Estimation apparatus

Abstract

To provide an estimation apparatus capable of accurately estimating a response from a customer to measures that have not been taken by a company concerned.SOLUTION: An estimation apparatus 1 includes: a target data acquisition unit 14 which acquires target data including user data and measure data related to a target measure; and a response estimation unit which acquires characteristics of the target measure and characteristics of a user from the target data, using a learning model 73 trained by machine learning so as to receive learning data including user data and measure data related to a measure and output characteristics of the measure and characteristics of the user, and estimates a response rate of the user with respect to the target measure on the basis of the acquired characteristics of the target measure and the characteristics of the user. The learning data includes at least user data and measure data related to measures of other companies, the measures have not been taken by a company concerned.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to an estimation device. [Background technology]

[0002] In services offered in the digital market, various measures are implemented to encourage users to use the services. In order to consider future measures, it is important to estimate whether or not these measures are actually producing the expected effect. As a related technique, Patent Document 1 discloses a direct marketing support device equipped with a scoring model for calculating a score of a customer's response to a direct marketing measure from the customer's attribute information. The direct marketing support device inputs attribute information of a customer who has not implemented a measure into the scoring model, thereby calculating a score of the response expected when the measure is implemented for the customer. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-135751 A Summary of the Invention [Problem to be solved by the invention]

[0004] For example, if a learning model for predicting customer responses to measures is generated using only information on measures that have been implemented by the company in the past, the prediction accuracy of the learning model may be reduced for customer responses to measures that the company has not implemented in the past.

[0005] The present invention provides an estimation device capable of accurately estimating a customer's reaction to a measure that the company has not implemented in the past. [Means for solving the problem]

[0006] The estimation device according to the present disclosure includes a target data acquisition unit that acquires target data including policy data related to a target policy and user data, and a response estimation unit that acquires characteristics of the target policy and characteristics of the user from the target data using a learning model trained to output characteristics of the policy and characteristics of the user when learning data including the policy data related to a policy and the user data is input, and estimates a user response rate to the target policy based on the acquired characteristics of the target policy and characteristics of the user. The learning data includes at least policy data and user data related to policies of other companies that are not implemented by the company. Effect of the Invention

[0007] The estimation device according to the present disclosure can accurately estimate customer responses to measures that a company has not implemented in the past. [Brief description of the drawings]

[0008] [Figure 1] FIG. 2 is a block diagram showing a configuration of an estimation device. [Diagram 2] 3 is a flowchart showing an overall flow of processing performed by the estimation device. [Diagram 3] 13 is a flowchart showing a learning data acquisition process performed by a learning data acquisition unit. [Figure 4] A figure showing an example of information input to a learning model and information output from the learning model. [Diagram 5] 13 is a flowchart showing a process of estimating a policy response rate performed by a response estimation unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In each drawing, the same or corresponding elements are denoted by the same reference numerals. For clarity of explanation, duplicate explanations will be omitted as necessary.

[0010] In related technology, the effect of a new measure is estimated based on measures previously implemented within the company's own service, but when considering a measure with no past examples, it is difficult to estimate the effect of the measure with high accuracy because there are no similar measures. Therefore, the estimation device according to the present disclosure improves the estimation accuracy of the customer response rate to the measure (hereinafter, sometimes referred to as the "measure response rate") by adding data on the measures of other services to the feature. Specifically, the estimation device according to the present disclosure generates a machine learning model using language information of the measures sent from SNS (Social Networking Service) and HP (Home Page) of other companies' services. In addition, the estimation device according to the present disclosure calculates the characteristics of the measure and the user's response rate to the measure by inputting information on the considered measure into the generated model.

[0011] (Configuration of Estimation Device 1) The configuration of the estimation device 1 according to the embodiment will be described with reference to FIG. 1. FIG. 1 is a block diagram showing the configuration of the estimation device 1. The estimation device 1 includes a learning data acquisition unit 11, a preprocessing unit 12, a learning unit 13, a target data acquisition unit 14, a reaction estimation unit 15, and a storage unit 17. The estimation device 1 includes a processor and a memory, which are not shown in the figure. The storage unit 17 can function as a storage device in which a program in which the processing according to the embodiment is implemented is stored. The processor loads the program from the storage unit 17 into the memory and executes the program. In this way, the processor realizes the functions of the learning data acquisition unit 11 and the like. The learning data acquisition unit 11 and the like may each be realized by dedicated hardware. In addition, some or all of the components of the estimation device 1 may be realized by a general-purpose or dedicated circuit, processor, or the like, or a combination of these. A CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or the like can be used as the processor.

[0012] The learning data acquisition unit 11 accesses various DBs (DataBases), SNS, websites, etc., and acquires learning data used for learning in the learning unit 13. The learning data acquisition unit 11 acquires, as learning data, policy data related to policies and user data related to the policies. The learning data includes at least policy data related to policies of other companies that have not been implemented by the company, and user data. The policy data may include, for example, language information and image information associated with policies implemented in services provided by the company or other companies. The policy data may include, for example, users participating in the policies, an overview of the policies, or content posted on SNS, websites, etc. related to the policies.

[0013] The user data may include, for example, user attribute information, language information transmitted on an SNS, or user behavior information within the company's service. The user attribute information is, for example, information such as the user's gender, age, or place of residence. The user attribute information may be represented by an attribute value indicating each attribute. The language information transmitted on an SNS is language information such as text transmitted by the user on an SNS or the like. The user behavior information is information indicating the user's behavior within the company's or another company's service. The user behavior information is, for example, engagement information for a campaign, image information, or posted content for a campaign. The engagement information is, for example, information on a reaction indicating a high evaluation made by a user to a post on an SNS regarding a campaign. The reaction indicating a high evaluation may be, for example, the spread of a post, the number of "likes," or a reply to a post. The image information is an image attached to a post, a profile image of a user or a service, or the like. The user behavior information may include a campaign in which the user participated, a purchased product, or a page viewed on an SNS or the like.

[0014] The learning data acquisition unit 11 acquires learning data from the company's DB or a server of another company. For example, for the company's own service, the learning data acquisition unit 11 acquires policy data and user data related to the service by referring to the company's policy DB 71. For the other company's service, the learning data acquisition unit 11 acquires data from the official SNS or HP of the other company's service via a network such as the Internet (not shown). The data may include, for example, engagement information, image information, or language information. Note that, although information on SNS and HP is used here as an example, any information may be used without being limited to these. The learning data acquisition unit 11 may acquire data from a server of another company that provides a predetermined service. The learning data acquisition unit 11 can acquire learning data from any service in which language information of a policy and a user who responds to the policy can be observed.

[0015] The preprocessing unit 12 performs preprocessing on the learning data acquired by the learning data acquisition unit 11 before applying the learning model to the data. For example, the preprocessing unit 12 performs preprocessing on language information included in the data. Specifically, the preprocessing unit 12 extracts character information included in the data and shapes the extracted character information. Shaping of the character information includes standardizing characters to full-width or half-width, or standardizing English or Japanese notation. The preprocessing unit 12 also performs word division on the extracted character information. The preprocessing unit 12 may also perform a process of removing stop words that are not to be processed. Furthermore, if characters exist in the acquired image information, the preprocessing unit 12 performs OCR (Optical Character Recognition) processing to extract character information. The preprocessing unit 12 performs preprocessing on the extracted character information.

[0016] The learning unit 13 performs learning of a machine learning model using the data preprocessed by the preprocessing unit 12. The learning unit 13 generates a learning model 73 using a known machine learning method, and stores it in the storage unit 17. The learning unit 13 may re-learn the learning model 73.

[0017] The target data acquisition unit 14 acquires target data including policy data and user data related to a target policy that is a target for estimating a policy response rate. The target data includes, for example, language information related to a new policy. The target data may include information related to a target user.

[0018] In order to consider a new measure, the reaction estimation unit 15 calculates a user reaction rate to the measure using the learning model 73 obtained by the learning unit 13. Specifically, the reaction estimation unit 15 inputs the target data acquired by the target data acquisition unit 14 to the learning model 73. The reaction estimation unit 15 acquires the characteristics of the target measure and the characteristics of the user output from the learning model 73. The reaction estimation unit 15 stores the acquired characteristics of the target measure and the characteristics of the user as policy characteristic information 74 and user characteristic information 75 in the storage unit 17. The reaction estimation unit 15 estimates a user reaction rate to the target measure based on the policy characteristic information 74 and the user characteristic information 75.

[0019] The storage unit 17 stores a company policy DB 71, a company user DB 72, a learning model 73, policy characteristic information 74, and user characteristic information 75. The company policy DB 71 and the company user DB 72 are accumulated data related to the company's services. The company policy DB 71 and the company user DB 72 include information on policies and users in the company's services, respectively.

[0020] The learning model 73 is a learning model trained by the learning unit 13. The learning model 73 is trained to output characteristics of a policy and characteristics of a user when learning data including policy data related to a policy and user data is input. The policy characteristic information 74 is an accumulation of topic distribution for each policy output from the learning model 73. The user characteristic DB 75 is an accumulation of topic distribution for each user output from the learning model 73. The storage unit 17 may store information other than the above. In addition, a part or all of the above information may be stored outside the estimation device 1.

[0021] The above-described configuration of the estimation device 1 is merely an example and may be modified as appropriate. For example, some or all of the components of the estimation device 1 may be realized by a plurality of information processing devices, circuits, etc. The plurality of information processing devices, circuits, etc. may be centrally arranged or distributed. For example, the information processing devices, circuits, etc. may be realized as a client-server system, a cloud computing system, or the like, each of which is connected via a communication network.

[0022] (Processing of Estimation Device 1) The processing performed by the estimation device 1 will be described with reference to Figs. 2 to 5. Fig. 2 is a flowchart showing the overall flow of the processing performed by the estimation device 1. Fig. 3 is a flowchart showing the learning data acquisition processing performed by the learning data acquisition unit 11. Fig. 4 is a diagram showing an example of information input to the learning model 73 and information output from the learning model 73. Fig. 5 is a flowchart showing the process of estimating a policy response rate performed by the response estimation unit 15. Below, the overall flow will be described with reference to Fig. 2, and each step will be described in detail with reference to Figs. 3 to 5.

[0023] First, the learning data acquisition unit 11 performs a learning data acquisition process (S1). The learning data acquisition unit 11 acquires data related to the company's policies by referring to the company's policy DB 71 and the company's user DB 72. The learning data acquisition unit 11 also acquires data related to other companies' policies from SNS, websites, etc.

[0024] Here, the process of step S1 will be described in detail with reference to Fig. 3. The learning data acquiring unit 11 repeats the processes of steps S11 to S13 for all measures of the company's own service and other company's services. The learning data acquiring unit 11 may limit the processing target by identifying other company's services that compete with the company's own service from among the other company's services.

[0025] First, the learning data acquisition unit 11 determines whether the policy to be estimated is the company's policy (S11). The learning data acquisition unit 11 may determine whether it is similar to the company's policy by a predetermined amount or more. If it is the company's policy (YES in S11), the learning data acquisition unit 11 extracts data from the company's policy DB 71 in the storage unit 17 (S12). If it is not the company's policy (NO in S11), the learning data acquisition unit 11 acquires data from the official SNS of the other company's service or the like (S13). When the above-mentioned processing is completed for all policies of the company's service and the other company's services, the learning data acquisition unit 11 terminates the learning data acquisition processing.

[0026] In the above-mentioned learning data acquisition process, the following examples are possible with regard to the handling of language information from SNS, websites, etc. in the implementation measures of other companies. (Example 1) The learning data acquisition unit 11 acquires language information directly from SNS, websites, etc. (Example 2) The learning data acquisition unit 11 acquires information on users who have responded to a campaign from their reactions (e.g., quotes, replies, "likes," etc.). The learning data acquisition unit 11 considers the language information of the campaign to have been generated by the user. For example, assume that an official account of a campaign posts on an SNS that a cashback will be given to users by using a specific restaurant. Assume that users a and b have responded to the post by quoting the post or replying to the post. The learning data acquisition unit 11 acquires information on users a and b who have responded to the cashback campaign. The learning data acquisition unit 11 considers the language information of the cashback campaign to have been generated by users a and b. In this way, when multiple users have responded to a campaign, the learning data acquisition unit 11 acquires the campaign in association with a group of users who responded.

[0027] Returning to FIG. 2, the explanation will be continued. The preprocessing unit 12 performs preprocessing on the learning data acquired by the learning data acquisition unit 11 (S2). Next, the learning unit 13 uses the preprocessed data to learn the policy characteristics and user characteristics (S3). When learning data including policy data related to a policy and user data is input, the learning unit 13 trains the learning model 73 to output the policy characteristics and user characteristics. For example, the learning unit 13 learns the policy and the characteristics (topic distribution) for each user using a machine learning model such as an author-topic model. The learning unit 13 stores the generated learning model 73 in the storage unit 17.

[0028] The learning unit 13 may generate the learning model 73 using different input data. In the following, an example of generating the learning model 73 using two different input data will be described with reference to Fig. 4. The following examples 1 and 2 correspond to the above-mentioned examples 1 and 2.

[0029] (Example 1) In Example 1, the learning unit 13 learns the characteristics (topics) of a measure directly from the linguistic information of the measure. The learning unit 13 receives the linguistic information of the measure and the linguistic information sent by the user as input, and trains the learning model 73 to output a topic distribution for each measure and a topic distribution for each user. The tables shown in the upper and middle parts of FIG. 4 show the topic distribution D1 for each measure and the topic distribution D2 for each user, which are output in Example 1. For example, in topic distribution D1, if the topic distribution for each measure is Φ, then the topic distributions for measures A, B, and C are Φ A , Φ B , Φ C In addition, in topic distribution D2, if the topic distribution for each user is θ, the topic distributions for users a, b, and c are θ a , θ b , θ c It is expressed as:

[0030] (Example 2) In Example 2, the learning unit 13 learns the characteristics of users who responded to the announcement of a policy on SNS or the like. When a user responded to a policy, the learning unit 13 considers that the linguistic information of the policy was generated by that user, and compiles the linguistic information of the policy for each user. The learning unit 13 trains the learning model 73 so that the linguistic information of the policy is input and the topic distribution for each user is output. The table shown in the lower part of Figure 4 shows the topic distribution D3 for each user that is output in Example 2. As with topic distribution D2, if the topic distribution for each user in topic distribution D3 is θ, the topic distributions for users a, b, and c are each θ a , θ b , θ c It is expressed as:

[0031] Returning to FIG. 2, the explanation will be continued. The target data acquisition unit 14 acquires target data to be estimated (S4). For example, the target data acquisition unit 14 acquires target data from an operator who operates the estimation device 1 via an input unit (not shown). The input unit is, for example, an input device such as a mouse or a keyboard. Here, it is assumed that the target data acquisition unit 14 acquires target data including information about a new measure X, which is a company service. Next, the response estimation unit 15 performs estimation processing to calculate a measure response rate for measure X using the learning model 73 and the characteristics (topic distribution) of each user (S5).

[0032] Now, the processing of step S5 will be described with reference to FIG. 5. First, the response estimation unit 15 acquires linguistic information of the measure X from the target data (S21). The response estimation unit 15 inputs the linguistic information of the measure X to the preprocessing unit 12 and performs preprocessing (S22). The response estimation unit 15 estimates the characteristics of the measure X from the target data after preprocessing using the learning model 73 (S23). The response estimation unit 15 estimates the characteristics of the measure X as a vector probability value Φ XThe response estimation unit 15 reads the user characteristic θ by using the user characteristic information 75 in order to calculate the response rate of each user to the campaign X (S24). The user characteristic, like the characteristic of campaign X, is a probability value for n topics (n is an integer equal to or greater than 1) and is expressed by using a vector.

[0033] Next, the response estimation unit 15 estimates the response rate of the measure X (S25). The response estimation unit 15 calculates a scalar value indicating a probability value by multiplying a column vector indicating the characteristics of the measure X by a row vector indicating the characteristics of the user. In this way, the response estimation unit 15 calculates the user's response rate to the measure X. For example, the response rate P Xa is the column vector Φ of policy X X and user a's row vector θ a Using these, it is expressed by the following formula (1): The response estimation unit 15 can estimate the policy response rate for other users in the same manner. P Xa =Φ X θ a (1)

[0034] Returning to Fig. 2, the description will be continued. The response estimation unit 15 presents the result calculated by the above formula (1) as an estimation result of the response rate of the user a to the campaign X (S6). For example, the response estimation unit 15 displays the estimation result on a display unit such as a display (not shown).

[0035] As described above, in the estimation device 1 according to the present embodiment, in addition to the feature quantities that can be extracted from the company's own service, information on measures implemented in other companies' services is acquired from official SNS, websites, etc., and added to the input feature quantities. This allows the estimation device 1 to estimate customer responses to measures that the company has not implemented in the past with higher accuracy than when only information on measures implemented by the company is used.

[0036] Furthermore, for new users of the company's service, the estimation device 1 estimates the characteristics of the user using the user's SNS information. By calculating the policy response rate for each user characteristic in advance based on the data of the company's service, the estimation device 1 can estimate the characteristics of the policy to which the new user will respond. This makes it possible to estimate the policy response rate with high accuracy when adding a new user. Furthermore, as described above, when estimating the policy response rate, the estimation device 1 learns a model for estimating the user characteristics, and therefore can estimate the user characteristics.

[0037] The above-mentioned program includes a set of instructions (or software code) for making the computer perform the functions described in the embodiment when the program is loaded into the computer. The program may be stored in a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, the non-transitory computer-readable medium or the tangible storage medium includes random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disc (DVD), Blu-ray (registered trademark) disk or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device. The program may also be transmitted on various types of transitory computer-readable medium or communication medium. By way of example and not limitation, the transitory computer-readable medium or communication medium includes electrical, optical, acoustic, or other forms of propagating signals.

[0038] The present disclosure is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit and scope of the present disclosure. [Explanation of symbols]

[0039] 1: Estimation device, 11: Learning data acquisition unit, 12: Preprocessing unit, 13: Learning unit, 14: Target data acquisition unit, 15: Response estimation unit, 17: Memory unit, 71: Company policy DB, 72: Company user DB, 73: Learning model, 74: Policy characteristic information, 75: User characteristic information, D1 to D3: Topic distribution

Claims

[Claim 1] a target data acquisition unit that acquires target data including policy data related to a target policy and user data; a response estimation unit that, when learning data including policy data related to a policy and user data is input, acquires characteristics of the target policy and characteristics of the user from the target data using a learning model that has been trained to output characteristics of the policy and characteristics of the user, and estimates a response rate of the user to the target policy based on the acquired characteristics of the target policy and characteristics of the user; The learning data includes at least policy data on policies of other companies that have not been implemented by the company and user data. Estimation device.

Images