JP2022011154A - Evaluation apparatus, evaluation method, and evaluation program - Google Patents

Abstract

To extract useful information from characteristics.SOLUTION: An evaluation apparatus according to the present application includes a generation unit and an evaluation unit. The generation unit generates first characteristic information that assumes an object having first characteristics has second characteristics, and second characteristic information that assumes the object does not have the second characteristics. The evaluation unit compares first probability that a predetermined event related to an object will occur, the first probability being predicted based on the first characteristic information generated by the generation unit, when the object has the second characteristics, with second probability that the predetermined event related to an object will occur, the second probability being predicted based on the second characteristic information generated by the generation unit, when the object does not have the second characteristics, to evaluate the effect of the second characteristics with respect to the probability that the predetermined event related to the object will occur.SELECTED DRAWING: Figure 3

Description

There is an application for exception of loss of novelty

The present disclosure relates to evaluation devices, evaluation methods and evaluation programs.

Technology for analyzing information is developing in a wide range of fields such as science, industry, and business. As the use of information progresses, new analytical techniques have been developed and used for various purposes.

The data to be analyzed is generated from, for example, the Internet of things (IoT) of things and things, and the generated data is used for predictive analysis. Predictive analytics predicts the properties of an analysis target from data that represent the characteristics of the analysis target. An example of predictive analytics is prediction using a machine learning model. In recent years, various techniques have been proposed to improve the accuracy of predictions using models. For example, it has been proposed to use the output of a model as an input of another model (Patent Document 1).

Japanese Unexamined Patent Publication No. 2019-36173

However, in the above prior art, useful information may not be derived from the feature of interest. For example, if the predictive analytics target itself is not appropriate, useful information may not be available from the forecast results.

The present application has been made in view of the above and aims to derive useful information from the features of the subject.

In the evaluation device according to the embodiment of the present disclosure, the first feature information assuming that the object having the first feature has the second feature and the subject do not have the second feature. The target when the target has the second feature, which is predicted based on the creation unit for creating the second feature information assuming that the second feature information and the first feature information created by the creation unit. When the object does not have the second feature, which is predicted based on the first possibility that a predetermined event related to the above occurs and the second feature information created by the creator. From the comparison with the second possibility that the predetermined event related to the subject occurs, the influence of the second feature on the possibility that the predetermined event related to the subject occurs is evaluated. It has an evaluation unit.

According to one aspect of the embodiment, useful information can be derived from the characteristics of the subject.

FIG. 1 is an explanatory diagram showing an example of comparison between a typical advertisement distribution and a preferable advertisement distribution. FIG. 2A is an explanatory diagram showing an example of a predicted lift extraction process for extracting a predicted lift from a predicted probability according to an exemplary embodiment of the present disclosure. FIG. 2B is an explanatory diagram showing an example of a predicted lift extraction process for extracting a predicted lift from a predicted probability according to an exemplary embodiment of the present disclosure. FIG. 2C is an explanatory diagram showing an example of a predicted lift extraction process for extracting a predicted lift from a predicted probability according to an exemplary embodiment of the present disclosure. FIG. 3 is a diagram showing an example of an information distribution system according to an embodiment. FIG. 4 is a flowchart showing an example of a process for delivering an advertisement based on a lift value of a predicted conversion probability, which is executed by the information distribution device according to the embodiment. FIG. 5 is a diagram showing an example of a hardware configuration.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. The present invention is not limited to this embodiment. Details of one or more embodiments are described in the following description and drawings. Further, the plurality of embodiments can be appropriately combined as long as the processing contents do not contradict each other. Further, in the following one or a plurality of embodiments, the same parts are designated by the same reference numerals, and duplicate description will be omitted.

[1. Exemplary embodiments]
First, exemplary embodiments of the present disclosure will be described in detail with reference to FIGS. 1 and 2A-2C.

[1-1. Outline of an exemplary embodiment]
The information distribution device according to the exemplary embodiment predicts the conversion probability when the user touches the advertisement and the conversion probability when the user does not touch the advertisement.

Then, the information distribution device compares the two predicted conversion probabilities, and based on this comparison, predicts the lift value of the conversion probability as the increase range of the conversion probability depending on the presence or absence of advertisement contact. The information distribution device determines the target of advertisement distribution from among the users based on the prediction of the lift value of each user.

As a result, the information distribution device can deliver the advertisement to the user who shows a high conversion probability increase in the advertisement contact, although the information distribution device does not have a high conversion probability. In addition, the information distribution device ensures that the ad is delivered to users who convert independently of the ad contact, such as users who originally have a high conversion probability and do not show a high increase in the conversion probability in the ad contact. Can be prevented.

Therefore, the information distribution device can increase the ratio of users who convert to all users.

[1-2. Introduction to exemplary embodiments]
The target of ad delivery is determined based on the ad delivery strategy. An example of typical ad delivery is ad delivery based on the predicted conversion rate that the ad is delivered to users who are likely to convert.

The term "conversion (CV)" may refer to a user action such as "to raise a conversion event" or "to perform a conversion action". For example, executing a conversion action may be "viewing an ad and then purchasing an ad product for a given content" or "clicking on an ad to purchase an ad product for linked content". That is included.

FIG. 1 is an explanatory diagram showing an example of comparison between a typical advertisement distribution and a preferable advertisement distribution. In the example of FIG. 1, the target candidates for advertisement distribution include users such as user U1, user U2, user U3, user U4, and user U5. User U1 and user U2 are users with high predicted conversion rates. On the other hand, users U4 and U5 are users who do not have a high predicted conversion rate.

A typical advertisement distribution TAD1 is an advertisement distribution based on a predicted conversion rate. As shown in FIG. 1, in a typical advertisement delivery TAD1, the advertisement is delivered to the user U1 and the user U2 having a high predicted conversion rate. As shown in FIG. 1, in a typical advertisement distribution TAD1, an advertisement hits a user who is likely to convert.

However, in general, potential target for ad delivery includes users who convert without contacting the ad. Users who convert without touching the ad are also included among those who are likely to convert. In the example of FIG. 1, it is assumed that the user U1 is a user who converts without touching the advertisement. In a typical advertisement distribution TAD1, a user U1 who converts without contacting an advertisement is included in the target of advertisement distribution.

Also, in general, potential target for ad delivery includes users who have a high probability of converting when the ad hits the user. Such users are included among the users who are not originally likely to convert. In the example of FIG. 1, it is assumed that the user U4 and the user U5 are users who will convert with a high probability when the advertisement hits the user. In a typical advertisement distribution TAD1, the user U4 and the user U5, which are likely to convert when the advertisement hits the user, are not included in the advertisement distribution target. However, when the advertisement hits the user U4 and the user U5, the conversion of the user U4 and the user U5 becomes very easy to be acquired.

As described above, the first problem of the typical advertisement distribution TAD1 is that the advertisement is distributed to the user U1 who converts without contacting the advertisement. Further, the second problem of the typical advertisement distribution TAD1 is that the advertisement is not delivered to the user U4 or the user U5 who will convert with a high probability when the advertisement hits the user. In other words, in a typical advertisement distribution TAD1, a user who converts without contacting an advertisement is targeted for advertisement distribution, and a user who converts due to contact with an advertisement is overlooked.

As an example of ideal advertisement distribution, the preferred advertisement distribution PAD1 distributes an advertisement to a user who converts by contact with the advertisement. As shown in FIG. 1, in the preferred ad delivery PAD1, the ad is not delivered to the user U1 who converts without contacting the ad. On the other hand, the advertisement is delivered to the user U4 and the user U5 who will convert with a high probability when the advertisement hits the user. As described above, the preferred advertisement distribution PAD1 is aimed at users who convert without contacting the advertisement and who show a high increase in the conversion rate due to the contact of the advertisement.

Therefore, the information distribution device according to the exemplary embodiment executes the predictive lift extraction process described below so that the advertisement is distributed to the user as in the case of the preferred advertisement distribution PAD1.

[1-3. Predictive lift extraction process]
Hereinafter, the predicted lift extraction process according to the exemplary embodiment will be described with reference to FIGS. 2A to 2C.

2A-2C are explanatory views showing an example of a predicted lift extraction process for extracting a predicted lift from a predicted probability according to an exemplary embodiment of the present disclosure. In an exemplary embodiment, the predictive lift extraction process is performed by the information distribution device 100, which will be described later with reference to FIG. The information distribution device 100 is an example of an evaluation device.

Referring to FIG. 2A, first, the information distribution device 100 acquires the training set TS1 (step S1). The training set TS1 includes an instance and a label associated with the instance. In the example of FIG. 2A, the instance corresponds to the explanatory variable (ie, input) of the model, and the label corresponds to the objective variable (ie, output or target) of the model.

As shown in the example of FIG. 2A, an instance shows a combination of a user's attribute and a value of this attribute as a user's feature (also referred to as a "feature amount"). Here, the objective variable is the presence or absence of conversion, and the explanatory variables are the attributes (for example, interests such as gender, age, domestic travel, overseas travel, etc.) and the presence or absence of advertising contact.

For illustration purposes, it is assumed that a particular airline has asked an internet company associated with the information distribution device 100 to carry out an advertising campaign to deliver advertising content related to the particular airline. In this example, the ad creative includes an ad text such as "If you go to New York, that plane (exemplary ad text)". In this case, the conversion action or conversion event is the purchase of a ticket to New York.

For example, training set TS1 shows that a 27-year-old man is interested in domestic travel and not overseas travel. Training set TS1 also shows that the 27-year-old man came into contact with the ad and did not convert thereafter.

Also, for example, the training set TS1 shows that a 35-year-old female is interested in domestic travel and is interested in overseas travel. In addition, Training Set TS1 shows that the 35-year-old woman was not in contact with the ad before the 35-year-old woman converted.

Next, the information distribution device 100 builds a conversion (CV) probability prediction model CPPM1 by training a machine learning model from the acquired training set TS1 (step S2). The machine learning model is, for example, a regression model such as logistic regression.

Referring to FIG. 2B, after step S2 of FIG. 2A, the information distribution device 100 acquires the data set DS1 (step S3). The data set DS1 contains attribute information of a new user (for example, interests such as gender, age, overseas travel, etc.) as a new instance. For example, dataset DS1 shows that a 26-year-old man is not interested in traveling abroad and a 22-year-old woman is not interested in traveling abroad.

Next, the information distribution device 100 creates a data set assuming the presence or absence of advertisement contact from the acquired data set DS1 (step S4).

More specifically, the information distribution device 100 adds presence / absence information indicating the presence / absence of advertisement contact to the attribute information included in the data set DS1. In the example of FIG. 2B, the information distribution device 100 creates the data set PDS1 as a data set assuming that there is an advertisement contact. Further, the information distribution device 100 creates a data set PDS2 as a data set assuming that there is no advertisement contact.

Referring to FIG. 2C, after step S4 of FIG. 2B, the information distribution device 100 predicts conversion (CV) probabilities before and after ad contact for each new user (step S5).

In the example of FIG. 2C, the information distribution device 100 has a data set PDS1 (that is, a data set assuming that there is no advertisement contact) and a data set PDS2 (that is, a data set that assumes that there is advertisement contact). By applying the conversion probability prediction model CPPM1, the conversion probability CPBA1 before the advertisement contact and the conversion probability CPAA1 after the advertisement contact are predicted.

In this way, the information distribution device 100 calculates two predicted conversion probabilities of the same user, which are different from each other depending on the presence or absence of advertisement contact. In the example of FIG. 2C, the conversion probability CPBA1 before contacting the advertisement shows the predicted value of the conversion probability when the new user does not contact the advertisement. On the other hand, the conversion probability CPAA1 after the advertisement contact indicates the predicted value of the conversion probability when the new user contacts the advertisement. The information distribution device 100 outputs a predicted value of the conversion probability before and after the advertisement contact for each new user (for example, a 26-year-old man and a 42-year-old female).

After that, the information distribution device 100 calculates the lift value of the conversion (CV) probability due to the advertisement contact (step S6). In the example of FIG. 2C, the information distribution device 100 calculates the difference between the conversion probability CPAA1 after the advertisement contact and the conversion probability CPBA1 before the advertisement contact as the lift value LVCP1 of the conversion probability due to the advertisement contact.

In the example of FIG. 2C, for example, the lift value of a 26-year-old man is a lift value of "0.15-0.10 = 0.05". Further, for example, the lift value of a 42-year-old female is a lift value of "0.45-0.20 = 0.25". As described above, the information distribution device 100 extracts the difference between the predicted conversion probability after the advertisement contact and the predicted conversion probability before the advertisement contact as the lift value of the predicted conversion probability due to the advertisement contact. Such a lift value indicates how much the user's conversion probability increases when the advertisement hits the user.

Then, the information distribution device 100 selects a user who is the target of advertisement distribution based on the lift value LVCP1 of the conversion probability due to the advertisement contact. For example, the information distribution device 100 sorts new users based on the lift value LVCP1 of the conversion probability due to advertisement contact, and determines the top n people (n is an arbitrary natural number) as the target of advertisement distribution.

[1-4. Effect of exemplary embodiment]
As described above, the information distribution device 100 according to the exemplary embodiment can search for a user who will convert with a high probability when the advertisement hits the user, as a desirable user as a target for advertisement distribution. .. Such ads to users are likely to lead to user conversions. The information distribution device 100 can increase the conversions of the acquired users without overlooking the users who show a high increase in the conversion rate due to the contact with the advertisement.

From the viewpoint of comparison with the typical advertisement distribution TAD1 described above with reference to FIG. 1, the information distribution device 100 can realize overall optimization of advertisement distribution while there are restrictions on the amount of advertisement distribution. ..

More specifically, the information distribution device 100 can determine to which user the advertisement is distributed based on the lift value of the predicted conversion probability due to the advertisement contact. For example, the information distribution device 100 distributes an advertisement to a user who is likely to convert when the user comes into contact with the advertisement (that is, a user who exhibits a higher lift value due to contact with the advertisement). In this way, the information distribution device 100 can reduce the opportunity loss that may occur when the advertisement is not delivered to the user who converts due to the contact with the advertisement.

On the other hand, even when the user has a high conversion rate, the information distribution device 100 does not deliver the advertisement to the user who does not show a high lift value due to contact with the advertisement. Therefore, the conversion rate of a user having a high conversion rate may be lower than that of a typical advertisement distribution TAD1. However, the sum or average of the lift values of all users is increased as compared with the typical advertisement distribution TAD1. That is, the ratio of users who convert to all users increases as compared with the typical advertisement distribution TAD1. In addition, some users who do not show a high lift value due to contact with the ad will convert without contacting the ad. As a result, the information distribution device 100 can increase the conversion acquisition efficiency of the user.

Therefore, the information distribution device 100 can increase the conversion of the user acquired by the advertisement while maintaining the conversion of the user who converts without touching the advertisement.

Hereinafter, the information distribution device 100 that performs such a predictive lift extraction process will be described in detail.

[2. Information distribution system]
Next, a configuration example of a system including the information distribution device 100 will be described with reference to FIG.

[2-1. Information distribution system components]
FIG. 3 is a diagram showing an example of the information distribution system 1 according to the embodiment. As shown in FIG. 3, the information distribution system 1 includes components such as an information distribution device 100, an information providing device 200, and a user device 300. Although not shown in FIG. 3, the information distribution system 1 may include a plurality of information distribution devices 100, a plurality of information providing devices 200, and a plurality of user devices 300. Further, the information distribution system 1 may include other components such as devices of an entity (for example, a trader, an end user) related to the information distribution device 100.

In the information distribution system 1, the information distribution device 100, the information providing device 200, and the user device 300 are each connected to the network N by wire or wirelessly. The network N is, for example, a network such as the Internet, a WAN (Wide Area Network), or a LAN (Local Area Network). The components of the information distribution system 1 can communicate with each other via the network N.

The information distribution device 100 (corresponding to an example of the evaluation device) is an information processing device that executes processing for information distribution. An example of information distribution is content distribution. For example, the information distribution device 100 can distribute an advertisement as content.

The information distribution device 100 can extract various information about the target from the characteristics of the target by using the data analysis method. The information distribution device 100 can select a wide variety of targets as targets for data analysis. For example, in Internet advertising, the target is an Internet user. In this case, user information (eg, user attributes) is used to analyze the effectiveness of the advertisement on the user.

The information distribution device 100 can give user information to the machine learning algorithm and analyze the effect of the advertisement on the user. For example, the information distribution device 100 can build a machine learning model that predicts the effect of an advertisement from user information. An example of such a machine learning model is a predictive model in which the user's characteristics are inputs and the user's properties (for example, the effect of the advertisement on the user) are outputs.

The information distribution device 100 may be any type of information processing device including a server. A plurality of information distribution devices 100 may each provide functions of various servers such as a web server, an application server, and a database server. A configuration example of the information distribution device 100 will be described in detail in the next section.

The information providing device 200 is an information processing device that provides various information to the information distribution device 100. The information providing device 200 can provide various Internet services. In addition, the information providing device 200 can collect various information via various Internet services. The information providing device 200 may be any type of information processing device including a server.

Various internet services include search, portal, shopping, auction, news, finance, travel, restaurant introduction, sports, TV, video, games, maps, route information, food, real estate, automobiles, blogs, marriage, cartoons, fortune-telling, It may be a service such as local information.

The information providing device 200 can provide various Internet services via a client program. The client program is, for example, a program such as a smartphone application or a web browser. The information providing device 200 can collect information on users of various Internet services through the provision of the various Internet services described above.

The information providing device 200 can collect user information such as a search history, a purchase history, a browsing history, and a movement history. The information providing device 200 can store such user information as event information and feature information.

Event information represents a specific event. For example, the event information can indicate that the user has generated a conversion event or executed a conversion action. As an example, the event information may indicate that the user has browsed the advertisement and then purchased the product of the advertisement in the predetermined content. As another example, the event information may indicate that the user clicks on the advertisement and purchases the product of the advertisement in the linked content.

The feature information represents the feature of the object. For example, the feature information can indicate a user's action history (for example, search history) and user's attribute information (for example, demographic attribute, psychographic attribute).

The user device 300 is an information processing device used by the user. The user device 300 can request access to various Internet services provided by the information providing device 200. The user device 300 may be any type of information processing device including a client device such as a smartphone, a desktop PC (Personal Computer), a notebook PC, and a tablet PC.

The user device 300 can request various information from the information providing device 200 via the above-mentioned various Internet services. Further, the user device 300 can receive various information from the information providing device 200 via the above-mentioned various Internet services. As described above, the user information of the user device 300 can be collected by the information providing device 200 via the various Internet services.

[2-2. Configuration of information distribution device]
As shown in FIG. 3, the information distribution device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The information distribution device 100 includes an input unit (for example, a keyboard, a mouse, etc.) that receives various operations from an administrator or the like who uses the information distribution device 100, and a display unit (liquid crystal display, etc.) for displaying various information. You may have.

(Communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 110 is connected to the network by wire or wirelessly. The communication unit 110 may be communicably connected to the information providing device 200 and the user device 300 via the network N. The communication unit 110 can transmit and receive information between the information providing device 200 and the user device 300 via a network network.

(Memory unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory (Flash Memory), or a storage device such as a hard disk or an optical disk. As shown in FIG. 3, the storage unit 120 includes an event information storage unit 121, a model storage unit 122, a feature information storage unit 123, and a distribution information storage unit 124.

(Event information storage unit 121)
The event information storage unit 121 stores event information. As mentioned above, event information represents a particular event. The event information can show the relationship between the occurrence of a predetermined event related to the object and the characteristics of the object. For example, event information can include training instances. In this case, the instance is a feature of interest and the label associated with the instance can be a given event.

In the exemplary embodiment described above, the event information is the training set TS1 described above with reference to FIG. 2A. In the example of FIG. 2A, the target features are the user's attributes (eg, gender, age, interests such as domestic travel, overseas travel, etc.) and the presence or absence of advertising contact, and the label is the presence or absence of conversion. Training set TS1 shows information about past user conversions.

In the example of FIG. 2A, the event information (ie, training set TS1) suggests that a 27-year-old man converted because the advertisement was delivered. The event information also indicates that a 35-year-old female converted, even though the ad was not delivered. In this case, this event information suggests that the 35-year-old woman originally planned to purchase the advertising product.

(Model storage unit 122)
The model storage unit 122 stores a model (that is, a machine learning model). The model storage unit 122 can store a model (for example, data of a machine learning model) generated by the generation unit 133, which will be described later. In some embodiments, the model received by the collector 131 described below may be stored.

In one example, the objective variable of the model is the presence or absence of conversion, and the explanatory variables of the model are user information including attribute data (eg, gender, age, interests, etc.) and the presence or absence of advertising contact.

(Characteristic information storage unit 123)
The feature information storage unit 123 stores feature information. As described above, the feature information represents the feature of the object. For example, feature information can include new instances. In this case, the new instance can be a feature of the new target.

In the exemplary embodiment described above, the feature information is the dataset DS1 described above with reference to FIG. 2B. In the example of FIG. 2B, the characteristic of the object (for example, a new object) is the attribute (for example, gender, age, interest of overseas travel, etc.) of the user (for example, a new user).

(Distribution information storage unit 124)
The distribution information storage unit 124 stores distribution information. The distribution information is various information distributed by the distribution unit 139, which will be described later. An example of various types of information is content (for example, an advertisement).

In the exemplary embodiment described above, the distribution information is an advertisement. In the example of FIGS. 2A-2C, the advertisement is advertising content related to a particular airline. In the example of FIGS. 2A to 2C, since this advertisement content (for example, an advertisement including an advertisement such as "If you go to New York, that airplane (example advertisement)") is delivered, the user who purchased the ticket There is. In other words, some users did not buy a ticket if this advertising content was not delivered. In this way, user conversions are influenced by ad creatives.

(Control unit 130)
The control unit 130 is a controller, and for example, various programs (evaluation programs) stored in a storage device inside the information distribution device 100 by a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). (Corresponding to one example) is realized by executing RAM or the like as a work area. Further, the control unit 130 is a controller, and may be realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a GPU GPU (General Purpose Graphic Processing Unit). ..

As shown in FIG. 3, the control unit 130 includes a collection unit 131, an event information acquisition unit 132, a generation unit 133, a reception unit 134, a feature information acquisition unit 135, a creation unit 136, and a prediction unit 137. It has an evaluation unit 138 and a distribution unit 139, and realizes or executes the functions and operations of information processing described below. Further, the control unit 130 can realize the above-mentioned predictive lift extraction process with reference to FIGS. 2A to 2C. One or more processors of the information distribution device 100 may realize the function of each control unit in the control unit 130 by executing an instruction stored in one or more memories of the information distribution device 100. can. The internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 3, and may be any other configuration as long as it is configured to perform information processing described later. For example, the evaluation unit 138 may perform all or part of the information processing described later with respect to the units other than the evaluation unit 138.

(Collecting unit 131)
The collection unit 131 can collect various types of information used to execute processing for information distribution. For example, the collecting unit 131 can collect the above-mentioned event information. The collecting unit 131 can receive event information from the information providing device 200. The collecting unit 131 may collect the above-mentioned model (that is, a machine learning model).

The collecting unit 131 can receive various information such as event information and a model from a predetermined information processing device (for example, a device of an entity (for example, a specific Internet company) related to the information distribution device 100). Further, for example, the collecting unit 131 can receive such information from the administrator who uses the information distribution device 100 via the user interface. The collecting unit 131 may store the received event information in the event information storage unit 121. Further, the collecting unit 131 may store the received model in the model storage unit 122.

(Event information acquisition unit 132)
The event information acquisition unit 132 can acquire various types of information used to execute a process for information distribution. The event information acquisition unit 132 can acquire various information from the storage unit 120. For example, the event information acquisition unit 132 can acquire the above-mentioned event information from the event information storage unit 121.

In one example, the event information acquisition unit 132 acquires a training set. As mentioned above with reference to FIG. 2A, the training set includes the instance and the label associated with the instance.

(Generation unit 133)
The generation unit 133 can generate a machine learning model used to execute a process for information distribution. The generation unit 133 can construct a machine learning model from the event information acquired by the event information acquisition unit 132.

The generation unit 133 indicates an event showing the relationship between the occurrence of a predetermined event related to the second target and the characteristics of the second target with respect to the prediction of the possibility that a predetermined event related to the first target occurs. Machine learning models learned from information can be generated. For example, the generation unit 133 gives event information to the machine learning model. Then, the generation unit 133 can train the machine learning model based on the event information. The generator 133 can search for model parameter values so that the generalization error is minimized. The generation unit 133 can implement various optimizers in order to search for model parameter values.

In one example, the generation unit 133 builds a conversion probability prediction model by training a machine learning model from the training set acquired by the event information acquisition unit 132. As described above with reference to FIG. 2A, the machine learning model is, for example, a regression model such as logistic regression.

(Receiver 134)
The receiving unit 134 can receive various types of information used to execute processing for information distribution. For example, the receiving unit 134 can receive the above-mentioned feature information. The receiving unit 134 can receive feature information from the information providing device 200.

The receiving unit 134 can receive various information such as feature information from a predetermined information processing device (for example, a device of an entity related to the information distribution device 100 (for example, a device of a specific Internet company)). Further, for example, the receiving unit 134 can receive such information from the administrator who uses the information distribution device 100 via the user interface. The receiving unit 134 may store the received feature information in the feature information storage unit 123.

(Characteristic information acquisition unit 135)
The feature information acquisition unit 135 can acquire various types of information used to execute a process for information distribution. The feature information acquisition unit 135 can acquire various information from the storage unit 120. For example, the feature information acquisition unit 135 can acquire the above-mentioned feature information from the feature information storage unit 123.

The feature information acquisition unit 135 can acquire feature information representing the feature of the target. The feature information acquisition unit 135 can acquire feature information indicating that the target has the first feature. For example, the feature information acquisition unit 135 can acquire feature information representing the value of the first attribute, assuming that it has the first feature.

The feature information acquisition unit 135 can acquire feature information indicating that it has the first feature, which is the first part of the plurality of features used for predicting the possibility that an event related to the target will occur. ..

In one example, the feature information acquisition unit 135 acquires a data set. As mentioned above with reference to FIG. 2B, the dataset contains, as a new instance, attribute information of the new user (eg, interests such as gender, age, overseas travel, etc.).

(Creation unit 136)
The creation unit 136 can create feature information. For example, the creation unit 136 can create new feature information from the feature information acquired by the feature information acquisition unit 135. The creation unit 136 can create a pseudo data set assuming the existence of a new feature from the data set for predicting a predetermined event.

In at least one embodiment, the preparation unit 136 has a first feature information assuming that the object having the first feature has the second feature, and the subject has this second feature. Create a second feature information assuming that it does not exist. For example, the creation unit 136 creates the first feature information and the second feature information from the feature information acquired by the feature information acquisition unit 135.

From the feature information acquired by the feature information acquisition unit 135, the creation unit 136 has the first feature information assuming that the target has the second feature in addition to the first feature, and the target is the first feature. It is possible to create a second feature information that has one feature but does not have this second feature. In other words, from the feature information acquired by the feature information acquisition unit 135, the creation unit 136 obtains the first feature information on the assumption that the target has the first feature and the second feature. It is possible to create the second feature information on the assumption that the object has the first feature but does not have the second feature. That is, it is assumed that the first feature information includes the second feature in addition to the first feature. On the other hand, in the second feature information, it is assumed that the first feature is the target but the second feature is not the target.

The creation unit 136 indicates that it has a first feature acquired by the feature information acquisition unit 135, which is the first part of a plurality of features used for predicting the possibility of an event related to an object occurring. From the feature information, it is assumed that the subject has, in addition to the first feature, a second feature, which is the second part of the plurality of features used to predict the likelihood that an event related to the subject will occur. It is possible to create the feature information of the first hypothesis made and the feature information of the second hypothesis assuming that the object has the first feature but does not have the second feature.

In at least one embodiment, the creation unit 136 has the first feature information representing the first value of the second attribute and the second feature as having the second feature. A second feature information representing the second value of this second attribute is created.

In at least one embodiment, the creation unit 136 has a first feature information indicating that the predetermined information has been delivered to the subject, and the subject does not have the second feature, as the subject has the second feature. As a matter of fact, a second feature information indicating that this predetermined information has not been delivered to the target is created.

In one example, the creation unit 136 creates a data set assuming the presence or absence of advertisement contact from the data set acquired by the feature information acquisition unit 135.

More specifically, as described above with reference to FIG. 2B, the creation unit 136 adds presence / absence information indicating the presence / absence of advertisement contact to the attribute information included in the data set. The creation unit 136 creates a first data set as a data set assuming that there is an advertisement contact. Further, the creation unit 136 creates a second data set as a data set assuming that there is no advertisement contact.

(Prediction unit 137)
The prediction unit 137 can predict the possibility that a predetermined event will occur. For example, the prediction unit 137 can predict the possibility that a predetermined event will occur by applying a machine learning algorithm to the feature information created by the creation unit 136.

In at least one embodiment, the prediction unit 137 learns so as to predict the possibility that a predetermined event will occur in the learning target from the feature information of the learning target having the first feature or the second feature. Using the predicted model, the first possibility that a predetermined event related to the subject occurs when the subject has the second feature, and the subject is related when the subject does not have the second feature. Predict a second possibility that a given event will occur. For example, the prediction unit 137 uses a prediction model trained using feature information showing the relationship between the occurrence of a predetermined event related to the learning target and the feature of the learning target, and the first possibility and the second. Predict the possibility of.

The prediction unit 137 has a second target for predicting the possibility that a predetermined event related to the first target will occur with respect to the first feature information and the second feature information created by the preparation unit 136. By applying a learning algorithm learned from the event information showing the relationship between the occurrence of a predetermined event related to the second feature and the feature of the second target, the predetermined event related to the subject when the subject has the second feature. It is possible to predict the first possibility that an event will occur and the second possibility that a predetermined event related to the subject will occur if the subject does not have the second feature.

Examples of learning algorithms include model-based learning such as logistic regression, support vector machines, neural networks, and instance-based learning such as k-nearest neighbors. The prediction unit 137 applies a model trained using the above-mentioned event information as a learning algorithm, so that a predetermined event related to the target occurs when the target has the second feature. It is possible to predict the possibility and the second possibility that a predetermined event related to the subject will occur if the subject does not have the second feature.

In one example, predictor 137 predicts conversion probabilities before and after ad contact for each new user. As described above with reference to FIG. 2C, the prediction unit 137 has a first data set (that is, a data set assuming that there is no advertisement contact) and a second data set (that is, that there is advertisement contact). By applying the conversion probability prediction model to the data set (data set assuming), the conversion probability before the advertisement contact and the conversion probability after the advertisement contact are predicted.

(Evaluation unit 138)
The evaluation unit 138 can evaluate the influence of the characteristics of the subject on the possibility that a predetermined event related to the subject occurs. For example, the evaluation unit 138 can evaluate the change in the possibility that a predetermined event related to the target occurs based on the possibility predicted by the prediction unit 137. For example, the evaluation unit 138 can evaluate the influence of the characteristics of the target on the possibility that the target causes a predetermined event based on the possibility predicted by the prediction unit 137.

In at least one embodiment, the evaluation unit 138 causes a predetermined event related to the subject when the subject has the second feature, which is predicted based on the first feature information created by the creation unit 136. This predetermined event related to the subject occurs when the subject does not have the second feature, predicted based on the first possibility of doing so and the second feature information created by the creator 136. From the comparison with the second possibility, the influence of the second feature on the possibility that this predetermined event related to the subject occurs is evaluated. For example, the evaluation unit 138 may generate this predetermined event related to the subject from the comparison between the first possibility predicted by the prediction unit 137 and the second possibility predicted by the prediction unit 137. The effect of the second feature on sex is evaluated.

In at least one embodiment, the evaluation unit 138 considers the first possibility and the second possibility as the influence of the second feature on the possibility that this predetermined event related to the subject occurs. Evaluate a given index for comparison. For example, the evaluation unit 138, as a predetermined index, determines which of the first possibility is the possibility that a predetermined event related to the subject occurs when the subject is not assumed to have the second characteristic. Calculate an evaluation index that indicates whether the degree has increased. For example, the evaluation unit 138, as an evaluation index, is based on at least one of the difference between the first possibility and the second possibility and the ratio of the first possibility and the second possibility. The lift value according to the feature of 2 is calculated.

In at least one embodiment, the evaluation unit 138 gives a second feature to the possibility that this predetermined event related to the subject occurs from the comparison between the first possibility and the second possibility. The effect comes from the difference between the first value of the second attribute represented by the first feature information and the second value of the second attribute represented by the second feature information. Evaluate changes in the likelihood that certain events related to the subject will occur.

In at least one embodiment, the evaluation unit 138 has a first possibility that the subject takes a predetermined action that is a predetermined event when the subject has a second feature, and the subject has a second feature. From the comparison with the second possibility that the subject takes this predetermined action, which is a predetermined event in the absence of the subject, a predetermined event related to the subject, which is the possibility that the subject takes this predetermined action, may occur. Evaluate the effect of the second feature on sex.

In at least one embodiment, the evaluation unit 138 compares the first possibility with the second possibility, and the influence of the second feature on the possibility that a predetermined event related to the subject occurs. As a result, the influence of the distribution of predetermined information on the possibility that a predetermined event related to the target occurs is evaluated.

In one example, the evaluation unit 138 calculates the lift value of the conversion probability due to the advertisement contact. As described above with reference to FIG. 2C, the evaluation unit 138 calculates the difference between the conversion probability after the advertisement contact and the conversion probability before the advertisement contact as the lift value of the conversion probability due to the advertisement contact.

As described above, the evaluation unit 138 does not simply evaluate the influence of the predetermined attribute on the prediction of the occurrence of the event by fluctuating the value of the predetermined attribute included in the feature information in advance. , Evaluate the effect of the new attribute on the prediction of the occurrence of the event from the new feature information assuming the existence of the new attribute that is not included in the feature information in advance and the existence of the value of this new attribute. can do.

(Distribution Department 139)
The distribution unit 139 can distribute predetermined information to a target. The distribution unit 139 can acquire distribution information as predetermined information from the distribution information storage unit 124. As described above, the distribution information is, for example, advertising content. The distribution unit 139 may determine the advertisement content to be distributed based on various data related to the advertisement content (for example, bid price, advertisement quality).

The distribution unit 139 can distribute the advertisement content acquired from the distribution information storage unit 124 to the user device 300. For example, the distribution unit 139 can distribute the advertisement content to the content slots in the various contents provided by the information providing device 200. The various contents are, for example, contents such as web pages and in-app contents related to the above-mentioned various Internet services.

In at least one embodiment, the distribution unit 139 distributes the predetermined information to the target when the change in the possibility that the predetermined event related to the target occurs, which is evaluated by the evaluation unit 138, satisfies the predetermined condition. do.

In one example, the distribution unit 139 selects a user to be targeted for advertisement distribution based on the lift value of the conversion probability due to the advertisement contact. As described above with reference to FIG. 2C, the distribution unit 139 sorts new users based on the lift value of the conversion probability due to advertisement contact, and targets the top n people (n is an arbitrary natural number) for advertisement distribution. To be determined as.

[3. Predictive lift extraction process flow]
Next, with reference to FIG. 4, the procedure of the predicted lift extraction process by the information distribution device 100 according to the embodiment will be described.

FIG. 4 is a flowchart showing an example of a process for delivering an advertisement based on a lift value of a predicted conversion probability, which is executed by the information distribution device 100 according to the embodiment.

As shown in FIG. 4, first, the event information acquisition unit 132 of the information distribution device 100 acquires the training set (step S101). For example, the event information acquisition unit 132 acquires a training set from the event information storage unit 121.

The training set, as a training instance, contains feature information that represents the characteristics of the user. The feature information is user information such as presence / absence of advertisement contact, action history (for example, search history), and attribute information (for example, demographic attribute, psychographic attribute). The label is associated with the training instance. The label indicates the presence or absence of conversion. The predictor is characteristic information such as user information, and the numerical value of the target is the conversion rate.

Next, the generation unit 133 of the information distribution device 100 constructs a conversion (CV) probability prediction model from the training set acquired by the event information acquisition unit 132 (step S102). The generation unit 133 gives a training set to the model and trains the model based on the training set. The generator 133 can train the model so that the generalization error is minimized.

Learning algorithms used as models are, for example, logistic regression and various machine learning systems (eg, XGBoost). The conversion probability prediction model can make it possible to understand the correlation between conversion and user characteristics and the causal relationship between conversion and user characteristics.

Next, the feature information acquisition unit 135 of the information distribution device 100 acquires a data set (step S103). The dataset is the data of the new instance. The dataset, as a new instance, contains feature information that represents the characteristics of the new user.

The new user is a candidate for an advertisement target (that is, an advertisement distribution target). Therefore, the feature information of the new user does not include information regarding the presence or absence of advertising contact at this point. The new user is, for example, a unique user. A unique user is a user who has used a predetermined service for a predetermined period. In this case, the unique user is a candidate target for the advertisement associated with the given service.

Next, the creation unit 136 of the information distribution device 100 creates two data sets having different presence / absence of advertisement contact from the data set acquired by the feature information acquisition unit 135 (step S104). More specifically, the creation unit 136 adds a feature assuming that "the user has an advertisement contact" to a series of features included in the data set acquired by the feature information acquisition unit 135. Further, the creation unit 136 adds a feature assuming that "the user has no advertising contact" to the series of features included in the data set acquired by the feature information acquisition unit 135.

In other words, the creation unit 136 adds the ad contact attribute and the value "yes" of this attribute to the data set acquired by the feature information acquisition unit 135, thereby adding the ad contact "yes" data set. create. Further, the creation unit 136 creates a data set of advertisement contact "none" by adding the attribute of the advertisement contact and the value "none" of this attribute to the data set acquired by the feature information acquisition unit 135. do.

Next, the prediction unit 137 of the information distribution device 100 applies the conversion (CV) probability prediction model generated by the generation unit 133 to the two data sets created by the creation unit 136 with different presence or absence of advertisement contact. By doing so, the conversion (CV) probability with ad contact and the conversion (CV) probability without ad contact are predicted (step S105).

That is, the prediction unit 137 predicts the conversion probability when the new user touches the advertisement and the conversion probability when the new user does not touch the advertisement by using the conversion probability prediction model. The prediction unit 137 outputs the conversion probability with advertisement contact and the conversion probability without advertisement contact for each new user.

Next, the evaluation unit 138 of the information distribution device 100 determines the conversion (CV) probability based on the conversion (CV) probability with advertisement contact and the conversion (CV) probability without advertisement contact predicted by the prediction unit 137. The lift value of is calculated (step S106). The evaluation unit 138 outputs the lift value of the conversion probability for each new user. In this way, the evaluation unit 138 can score the candidate advertisement target based on the increase in the conversion probability caused by the advertisement contact.

The lift value of the conversion probability is defined by a predetermined formula used for the conversion probability with ad contact and the conversion probability without ad contact. The form of this predetermined expression may correspond to the form of the lift value expression used in association rule learning. Alternatively, the form of this predetermined formula corresponds to the difference between the conversion probability with ad contact and the conversion probability without ad contact, or the ratio between the conversion probability with ad contact and the conversion probability without ad contact. May be good.

As an example, the evaluation unit 138 extracts the difference between the conversion probability with ad contact and the conversion probability without ad contact, and uses the extracted difference as the increase value of the conversion probability when the advertisement hits a new user. Determined as a certain score. Such a score is an example of a conversion probability lift value.

Next, the distribution unit 139 of the information distribution device 100 determines the advertisement target based on the lift value of the conversion (CV) probability calculated by the evaluation unit 138 (step S107). For example, the distribution unit 139 sorts a plurality of new users based on the lift value of the conversion probability, and determines the top n new users (n is an arbitrary natural number) as the advertisement target. Alternatively, if the lift value of the conversion probability of the new user satisfies the threshold value, the distribution unit 139 may determine this new user as an advertisement target. In some implementations, if the new user's "conversion probability with ad contact" does not meet the threshold, distribution unit 139 does not have to determine this new user as an ad target.

[4. Other embodiments]
The information distribution device 100 according to the above-described embodiment may be implemented in various different forms other than the above-mentioned embodiment. Therefore, in the following, another embodiment of the above-mentioned information distribution device 100 will be described.

[4-1. Information for multiple types of distribution]
When there are a plurality of types of distribution information, the information distribution device 100 may consider a combination of the plurality of types of distribution information. For example, in the above exemplary embodiment, the information distribution device 100 may consider the presence of a first advertising contact and the presence of a second advertising contact. As an example, the first advertisement may include an advertisement text such as "If you go to New York, that airplane (exemplary advertisement text)". Further, the second advertisement may include an advertisement text such as "mileage campaign (exemplary advertisement text)".

For example, the explanatory variables of the conversion probability prediction model may be attributes (for example, interests such as gender, age, domestic travel, overseas travel, etc.), the presence or absence of the first advertisement contact, and the presence or absence of the second advertisement contact. .. In this case, the creation unit 136 of the information distribution device 100 may create a data set assuming the presence / absence of the first advertisement contact or the presence / absence of the second advertisement contact. For example, the creation unit 136 has a first data set assuming that there is no first ad contact, no second ad contact, a first ad contact, and a second ad. The second dataset, which assumes no contact, the third dataset, which assumes that there is no first ad contact, and the second ad contact, and the first ad contact You may create a fourth dataset that assumes that there is and there is a second ad contact.

The information distribution device 100 may consider the order of advertisement contact. For example, the attribute corresponding to the first advertisement contact may correspond to the first timing. Further, the attribute corresponding to the second advertisement contact may correspond to the second timing.

Then, the prediction unit 137 of the information distribution device 100 may calculate four predicted conversion probabilities of the same user, which are different from each other depending on the presence / absence of the first advertisement contact and the presence / absence of the second advertisement contact. As a result, the evaluation unit 138 of the information distribution device 100 can calculate four lift values corresponding to the presence / absence of the first advertisement contact and the presence / absence of the second advertisement contact. The distribution unit 139 of the information distribution device 100 may select a user who is a target of advertisement distribution based on these four lift values.

[4-2. Applicable domain]
In the exemplary embodiment shown in FIGS. 2A to 2C, the information distribution device 100 selects a user to be targeted for advertisement distribution based on the lift value of the conversion probability due to the advertisement contact, but is limited to this. It is not something that will be done. In some implementations, the information distribution device 100 may be applied to domains other than advertisement distribution. It is possible to apply the above-mentioned "effect of the second feature of the target when the target has the first feature on the possibility that an event related to the target occurs" to the domain other than the advertisement distribution. Is. For example, when the first feature of the target is the user information of the user (for example, attribute information, behavior information), the information distribution device 100 can be applied to the following domains.

(coupon)
As described above, when the first feature of the target is the user information of the user, the second feature of the target may be whether various coupons have been delivered. In this case, the predetermined event related to the target may be a conversion event such as the use of various coupons.

(Push Notification)
The second feature of the subject may be whether the push notification has been delivered. In this case, the predetermined event related to the target may be a conversion event such as a purchase of a product related to the push notification.

(questionnaire)
The second feature of the subject may be whether the questionnaire was delivered. In this case, the predetermined event related to the target may be a conversion event such as an answer to a questionnaire.

(Incentive)
The second feature of the subject may be whether incentives such as points have been delivered. In this case, the predetermined event related to the subject may be a conversion event such as the use of an incentive.

[4-3. Target entity]
The target is, for example, a human user, but is not limited to this. In some embodiments, the subject may be a machine agent (eg, a reinforcement learning agent). In this case, the characteristics of the agent may be determined based on the behavior and attributes of the agent. Such features may be created by various feature quantity engineering. A predetermined event related to an agent may be the execution of a predetermined task. The conversion event may be read as a success episode or a failure episode when the target is an agent of reinforcement learning.

[5. others〕
Further, among the processes described in the above-described embodiment, a part of the processes described as being automatically performed can also be performed manually. Alternatively, all or part of the process described as being performed manually can be automatically performed by a known method. In addition, information including processing procedures, specific names, various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the information shown in the figure.

Further, each component of each of the illustrated devices is a functional concept, and does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or part of them may be functionally or physically distributed / physically in any unit according to various loads and usage conditions. Can be integrated and configured.

For example, a part or all of the storage unit 120 shown in FIG. 3 may not be held by the information distribution device 100, but may be held by a storage server or the like. In this case, the information distribution device 100 accesses the storage server to provide event information (for example, training instance), model data (for example, trained model data), feature information (for example, new instance), and distribution information. (For example, advertising content) and other information are acquired.

[6. Hardware configuration]
Further, the information distribution device 100 according to the above-described embodiment is realized by, for example, a computer 1000 having a configuration as shown in FIG. FIG. 5 is a diagram showing an example of a hardware configuration. The computer 1000 is connected to the output device 1010 and the input device 1020, and the arithmetic unit 1030, the primary storage device 1040, the secondary storage device 1050, the output IF (Interface) 1060, the input IF 1070, and the network IF 1080 are connected by the bus 1090. Have.

The arithmetic unit 1030 operates based on a program stored in the primary storage device 1040 or the secondary storage device 1050, a program read from the input device 1020, or the like, and executes various processes. The primary storage device 1040 is a memory device such as a RAM that temporarily stores data used by the arithmetic unit 1030 for various operations. Further, the secondary storage device 1050 is a storage device in which data used by the calculation device 1030 for various calculations and various databases are registered, such as a ROM (Read Only Memory), an HDD (Hard Disk Drive), and a flash memory. Is realized by.

The output IF 1060 is an interface for transmitting information to be output to an output device 1010 that outputs various information such as a monitor and a printer. For example, USB (Universal Serial Bus), DVI (Digital Visual Interface), and the like. It is realized by a connector of a standard such as HDMI (registered trademark) (High Definition Multimedia Interface). Further, the input IF 1070 is an interface for receiving information from various input devices 1020 such as a mouse, a keyboard, a scanner, and the like, and is realized by, for example, USB.

The input device 1020 is, for example, an optical recording medium such as a CD (Compact Disc), a DVD (Digital Versatile Disc), a PD (Phase change rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), or a tape. It may be a device that reads information from a medium, a magnetic recording medium, a semiconductor memory, or the like. Further, the input device 1020 may be an external storage medium such as a USB memory.

The network IF 1080 receives data from another device via the network N and sends it to the arithmetic unit 1030, and also transmits the data generated by the arithmetic unit 1030 to the other device via the network N.

The arithmetic unit 1030 controls the output device 1010 and the input device 1020 via the output IF 1060 and the input IF 1070. For example, the arithmetic unit 1030 loads a program from the input device 1020 or the secondary storage device 1050 onto the primary storage device 1040, and executes the loaded program.

For example, when the computer 1000 functions as the information distribution device 100, the arithmetic unit 1030 of the computer 1000 realizes the function of the control unit 130 by executing the program loaded on the primary storage device 1040.

[7. effect〕
As described above, the information distribution device 100 according to the embodiment has a creation unit 136 and an evaluation unit 138.

In the information distribution device 100 according to the embodiment, the creation unit 136 describes the first feature information on the assumption that the object having the first feature has the second feature, and the target is the second feature. Create a second feature information on the assumption that it does not have. Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 is related to the target when the target has the second feature, which is predicted based on the first feature information created by the creation unit 136. This prescription related to the subject, predicted based on the first possibility that a predetermined event will occur and the second feature information created by the creator 136, if the subject does not have the second feature. From the comparison with the second possibility that the event of the above occurs, the influence of the second feature on the possibility that this predetermined event related to the subject occurs is evaluated.

Further, the information distribution device 100 according to the embodiment learns from the feature information of the learning target having the first feature or the second feature so as to predict the possibility that a predetermined event occurs in the learning target. Using the predicted model, the first possibility that a predetermined event related to the subject occurs when the subject has the second feature, and the subject is related when the subject does not have the second feature. It has a prediction unit 137 that predicts a second possibility that a predetermined event will occur. Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 is targeted from the comparison between the first possibility predicted by the prediction unit 137 and the second possibility predicted by the prediction unit 137. Evaluate the impact of the second feature on the likelihood that this predetermined event will occur.

Further, in the information distribution device 100 according to the embodiment, the prediction unit 137 uses a prediction model trained using the feature information showing the relationship between the occurrence of a predetermined event related to the learning target and the feature of the learning target. Then, the first possibility and the second possibility are predicted.

Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 has the first possibility and the second possibility as the influence of the second feature on the possibility that this predetermined event related to the target occurs. Evaluate a given index to compare with the possibility of.

Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138, as a predetermined index, has a predetermined event related to the target when the first possibility is not assumed that the target has the second feature. Calculate an evaluation index that indicates how much the increase has occurred with respect to the possibility of occurrence of.

Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 evaluates the difference between the first possibility and the second possibility and the ratio between the first possibility and the second possibility as an evaluation index. The lift value according to the second feature is calculated based on at least one of them.

Further, in the information distribution device 100 according to the embodiment, the creating unit 136 has the first feature information representing the first value of the second attribute and the second feature, as having the second feature. As a non-existence, a second feature information representing the second value of this second attribute is created. Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 determines the possibility that this predetermined event related to the target occurs from the comparison between the first possibility and the second possibility. As the influence of the second feature, the first value of the second attribute represented by the first feature information and the second value of the second attribute represented by the second feature information. Evaluate changes in the likelihood of a given subject-related event resulting from the difference.

Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 has a first possibility that the target takes a predetermined action, which is a predetermined event, when the target has the second feature, and the target is the second. By comparison with the second possibility that the subject takes this predetermined action, which is a predetermined event when the subject does not have the characteristics of, the predetermined action related to the subject, which is the possibility that the subject takes this predetermined action. Evaluate the impact of the second feature on the likelihood of an event occurring.

Further, in the information distribution device 100 according to the embodiment, the creation unit 136 has the first feature information indicating that the predetermined information has been delivered to the target, and the target has the second feature. Assuming that there is no feature of 2, a second feature information indicating that this predetermined information has not been delivered to the target is created. Further, in the information distribution device 100 according to the embodiment, the evaluation unit 138 has a second possibility that a predetermined event related to the target occurs from the comparison between the first possibility and the second possibility. As an effect of the characteristics of, the effect of the distribution of predetermined information on the possibility that a predetermined event related to the target occurs is evaluated.

Further, the information distribution device 100 according to the embodiment distributes the predetermined information to the target when the change in the possibility that the predetermined event related to the target occurs, which is evaluated by the evaluation unit 138, satisfies the predetermined condition. It has a distribution unit 139.

By each of the above-mentioned processes, the information distribution device 100 can extract useful information from the features.

Although some of the embodiments of the present application have been described in detail with reference to the drawings, these are examples, and various modifications are made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure column of the invention. It is possible to carry out the present invention in other modified forms.

Further, the above-mentioned information distribution device 100 may be realized by a plurality of server computers, and depending on the function, it may be realized by calling an external platform or the like by API (Application Programming Interface), network computing, or the like. Can be changed flexibly.

Further, the above-mentioned "section, module, unit" can be read as "means" or "circuit". For example, the collecting unit can be read as a collecting means or a collecting circuit.

1 Information distribution system 100 Information distribution device 110 Communication unit 120 Storage unit 121 Event information storage unit 122 Model storage unit 123 Feature information storage unit 124 Distribution information storage unit 130 Control unit 131 Collection unit 132 Event information acquisition unit 133 Generation unit 134 Reception Part 135 Feature information acquisition part 136 Creation part 137 Prediction part 138 Evaluation part 139 Distribution part 200 Information provision device 300 User device

Claims (12)

The first feature information assuming that the object having the first feature has the second feature and the second feature information assuming that the target does not have the second feature. The creation part to create and
The first possibility that a predetermined event related to the target occurs when the target has the second feature, which is predicted based on the first feature information created by the creation unit, and the above-mentioned. With the second possibility that the predetermined event related to the target occurs when the target does not have the second feature, which is predicted based on the second feature information created by the creation unit. An evaluation device comprising an evaluation unit for evaluating the influence of the second feature on the possibility that the predetermined event related to the target may occur. Using a prediction model that has been trained to predict the possibility that the predetermined event will occur in the learning target from the feature information of the learning target having the first feature or the second feature. The first possibility that the predetermined event related to the subject occurs when the subject has the second feature, and the subject related to the subject when the subject does not have the second feature. Further provided with a prediction unit for predicting the second possibility that the predetermined event will occur.
The evaluation unit compares the first possibility predicted by the prediction unit with the second possibility predicted by the prediction unit, and the predetermined event related to the target occurs. The evaluation device according to claim 1, wherein the influence of the second feature on the possibility is evaluated. The prediction unit uses the prediction model trained using the feature information indicating the relationship between the occurrence of the predetermined event related to the learning target and the feature of the learning target, and the first possibility. The evaluation device according to claim 2, wherein the property and the second possibility are predicted. In order to compare the first possibility and the second possibility as the influence of the second feature on the possibility that the predetermined event related to the object occurs. The evaluation device according to any one of claims 1 to 3, characterized in that the predetermined index of the above is evaluated. As the predetermined index, the evaluation unit determines that the first possibility is the possibility that a predetermined event related to the target occurs when the target is not assumed to have the second characteristic. The evaluation device according to claim 4, further comprising calculating an evaluation index indicating how much the increase has occurred. The evaluation unit sets the evaluation index as at least one of the difference between the first possibility and the second possibility and the ratio between the first possibility and the second possibility. The evaluation device according to claim 5, wherein the lift value according to the second feature is calculated based on the above. The creating unit has the first feature information representing the first value of the second attribute as having the second feature, and the second feature as having no second feature. Create the second feature information that represents the second value of the attribute,
From the comparison between the first possibility and the second possibility, the evaluation unit determines that the influence of the second feature on the possibility that the predetermined event related to the target occurs. The difference between the first value of the second attribute represented by the first feature information and the second value of the second attribute represented by the second feature information. The evaluation device according to any one of claims 1 to 6, wherein the change in the possibility of occurrence of the predetermined event related to the object is evaluated. The evaluation unit has the first possibility that the target takes a predetermined action which is the predetermined event when the target has the second feature, and the target has the second feature. From the comparison with the second possibility that the subject takes the predetermined action, which is the predetermined event in the absence of the present, the predetermined behavior related to the subject, which is the possibility that the subject takes the predetermined action. The evaluation device according to any one of claims 1 to 7, wherein the influence of the second feature on the possibility of occurrence of the above-mentioned event is evaluated. The creating unit has the first feature information indicating that the predetermined information has been delivered to the target, and the target does not have the second feature, as the target has the second feature. As the second feature information indicating that the predetermined information has not been delivered to the target, the second feature information is created.
From the comparison between the first possibility and the second possibility, the evaluation unit determines that the second feature has an effect on the possibility that the predetermined event related to the target occurs. The evaluation according to any one of claims 1 to 8, wherein the influence of the distribution of the predetermined information on the possibility that the predetermined event related to the target occurs is evaluated. Device. A distribution unit that distributes the predetermined information to the target when the change in the possibility of the predetermined event related to the target satisfied by the evaluation unit satisfies the predetermined condition is further provided. The evaluation device according to claim 9. It ’s an evaluation method performed by a computer.
The first feature information assuming that the object having the first feature has the second feature and the second feature information assuming that the target does not have the second feature. The creation process to create and
The first possibility that a predetermined event related to the target occurs when the target has the second feature, which is predicted based on the first feature information created by the preparation step, and the above-mentioned. With the second possibility that the predetermined event related to the target occurs when the target does not have the second feature, which is predicted based on the second feature information created by the creation step. An evaluation method comprising an evaluation step of evaluating the influence of the second feature on the possibility that the predetermined event related to the target occurs from the comparison of the above. The first feature information assuming that the object having the first feature has the second feature and the second feature information assuming that the target does not have the second feature. The creation procedure to create and
The first possibility that a predetermined event related to the target occurs when the target has the second feature, which is predicted based on the first feature information created by the preparation procedure, and the above-mentioned. With the second possibility that the predetermined event related to the target occurs when the target does not have the second feature, which is predicted based on the second feature information created by the creation procedure. An evaluation program characterized by having a computer execute an evaluation procedure for evaluating the influence of the second feature on the possibility that the predetermined event related to the target occurs from the comparison of the above.

Images