JP6924303B2 - Generation device, generation method, and generation program - Google Patents

Description

The present invention relates to a generator, a generator, and a generator.

Conventionally, a technique for predicting a predetermined target based on a generated prediction model (hereinafter, simply referred to as “model”) has been provided. For example, a technique for predicting a predetermined demand using a model is provided. In such a technique, when a model for predicting interest in content to be delivered to a user is generated, a model is generated for all users or a model is generated for each user.

Japanese Unexamined Patent Publication No. 2013-196037

However, with the above-mentioned prior art, it may be difficult to generate a model that appropriately predicts a predetermined target. For example, when a model is generated for all of a large number of users, it is difficult to generate an appropriately predictable model for each user. Further, for example, when a model is generated for each of a large number of users, it is difficult to generate an appropriately predictable model for a user whose amount of information used for model generation is not sufficient.

The present application has been made in view of the above, and an object of the present application is to provide a generation device, a generation method, and a generation program for generating a model that appropriately predicts a predetermined object based on a context.

The generation device according to the present application generates an acquisition unit that acquires behavior information related to a user's behavior for each category related to the user from a terminal device used by the user, and division information that is behavior information divided based on the context. , Generates union information which is divided information combined based on the tendency of action presence / absence indicated by the action information included in each generated divided information, and predicts the content that the user is interested in using the generated union information. It is characterized by having a generation unit for generating a model.

According to one aspect of the embodiment, it is possible to generate a model that appropriately predicts a predetermined object based on the context.

FIG. 1 is a diagram showing an example of a generation process according to an embodiment. FIG. 2 is a diagram showing an example of distribution processing according to the embodiment. FIG. 3 is a diagram showing a configuration example of the distribution device according to the embodiment. FIG. 4 is a diagram showing an example of a user information storage unit according to the embodiment. FIG. 5 is a diagram showing an example of the behavior information storage unit according to the embodiment. FIG. 6 is a diagram showing an example of the divided information storage unit according to the embodiment. FIG. 7 is a diagram showing an example of the cluster information storage unit according to the embodiment. FIG. 8 is a diagram showing an example of the learning information storage unit according to the embodiment. FIG. 9 is a diagram showing an example of the content information storage unit according to the embodiment. FIG. 10 is a flowchart showing an example of the generation process according to the embodiment. FIG. 11 is a flowchart showing an example of the generation process according to the embodiment. FIG. 12 is a flowchart showing an example of the distribution process according to the embodiment. FIG. 13 is a diagram showing an example of a classification regarding a user according to an embodiment. FIG. 14 is a diagram showing an example of using the divided information in other categories according to the embodiment. FIG. 15 is a diagram showing an example of using the model in other categories according to the embodiment. FIG. 16 is a diagram showing an example of a variation in the context according to the embodiment. FIG. 17 is a diagram showing an example of context integration according to an embodiment. FIG. 18 is a diagram showing an example of estimation of user attributes based on the cluster and the model according to the embodiment. FIG. 19 is a hardware configuration diagram showing an example of a computer that realizes the function of the distribution device.

Hereinafter, a generator, a generation method, and a mode for carrying out the generation program (hereinafter referred to as “the embodiment”) according to the present application will be described in detail with reference to the drawings. It should be noted that this embodiment does not limit the generator, the generation method, and the generation program according to the present application. Further, in each of the following embodiments, the same parts are designated by the same reference numerals, and duplicate description is omitted.

(Embodiment)
[1-1. Generation process]
First, an example of the generation process according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of a generation process according to an embodiment. In the example shown in FIG. 1, the distribution device 100 generates a model for predicting a user's interest in content as a predetermined target. Further, the distribution device 100 provides a service of distributing content to the user using the generated model, and this point will be described with reference to FIG. In addition, in FIG. 1 and FIG. 2, the classification regarding a user will be described as a user. 1 and 2 show a case where users U1 to U4 use terminal devices 10-1 to 10-4, respectively. In the case where the terminal devices 10-1 to 10-4 are described without distinction, they are collectively referred to as the terminal device 10.

As shown in FIGS. 1 and 2, the distribution system 1 includes a terminal device 10 and a distribution device 100. The terminal device 10 and the distribution device 100 are connected to each other so as to be communicable by wire or wirelessly via a predetermined communication network (not shown). The distribution system 1 shown in FIG. 1 may include a plurality of terminal devices 10 and a plurality of distribution devices 100.

The terminal device 10 is an information processing device used by the user. The terminal device 10 is realized by, for example, a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. FIG. 1 shows a case where the terminal device 10 is a smartphone.

In addition, the terminal device 10 accepts operations by the user. In the example shown in FIG. 1, the terminal device 10 requests the distribution device 100 for the content to be displayed on a predetermined application (for example, a browser or the like). In the following, the terminal device 10 may be referred to as a user. That is, in the following, the user can be read as the terminal device 10.

The distribution device 100 is a generation device that generates a model for predicting a predetermined target for each union information in which the division information, which is the behavior information divided based on the context, is combined. Further, the distribution device 100 is an information processing device that provides a distribution service that distributes content in response to a content distribution request. For example, the distribution device 100 is an information processing device that distributes content to the terminal device 10 according to a user who uses the terminal device 10 that requests the distribution of the content.

In FIG. 1, the distribution device 100 distributes content to each user and collects behavior information indicating the presence or absence of each user's behavior with respect to the distributed content. For example, the distribution device 100 distributes the content to the users U1 to U4, and collects the action information indicating the presence or absence of the action of each user with respect to the distributed content. For example, the distribution device 100 collects information indicating whether or not the user has an action (for example, a click of the content) with respect to the content (for example, the content CN11) distributed to the user as the action information.

First, the distribution device 100 distributes the content to the terminal device 10-1 used by the user U1 (step S11-1). Then, the distribution device 100 acquires the action information indicating the presence / absence of the action of the user U1 with respect to the distributed content from the terminal device 10-1 (step S12-1). Further, the distribution device 100 distributes the content to the terminal device 10-2 used by the user U2 (step S11-2). Then, the distribution device 100 acquires the action information indicating the presence / absence of the action of the user U2 with respect to the distributed content from the terminal device 10-2 (step S12-2). Further, the distribution device 100 distributes the content to the terminal device 10-3 used by the user U3 (step S11-3). Then, the distribution device 100 acquires the action information indicating the presence / absence of the action of the user U3 with respect to the distributed content from the terminal device 10-3 (step S12-3). Further, the distribution device 100 distributes the content to the terminal device 10-4 used by the user U4 (step S11-4). Then, the distribution device 100 acquires the action information indicating the presence / absence of the action of the user U4 with respect to the distributed content from the terminal device 10-4 (step S12-4).

It should be noted that steps S11-1 to S11-4 are for explaining the process, and any of steps S11-1 to S11-4 may be performed first, and each step S11-1 to S11-4 may be performed first. May be performed multiple times. Hereinafter, when steps S11-1 to S11-4 will be described without distinction, they are collectively referred to as step S11. Further, steps S12-1 to S12-4 are for explaining the processing, and if each is after the corresponding step S11, any of steps S12-1 to S12-4 is performed first. Each step S12-1 to S12-4 may be performed a plurality of times. Hereinafter, when steps S12-1 to S12-4 will be described without distinction, they are collectively referred to as step S12. For example, each time each step S11 is performed, step S12 may be performed to acquire action information. Further, for example, each step S11 may be performed a plurality of times, and step S12 may be performed at a predetermined timing to acquire action information.

Then, the distribution device 100 divides the information regarding the behavior of each user collected in steps S11 to S12 (step S13). Specifically, the distribution device 100 divides the collected behavior information of each user based on the context.

Here, the context means the situation of the user or the terminal device 10 or the environment (background) of the user or the terminal device 10. For example, the context is information corresponding to a situation in which the terminal device 10 is used by the user or a state in which the user who owns the terminal device 10 is placed. That is, the context in FIGS. 1 and 2 corresponds to the situation of users U1 to U4, for example, the environment where the users U1 to U4 are located. Specifically, the contexts in FIGS. 1 and 2 correspond to the situations of users U1 to U4 when the content is distributed (displayed) to each user. For example, the context may be estimated based on the time when the content is delivered or the user's position information acquired from the terminal device 10. Note that each context may be exclusive or duplicated.

Further, in FIGS. 1 and 2, for simplification of explanation, a context based on a time zone and a position is shown. Specifically, the contexts in the examples of FIGS. 1 and 2 are four: morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), and midnight (22 to 4 o'clock). Eight contexts CT11-CT18 are shown based on the time of day and two locations of home (home) or office (work location).

Here, the divided information storage unit 123 in FIG. 1 shows a state in which the action information of each user is divided for each context in step S13. The divided information storage unit 123 shows a case where the user who is a division is a row and the context is a column. The area where one user and one context intersect in the divided information storage unit 123 is defined as a mass. For example, in FIG. 1, the region where the user U4 and the context CT12 intersect is defined as the mass MS42. When the position of each cell is not distinguished, it may be referred to as mass MS. Further, even in the cluster information storage unit 124 or the like, if there is an area where one user and one context intersect, that area may be a mass MS. The information included in each mass MS shown in FIG. 1 is divided information in which the behavior information regarding the user's behavior for each division regarding the user is divided based on the context.

For example, the mass MS at the position where the user U1 and the context CT11 intersect (the mass MS on the upper left) includes division information indicating the presence or absence of actions such as actions AT11, AT12, and AT13. For example, the mass MS at the position where the user U4 and the context CT18 intersect (the mass MS at the lower right) includes division information indicating the presence or absence of actions such as actions AT48 and AT11.

Further, “◯” shown on the right side of the action AT11 or the like in each mass MS in the divided information storage unit 123 indicates that the user has performed the corresponding action. Further, "x" shown on the right side of the action AT11 or the like in each mass MS in the divided information storage unit 123 indicates that the user did not perform the corresponding action. For example, in the mass MS at the position where the user U1 and the context CT11 intersect, the action AT11 is “◯”, indicating that the user U1 has performed the action AT11 at the morning house. Further, for example, in the mass MS at the position where the user U1 and the context CT11 intersect, the action AT12 is “x”, indicating that the user U1 did not perform the action AT12 in the morning house. Further, for example, in the mass MS at the position where the user U1 and the context CT11 intersect, the action AT13 is “◯”, indicating that the user U1 has performed the action AT13 in the morning house.

Then, the distribution device 100 generates a model by using the division information of each mass MS in the division information storage unit 123 (step S14). For example, the distribution device 100 clusters the division information of each mass MS in the division information storage unit 123, and generates a model for each cluster. For example, the distribution device 100 generates a model that predicts a user's interest in content.

For example, the distribution device 100 may perform clustering by a clustering method in which the number of clusters is not set. Further, for example, the distribution device 100 may perform clustering by a clustering method for setting the number of clusters. For example, the distribution device 100 clusters the division information of each mass MS by using the k-means method or the logistic regression method using the Dirichlet process, and generates a model using the division information for each cluster. In FIG. 1, the distribution device 100 will be described below assuming that the distribution device 100 clusters the division information of each mass MS by using the logistic regression method using the Dirichlet process and generates a model using the division information for each cluster. .. The details of the clustering method will be described later.

The distribution device 100 may target mass MS including a predetermined amount or more of division information as a model generation target in step S14. For example, the distribution device 100 does not target the mass MS corresponding to the context CT17 of the user U4 and the mass MS corresponding to the context CT11 of the user U2 that do not include the division information to be clustered. Further, for example, in the distribution device 100, the mass MS corresponding to the contexts CT13 and CT18 of the user U1 containing less than a predetermined amount (for example, two) of divided information may not be the target of clustering.

In step S14, the distribution device 100 generates model information MD11. The model information MD 11 includes information related to clustering of divided information included in each mass MS as shown in the cluster information storage unit 124. Further, the model information MD 11 includes information about models M1 to M6 generated for each cluster as shown in the learning information storage unit 125.

In FIG. 1, the distribution device 100 classifies the division information of each mass MS in the division information storage unit 123 into six clusters (M1 to M6 in the cluster information storage unit 124). Note that M1 to M6 in the cluster information storage unit 124 indicate the cluster to which each division information belongs and the model to be generated. Further, in the mass MS in the cluster information storage unit 124, a model generated by a predetermined method is assigned to the mass MS whose symbols M1 to M6 are not shown. For example, the distribution device 100 may assign a model to a mass MS to which a model is not assigned in the cluster information storage unit 124 based on a model assigned to another mass MS. For example, the distribution device 100 assigns the model M4 assigned to the mass MS corresponding to the context CT15 of the user U1 of the same male in his twenties to the mass MS corresponding to the context CT15 of the user U4 to which the model is not assigned. May be good. Further, for example, in the distribution device 100, the model assigned to the mass MS corresponding to the context CT17 of the user U4 to which the model is not assigned is assigned to the mass MS corresponding to the context CT17 of the user U1 or the user U2 in the same 20s. You may assign a model based on. In this case, even if the distribution device 100 assigns a model that combines the model M2 assigned to the mass MS corresponding to the context CT17 of the user U1 and the model M1 assigned to the mass MS corresponding to the context CT17 of the user U2. good. For example, the distribution device 100 has a weight of each feature of the model M2 assigned to the mass MS corresponding to the context CT17 of the user U1 and a weight of each feature of the model M1 assigned to the mass MS corresponding to the context CT17 of the user U2. A model in which and is synthesized at a predetermined ratio may be assigned. For example, the distribution device 100 assigns a model in which the ratio of the model M2 assigned to the mass MS corresponding to the context CT17 of the male user U1 in his twenties is larger to the mass MS corresponding to the context CT17 of the user U4. You may. For example, the distribution device 100 may assign a model M1-2 synthesized at a ratio of 0.3 for the model M1 and 0.7 for the model M2 to the mass MS corresponding to the context CT17 of the user U4. That is, the distribution device 100 generated the model M1- by adding the weight obtained by multiplying the weight of each feature of the model M1 by 0.3 and the weight obtained by multiplying the weight of each feature of the model M2 by 0.7. 2 may be assigned to the mass MS corresponding to the context CT17 of the user U4.

In FIG. 1, the distribution device 100 corresponds to the mass MS corresponding to the context CT11 of the user U1, the mass MS corresponding to the context CT15 of the user U2, the mass MS corresponding to the context CT17 of the user U2, and the context CT13 of the user U4. The division information included in the mass MS is classified into the same cluster. Further, it is shown that the distribution device 100 generated the model M1 by using the union information obtained by combining the division information included in the above four mass MSs. Further, as shown in the learning information storage unit 125, in the model information regarding the model M1, the weight of the feature 1 is "0.5", the weight of the feature 2 is "-0.4", and the weight of the feature 3 is "0. 2 ”etc. The features may be set by the administrator of the distribution device 100 or the like, or may be extracted in the model generation process. For example, the feature may be a character string (keyword) such as "economy", "smartphone", or "soccer". In addition, the features of each model may be the same or different.

Further, the distribution device 100 shares the divided information included in the mass MS corresponding to the context CT12 and CT17 of the user U1, the mass MS corresponding to the context CT14 of the user U2, and the mass MS corresponding to the context CT12 of the user U3 in the same cluster. Classify into. Further, the distribution device 100 generated the model M2 by using the union information obtained by combining the mass MS corresponding to the context CT14 of the user U2 and the divided information included in the mass MS corresponding to the context CT12 of the user U3. Is shown. Further, as shown in the learning information storage unit 125, the model information regarding the model M2 has a weight of feature 1 of "-0.3", a weight of feature 2 of "2.1", and a weight of feature 3 of "-0". .2 ”etc.

As described above, in FIG. 1, the distribution device 100 classifies the division information of each mass MS in the division information storage unit 123 into six clusters (union information), and uses the information included in each cluster (union information). Models M1 to M6 are generated.

As described above, even if the distribution device 100 divides the behavior information of each division based on the context, it suppresses the shortage of the amount of information used for generating each model by performing clustering across the divisions. can do. Further, the distribution device 100 can generate an appropriate model according to each context of each division by clustering the information divided based on the division and the context. Therefore, the distribution device 100 can generate a model that appropriately predicts a predetermined target based on the context.

Note that FIG. 1 shows an example in which information regarding the presence or absence of the user's behavior in the distributed content is used as the behavior information, but what if the behavior information is the behavior performed by the user and can be collected? Information may be used. For example, the behavior information may be information related to the user's electronic commerce, or may be behavior related to the delivered advertisement (for example, click, conversion, etc.). In addition, the location is not limited to the home (home) or office (work location), and may be various information such as a waypoint during commuting, a frequently visited location, or moving.

[1-2. Delivery process]
Next, an example of the distribution process according to the embodiment will be described with reference to FIG. FIG. 2 is a diagram showing an example of distribution processing according to the embodiment. FIG. 2 shows a case where the user is the user U1. In FIG. 2, the case where the distribution device 100 has already acquired the user information regarding the user attribute of the user U1 and the user information such as the action history has been acquired will be described as an example.

First, the distribution device 100 acquires a content distribution request from the terminal device 10-1 used by the user U1 (step S15). In FIG. 2, in step S15, the terminal device 10-1 transmits information for identifying the user U1 (for example, a login ID or the like), position information indicating the location of the terminal device 10-1, or the like to the distribution device 100. The terminal device 10 acquires position information by a GPS (Global Positioning System) sensor or the like. In FIG. 2, it is assumed that the step S15 is performed in a situation where the time is 9:00 am and the location (position) of the terminal device 10-1 is the position LC12-1. The position LC12-1 or the like, which is information indicating the position and has a hyphen (“-”), means the position information within a predetermined range from the position LC12. Further, the positions LC11-1, LC11-2, etc. (see FIG. 5) mean position information within a predetermined range from the position LC11. That is, when the position LC12-1 indicates the position information of the user U1, it is estimated that the user U1 is at the work location at the position LC12.

The distribution device 100 that has acquired the content distribution request from the terminal device 10-1 determines a model for determining the content to be distributed to the terminal device 10-1 (step S16). The distribution device 100 estimates that the distribution request source is the user U1 based on the information for identifying the user U1 acquired from the terminal device 10-1. Then, the distribution device 100 estimates the location of the user U1 based on the user information of the user U1 stored in the user information storage unit 121 and the position LC12-1 of the terminal device 10-1. In FIG. 2, as shown in the user information storage unit 121, the work location of the user U1 is the position LC12, and the position of the terminal device 10-1 is the position LC12-1. Therefore, in the distribution device 100, the user U1 is in the office ( Estimated to be at work location).

Then, since the time when the distribution request is made in step S15 is 9:00 am, the distribution device 100 estimates that the context is the context CT12 corresponding to the office in the morning. After that, the distribution device 100 determines the model to be used for determining the content to be distributed in the model M2 based on the information indicating the model M2 stored in the mass MS corresponding to the user U1 and the context CT12 in the learning information storage unit 125. do.

For example, the distribution device 100 calculates the score of each content stored in the content information storage unit 126 (see FIG. 9) using the determined model M2. Specifically, the distribution device 100 includes character information and image information such as titles and article texts of each content CN11 (article A) to content CN18 (article H) stored in the content information storage unit 126, and model M2. Is used to calculate the scores of each content CN11 to CN18 and the like. In FIG. 2, the distribution device 100 calculates the scores of the contents CN11 to CN18 and the like as shown in the contents list CL11. For example, the distribution device 100 calculates the score of the content CN11 which is the article A as "3.6". Further, for example, the distribution device 100 calculates the score of the content CN12, which is the article B, as “2.5”.

Then, the distribution device 100 determines the ranking of each content CN11 to CN18 or the like based on the score shown in the content list CL11 (step S17). In FIG. 2, the distribution device 100 determines the order of the contents CN11 to CN18 and the like in order from the one with the highest score shown in the contents list CL11. Specifically, the distribution device 100 determines the ranking of the article E, which is the content CN15 having the highest score, to be the first. Further, the distribution device 100 determines the ranking of the article G, which is the content CN17 having the highest score next to the article E, to be second. As a result, as shown in the ranking list LL11, the distribution device 100 has the ranking 1st place is the article E, the ranking 2nd place is the article G, the ranking 3rd place is the article A, and the ranking 4th place is the article F. Decide that there is.

After that, the distribution device 100 distributes the content shown in the ranking list LL11 to the terminal device 10-1 (step S18). Specifically, the content displayed in the order of article E, article G, article A, article F, etc. is distributed to the terminal device 10-1. Then, the terminal device 10-1 that has received the content shown in the ranking list LL11 displays the content (step S19). In FIG. 2, the terminal device 10-1 displays the article E, the article G, the article A, and the article F, which are the contents having the higher rank among the contents shown in the rank list LL11. The user U1 who uses the terminal device 10-1 touches the screen of the terminal device 10-1 and performs a scroll operation or the like to sequentially display the articles B, the articles H, and the like from the article F to the bottom.

As described above, the distribution device 100 determines the model according to the context when the distribution request is acquired. Then, the distribution device 100 determines the ranking of the contents based on the score of the contents calculated using the determined model. As a result, the distribution device 100 can distribute appropriate content according to the user and the context at the time of the distribution request. Therefore, the distribution device 100 can extract appropriate contents. In the above example, the distribution device 100 calculates the score of each content calculated using the model selected based on the user's context, and ranks the calculated contents in order from the highest score. Indicated. As described above, in the above-described example, the distribution device 100 shows a case where the distribution device 100 uses a model that calculates the possibility of viewing each content as a continuous numerical value (score) in the corresponding context. That is, in the above example, the higher the score of each content calculated by the distribution device 100 using the model selected based on the user's context, the higher the possibility that the content will be viewed. , The distribution device 100 may make any prediction using the selected model.

For example, the distribution device 100 may predict that the user will browse the content having a score equal to or higher than a predetermined threshold value, and predict that the user will not browse the content having a score lower than the predetermined threshold value. In this way, the distribution device 100 may or may not allow the user to browse the content (for example, the value "1") based on the score of each content calculated using the model selected based on the context (for example, the value "0"). ”) The binary prediction may be performed. The distribution device 100 is not limited to binary prediction, but may generate a model for predicting continuous value or discrete value, and predict continuous value or discrete value based on the generated model. For example, the distribution device 100 may make a prediction corresponding to a so-called multi-label problem in which not only two values of "0" and "1" but also three or more values are handled. For example, the distribution device 100 may predict at which stage of three or more stages the user is present, instead of the two stages of whether or not the user browses the content. For example, the distribution device 100 may use a plurality of threshold values to predict at which stage the user is likely to browse the content. Specifically, the prediction device 100 uses a first threshold value and a second threshold value smaller than the first threshold value, and when the score is equal to or higher than the first threshold value, the user may browse the content. If it is judged to be high and the score is less than the first threshold value and is equal to or higher than the second threshold value, it is judged that the user has a medium possibility of viewing the content, and if the score is less than the second threshold value, the content is judged to be high. May be predicted to be unlikely to be viewed by the user. Further, for example, the distribution device 100 may generate a model for predicting the number of reactions or predicting the score of a predetermined target, and make a prediction based on the generated model. That is, any prediction may be made using the model selected by the distribution device 100 based on the user's context. Further, the distribution device 100 may generate a model according to the target to be predicted as described above.

[2. Distribution device configuration]
Next, the configuration of the distribution device 100 according to the embodiment will be described with reference to FIG. FIG. 3 is a diagram showing a configuration example of the distribution device 100 according to the embodiment. As shown in FIG. 3, the distribution device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The distribution device 100 has an input unit (for example, a keyboard, a mouse, etc.) that receives various operations from the administrator of the distribution device 100, and a display unit (for example, a liquid crystal display, etc.) for displaying various information. You may.

(Communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. Then, the communication unit 110 is connected to the network by wire or wirelessly, and transmits / receives information to / from the terminal device 10.

(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 according to the embodiment includes a user information storage unit 121, an action information storage unit 122, a division information storage unit 123, a cluster information storage unit 124, and a learning information storage unit 125. , Content information storage unit 126.

(User information storage unit 121)
The user information storage unit 121 according to the embodiment stores various information about the user. For example, the user information storage unit 121 stores various information related to user attributes. FIG. 4 is a diagram showing an example of a user information storage unit according to the embodiment. The user information storage unit 121 shown in FIG. 4 includes items such as "user ID", "age", "gender", "home", "work location", and "interest".

The "user ID" indicates identification information for identifying a user. For example, the user identified by the user ID "U1" corresponds to the user U1 shown in the example of FIG. Further, "age" indicates the age of the user identified by the user ID. The "age" may be the specific age of the user identified by the user ID, such as 35 years old. Further, "gender" indicates the gender of the user identified by the user ID.

Further, "home" indicates the location information of the user's home identified by the user ID. In the example shown in FIG. 4, "home" is illustrated with an abstract code such as "LC11", but may be information indicating latitude or longitude. Further, for example, "home" may be an area name or an address.

Further, the "work location" indicates the location information of the user's work location identified by the user ID. In the example shown in FIG. 4, the "work location" is shown by an abstract code such as "LC12", but may be information indicating latitude or longitude. Further, for example, "home" may be an area name or an address.

Further, "interest" indicates the interest of the user identified by the user ID. That is, "interest" indicates an object that the user identified by the user ID is highly interested in. In the example shown in FIG. 4, one "interest" is shown for each user, but there may be a plurality of "interests".

For example, in the example shown in FIG. 4, the age of the user identified by the user ID "U1" is "20's", and the gender is "male". Further, for example, the user identified by the user ID "U1" indicates that his / her home is "LC11". Further, for example, the user identified by the user ID "U1" indicates that the work location is "LC12". Further, for example, the user identified by the user ID "U1" indicates that he / she is interested in "sports".

The user information storage unit 121 is not limited to the above, and may store various information depending on the purpose. For example, the user information storage unit 121 may store information such as a name, family structure, and income.

(Behavior information storage unit 122)
The behavior information storage unit 122 according to the embodiment stores various information related to the user's behavior. FIG. 5 is a diagram showing an example of the behavior information storage unit according to the embodiment. For example, the behavior information storage unit 122 stores the user's behavior information for the content delivered to the terminal device 10 of each user. The action information storage unit 122 shown in FIG. 5 includes items such as "user ID", "behavior", and "situation". In addition, the "action" includes items such as "action content" and "presence / absence". In addition, the "situation" includes items such as "time" and "position".

The "user ID" indicates identification information for identifying a user. For example, the user identified by the user ID "U1" corresponds to the user U1 shown in the example of FIG. Further, "behavior" indicates information about the behavior of the corresponding user. Further, "status" indicates information regarding the status of the corresponding user.

"Behavior content" indicates identification information for identifying each action. For example, the action identified by the action content "AT11" (action AT11) corresponds to the user's action (for example, clicking the content) with respect to the content (for example, content CN11) delivered to the user. The action content is not limited to the identification information, and a specific action content may be stored. In addition, "presence / absence" indicates the presence / absence of the corresponding action content. When "presence / absence" is "1", it indicates that the user has performed the corresponding action. Further, when "presence / absence" is "0", it indicates that the user has not performed the corresponding action.

"Time" indicates the time for each action. The time "TM11" indicates the time related to the action AT11. For example, the time "TM11" may be the time when the content (for example, the content CN11) is delivered to the user. Further, for example, the time "TM11" may be the time when the user clicks the content (for example, the content CN11) delivered to the user. The time “TM12” related to the action (action AT12) identified by the action content “AT12” which is “0” may be, for example, the time when the content is delivered to the user. Further, for example, the time "TM12" may be a time after a predetermined time has elapsed without being clicked after the content is delivered to the user.

In addition, "position" indicates a position related to each action. The position "LC11-1" indicates the position regarding the action AT11. For example, the position "LC11-1" may be the position information when the user requests the distribution of the content. Further, for example, the position "LC11-1" may be the position information when the user clicks the distributed content.

For example, in the example shown in FIG. 5, the user identified by the user ID “U1” has an action (action AT13) identified by the action content “AT13” with or without “1”, and has performed the action AT13. show. Further, the user identified by the user ID "U1" indicates that the time related to the action AT13 is the time "TM13" and the position is the position "LC11-3".

The behavior information storage unit 122 is not limited to the above, and may store various information depending on the purpose. For example, the behavior information storage unit 122 may store information regarding the time from when the content is delivered until the user acts. Further, although FIG. 5 shows a case where the action information is stored in the action information storage unit 122 for each user ID, the action information is not limited to each user ID and may be stored, for example, in chronological order.

(Divided information storage unit 123)
The divided information storage unit 123 according to the embodiment stores various information related to the divided information. FIG. 6 is a diagram showing an example of the divided information storage unit according to the embodiment. The divided information storage unit 123 shown in FIG. 6 stores the action information divided based on the "division" and the "context". Specifically, the action information corresponding to each "context" is stored for each "classification" of the user. For example, the information stored in the divided information storage unit 123 is generated from the behavior information stored in the behavior information storage unit 122.

"Category" indicates a classification related to a user. FIG. 6 shows users U1 to U4, that is, a case where each user is a division. The "classification" is not limited to each user, but may be an abstract user classification or the like, but this point will be described later. Further, "context" indicates a context. FIG. 6 shows a context based on time zone and position for simplicity. Specifically, the "context" consists of four time zones: morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), midnight (22 to 4 o'clock), and home (22:00 to 4 o'clock). Eight contexts CT11-CT18 are shown based on two locations, home) or office (work location). The "○" shown on the right side of the action content such as "AT11" in each square shown in FIG. 6 corresponds to the presence / absence "1", and the "x" corresponds to the presence / absence "0". That is, in the case of "○", it indicates that the user has performed the corresponding action. In addition, in the case of "x", it indicates that the user did not perform the corresponding action.

For example, in the example shown in FIG. 6, the user U1 indicates that the action AT11 is “◯” in the context CT11 and the action AT11 is performed at home (home) in the time zone “morning”. Further, the user U1 indicates that the action AT11 is “x” in the context CT12 and the action AT11 is not performed in the office (work location) in the time zone “morning”.

The divided information storage unit 123 is not limited to the above, and various information may be stored depending on the purpose. Further, in FIG. 6, for the sake of explanation, a case where information regarding the action content and presence / absence of each action information is stored in the divided information storage unit 123 is illustrated, but the divided information storage unit 123 stores the information in the action information storage unit 122. Information on which cell the action information given corresponds to may be stored.

(Cluster Information Storage Unit 124)
The cluster information storage unit 124 according to the embodiment stores various information regarding the combination of the divided information. FIG. 7 is a diagram showing an example of the cluster information storage unit according to the embodiment. The cluster information storage unit 124 shown in FIG. 7 stores information indicating how the division information of each cell shown in the division information storage unit 123 in FIG. 6 is combined and information identifying a model corresponding to each combination. Will be done. Specifically, information for identifying the model corresponding to each cell according to the user's "classification" and "context" is stored. For example, the information stored in the cluster information storage unit 124 is generated by a process of generating a model.

"Category" indicates a classification related to a user. FIG. 7 shows a case where users U1 to U4, that is, each user is a division, as in FIG. Further, "context" indicates a context. FIG. 7 shows a context based on time zone and position, as in FIG. Specifically, the "context" consists of four time zones: morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), midnight (22 to 4 o'clock), and home (22:00 to 4 o'clock). Eight contexts CT11-CT18 are shown based on two locations, home) or office (work location). In addition, "M1", "M2", etc. in each cell shown in FIG. 7 indicate information for identifying a model generated by union information combining division information of cells in which the same code is stored.

For example, in the example shown in FIG. 7, the model identified by "M1" (hereinafter, may be referred to as "model M1"; the same applies to the other "M2" to "M6") is the context CT11 of the user U1. From the union information that combines the mass MS corresponding to the mass MS corresponding to the user U2, the mass MS corresponding to the context CT15 of the user U2, the mass MS corresponding to the context CT17 of the user U2, and the mass MS corresponding to the context CT13 of the user U4. Indicates that it was generated. Further, for example, the model M2 combines the division information included in the mass MS corresponding to the context CT12 and CT17 of the user U1, the mass MS corresponding to the context CT14 of the user U2, and the mass MS corresponding to the context CT12 of the user U3. Indicates that it was generated from the union information.

The cluster information storage unit 124 is not limited to the above, and may store various information depending on the purpose.

(Learning information storage unit 125)
The learning information storage unit 125 according to the embodiment stores information related to learning. For example, the learning information storage unit 125 stores the model information generated by the generation process. FIG. 8 is a diagram showing an example of the learning information storage unit according to the embodiment. The learning information storage unit 125 shown in FIG. 8 has items such as "feature 1" to "feature 3" corresponding to each model M1 to M6 as model information.

For example, in the example shown in FIG. 8, the model information regarding the model M1 is such that the weight of the feature 1 is "0.5", the weight of the feature 2 is "-0.4", the weight of the feature 3 is "0.2", and the like. Indicates that. For example, when the features (features) of a model are represented by an m-dimensional vector, the number of features is m, and the weights of features 1 to m are stored.

The learning information storage unit 125 is not limited to the above, and may store various model information depending on the purpose.

(Content information storage unit 126)
The content information storage unit 126 according to the embodiment stores various information related to the content (article). FIG. 9 is a diagram showing an example of the content information storage unit according to the embodiment. The content information storage unit 126 shown in FIG. 9 has items such as "content ID", "content", and "category".

The "content ID" indicates identification information for identifying the content. "Content" indicates an article that is content acquired from a provider such as a so-called content provider. In FIG. 9, an example in which conceptual information such as "article A" is stored in "content" is shown, but in reality, character information, a combination of character information and an image, or a file showing the storage location thereof is shown. The path name etc. are stored. In addition, "category" indicates classification information based on the content.

For example, in the example shown in FIG. 9, the article A (content CN11) identified by the content ID "CN11" is classified into the category "economy". Further, in the example shown in FIG. 9, the article B (content CN12) identified by the content ID “CN12” is classified into the category “sports”.

The content information storage unit 126 is not limited to the above, and may store various information depending on the purpose. For example, the content information storage unit 126 may store information regarding the date and time when the content was acquired and the date and time when the content was created. The content information storage unit 126 may store information regarding the evaluation value of each provider. Further, the content information storage unit 126 may store identification information for identifying the content provider. Further, the content information storage unit 126 may store the score of the content calculated by each model.

(Control unit 130)
Returning to the description of FIG. 3, the control unit 130 is an example of various programs (an example of a distribution program) stored in a storage device inside the distribution device 100 by, for example, a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). Is realized by executing RAM as a work area. Further, the control unit 130 is realized by, for example, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 3, the control unit 130 includes an acquisition unit 131, an estimation unit 132, a generation unit 133, a determination unit 134, and a distribution unit 135, and functions and operations of information processing described below. To realize or execute. The internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 3, and may be another configuration as long as it is a configuration for performing information processing described later. Further, the connection relationship of each processing unit included in the control unit 130 is not limited to the connection relationship shown in FIG. 3, and may be another connection relationship.

(Acquisition unit 131)
The acquisition unit 131 acquires the user's behavior information. The acquisition unit 131 acquires the user's action information from the terminal device 10 used by the user. In addition, the acquisition unit 131 acquires user information about the user. In addition, the acquisition unit 131 acquires the division information which is the behavior information regarding the user's behavior for each division regarding the user and which is the behavior information divided based on the context. For example, the acquisition unit 131 acquires the division information from the division information storage unit 123. In addition, the acquisition unit 131 acquires the division information in which the behavior information regarding the behavior of the user or the user for each classification based on the user attribute is divided based on the context as the division.

In addition, the acquisition unit 131 acquires a content distribution request from the terminal device 10. In addition, the acquisition unit 131 acquires information that identifies the user (for example, login ID, terminal identification information, etc.), location information of the terminal device 10, and the like.

(Estimation unit 132)
The estimation unit 132 estimates various information from the various acquired information. For example, the estimation unit 132 estimates various information based on the time when the content is delivered and the user's position information acquired from the terminal device 10. For example, the estimation unit 132 estimates the user of the distribution request source from the information for identifying the user acquired from the terminal device 10. For example, the estimation unit 132 estimates the location of the user based on the user information of the user stored in the user information storage unit 121 and the position of the terminal device 10. In FIG. 2, in the estimation unit 132, as shown in the user information storage unit 121, the work location of the user U1 is the position LC12, and the position of the terminal device 10-1 is the position LC12-1. Estimated to be at work location). Further, in FIG. 2, since the delivery request is made at 9:00 am, the estimation unit 132 estimates the context to be the context CT12 corresponding to the office in the morning.

Further, in the case where one division is a user, the estimation unit 132 uses the user based on the degree of similarity between the union information including each division information of the user and the union information including each division information of the other division. The user attributes of the above are estimated, but the details will be described later.

(Generator 133)
The generation unit 133 may generate division information. Further, the generation unit 133 may store the generated division information in the division information storage unit 123. For example, the generation unit 133 may divide the collected information regarding the behavior of each user. For example, the generation unit 133 divides the collected behavior information of each user based on the context estimated by the estimation unit 132.

The generation unit 133 generates a model for predicting a predetermined target for each union information which is the division information combined based on the tendency of the behavior information included in each division information. For example, in FIG. 1, the mass MS at the position where the user U1 and the context CT11 intersect and the mass MS at the position where the user U4 and the context CT13 intersect are classified into the same cluster. The mass MS at the position where the user U1 and the context CT11 intersect and the mass MS at the position where the user U4 and the context CT13 intersect both include division information in which the action AT11 is “◯”. The mass MS at the position where the user U1 and the context CT11 intersect, and the mass MS at the position where the user U4 and the context CT13 intersect include division information showing the same tendency. Then, the generation unit 133 uses the union information including the division information of the mass MS at the position where the user U1 and the context CT11 intersect and the division information of the mass MS at the position where the user U4 and the context CT13 intersect, and the model M1 To generate.

In addition, the generation unit 133 generates a model that predicts a predetermined target regarding the user included in the division corresponding to the division information in the context corresponding to each of the division information included in the union information. The generation unit 133 generates a model for predicting a predetermined target for each union information including a combination that straddles the divisions. For example, the generation unit 133 uses the union information including the division information of the mass MS at the position where the user U2 and the context CT12 intersect and the division information of the mass MS at the position where the user U4 and the context CT14 intersect, and the model M6 To generate. That is, the generation unit 133 generates the model M6 by using the union information straddling the divisions between the user U2 and the user U4.

For example, the generation unit 133 determines for each union information in which the division information is combined so that the prediction accuracy of the predetermined target in the combination of the user classification and the context corresponding to each division information included in the union information is high. Generate a model that predicts the target. For example, the generation unit 133 generates a model for predicting a predetermined target for each union information in which the division information in which the behavior information included in each division information is similar is combined. For example, the generation unit 133 generates a model that predicts the user's interest in the content as a predetermined target.

For example, the generation unit 133 generates a model for predicting a predetermined target for each union information in which the division information is combined by a method in which the number of union information becomes a predetermined number. For example, the generation unit 133 generates a model for each union information in which division information is combined using the k-means method. In this case, the generation unit 133 generates the union information by combining the division information so as to be the number of clusters (the number of union information) determined (set) by the determination unit 134, and generates a model for each union information.

For example, the generation unit 133 generates a model for predicting a predetermined target for each union information in which the division information is combined by a method in which the number of union information can fluctuate. For example, the generation unit 133 generates a model for each union information in which division information is combined by using a logistic regression method using a Dirichlet process. For example, the generation unit 133 generates union information by sequentially combining the division information, and generates a model for each union information.

For example, the generation unit 133 generates union information by combining a plurality of division information, and generates a model for the union information. For example, the generation unit 133 randomly selects a plurality of division information to generate union information, and generates a model for the union information. Then, the generation unit 133 verifies the prediction accuracy of the generated model. The prediction accuracy here means the probability that the presence or absence of user behavior can be predicted. For example, the prediction accuracy referred to here means the probability that a predetermined score or higher will be calculated for the content that is "○" in the behavior information when the model is used. Then, when the prediction accuracy of the generated model is equal to or higher than a predetermined threshold value, the generation unit 133 may perform a process of maintaining the combination and further combining the division information. Further, the generation unit 133 may use a predetermined hyperparameter such as a penalty term. For example, the generation unit 133 may use a penalty term whose value decreases as the number of clusters decreases. For example, the generation unit 133 may determine whether to maintain or cancel the combination based on a numerical value calculated by subtracting a value obtained by multiplying the number of clusters by a predetermined coefficient from the prediction accuracy of the model. In this case, the generation unit 133 generates union information and a model classified into a predetermined number of clusters based on a trade-off between the decrease in the number of clusters and the prediction accuracy of the model.

For example, the generation unit 133 may verify the prediction accuracy of the generated model by cross-validating a part of the model for generation and the rest for verification. For example, the generation unit 133 may use 80% of the union information for model generation and the remaining 20% for verification. For example, when the verified prediction accuracy satisfies a predetermined criterion, the generation unit 133 may perform a process of maintaining the combination and further combining the division information. Further, when the verified prediction accuracy does not satisfy a predetermined criterion, the generation unit 133 may release the combination and perform a process of combining another division information. Further, each time the division information is combined, the generation unit 133 maintains or cancels the combination by averaging the information used for model generation and the information used for verification and performing cross-validation a plurality of times. May be determined.

For example, in the generation unit 133, when one division is a user and the division information of the user in one context does not satisfy a predetermined condition, the generation unit 133 divides another division similar to the user attribute of the user in one context. Generate a model by adding information to the user's split information in one context. For example, when the amount of division information of the user in one context is less than a predetermined amount, the generation unit 133 adds the division information of the other division in one context to the division information of the user in one context, thereby performing a model. To generate. The details of these points will be described later.

(Decision unit 134)
The determination unit 134 determines a model for determining the content to be distributed to the terminal device 10. In FIG. 2, the determination unit 134 uses a model M2 as a model used for determining the content to be distributed based on the information indicating the model M2 stored in the mass MS corresponding to the user U1 and the context CT12 in the cluster information storage unit 124. To decide.

For example, the determination unit 134 calculates the score of the content based on the determined model and the content. Further, for example, the determination unit 134 determines the ranking of each content based on the calculated score. In FIG. 2, the determination unit 134 determines the ranking of each content CN11 to CN18 or the like based on the score shown in the content list CL11. Further, the determination unit 134 determines the ranking of the article E, which is the content CN15 having the highest score, to be the first place. In addition, the determination unit 134 determines the ranking of the article G, which is the content CN17 having the highest score next to the article E, to be second. As a result, as shown in the ranking list LL11, the determination unit 134 has the ranking 1st place is the article E, the ranking 2nd place is the article G, the ranking 3rd place is the article A, and the ranking 4th place is the article F. Decide that there is.

(Distribution unit 135)
The distribution unit 135 distributes various information to the terminal device 10. For example, the distribution unit 135 distributes the content to the terminal device 10. In addition, the distribution unit 135 distributes the content determined by the determination unit 134. In FIG. 2, the distribution unit 135 distributes the content shown in the ranking list LL11 to the terminal device 10-1. Further, the distribution unit 135 distributes the contents displayed in the order of article E, article G, article A, article F, etc. to the terminal device 10-1.

[3. Generation process flow]
[3-1. Method without setting the number of clusters]
Next, the procedure of the generation process by the distribution system 1 according to the embodiment will be described with reference to FIG. FIG. 10 is a flowchart showing an example of the generation process according to the embodiment. Specifically, FIG. 10 is a flowchart showing an example of generation processing by a method that does not determine the number of clusters (number of union information).

As shown in FIG. 10, the acquisition unit 131 of the distribution device 100 acquires the behavior information of each user (step S101). For example, the acquisition unit 131 acquires user behavior information from the terminal device 10 used by each user.

Further, the generation unit 133 of the distribution device 100 divides the information according to the context (step S102). For example, the generation unit 133 generates divided information by dividing the action information according to the context when the action information is acquired. In addition, step S101 and step S102 may be performed a plurality of times.

After that, the generation unit 133 clusters the division information and generates a model by a method that does not determine the number of clusters (step S103). For example, the generation unit 133 generates a model for each union information in which division information is combined by using a logistic regression method using a Dirichlet process.

[3-2. Method to set the number of clusters]
Next, the procedure of the generation process by the distribution system 1 according to the embodiment will be described with reference to FIG. FIG. 11 is a flowchart showing an example of the generation process according to the embodiment. Specifically, FIG. 11 is a flowchart showing an example of a generation process by a method of presetting (determining) the number of clusters (number of union information).

As shown in FIG. 11, the acquisition unit 131 of the distribution device 100 acquires the behavior information of each user (step S201). For example, the acquisition unit 131 acquires user behavior information from the terminal device 10 used by each user.

Further, the generation unit 133 of the distribution device 100 divides the information according to the context (step S202). For example, the generation unit 133 generates divided information by dividing the action information according to the context when the action information is acquired. In addition, step S201 and step S202 may be performed a plurality of times.

Further, the determination unit 134 of the distribution device 100 sets (determines) the number of clusters (the number of union information) (step S203). For example, the determination unit 134 sets (determines) the number of clusters (the number of union information) to "6". The process of step S203 may be performed at any timing as long as it is before step S204.

After that, the generation unit 133 performs clustering of division information and generation of a model by a method of determining the number of clusters (step S204). For example, the generation unit 133 generates a model for each union information in which division information is combined using the k-means method.

[4. Delivery processing flow]
Next, the procedure of the distribution process by the distribution system 1 according to the embodiment will be described with reference to FIG. FIG. 12 is a flowchart showing an example of the distribution process according to the embodiment.

As shown in FIG. 12, the acquisition unit 131 of the distribution device 100 acquires the distribution request (step S301). For example, the acquisition unit 131 acquires a content distribution request from the terminal device 10 used by each user.

Then, the estimation unit 132 of the distribution device 100 estimates the user and the context (step S302). For example, the estimation unit 132 estimates the user of the distribution request source from the information for identifying the user acquired from the terminal device 10. Further, for example, the estimation unit 132 estimates the context based on the time when the distribution request is made and the position of the terminal device 10.

After that, the determination unit 134 of the distribution device 100 determines the model (step S303). For example, the determination unit 134 determines a model for determining the content to be distributed to the terminal device 10.

Then, the determination unit 134 determines the order of the contents (step S304). For example, the determination unit 134 calculates the score of the content based on the model and the content determined in step S303. Further, for example, the determination unit 134 determines the ranking of each content based on the calculated score.

After that, the distribution unit 135 of the distribution device 100 distributes the content based on the order (step S305). For example, the distribution unit 135 distributes the content based on the order determined in step S304.

[5. Classification and use of information]
In the above-mentioned example, the classification related to the user has been described as the user, but the classification related to the user is not limited to the user, and various classifications may be used depending on the purpose. This point will be described with reference to FIG. FIG. 13 is a diagram showing an example of a classification regarding a user according to an embodiment.

Here, the division information list UL11 in FIG. 13 shows a state in which the behavior information of each user is divided for each context. Similar to the divided information storage unit 123, the divided information list UL11 shows a case where the user who is the division is a row and the context is a column. The area where one user and one context intersect in the divided information list UL11 is defined as a mass. For example, in FIG. 13, the region where the user U4 and the context CT12 intersect is defined as the mass MS42-1.

The information included in each mass MS shown in FIG. 13 is divided information in which the behavior information regarding the user's behavior for each division regarding the user is divided based on the context. Note that in FIG. 13, the behavior AT11 and the like are not shown. For example, in FIG. 13, “◯” and “x” in each mass MS indicate the amount of division information included in each mass MS. Further, the meanings of "◯" and "x" in each mass MS in FIG. 13 are the same as those in FIG. For example, it indicates that the amount of divided information contained in the mass MS at the position where the user U1 and the context CT11 intersect is larger than the amount of the divided information contained in the mass MS at the position where the user U1 and the context CT12 intersect. .. Further, for example, the amount of divided information included in the mass MS at the position where the user U1 and the context CT12 intersect is larger than the amount of the divided information included in the mass MS at the position where the user U1 and the context CT13 intersect. Is shown. Further, for example, it is shown that the mass MS at the position where the user U2 and the context CT11 intersect does not include the division information.

In FIG. 13, division information for each classification based on a predetermined user attribute is generated based on the division information of each user (step S21). For example, the distribution device 100 generates classification information based on predetermined user attributes, such as classification G1, classification G2, and classification G3, as shown in the classification information list GL11 for each classification in FIG. Specifically, the distribution device 100 generates division information of classification G1 for men in their twenties. In this case, the distribution device 100 integrates the divided information of the user U1 and the user U4 who are men in their twenties to generate the divided information of the classification G1. In addition, the distribution device 100 generates division information of classification G2 for women in their twenties. In this case, the distribution device 100 integrates the division information of the user U2 and the like who are women in their twenties to generate the division information of the classification G2. In addition, the distribution device 100 generates division information of the classification G3 of the sport of interest. In this case, the distribution device 100 integrates the divided information of the user U1 or the like whose interest is sports as shown in the user information storage unit 121 to generate the divided information of the classification G3.

Then, the distribution device 100 may add the division information of each mass MS of the division information list GL11 to the mass MS of the user. This point will be described with reference to FIG. FIG. 14 is a diagram showing an example of using the divided information in other categories according to the embodiment.

The division information list GL11 shown in FIG. 14 is the same as the division information list GL11 shown in FIG. Here, in the division information list UL11 in FIG. 13, the division information is not included in the mass MS42-1 at the position where the user U4 and the context CT12 intersect. That is, the mass MS42-1 at the position where the user U4 and the context CT12 intersect is a mass MS lacking division information.

Therefore, the distribution device 100 uses the division information for each classification (step S22). For example, the distribution device 100 adds the division information included in the corresponding mass MS in the division information list GL11 to the mass MS42-2 as shown in the division information list UL12. Specifically, the distribution device 100 adds the division information included in the mass MS of the context CT12 in the classification G1 in the division information list GL11 to the mass MS42-2. As described above, the distribution device 100 adds the division information included in the corresponding mass MS in the division information list GL11 to the mass MS lacking the division information in the division information list UL11, so that the division information can be obtained in the division information list UL11. The shortage of mass MS can be reduced. Further, for example, the distribution device 100 adds the division information included in the mass MS of the context CT13 in the user U1 to the mass MS of the context CT13 in the classification G3 corresponding to the interest "sports" of the user U1 to the mass MS42-2. May be good. Further, the distribution device 100 may add the division information included in the corresponding mass MS in the user similar to the user attribute of the user to the mass MS in which the division information of a certain user is insufficient. For example, the distribution device 100 may add the division information included in the mass MS of the context CT12 in the user U1 having similar user attributes to the mass MS42 of the context CT12 in the user U4.

Further, the distribution device 100 may generate a model using the generated division information list GL11 and use the generated model. This point will be described with reference to FIG. FIG. 15 is a diagram showing an example of using the model in other categories according to the embodiment. The distribution device 100 may generate a model by combining the divided information list UL11 and the divided information list GL11. For example, the distribution device 100 may perform the generation process using the divided information list UL11 and the divided information list GL11. For example, the distribution device 100 may cluster the division information included in the division information list UL11 and the division information included in the division information list GL11 to generate a model for each cluster.

The cluster information list ML11 in FIG. 15 includes information related to clustering of divided information included in each mass MS as shown in the cluster information storage unit 124. The cluster information list ML11 in FIG. 15 indicates that the cluster information list ML11 includes the six clusters indicated by M1 to M6.

Further, the cluster information list ML12 in FIG. 15 includes information on clustering of the division information generated from the division information list GL11 for each classification. The cluster information list ML12 in FIG. 15 shows that it includes 6 clusters indicated by M11 to M13 and the like. That is, the distribution device 100 generates models such as models M11 to M13 as shown in the cluster information list ML12 from the divided information list GL11.

Here, in the cluster information list ML11 in FIG. 15, the mass MS at the position where the user U4 and the context CT12 intersect does not belong to any cluster. That is, the mass MS at the position where the user U4 and the context CT12 intersect is a mass MS to which a model is not assigned.

Therefore, the distribution device 100 uses the model for each classification (step S31). Specifically, the distribution device 100 assigns a model included in the corresponding mass MS in the cluster information list ML12 to the mass MS at the position where the user U4 and the context CT12 intersect. In this case, the distribution device 100 assigns the model M12 included in the mass MS at the position where the classification G1 and the context CT12 intersect to the mass MS at the position where the user U4 and the context CT12 intersect. In this way, the distribution device 100 assigns the model included in the corresponding mass MS in the cluster information list ML12 to the mass MS to which the model is not assigned in the cluster information list ML11, so that the model is assigned in the cluster information list ML11. It is possible to reduce the mass MS that has not been used.

[6. Context change and integration]
In the above example, the context is four time zones of morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), midnight (22 to 4 o'clock), and home (home). ) Or eight contexts CT11-CT18 based on two locations in the office (work location). However, the context is not limited to the above eight, and may be varied depending on the purpose. For example, the distribution device 100 may change the context according to the amount of information in each category and the like. This point will be described with reference to FIG. FIG. 16 is a diagram showing an example of a variation in the context according to the embodiment.

The division information list UL11 shown in FIG. 16 is the same as the division information list UL11 shown in FIG. That is, the divided information list UL11 has four time zones of morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), and midnight (22 to 4 o'clock), and home (home). Or, the case where it is divided into eight contexts CT11 to CT18 based on two positions of the office (work location) is shown. The divided information list UL11 includes users U1 to U4 and the like.

Here, it is assumed that the user U111 is a user who collects a large amount of behavior information as compared with other users U1 to U4 and the like. Therefore, the distribution device 100 generates the divided information for the user U111 based on the contexts that are more subdivided than the eight contexts shown in the divided information list UL11.

In FIG. 16, the distribution device 100 has 48, as shown in the divided information list UL111, based on the 24 hours from 0:00 to 23:00 and the two positions of the house (home) or the office (work location). The division information is generated based on the contexts CT111 to CT122 and the like. That is, the distribution device 100 generates divided information for the user U111, which has a large amount of collected behavior information, based on 48 contexts that are subdivided from eight contexts. In addition, "0 o'clock" in the division information list UL111 shown in FIG. 16 means 0 o'clock, and corresponds to, for example, from 00:00 to 0:59:59. The same applies to "1 o'clock" to "23:00" in the division information list UL111 shown in FIG.

In this way, the distribution device 100 may change the context for generating the divided information based on the amount of behavioral information collected for each user. The distribution device 100 may generate a model by combining the divided information list UL11 and the divided information list UL111. Further, for a user who includes sufficient division information in each context by himself / herself, such as the user U111 shown in the division information list UL111, the distribution device 100 may generate a model for each of the users. For example, the distribution device 100 may generate a model corresponding to each context of the user U111 by using only the behavior information of the user U111 itself. Further, as shown in FIG. 13, the distribution device 100 may add the division information of the user U111 to the mass MS of the user who lacks the division information. That is, the distribution device 100 may generate a model corresponding to each context of the user U111 and allow other users to use the information of the user U111.

Further, when the amount of information of the mass MS in a predetermined division is small, the distribution device 100 may integrate the contexts in that division. This point will be described with reference to FIG. FIG. 17 is a diagram showing an example of context integration according to an embodiment.

The division information list UL11 shown in FIG. 17 is the same as the division information list UL11 shown in FIG. That is, the divided information list UL11 has four time zones of morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), and midnight (22 to 4 o'clock), and home (home). Or, the case where it is divided into eight contexts CT11 to CT18 based on two positions of the office (work location) is shown. The divided information list UL11 includes users U1 to U4 and the like.

Here, it is assumed that the user U112 is a user who has very little action information collected at midnight as compared with other users U1 to U4 and the like. Therefore, the distribution device 100 generates division information for the user U112 based on seven contexts in which the contexts related to midnight are integrated among the eight contexts shown in the division information list UL11. Specifically, the distribution device 100 distributes the divided information based on seven of the eight contexts shown in the divided information list UL11, including the context CT121 corresponding to the midnight in which the context CT17 and the context CT18 related to the midnight are integrated. Generate. That is, the distribution device 100 generates divided information for the user U112, which has very little behavior information collected at midnight, based on the context in which the distinction between the home (home) and the office (work location) at midnight is lost. do.

In this way, the distribution device 100 may integrate the context for generating the divided information based on the amount of behavioral information collected for each user. The distribution device 100 may generate a model by combining the divided information list UL11 and the divided information list UL112.

[7. User attribute estimation]
Further, the distribution device 100 may estimate the user attributes of the user based on the arrangement of clusters in each division and the like. This point will be described with reference to FIG. FIG. 18 is a diagram showing an example of estimation of user attributes based on the cluster and the model according to the embodiment.

The division information list UL21 shown in FIG. 18 is obtained by adding the division information of the user U21 to the division information list UL11 shown in FIG. That is, the divided information list UL21 has four time zones of morning (4 to 10 o'clock), noon (10 to 16 o'clock), night (16 to 22 o'clock), and midnight (22 to 4 o'clock), and home (home). Or, the case where it is divided into eight contexts CT11 to CT18 based on two positions of the office (work location) is shown. Further, the divided information list UL21 includes users U1 to U4, U21 and the like. Further, it is assumed that the user U21 is a user whose gender and age, which are user attributes, are unknown.

First, the distribution device 100 generates a model using the division information of each mass MS in the division information list UL21 (step S41). For example, the distribution device 100 clusters the division information of each mass MS in the division information list UL21, and generates a model for each cluster as shown in the cluster information list ML21. For example, the distribution device 100 generates a model by the same processing as in step S14 in FIG.

After that, the distribution device 100 estimates the gender and age of the user U21 whose gender and age are unknown by arranging the clusters (step S42). For example, as shown in the user attribute information UA21, the distribution device 100 estimates the gender and age of the user whose cluster arrangement is similar to that of the user U21 in the divided information list UL21 as the gender and age of the user U21. ..

Here, in FIG. 18, the distribution device 100 classifies the mass MS corresponding to the context CT11 of the user U4 and the mass MS corresponding to the context CT12 of the user U21 into the same cluster. That is, the same model M3 is applied to the user U4 and the user U21 in the morning time zone. Further, the distribution device 100 classifies the mass MS corresponding to the context CT13 of the user U4 and the mass MS corresponding to the context CT13 of the user U21 into the same cluster. That is, when the user U4 and the user U21 are at home (home) in the daytime, the same model M1 is applied. Further, the distribution device 100 classifies the mass MS corresponding to the context CT16 of the user U4 and the mass MS corresponding to the context CT17 of the user U21 into the same cluster. That is, the same model M6 is applied to the user U4 and the user U21 during the time zone from night to midnight.

As described above, since the arrangement of the clusters, that is, the arrangement of the applied models is similar between the user U4 and the user U21, the distribution device 100 makes the gender and age of the user U21 the same as the gender and age of the user U4. Presumed to be. Specifically, the distribution device 100 estimates that the gender of the user U21 is the same "male" as the user U4, and the age is the same "20s" as the user U4. In this way, the distribution device 100 can estimate the user attributes of a user whose user attributes are unknown, based on the similarity of the cluster arrangement, that is, the arrangement of the model to be applied.

[8. effect〕
As described above, the distribution device 100 according to the embodiment has an acquisition unit 131 and a generation unit 133. The acquisition unit 131 acquires the division information which is the behavior information regarding the user's behavior for each division regarding the user and which is the behavior information divided based on the context. In addition, the generation unit 133 generates a model for predicting a predetermined target for each union information which is the division information combined based on the tendency of the behavior information included in each division information.

As a result, the distribution device 100 according to the embodiment can generate an appropriate model according to each context of each division by clustering the information divided based on the division and the context. Therefore, the distribution device 100 can generate a model that appropriately predicts a predetermined target based on the context.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 uses a model for predicting a predetermined target regarding a user included in the division corresponding to the division information in the context corresponding to each division information included in the union information. Generate.

As a result, the distribution device 100 according to the embodiment can generate an appropriate model according to each context of each division by clustering the information divided based on the division and the context. Therefore, the distribution device 100 can generate a model that appropriately predicts a predetermined target based on the context.

Further, in the distribution device 100 according to the embodiment, the acquisition unit 131 acquires the division information in which the behavior information regarding the behavior of the user or the user for each classification based on the user attribute is divided based on the context as a division.

As a result, the distribution device 100 according to the embodiment can adjust the amount of division information included in each cell by appropriately changing the classification. Therefore, it is possible to generate a model that appropriately predicts a predetermined target based on the context.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 generates a model for predicting a predetermined target for each union information including a combination straddling the divisions.

As a result, even if the distribution device 100 according to the embodiment divides the behavior information of each division based on the context, the amount of information used for generating each model is insufficient by performing clustering across the divisions. Can be suppressed. Therefore, the distribution device 100 can generate a model that appropriately predicts a predetermined target based on the context.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 combines the division information so that the prediction accuracy of the predetermined target in the combination of the user classification and the context corresponding to each division information included in the union information is high. For each union information provided, a model that predicts a predetermined target is generated.

As a result, the distribution device 100 according to the embodiment combines the division information so that the prediction accuracy of the predetermined target in the combination of the user classification and the context corresponding to each division information included in the union information becomes high. It is possible to generate a model that appropriately predicts a predetermined target based on the above.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 generates a model for predicting a predetermined target for each union information in which the division information in which the behavior information included in each division information is similar is combined.

As a result, the distribution device 100 according to the embodiment can generate a model that appropriately predicts a predetermined target based on the context by combining the division information in which the behavior information included in each division information is similar.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 generates a model for predicting a predetermined target for each union information in which the division information is combined by a method in which the number of union information becomes a predetermined number.

As a result, the distribution device 100 according to the embodiment can generate a model that appropriately predicts a predetermined target based on the context by using a method in which the number of union information becomes a predetermined number.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 generates a model for predicting a predetermined target for each union information in which the division information is combined by a method in which the number of union information can fluctuate.

As a result, the distribution device 100 according to the embodiment can generate a model that appropriately predicts a predetermined target based on the context by using a method in which the number of union information can fluctuate.

Further, in the distribution device 100 according to the embodiment, the generation unit 133 generates a model for predicting the user's interest in the content as a predetermined target.

Thereby, the distribution device 100 according to the embodiment can generate a model that appropriately predicts the user's interest in the content based on the context.

Further, in the distribution device 100 according to the embodiment, when the generation unit 133 indicates that one division is a user and the division information of the user in one context does not satisfy a predetermined condition, the user of the user in one context. A model is generated by adding the division information of other divisions similar to the attributes to the division information of the user in one context.

As a result, the distribution device 100 according to the embodiment can generate a model that appropriately predicts a predetermined target based on the context, regardless of the user.

Further, in the distribution device 100 according to the embodiment, when the amount of division information of the user in one context is less than a predetermined amount, the generation unit 133 transfers the division information of another division in one context to the user in one context. A model is generated by adding to the division information of.

As a result, the distribution device 100 according to the embodiment can generate a model that appropriately predicts a predetermined target based on the context even for a user who lacks division information.

Further, the distribution device 100 according to the embodiment has an estimation unit 132. When one division is a user, the estimation unit 132 uses the user of the user based on the degree of similarity between the union information including each division information of the user and the union information including each division information of the other division. Estimate the attributes.

As a result, the distribution device 100 according to the embodiment can appropriately estimate the user attribute even when the user attribute of the user is unknown.

[9. Hardware configuration]
The 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. 19 is a hardware configuration diagram showing an example of a computer that realizes the function of the distribution device. The computer 1000 has a CPU 1100, a RAM 1200, a ROM 1300, an HDD 1400, a communication interface (I / F) 1500, an input / output interface (I / F) 1600, and a media interface (I / F) 1700.

The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400, and controls each part. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started, a program depending on the hardware of the computer 1000, and the like.

The HDD 1400 stores a program executed by the CPU 1100, data used by such a program, and the like. The communication interface 1500 receives data from another device via the network N and sends it to the CPU 1100, and transmits the data generated by the CPU 1100 to the other device via the network N.

The CPU 1100 controls an output device such as a display or a printer, and an input device such as a keyboard or a mouse via the input / output interface 1600. The CPU 1100 acquires data from the input device via the input / output interface 1600. Further, the CPU 1100 outputs the generated data to the output device via the input / output interface 1600.

The media interface 1700 reads a program or data stored in the recording medium 1800 and provides the program or data to the CPU 1100 via the RAM 1200. The CPU 1100 loads the program from the recording medium 1800 onto the RAM 1200 via the media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase change rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. And so on.

For example, when the computer 1000 functions as the distribution device 100 according to the embodiment, the CPU 1100 of the computer 1000 realizes the function of the control unit 130 by executing the program loaded on the RAM 1200. The CPU 1100 of the computer 1000 reads and executes these programs from the recording medium 1800, but as another example, these programs may be acquired from another device via the network N.

Although some of the embodiments and modifications of the present application have been described in detail with reference to the drawings, these are examples and vary based on the knowledge of those skilled in the art, including the embodiments described in the disclosure line of the invention. It is possible to carry out the present invention in another form in which the above is modified or improved.

[10. others〕
Further, among the processes described in the above-described embodiments and modifications, all or part of the processes described as being automatically performed can be manually performed, or are described as being manually performed. It is also possible to automatically perform all or part of the processed processing by a known method. In addition, the processing procedure, specific name, and information including various data and parameters shown in the above document and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

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 the device is functionally or physically dispersed / physically distributed in any unit according to various loads and usage conditions. Can be integrated and configured.

Further, the above-described embodiments and modifications can be appropriately combined as long as the processing contents do not contradict each other.

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

1 Distribution system 100 Distribution device (generation device)
121 User information storage unit 122 Behavior information storage unit 123 Divided information storage unit 124 Cluster information storage unit 125 Learning information storage unit 126 Content information storage unit 130 Control unit 131 Acquisition unit 132 Estimating unit 133 Generation unit 134 Decision unit 135 Distribution unit 10 Terminals Equipment N network

Claims (13)

An acquisition unit that acquires behavior information related to user behavior for each category related to the user from the terminal device used by the user, and an acquisition unit.
Divided information that is divided action information based on the context is generated, and union information that is combined based on the tendency of the presence or absence of action indicated by the action information included in each generated divided information is generated and generated. a model for each union information, a generation unit for generating a model Ru used to obtain the score by applying the respective content,
A generator characterized by being equipped with. The generator
The generator according to claim 1, wherein a model for predicting a predetermined target for a user included in the division corresponding to the division information in a context corresponding to each of the division information included in the union information is generated. .. The generator
The generation according to claim 1 or 2, wherein as the division, the behavior information regarding the behavior of the user or the user for each classification based on the user attribute generates the division information divided based on the context. Device. The generator
The generator according to any one of claims 1 to 3, wherein a model for predicting a predetermined target is generated for each union information including a combination straddling the divisions. The generator
For the model generated using the union information, the model is generated based on whether the prediction accuracy of a predetermined target in the combination of user classification and context corresponding to each division information included in the union information satisfies a predetermined condition. The generator according to any one of claims 1 to 4, wherein the process is repeated. The generator
The invention according to any one of claims 1 to 5, wherein a model for predicting a predetermined target is generated for each union information in which the behavioral information contained in each division information is similar to the division information. Generator. The generator
The invention according to any one of claims 1 to 6, wherein a model for predicting a predetermined target is generated for each union information in which the division information is combined by a method in which the number of union information becomes a predetermined number. Generator. The generator
The generation according to any one of claims 1 to 6, wherein a model for predicting a predetermined target is generated for each union information in which the division information is combined by a method in which the number of union information can fluctuate. Device. The generator
When one division is a user and the division information of the user in one context does not satisfy a predetermined condition, the division information of another division similar to the user attribute of the user in the one context is input to the one. The generator according to any one of claims 1 to 8, wherein a model is generated by adding to the division information of the user in the context of. The generator
When the amount of the divided information of the user in the one context is less than a predetermined amount, the model is added by adding the divided information of the other division in the one context to the divided information of the user in the one context. The generator according to claim 9, wherein the generator is generated. When one division is a user, the user attribute of the user is estimated based on the similarity between the union information including each division information of the user and the union information including each division information of the other division. Estimator,
The generator according to any one of claims 1 to 10, further comprising. It ’s a computer-executed generation method.
The acquisition process of acquiring the behavior information related to the user's behavior for each category related to the user from the terminal device used by the user, and the acquisition process.
Divided information that is divided action information based on the context is generated, and union information that is combined based on the tendency of the presence or absence of action indicated by the action information included in each generated divided information is generated and generated. a model for each union information, a generation step of generating a model Ru used to obtain the score by applying the respective content,
A generation method characterized by including. The acquisition procedure for acquiring the behavior information related to the user's behavior for each category related to the user from the terminal device used by the user, and the acquisition procedure.
Divided information that is divided action information based on the context is generated, and union information that is combined based on the tendency of the presence or absence of action indicated by the action information included in each generated divided information is generated and generated. a model for each union information, a generating step of generating a model Ru used to obtain the score by applying the respective content,
A generator characterized by having a computer execute.

Images