each content 210 (page) of the poster. The statistical processing described later is performed (S231). The session data 41 is data within the period received from the processing target selection unit 21.

Next, the range conversion unit 24 combines the page attribute data 43, the poster attribute data 44, and the poster data 520 including the poster target data 45 with the processing result of the target determination item processing (S232). The page attribute data 43 uses the data corresponding to the visit page 413 included in the session data 41 within the period received from the processing target selection unit 21.

Then, the data obtained by combining the target determination item processing result of the user data 510 and the poster data 520 is given to the target determination model 25 to determine the item to be extracted and the range of values.

FIG. 12 is a flowchart showing an example of processing performed by the target determination item processing unit 28. This process is executed in step S231 of FIG. 11 above.

The target determination item processing unit 28 acquires the user data 510 shown in FIG. 11 (S32). The target determination item processing unit 28 determines whether or not to use the user attribute data 42 (S32). Whether or not to use the user attribute data 42 can be set in advance for each poster identifier in the poster target data 45, for example.

The target determination item processing unit 28 refers to the poster target data 45 and proceeds to step S33 when using the user attribute data 42, and proceeds to step S36 when not using it.

In step S33, the target determination item processing unit 28 acquires the tags of the industry 423 of the user attribute data 42, the visit page 413 of the session data 41, and the page attribute data 43, and calculates the feature amount of the industry 423. Then, the target determination item processing unit 28 calculates the distance between the user's industry 423s in the space of the calculated feature amount by using a multidimensional scaling (MDS: Multi-Dimensional Scaling) or the like, and calculates this distance. Let it be similar.

As shown in FIG. 19, this process aggregates the number of views of the user attribute data 42 for each industry 423 for each tag of the page attribute data 43 for each contributor, and generates the number of views data 530. The number of views data 530 in FIG. 19 is information obtained by calculating the total number of views of the user attribute data 42 for each industry 423 for each tag of the content 210.

In FIG. 19, for the page of tag A of the content 210 of the poster A, the aggregated value of the number of views of each of the users of the industry a to the industry d is stored. The number of views data 530 in FIG. 19 can express the amount of interest of the user's industry 423 for each tag of the poster A.

The target determination item processing unit 28 calculates the feature amount 1 and the feature amount 2 from the browsing number data 530 of FIG. 19 by using the multidimensional scaling analysis method, and as shown in FIG. 20, the feature amount 1 and the feature amount 2 Industry 423 is arranged in the space. Note that FIG. 20 is a map expressing the distance between the industry 423 represented by the features 1 and 2 as the degree of similarity. In the illustrated example, an example in which the degree of similarity is calculated from the number of views data 530 for the content 210 of the poster A is shown.

Next, in step S34 of FIG. 12, the target determination item processing unit 28 determines whether or not to utilize the characteristics of the session. Whether or not to use the session feature can be set in advance for each poster identifier in the poster target data 45, for example.

The target determination item processing unit 28 refers to the poster target data 45 and proceeds to step S35 when using the characteristics of the session, and ends the process when not using it. In step S35, the target determination item processing unit 28 performs statistical processing of the user data 510 and the poster data 520 for each page of the poster. FIG. 21 is a diagram showing an example of statistical data 540 generated by statistical processing.

The statistical data is a diagram showing the result of the target determination item processing unit 28 totaling the number of views of the content 210 for each contributor by the user's industry 423. In FIG. 21, for the content 210 of the poster A, the aggregated value of the number of views of each of the users of the industry a to the industry d is stored. The statistical data 540 of FIG. 21 can express the amount of interest of the user's industry 423 for each contributor.

FIG. 22 is a diagram showing an example of a similarity map in which the result of statistical processing is added to the map of FIG. 21. In the figure, the size of the circle for each industry is proportional to the number of views of users in each industry for poster A.

As described above, the target determination item processing unit 28 outputs information that aggregates the distance between the industry 423 that has been statistically processed by the user attribute data 42 and the page attribute data 43 for each contributor.

In step S36 when the user attribute data 42 is not used, the data processing unit 30 used by the session feature calculation unit 22 performs data processing such as the staying time for each visit page 413 to reach the target determination model 25. Output.

As described above, when the target calculation unit 23 uses the target determination model 25, there is no range conversion information 46 by inputting the data generated by the processes of FIGS. 10 to 12 into the target determination model 25. Also, the range of items and values can be determined.

<Learning process>
Next, the learning process for constructing the target determination model 25 performed by the learning unit 31 will be described. FIG. 13 is a diagram showing an example of selection data 550 that defines data for learning the target determination model 25 when the user attribute data 42 and the poster attribute data 44 are not used.

The selection data 550 is a table that includes the ID 5501, the target customer 5502, the average stay time 5503, and the number of views 5504 in one record. The contributor's identifier is stored in ID5501. The target customer 5502 stores the target type selected by each contributor. The target type may be selected for each contributor from preset qualitative information.

In the average stay time 5503, the condition of the average stay time that the user stays (views) is stored in the page provided by the poster of the ID 5501. The number of views 5504 stores the condition of the total number of views by the user on the page provided by the poster of the ID 5501.

The selection data 550 may be generated by the administrator of the target user feature extraction server 1 based on the target type received from the poster, or may be input from the posting terminal 300.

In the illustrated example, there are two types of target types for constructing the target determination model 25: "new" that prioritizes new customers and "existing" that prioritizes existing customers. An example of using 5503 and 5504 views is shown.

The area selected by the poster in the space of the average stay time 5503 and the number of views 5504 of the selection data 550 in FIG. 13 is as shown in FIG. FIG. 14 is a graph showing an example of selection data 550.

In FIG. 14, the area of the user feature targeted by the posters A and C who selected "existing" is shown by a solid line, and the area of the user feature targeted by the posters B and D who selected "new" is shown by a broken line. Indicated.

The learning unit 31 generates learning data under the conditions set in the selection data 550 and gives it to the target determination model 25 for learning. The learning data given to the target determination model 25 may be generated from the actual session data 41 and the user attribute data 42, but dummy data may be used.

The characteristics of the session of the target type (new or existing) do not have to be processed from the actual data, and some characteristics of the session are shown by dummy data, and the target type is selected for multiple contributors. It is possible to use the data in which the results of the trials are retained. Further, the area corresponding to the target type is obtained by converting the characteristics of the target type into the items of the selection data 550 in advance, and which item is selected for each target type may be output as shown in the graph of FIG. ..

Note that the learning data may be defined in the category table 560 and the condition table 570 shown in FIGS. 15A and 15B. FIG. 15A shows an example of a category table 560 that reflects the poster's target type (preference). FIG. 15B shows an example of a condition table for setting an item and a range of values for each category.

The category table 560 of FIG. 15A includes the ID 5601, the target customer 5602, and the category number 5603 in one record. The contributor's identifier is stored in ID5601. The target customer 5602 stores the target type selected by each contributor. The target type may be selected for each contributor from preset qualitative information. In the category number 5603, the number of the area indicating the characteristics of the session selected by each contributor is set. The category number 5603 stores a number selected by the poster from the preset numbers.

The condition table 570 of FIG. 15B includes the target customer 5701, the average stay time 5702, and the number of views 5703 in one record. The target customer 5701 stores a number corresponding to the category number 5603 of the category table 560.

The average stay time 5702 stores conditions related to the average stay time (viewed) by the user on the page provided by the poster. The number of views 5703 stores a condition regarding the total number of pages viewed by the user provided by the poster.

The selection data 550 may be generated by the administrator of the target user feature extraction server 1 based on the target type received from the poster, or may be input from the posting terminal 300.

In the example of FIG. 15A, the contributors A and C who selected "existing" as the target type selected the category number 5603 = "1", and the contributors B and D who selected "new" each had the category number 5603. An example in which = 2 and 3 are selected is shown.

The area corresponding to the category number 5603 is limited by the average stay time 5702 and the number of views 5703 of the condition table 570, and is as shown in FIG. In FIG. 16, data having an average stay time of less than 100 hours is classified into category “2” regardless of the number of views 5703. Further, the data in which the number of views 5703 is less than 50 and the average staying time is less than 100 hours is classified into category "3", and the other areas are classified into category "1".

As described above, the data for learning may be generated by the category table 560 that stores the preference of the poster and the condition table 570 that determines the range of the data.

Next, when determining the learning data of the target determination model 25, the user attribute data 42 and the poster attribute data 44 are used to be used between the poster and the user's industry in the same manner as in FIGS. 21 and 22 above. An example using distance is shown below.

FIG. 17 is a diagram showing an example of selection data 580 that defines data for learning of the target determination model 25. The selection data 580 is a table that includes the ID 5801, the target customer 5802, the industry 5803, the selected industry 5804, the distance 5805, and the number of views 5806 in one record. The identifier of the poster is stored in the ID 5801. The target customer 5802 stores the target type selected by each contributor. The target type may be selected for each contributor from preset qualitative information.

The industry 5803 stores the industry of the poster set in the poster attribute data 44. In the selected industry 5804, the industry of the user selected by the poster is stored. The distance 5805 stores the distance of the degree of similarity between the poster and the user's industry. The number of views 5806 stores the total number of pages viewed by the user provided by the poster of the ID 5801.

The selection data 580 may be generated based on the target type received from the poster by the administrator of the target user feature extraction server 1, or may be input from the posting terminal 300.

In the illustrated example, in the record of ID5801 = contributor A, "existing" is selected as the target customer 5802, the industry of poster A is 5803 = a, and the industry of the target user selected by poster A is 5804 = b. Data with a distance of 5805 = Lab between industries and a number of views of 5806 of 100 or more is defined as learning data.

FIG. 18 shows a map of the degree of similarity between the industry 5803 = a of the poster A of the selection data 5801 shown in FIG. 17 and the selection industry 5804. In the figure, the size of the circle for each industry is proportional to the number of views of the users of each industry a to d for the content 210 of the poster A.

In this example, the similarity between industries is calculated from the session data 41, the user attribute data 42, and the poster attribute data 44, and the distance to the selected industry is calculated from the poster attributes and the target type with reference to the similarity. And extract information about the industry.

In the illustrated example, the similarity of the user's attributes is applied as the similarity of the poster's attributes. However, if the type of business is similar regardless of whether the user or the poster is used, We use the assumption that the behavior for the tag of interest is similar.

Further, for example, the similarity is calculated from the session data 41 (access history) for the user's search word, and the like is not limited to using the data to the tag. Using the data extracted from these, the target determination model 25 is made to learn the selected items, with the explanatory variables as attributes and target types or preferences, and the objective variables as the distance between attributes and the number of visits (number of views). As the learning method, for example, a machine learning method such as Random Forest can be used.

<Conclusion>
As described above, the target user feature extraction server 1 of this embodiment extracts from the session data 41, the user attribute data 42, the page attribute data 43, and the poster attribute data 44 based on the target type desired by the poster. The items and the range of values of the data 50 are determined to generate the data 50 to be extracted. Then, by inputting the value range and the extraction target data 50 into the access feature extraction unit 27, the user who accessed the web server 200 (user terminal) obtains the user features that the poster who provides the content 210 to the web server 200 wants to acquire. It is possible to extract from the history of 100). In addition, the target user feature extraction server 1 can extract new users who are different from the poster's intention, and can also create a new business.

Further, since the target user feature extraction server 1 can extract the features of the session data 41 of the extracted user features from the page attribute data 43, what kind of content (tag) of the poster's content 210 shows the user's interest. Can be narrowed down and marketing can be supported.

In the above embodiment, the user's industry is used as the user attribute data 42, and the poster's industry is used as the poster attribute data 44, but the present invention is not limited to this. For example, the hobbies and tastes of the user and the hobbies and tastes of the poster can be used as attribute data, and the target user characteristics can be extracted from such attribute data.

Further, when determining the items of the extraction target data 50 and the range of values, the target determination model 25 should be used even if the range conversion information 46 corresponding to the target type reflecting the preference of the poster does not exist. Therefore, it is possible to extract the user characteristics of the target type that the poster wants to acquire from the session data 41 and the like.

As described above, the target user feature extraction server 1 of the above embodiment can have the following configuration.

(1) A target in which a computer (1) having a processor 11 and a memory 12 extracts a user feature to be acquired by a poster from history information (session data 41) that accesses the content (210) of the web server (200). It is a user feature extraction method

Session data (41) in which the computer stores the history information of the user terminal (100) that has accessed the content (210) of the web server (200), and the attribute information of the user who uses the user terminal (100). A user data acquisition step of acquiring the stored user attribute data (42) as user data (510), a page attribute data (43) in which the computer (1) stores the attributes of the content (210), and the like. The poster data acquisition step of acquiring the poster attribute data (44) storing the attributes of the poster who provided the content (210) as the poster data (520), and the computer (1) extract the data. The preference acquisition step of accepting the target contributor and acquiring the user information targeted by the contributor as the target type (461), and the computer (1) are the target type (461) of the contributor. The target calculation step (target calculation unit 23) for calculating the item of the data to be extracted from and the range of the value of the item, and the computer (1) from the user data (510) and the poster data (520). The session feature calculation step (session feature calculation unit 22) for calculating the extraction target data corresponding to the item, and the computer (1) range the value of the item from the extraction target data and the poster data (520). A target user feature extraction method comprising an access feature extraction step (access feature extraction unit 27) for calculating an access feature amount based on the above.

With the above configuration, as described above, the target user feature extraction server 1 of this embodiment changes from the session data 41, the user attribute data 42, the page attribute data 43, and the poster attribute data 44 to the target type desired by the poster. Based on this, the items and the range of values of the extraction target data 50 are determined to generate the extraction target data 50. Then, by inputting the value range and the extraction target data 50 into the access feature extraction unit 27, the user who accessed the web server 200 (user terminal) obtains the user features that the poster who provides the content 210 to the web server 200 wants to acquire. It is possible to extract from the history of 100). In addition, the target user feature extraction server 1 can extract new users who are different from the poster's intention, and can also create a new business.

Further, since the target user feature extraction server 1 can extract the features of the session data 41 of the extracted user features from the page attribute data 43, what kind of content (tag) of the poster's content 210 shows the user's interest. Can be narrowed down and marketing can be supported.

In the above embodiment, the user's industry is used as the user attribute data 42, and the poster's industry is used as the poster attribute data 44, but the present invention is not limited to this. For example, the hobbies and tastes of the user and the hobbies and tastes of the poster can be used as attribute data, and the target user characteristics can be extracted from such attribute data.

(2) In the target user extraction method according to (1) above, in the target calculation step (23), when the target type (461) is qualitative information, the qualitative information is data. A target user feature extraction method, which comprises a range conversion step (range conversion unit 24) for converting the item of

With the above configuration, it is possible to calculate the item of the data to be extracted and the range of the value of the item from the qualitative information, and it is possible to extract the user characteristics that match the taste of the target poster. It becomes.

(3) In the target user feature extraction method according to (2) above, in the range conversion step (23), the target type (461) and the user are added to a preset determination model (target determination model 25). A target user feature extraction method characterized in that data (510) and poster data (520) are input and an item of data to be extracted and a range of values of the item are output.

With the above configuration, the target type (461), user data (510), and poster data (520) are input to the preset target determination model 25, and the data items to be extracted from the qualitative information and the above items. It is possible to calculate the range of values of.

(4) In the target user feature extraction method according to (3) above, the computer (1) determines the user data (510), the poster data (520), and the target type (461). A target user feature extraction method characterized by further including a learning step (learning unit 31) given to a model (25) for learning.

With the above configuration, the learning unit 31 generates the target determination model 25 by machine learning the session data 41, the user attribute data 42, the page attribute data 43, the poster attribute data 44, and the target type acquired from the web server 200. be able to.

(5) In the target user feature extraction method according to (4) above, the learning step (31) uses the user attribute data (42) to calculate the similarity between user attributes. A target user feature extraction method comprising a calculation step (similarity calculation unit 29).

With the above configuration, the access feature extraction unit 27 can calculate the distance of the feature amount (similarity) of the industry between a plurality of users who have accessed the visit page 413 using the industry 423 of the user attribute data 42. , The content 210 can be presented as a user feature 51 as a group of users according to the distance for each poster.

In addition, the similarity calculation unit 29 calculates the similarity between industries from the session data 41, the user attribute data 42, and the poster attribute data 44, and the access feature extraction unit 27 refers to this similarity to the poster. Information about the distance to the selected industry and the industry can be extracted from the attributes and target type of.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment is described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the configurations described. Further, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Further, for a part of the configuration of each embodiment, any of addition, deletion, or replacement of other configurations can be applied alone or in combination.

Further, each of the above configurations, functions, processing units, processing means, etc. may be realized by hardware by designing a part or all of them by, for example, an integrated circuit. Further, each of the above configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function can be placed in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

Also, the control lines and information lines indicate what is considered necessary for explanation, and not all control lines and information lines are necessarily shown on the product. In practice, it can be considered that almost all configurations are interconnected.

Claims (15)

1. This is a target user feature extraction method in which a computer having a processor and a memory extracts the user features that the poster aims to acquire from the history information of accessing the contents of the web server.
   User data in which the computer acquires session data that stores history information of a user terminal that has accessed the contents of the web server and user attribute data that stores attribute information of a user who uses the user terminal as user data. Acquisition steps and
   A poster data acquisition step in which the computer acquires page attribute data storing the attributes of the content and poster attribute data storing the attributes of the poster who provided the content as poster data.
   The preference acquisition step in which the computer accepts the poster to be extracted and acquires the information of the user targeted by the poster as the target type.
   A target calculation step in which the computer calculates an item of data to be extracted from the target type of the poster and a range of values of the item.
   A session feature calculation step in which the computer calculates extraction target data corresponding to the item from the user data and the poster data, and
   An access feature extraction step in which the computer calculates an access feature amount based on the range of values of the item from the extraction target data and the poster data.
   A target user feature extraction method characterized by including.
2. The target user feature extraction method according to claim 1.
   The target calculation step is
   A target user feature extraction method comprising: when the target type is qualitative information, a range conversion step of converting the qualitative information into a data item and a range of values of the item.
3. The target user feature extraction method according to claim 2.
   The range conversion step
   A target user feature extraction method characterized in that the target type, the user data, and the poster data are input to a preset determination model, and the item of the data to be extracted and the range of the value of the item are output. ..
4. The target user feature extraction method according to claim 3.
   A target user feature extraction method, wherein the computer further includes a learning step in which the user data, the poster data, and the target type are given to the determination model for learning.
5. The target user feature extraction method according to claim 4.
   The learning step
   A target user feature extraction method comprising a similarity calculation step of calculating the similarity between user attributes using the user attribute data.
6. An extraction server with a processor and memory,
   A web server that provides content to user terminals and
   A target user feature extraction system having a posting terminal that provides the content to the web server.
   The web server collects history information that the user terminal has accessed the content, and the web server collects the history information.
   The posting terminal notifies the extraction server of the user information targeted by the poster of the content as the target type.
   The extraction server
   Session data storing the history information of the user terminal that has accessed the content of the web server and user attribute data storing the attribute information of the user who uses the user terminal are acquired as user data, and the attributes of the content are obtained. The stored page attribute data and the poster attribute data storing the attributes of the poster who provided the content are acquired as poster data, and the target type of the poster to be extracted is acquired. Department and
   A target calculation unit that calculates the item of data to be extracted from the target type of the poster and the range of values of the item, and
   A session feature calculation unit that calculates extraction target data corresponding to the item from the user data and the poster data, and
   An access feature extraction unit that calculates an access feature amount based on the range of values of the item from the extraction target data and the poster data, and an access feature extraction unit.
   A target user feature extraction system characterized by having.
7. The target user feature extraction system according to claim 6.
   The target calculation unit
   A target user feature extraction system comprising a range conversion unit that converts the qualitative information into a data item and a range of values of the item when the target type is qualitative information.
8. The target user feature extraction system according to claim 7.
   The range conversion unit
   A target user feature extraction system characterized in that the target type, the user data, and the poster data are input to a preset determination model, and the item of the data to be extracted and the range of the value of the item are output. ..
9. The target user feature extraction system according to claim 8.
   A target user feature extraction system, further comprising a learning unit that gives the user data, the poster data, and the target type to the determination model for learning.
10. The target user feature extraction system according to claim 9.
    The learning unit
    A target user feature extraction system including a similarity calculation step for calculating the similarity between user attributes using the user attribute data.
11. It is a target user feature extraction server that has a processor and memory and extracts the user features that the poster aims to acquire from the history information of accessing the contents of the web server.
    Session data storing the history information of the user terminal that has accessed the content of the web server and user attribute data storing the attribute information of the user who uses the user terminal are acquired as user data, and the attributes of the content are obtained. The stored page attribute data and the poster attribute data storing the attributes of the poster who provided the content are acquired as poster data, and the poster to be extracted and the poster of the content are the acquisition targets. The processing target selection unit that acquires the information of the user to be the target type, and
    A target calculation unit that calculates the item of data to be extracted from the target type of the poster and the range of values of the item, and
    A session feature calculation unit that calculates extraction target data corresponding to the item from the user data and the poster data, and
    An access feature extraction unit that calculates an access feature amount based on the range of values of the item from the extraction target data and the poster data, and an access feature extraction unit.
    A target user feature extraction server characterized by having.
12. The target user feature extraction server according to claim 11.
    The target calculation unit
    A target user feature extraction server comprising a range conversion unit that converts the qualitative information into a data item and a range of values of the item when the target type is qualitative information.
13. The target user feature extraction server according to claim 12.
    The range conversion unit
    A target user feature extraction server characterized in that the target type, the user data, and the poster data are input to a preset determination model, and the item of the data to be extracted and the range of the value of the item are output. ..
14. The target user feature extraction server according to claim 13.
    A target user feature extraction server further comprising a learning unit that gives the user data, the poster data, and the target type to the determination model for learning.
15. The target user feature extraction server according to claim 14.
    The learning unit
    A target user feature extraction server including a similarity calculation step for calculating the similarity between user attributes using the user attribute data.

Images