Information processing method, information processing device, and computer program

Abstract

To provide an information processing method, information processing device and computer program, which enable estimation of user funnels.SOLUTION: An information processing method disclosed herein comprises acquiring history information regarding given notification means used for making notifications to a user, and estimating a funnel the user belongs to on the basis of the acquired history information.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing method, an information processing device, and a computer program. [Background technology]

[0002] Customer contact points are opportunities for companies to communicate with customers and understand their needs and information. For example, there are a variety of ways to interact with customers, such as customer service at company stores or customer sites, product displays, posting magazines and flyers, advertising on the street, websites, social media, internet advertising, telephone, and email.

[0003] Customer contact is important for a company's brand power and sales, and strengthening customer contact is becoming increasingly important in order to approach not only existing customers but also new customers.

[0004] Patent Document 1 discloses an information distribution method that can appropriately grasp the needs of a user by determining the appropriate delivery time for the user based on information such as whether the emails delivered to the user have been opened and the history of click information on links in the emails, and by distributing the emails based on the determination results. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-23039 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the method of Patent Document 1 determines the delivery period based on the opening time or click time of existing users, and does not take into consideration the stage (funnel) at which a user is, for example, from becoming aware of a product to becoming interested in it and purchasing it. For this reason, it can be difficult to estimate a customer's funnel.

[0007] The present invention has been made in view of the above circumstances, and has an object to provide an information processing method, an information processing device, and a computer program that are capable of estimating a user's funnel. [Means for solving the problem]

[0008] The present application includes multiple means for solving the above problem, and one example is an information processing method that acquires historical information regarding a specified notification means that notified a user, and estimates the funnel to which the user belongs based on the acquired historical information. [Effects of the Invention]

[0009] According to the present invention, it is possible to estimate a user's funnel. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating an example of a configuration of an information processing system according to an embodiment of the present invention. [Figure 2] FIG. 10 is a diagram showing an example of user information recorded in a user information DB. [Figure 3] FIG. 10 is a diagram showing an example of survey information recorded in a survey information DB. [Figure 4] FIG. 10 is a diagram showing an example of questions and options for the questions included in the survey information. [Figure 5] 10 shows an example of item information recorded in the item information DB. [Figure 6] FIG. 10 is a diagram showing an example of email history information recorded in an email history information DB. [Figure 7] FIG. 10 is a diagram illustrating an example of a method for estimating a user funnel using a first estimation model. [Figure 8] FIG. 10 is a diagram showing an example of an email delivery history input to a first estimation model. [Figure 9] FIG. 10 is a diagram illustrating an example of a hidden Markov model implemented by the first estimation model. [Figure 10]FIG. 10 is a diagram illustrating an example of selection of candidate emails. [Figure 11] FIG. 10 is a diagram illustrating an example of a method for estimating an email open rate. [Figure 12] FIG. 10 is a diagram showing an example of an estimation result of an open rate output by a second estimation model. [Figure 13] FIG. 10 is a diagram illustrating an example of an estimated result of an open rate of a candidate email. [Figure 14] FIG. 10 is a diagram illustrating an example of a method for estimating a funnel based on event history information. [Figure 15] FIG. 10 is a diagram illustrating an example of a method for estimating emails to be delivered based on event history information. [Figure 16] FIG. 10 is a diagram illustrating a first example of a method for estimating a delivery email by an information processing device. [Figure 17] FIG. 10 is a diagram illustrating a second example of a method for estimating a delivery email by an information processing device. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing an example of the configuration of an information processing system according to this embodiment. The information processing system includes an information processing device 50. A plurality of terminal devices 10 and a data server 30 are connected to the information processing device 50 via a communication network 1.

[0012] The terminal device 10 may be configured as a smartphone or tablet terminal carried by a user (also referred to as a customer), or may be configured as a personal computer used by the user. The user can use the terminal device 10 to send a user ID and a password to the information processing device 50.

[0013] The data server 30 may be configured as a computer such as a server, and includes a user information DB (database) 31, a survey information DB 32, an item information DB 33, an email history information DB 34, and an event history information DB 35. The data server 30 is a server that allows an external device, such as an information processing device 50, to access the information recorded in the user information DB 31, the survey information DB 32, the item information DB 33, the email history information DB 34, and the event history information DB 35. The data server 30 has functions such as a function to write information to the user information DB 31, the survey information DB 32, the item information DB 33, the email history information DB 34, and the event history information DB 35, a function to read information from each of the aforementioned DBs, and a function to update each of the aforementioned DBs. Details of the user information DB 31, the survey information DB 32, the item information DB 33, the email history information DB 34, and the event history information DB 35 will be described later.

[0014] 1 shows a configuration including one data server 30, but is not limited to the example of FIG. 1. For example, the user information DB 31, the survey information DB 32, the item information DB 33, the email history information DB 34, and the event history information DB 35 may be distributed among multiple data servers. Also, the information processing device 50 may be configured to include each of the above-mentioned DBs.

[0015] The information processing device 50 can be configured as a computer such as a server, and includes a control unit 51 that controls the entire device, a communication unit 52, a memory 53, a display unit 54, an operation unit 55, a user information distribution estimation unit 56, an item information distribution estimation unit 57, and a storage unit 58.

[0016] The control unit 51 may be configured by incorporating a required number of central processing units (CPUs), micro-processing units (MPUs), graphics processing units (GPUs), etc. The control unit 51 may also be configured by combining digital signal processors (DSPs), field-programmable gate arrays (FPGAs), etc.

[0017] The communication unit 52 includes a communication module and has a function of communicating with the terminal device 10 and the data server 30 via the communication network 1. The communication unit 52 can receive (acquire) user information, survey information, item information, email history information, and event history information from the user information DB 31, survey information DB 32, item information DB 33, email history information DB 34, and event history information DB 35 of the data server 30, respectively.

[0018] The display unit 54 can be configured with a liquid crystal panel, an organic EL (Electro Luminescence) display, etc. Instead of the display unit 54, an external display device may be provided.

[0019] The operation unit 55 is configured with a touch panel or the like, and can be used to operate icons displayed on the display unit 54, move and operate a cursor, input characters, etc. The operation unit 55 may also be a mouse or a keyboard.

[0020] The storage unit 58 can be configured with a semiconductor memory, a hard disk, or the like, and stores a computer program 60 (program product), a first estimation model 61, a second estimation model 62, and required information. The required information may include, for example, user registration information (e.g., user ID, password, etc.) of existing users and new users.

[0021] The computer program 60 can be stored in the storage unit 58 by reading the computer program 60 recorded on a recording medium (for example, an optically readable disk storage medium such as a CD-ROM) M using a recording medium reading unit (not shown). The computer program 60 may also be read from a recording medium such as a storage device (semiconductor memory such as a solid state drive (SSD)) connected according to a standard for connecting to a computer (for example, Universal Serial Bus (USB) or other standard) and stored in the storage unit 58. The computer program 60 may also be downloaded from an external device via the communication unit 52 and stored in the storage unit 60. The first estimation model 61 and the second estimation model 62 will be described in detail later.

[0022] The memory 53 can be configured with semiconductor memory such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), flash memory, etc. A computer program 60 can be loaded into the memory 53, and the control unit 51 can execute the computer program 60. The control unit 51 can execute processing defined by the computer program 60. In other words, processing by the control unit 51 is also processing by the computer program 60.

[0023] The information processing device 50 may be configured from multiple devices (servers, etc.) so that the control unit 51, communication unit 52, memory 53, display unit 54, operation unit 55, user information distribution estimation unit 56, item information distribution estimation unit 57, and storage unit 58 are distributed, and each device is equipped with the aforementioned distributed functional units.

[0024] In this embodiment, the predetermined notification means is mainly assumed to be email, but the predetermined notification means is not limited to email. For example, any notification means that can transmit (deliver) required information to a user, such as a telephone call, SMS (Short Message Service), message, or push notification, may be used. In the following, this specification will describe the case where email is used as the predetermined notification means.

[0025] Next, each piece of information recorded in the user information DB 31, the survey information DB 32, the item information DB 33, the mail history information DB 34, and the event history information DB 35 in this embodiment will be described.

[0026] 2 is a diagram showing an example of user information recorded in the user information DB 31. The user information includes information such as the user's name, address, age, sex, occupation, marital status (whether married or single), telephone number, email address, annual income, and bank balance, which are associated with the user ID. The user information is generated based on user information provided by the user when the user registers a user ID from the terminal device 10 or when the user registers a user ID at a company store or the like. The user information includes user information of not only existing users but also new users.

[0027] FIG. 3 is a diagram showing an example of survey information recorded in the survey information DB 32. Survey information (also referred to as a survey) includes questions (question content) associated with a plurality of question IDs, each associated with a user ID, and the user's answers to the questions. As shown in FIG. 3, for example, assume that a survey is sent to a user with a user ID of ID=1001. The survey contains n questions with question IDs of ID=1, ID=2, ..., ID=n. For example, the n questions with question IDs of ID=1, ID=2, ..., ID=n are represented as q11, q12, ..., q1n. The user's answers to questions q11, q12, ..., q1n are represented as a11, a12, ..., a1n, respectively. The same applies to other users. Survey information can be sent from the information processing device 50 to the terminal device 10 at a desired timing. For example, for a new user, the required timing may include when the user registers or after a predetermined time has passed after registration (e.g., one day, three days, one week, etc.). For an existing user, the required timing may be when an email is sent (also called distribution), or when a predetermined time has passed since the email was sent (e.g., one day, two days, three days, etc.).

[0028] FIG. 4 shows an example of questions and options included in the survey information. As shown in FIG. 4, the questions include, for example, "Please choose the option that best suits your values," "Please tell us your hobbies," and "Are you familiar with XX Bank's product?" The options for the questions are, for each of the above questions, "1. Those who value trends, 2. Those who value individuality, 3. Those who spend time for others, 4. Those who spend time for themselves, 5. Those who value high risk and high return, 6. Those who value low risk and low return," "1. Music, 2. Driving, 3. Fishing, 4. Golf, 5. Skiing, 6. Mahjong and Pachinko," and "1. Already purchased, 2. Well-known, 3. Not very familiar, 4. Not interested." A user who receives (acquired) the survey information can answer the questions by selecting an option for each question. The survey information with selected answers is referred to as the survey result. Note that the survey information is merely an example and is not limited to the example shown in FIG. 4.

[0029] As mentioned above, the questionnaire information can include questions about the user's subjective information. By obtaining the questionnaire results, it is possible to obtain highly subjective (personal) information about the user, which can be used to estimate the needs, behavior, and psychological state of not only existing users but also new users.

[0030] By analyzing a user's funnel, it is possible to estimate the user's needs, behavior, and psychological state. A funnel represents the behavioral steps (behavioral phases) a user takes after becoming aware of a product. There are three types of funnels: a top funnel (purchase funnel), a bottom funnel (influence funnel), and a double funnel. A top funnel represents the changes in the user's psychological process and behavioral steps from when the user becomes aware of a product to when they purchase it. The top funnel has four funnels: awareness, interest, comparison and consideration, and purchase, and can represent the flow from product awareness to purchase. The bottom funnel represents the user's behavior after purchasing a product. The bottom funnel has three funnels: continuation, referral, and transmission, for example, and can represent the user's behavior after purchase. A double funnel is a combination of a top funnel and a bottom funnel. In this embodiment, a top funnel is used, but a bottom funnel or a double funnel may also be used.

[0031] 5 shows an example of item information recorded in the item information DB 33. The item information is information that indicates the attributes of an e-mail sent to a user. The item information includes information such as the product, purpose, subject, and body of the e-mail.

[0032] For example, a product is a product or commodity that a company handles, and if the company is a bank, its products include card loans, investment trusts, etc. Although not shown in Figure 5, products can include home loans, reverse mortgages (real estate utilization loans), apartment loans, NISA (Nippon Individual Savings Account: a small investment tax exemption system), time deposits, foreign currency deposits, government bonds for individuals, asset management, employee savings, individual-type defined contribution pension plans, inheritance, life insurance, etc.

[0033] The purpose of the email sent to the user can be categorized into, for example, awareness emails to make the user aware of the product, interest emails to get the user interested in the product they have become aware of, consideration emails to get the user to compare and consider products they have become interested in, and application emails to ask the user to apply for or purchase the product.

[0034] The subject is the subject of the email to be sent to the user. The subject can be, for example, a summary of the content you want to convey to the user in the email. The body is the body of the email and can include images, URLs, etc. You can also attach a questionnaire to be answered by the user to the email. Note that the item information is an example and is not limited to the example in FIG. 5.

[0035] FIG. 6 is a diagram showing an example of email history information recorded in the email history information DB 34. The email history information, which is history information related to a specific communication method, includes information such as an email ID, item information, delivery date and time, and whether or not the email has been opened by the user, all associated with the user ID. The email ID is an identifier for identifying the email. The item information is information indicating the attributes of the email, and includes information such as the product, purpose, and subject, as shown in FIG. 5. The item information in the email history information may also include the body of the email. The delivery date and time is the date and time the email was sent to the user.

[0036] In the example of Figure 6, two emails with email IDs "M001" and "M002" have been delivered to a user with a user ID of "1003," and it can be seen that the user has opened the email with email ID "M001," but has not opened the email with email ID "M002." The same is true for other user IDs. Note that the email history information is an example and is not limited to the example of Figure 6.

[0037] The event history information will be described later.

[0038] Next, a method for estimating a user's funnel, that is, a funnel to which the user belongs, will be described. In this embodiment, as described above, the funnel is divided into four stages: awareness, interest / concern, comparison / consideration, and purchase.

[0039] The control unit 51 acquires email history information (history information related to the predetermined notification means used) of emails sent from the data server 30 to the user via the communication unit 52. Note that the user can be identified, for example, by acquiring a user ID in advance from the terminal device 10. The control unit 51 can estimate the funnel to which the user belongs based on the acquired history information.

[0040] The first estimation model 61 can estimate the funnel of the user. The control unit 51 can use the first specified model 61 to estimate the funnel to which the user belongs.

[0041] FIG. 7 illustrates an example of a method for estimating a user funnel using a first estimation model 61. The first estimation model 61 is an estimation model that uses a method for data analysis using Bayesian inference. It uses a statistical method that probabilistically models the data generation process and calculates the information contained in the generation model as a probability distribution based on the obtained data. The first estimation model 61 treats input data as random variables and obtains output data as a probability distribution (or a vector, which is a group of multiple numbers sampled from the probability distribution). Bayesian inference using the first estimation model 61 is performed in three steps: (1) setting a prior probability, (2) correcting the prior probability using likelihood, and (3) calculating the posterior probability. The prior probability is a probability set based on past experience, etc. The likelihood represents the degree of accuracy of a given assumption when a result is obtained under that assumption, and indicates the accuracy of the prior probability. The posterior probability is calculated by multiplying the likelihood by the prior probability. In other words, the first estimation model 61 performs a process in which the set prior probability is corrected by the likelihood based on the actually obtained observation value.

[0042] As shown in FIG. 7A , when a user's survey information and the delivery history of emails delivered to the user are input to the first estimation model 61, the first estimation model 61 estimates the funnel to which the user belongs. In this case, the prior probability corresponds to the funnel to which the user belongs, and a pre-initialized funnel may be used, or an initial funnel may be set based on the input survey information. When the email delivery history is input as an observation value, the first estimation model 61 estimates the funnel to which the user belongs as a posterior probability. When the delivery history of the next email is input, the first estimation model 61 further estimates the funnel to which the user belongs as a posterior probability, using the estimated funnel as the prior probability. In this way, by inputting the email delivery history as time-series data to the first estimation model 61, the first estimation model 61 can output time-series data of the funnel to which the user belongs.

[0043] 7B, when the delivery history of emails delivered to a user is input to the first estimation model 61, the first estimation model 61 estimates the funnel to which the user belongs. In this case, the prior probability may be a funnel initially set in advance or a funnel estimated by the first estimation model 61.

[0044] As described above, the information processing device 50 of this embodiment can estimate a user funnel.

[0045] Furthermore, the control unit 51 can acquire the survey results (survey information) answered by the user and estimate the funnel to which the user belongs based on the acquired survey results. The survey results can provide information that is highly subjective (personality-oriented) about the user, making it possible to estimate the user funnel not only for existing customers but also for new customers.

[0046] Fig. 8 is a diagram showing an example of the delivery history of emails input to the first estimation model 61. As shown in Fig. 8, it is assumed that emails with email IDs M001, M002, ..., M00n were delivered to a certain user at times t1, t2, ..., tn. Whether or not each email was opened by the user is represented by X1, X2, ..., Xn. X1=1 means opened (opened), X2=0 means not opened (unopened), ..., Xn=1 means opened.

[0047] FIG. 9 is a diagram showing an example of a hidden Markov model implemented by the first estimation model 61. A hidden Markov model is a model that represents probabilistic state transitions, and outputs for each transitioned state are generated probabilistically. In FIG. 9, π is the probability distribution of the initial state, which is the probability distribution of the user's initial funnel. The probability distribution of the initial state π is, for example, the probability distribution of the initial funnel state estimated based on input survey information. Xi is an observed value, which represents a set of observed signal sequences. Fi is a hidden variable, which corresponds to the user's funnel. Fi represents a set of states. A is a transition probability matrix, which is the transition probability from one state (e.g., F1) to another state (e.g., F2). E is an output probability matrix, which is the probability of outputting an observed value (e.g., X1) in a certain state.

[0048] As shown in FIG. 9, states (user funnels as latent variables) at times t1, t2, ..., tn are denoted as F1, F2, ..., Fn, and observed values ​​at each state are denoted as X1, X2, ..., Xn. Here, observed value Xi indicates whether or not the email illustrated in FIG. 8 has been opened. That is, if observed value X1 is observed in state F1 at time t1, the first estimation model 61 can estimate state F2 at time t2. Similarly, if observed value Xn is observed in state Fn at time tn, the first estimation model 61 can estimate state F(n+1) at time t(n+1). The first estimation model 61 can also estimate state F(n+1) at time t(n+1) based on states F1, F2, ..., F(n-1) at times t1, t2, ..., t(n-1).

[0049] During learning, parameters such as the initial state probability distribution π, the transition probability matrix A, and the output probability matrix E are learned using statistical processing or neural networks, and the learned parameters can be used to estimate the user's funnel.

[0050] The control unit 51 acquires an email delivery history including time-series information indicating whether or not a user is aware of a predetermined notification means. The user's awareness is information indicating whether or not the user is aware of the notification means that notified the user, and in the case where the notification means is email, it is information indicating whether or not the email was opened. When the time-series information is input as an observation value, the control unit 51 can estimate the funnel to which the user belongs using a first estimation model 61 that outputs the funnel to which the user belongs in time series as a latent variable.

[0051] As described above, according to this embodiment, it is possible to estimate the transition of a user's funnel by Bayesian inference taking time series into consideration, and it is possible to estimate needs taking the user's funnel into consideration. Furthermore, since it is possible to estimate the user's funnel in time series, it is possible to estimate changes in the user's funnel and to estimate changes in the user's interests and concerns in a timely manner.

[0052] Once the funnel to which a user belongs can be estimated, the email (or candidate emails) to be delivered to the user can be selected according to the estimated funnel.

[0053] Figure 10 shows an example of candidate email selection. As shown in Figure 10, the funnels are "awareness," "interest / concern," "comparison / consideration," and "purchase / application." If the user's funnel is estimated to be "awareness," emails whose purpose is "awareness emails" can be selected (identified) as candidate emails. Note that candidate emails whose purpose is "awareness emails" include multiple emails with various products, titles, and text bodies. If the user's funnel is estimated to be "interest / concern," emails whose purpose is "interest emails" can be selected as candidate emails. If the user's funnel is estimated to be "comparison / consideration," emails whose purpose is "consideration emails" can be selected as candidate emails. If the user's funnel is estimated to be "purchase / application," emails whose purpose is "application emails" can be selected as candidate emails. Note that the example in Figure 10 is merely an example, and candidate emails are not limited to the example in Figure 10.

[0054] As described above, the control unit 51 can identify candidates for emails to be sent next to the user, or emails (notification content to be sent by the next notification means) according to the estimated funnel. When identifying emails (or candidate emails), the purpose of the emails corresponding to the funnel can be estimated using the method illustrated in FIG. 10. Furthermore, by acquiring a history of item information, which is an attribute of emails sent in the past, the control unit 51 can identify the product, title, body, and so on of the next email (or candidate email) to be sent based on the past item information.

[0055] Next, the estimation of the open rate of emails delivered to users will be described.

[0056] 11 is a diagram showing an example of a method for estimating the email open rate. When the control unit 51 acquires the user ID of the target user, it calculates the similarity between an ID vector obtained by vectorizing the acquired user ID and a user information vector obtained by vectorizing user information recorded in the user information DB 31.

[0057] An ID vector obtained by vectorizing a user ID is a vector that corresponds one-to-one with an ID. For example, for user ID=1001 shown in FIG. 2, the ID vector can be expressed as (0.1, 0.3, . . . , −0.3).

[0058] The user information vector obtained by vectorizing user information is a vector that corresponds one-to-one to each piece of attribute information in the user information. For example, there is a vector corresponding to each of the attribute information of the user information illustrated in Figure 2, namely, "Name," "Address," "Age," "Sex," "Occupation," "Marital status," "Telephone number," "Email address," "Annual income," and "Bank balance." For example, a vector (also referred to as a "Name vector") where "Name" is "XXX" can be expressed as (0.2, 0.1, ..., -0.1), and a vector where "Name" is "OXX" can be expressed as (0.4, -0.8, ..., -0.5). The same applies to other attribute information.

[0059] Parameter learning is performed so as to maximize the similarity between the ID vector and the user information vector. The similarity can be calculated, for example, by cosine similarity, but is not limited to this. By calculating the user similarity, the attributes of the target user can be determined. The control unit 51 outputs the user information vector corresponding to the user ID, the previously acquired questionnaire information (vectorized questionnaire information), and the target user's funnel estimated by the first estimation model 61 (funnel estimation result) to the user information distribution estimation unit 56.

[0060] The user information vector corresponding to a user ID is a vectorized version of the user information obtained for the user ID, and in the case of user ID=1001 shown in Fig. 2, it is composed of the following vectors: ID vector, name vector, address vector, age vector, gender vector, occupation vector, marital status vector, telephone number vector, email address vector, annual income vector, and bank balance vector. Each vector is output separately to the user information distribution estimation unit 56.

[0061] Furthermore, when the control unit 51 acquires the email ID of a candidate email selected (identified) based on the funnel estimation results, it calculates the similarity between an ID vector, which is a vectorization of the acquired email ID, and an item information vector, which is a vectorization of the item information recorded in the item information DB 33. Parameter learning is performed so as to maximize the similarity. Calculating the user similarity makes it possible to determine the attributes of the candidate email. The control unit 51 outputs the item information vector corresponding to the email ID (item ID) to the item information distribution estimation unit 57. Like the user information vector, the item information vector is composed of an ID vector and vectors such as a product vector, a purpose vector, a subject vector, and a body vector, which are vectorizations of the attribute information of the item information exemplified in FIG. 5.

[0062] The user information distribution estimation unit 56 calculates the posterior probability of the mean and standard deviation of the distribution (e.g., normal distribution) of an n-dimensional random variable of a user vector composed of vectorized user information, survey information, and funnels associated with the acquired email ID using Bayesian inference, samples a user vector from the obtained mean and standard deviation, and outputs the sampled user vector to the second estimation model 62.

[0063] Item information distribution estimation unit 57 uses Bayesian inference to calculate the posterior probability of the mean and standard deviation of the distribution (e.g., normal distribution) of an n-dimensional random variable of an item vector obtained by vectorizing the item information associated with the acquired email ID, samples an item vector from the obtained mean and standard deviation, and outputs the sampled item vector to second estimation model 62. Note that the number of dimensions n of the vector input to user information distribution estimation unit 56 is the same as the number of dimensions n of the vector input to item information distribution estimation unit 57, but may be different.

[0064] The second estimation model 62 calculates the open rate of candidate emails based on the input user vector and item vector. The open rate can be calculated, for example, by calculating the dot product of the user vector and the item vector. The second estimation model 62 quantifies the similarity between the user vector and the item vector using the dot product and converts the quantified value into a probability value between 0 and 1 using a sigmoid function. The user similarity calculation, the item similarity calculation, the user information distribution estimation unit 56, and the item information distribution estimation unit 57 may be integrated into the second estimation model 62. When integrated into the second estimation model 62, the second estimation model 62 can be trained based on the three calculation results of the user similarity calculation, the item similarity calculation, and the open rate prediction. That is, the second estimation model 62 learns parameters so that the vector based on the user ID and the vector based on the user information approach each other. The second estimation model 62 also learns parameters so that the vector based on the item ID and the vector based on the item information approach each other. Furthermore, the second estimation model 62 performs parameter learning so that the predicted opening rate approaches the correct value.

[0065] FIG. 12 is a diagram showing an example of the estimation result of the opening rate output by the second estimation model 62. In FIG. 12, the horizontal axis represents the probability of the opening rate, and the vertical axis represents the probability density. As shown in FIG. 12, this embodiment differs from the so-called point estimation model in that it uses a distribution estimation model that estimates the probability density according to the probability of the opening rate, and therefore it is possible to take into account the uncertainty of the output result (i.e., the magnitude of the probability). This makes it possible to mitigate the over-learning and cold start problems with a small amount of data, and improve estimation accuracy even for unknown data.

[0066] As described above, the control unit 51 uses the funnel to which the target user belongs (the estimated funnel result), the survey results answered by the user, and the user information of the user as random variables, and estimates a user information distribution representing the probability density of the random variables using the user information distribution estimation unit 56. The control unit 51 also uses item information indicating the notification content of candidate emails (predetermined notification means) to be sent to the user as random variables, and estimates an item information distribution representing the probability density of the random variables using the item information distribution estimation unit 57. Based on the estimated user information distribution and item information distribution, the control unit 51 can estimate the probability distribution of the user's email open rate (recognition rate) for the next candidate email (candidate notification content to be sent by the notification means) to the user. Note that there may be multiple candidate emails, or just one.

[0067] FIG. 13 is a diagram showing an example of the estimated results of the open rate of candidate emails. The user funnel estimated by the first estimation model 61 is assumed to be "Interests and Concerns." The email IDs for which the purpose of candidate emails is "Awareness Email" are assumed to be M001, M002, .... The product with email ID M001 is a card loan, and the predicted open rate is 50%. Furthermore, the product with email ID M002 is a mortgage, and the predicted open rate is 40%.

[0068] The email IDs for which the purpose of the candidate email is "Interested Email" are M0051, M0052, .... The product for email ID M0051 is asset management, and the predicted open rate is 80%. Also, the product for email ID M0052 is investment trusts, and the predicted open rate is 90%.

[0069] The email IDs for candidate emails with the purpose of "consideration email" are M0101, M0102, .... The product for email ID M0101 is NISA, and the predicted open rate is 70%. Also, the product for email ID M0102 is reverse mortgage, and the predicted open rate is 60%.

[0070] The email IDs for candidate emails with the purpose of "application email" are M0201, M0202, .... The product for email ID M0201 is a card loan, and the predicted open rate is 45%. Also, the product for email ID M0202 is an investment trust, and the predicted open rate is 55%.

[0071] The control unit 51 can identify the next email to be delivered to the target user according to the probability distribution of the estimated email open rate (recognition rate). Specifically, the control unit 51 can identify candidate emails whose predicted open rate is equal to or greater than a predetermined threshold as the next email to be delivered. Furthermore, the control unit 51 can determine in advance the number of emails to be delivered (e.g., 3), and identify candidate emails with the highest predicted open rate (e.g., top 3) as the next email to be delivered.

[0072] Next, a method for estimating a user's funnel and a method for specifying an email to be delivered to a user based on event history information will be described.

[0073] FIG. 14 illustrates an example of a method for estimating a funnel based on event history information. Event history information recorded in the event history information DB 35 can include, for example, a user's website browsing history. As shown in FIG. 14, the website browsing history of a user with user ID 2001 includes information such as the URLs viewed (URL descriptions) and the date and time of the browsing. In the example of FIG. 14, it can be seen that the user sequentially browsed site information such as "About the NISA System," "NISA Products," "Opening a NISA Account," and "Frequently Asked Questions." In such a case, the control unit 51 acquires the user's event history information and can estimate the funnel to which the user belongs based on the website history included in the acquired event history information. In the example of FIG. 14, it is estimated that the user has an interest in NISA, so the user's funnel can be estimated as "Interests."

[0074] Furthermore, although not shown, the control unit 51 can acquire event history information of the user and identify an email to be delivered to the user based on the website history included in the acquired event history information. In the example of Fig. 14, the attributes of the email to be delivered can be set as "interesting email" for the purpose and "NISA" for the product.

[0075] FIG. 15 illustrates an example of a method for estimating an email to be delivered based on event history information. Event history information recorded in the event history information DB 35 can include, for example, log information of apps used by users. Apps used by users are apps provided by companies to registered users, including apps that can be used when receiving services from companies (e.g., banks). As shown in FIG. 15, the app log information for a user with user ID 3001 includes information indicating the screen transitions used in the app (screen transition history) and information such as the transition dates and times. In the example of FIG. 15, it can be seen that the user transitioned through the securities application initial screen, input screen 1, input screen 2, and registration screen, and then exited the app without pressing the registration button. In such a case, the control unit 51 acquires the user's event history information and can estimate an email to be delivered to the user based on the app log information included in the acquired event history information. In the example of FIG. 15, the user stopped operating on the registration screen, which is the final stage of the securities application, so an email can be delivered to the user regarding how to register for the securities application. This allows the user to learn how to register.

[0076] Furthermore, although not shown, the control unit 51 can acquire event history information of the user and estimate the funnel to which the user belongs based on the application log information included in the acquired event history information. In the example of Fig. 15, it can be estimated that the user's funnel is "purchase / application."

[0077] As described above, the control unit 51 acquires event history information including at least one of the user's website browsing history and the user's app log history, and can perform at least one of estimating the funnel to which the user belongs and specifying the notification content to be sent to the user in the next email (notification means) based on the event history information.

[0078] The event information is not limited to website history or app log information, but may also include information that reveals the user's behavior history in response to external information such as news provided by news sites in Japan and overseas, or chronological information about a certain event.

[0079] FIG. 16 is a diagram showing a first example of a method for estimating email delivery by the information processing device 50. The control unit 51 performs initial setup of the funnel for the target user (S11). If there is user survey information, the initial setup of the user funnel can be performed based on the survey information. If there is no user survey information, the model operator can set in advance a uniform distribution (a distribution in which all events in a certain interval occur with equal probability) or the like as the initial setup of the funnel. The control unit 51 acquires history information of emails previously delivered to the user from the email history information DB 34 (S12), and determines whether or not the emails included in the acquired history information have been opened (S13).

[0080] The control unit 51 inputs time-series data indicating whether or not the package has been opened as an observation value into the estimation model (first estimation model 61) (S14), and determines whether or not there is user survey information (survey results) (S15). If there is user survey information (YES in S15), the control unit 51 inputs the user survey information into the estimation model (first estimation model 61) as an observation value (S16), and acquires latent variables output by the estimation model (S17).

[0081] If there is no user survey information (NO in S15), the control unit 51 performs the process of step S17. The control unit 51 estimates the user's funnel based on the acquired latent variables (S18), identifies emails to be delivered to the user based on the estimated funnel (S19), and ends the process. Note that the emails to be delivered can be identified by identifying item information (product, purpose, title, body, etc.), which are attributes of the emails.

[0082] 17 is a diagram showing a second example of a method for estimating a delivery email by information processing device 50. Control unit 51 acquires the user ID of the target user (S31), vectorizes the user ID, and vectorizes the user information recorded in user information DB 31 (S32). By vectorizing the user information, a user information vector (vectorized user information) corresponding to the user ID can be obtained.

[0083] The control unit 51 acquires the email IDs of candidate emails to be delivered to the user (S33). The candidate emails can be estimated based on the estimated user's funnel, as illustrated in FIG. 10. The control unit 51 vectorizes the item information associated with the email ID, and vectorizes the item information recorded in the item information DB 33 (S34). By vectorizing the item information, it is possible to obtain an item information vector (vectorized item information) corresponding to the email ID (item ID).

[0084] The control unit 51 estimates the user information distribution using Bayesian inference based on the vectorized user information, the user questionnaire information, and the user funnel (S35).The control unit 51 estimates the item information distribution using Bayesian inference based on the vectorized item information (S36).

[0085] The control unit 51 samples a user vector from the user information distribution (S37), and samples an item vector from the item information distribution (S38). The control unit 51 estimates the open rate of the candidate email based on the sampled user vector and item vector (S39). If there are multiple candidate emails, the open rate is estimated for each candidate email.

[0086] The control unit 51 identifies an email to be delivered to the user based on the estimated open rate (S40), and ends the process. Note that the email to be delivered can be identified by identifying item information (product, purpose, title, body, etc.), which is the attribute of the email.

[0087] As described above, according to this embodiment, it is possible to estimate a user's funnel or changes in the funnel by Bayesian inference taking time series into account, and to estimate needs taking the estimated funnel into account. Furthermore, it is possible to select and deliver appropriate emails to users based on the estimated funnel.

[0088] Furthermore, according to this embodiment, by using questionnaire information from users, it is possible to take into account the user's subjective information, making it easier to estimate which funnel the user belongs to. In email distribution, different emails must be distributed depending on the funnel, even for the same product. However, according to this embodiment, it is possible to estimate the funnel, so that appropriate emails can be distributed according to the estimated funnel.

[0089] (Supplementary Note 1) The information processing method acquires history information relating to a predetermined notification means that notifies a user, and estimates a funnel to which the user belongs based on the acquired history information.

[0090] (Supplementary Note 2) The information processing method according to Supplementary Note 1 further includes acquiring a survey result to which the user has answered, and estimating a funnel to which the user belongs based on the acquired survey result.

[0091] (Supplementary Note 3) In the information processing method according to Supplementary Note 2, the questionnaire includes questions regarding subjective information of the user.

[0092] (Appendix 4) In the information processing method of any one of Appendices 1 to 3, the history information includes time-series information indicating whether or not the user is aware of the specified notification means, and when the time-series information is input as an observation value, the funnel to which the user belongs is estimated using an estimation model that outputs the funnel to which the user belongs in time series as a hidden variable.

[0093] (Supplementary Note 5) The information processing method according to any one of Supplementary Note 1 to Supplementary Note 4 specifies notification content to be notified to the user by the next notification means according to the estimated funnel.

[0094] (Appendix 6) An information processing method in any one of Appendices 1 to 5, wherein a funnel to which the user belongs, a questionnaire result answered by the user, and user information of the user are set as random variables, a user information distribution representing the probability density of the random variables is estimated, item information indicating notification content to be notified to the user by a predetermined notification means is set as a random variable, an item information distribution representing the probability density of the random variables is estimated, a probability distribution of the user's recognition rate for candidate notification content to be notified to the user by a next notification means is estimated based on the estimated user information distribution and item information distribution, and notification content to be notified to the user by the next notification means is specified according to the estimated probability distribution of the recognition rate.

[0095] (Appendix 7) An information processing method according to any one of Appendices 1 to 6, obtains event history information including at least one of the user's website browsing history and the user's app log history, and performs at least one of estimating a funnel to which the user belongs and specifying notification content to be notified to the user by a next notification means based on the event history information.

[0096] (Supplementary Note 8) The information processing device includes a control unit, which acquires historical information regarding a predetermined notification means that notified a user, and estimates the funnel to which the user belongs based on the acquired historical information.

[0097] (Supplementary Note 9) The computer program causes a computer to execute a process of acquiring historical information regarding a predetermined notification means that notified a user, and estimating a funnel to which the user belongs based on the acquired historical information.

[0098] The matters described in each embodiment can be combined with each other. In addition, the independent claims and dependent claims described in the claims can be combined with each other in any combination, regardless of the reference format. Furthermore, although the claims use a format in which a claim references two or more other claims (multiple claim format), this is not limited to this format. A multiple claim (multi-multi claim) that references at least one other multiple claim may also be used. [Explanation of symbols]

[0099] 1 Network 10 Terminal Equipment 30 Data Server 31 User Information DB 32 Survey Information DB 33 Item Information DB 34 Email history information DB 35 Event history information DB 50 Information processing equipment 51 Control section 52 Communications Department 53 Memory 54 Display section 55 Operation section 56 User information distribution estimation unit 57 Item information distribution estimation unit 58 Memory section 60 Computer Programs 61 1st estimation part 62 Second estimation part

Claims

1. Acquire history information regarding a predetermined notification means used to notify the user; estimating a funnel to which the user belongs based on the acquired history information; Information processing methods.

2. Obtaining the survey results answered by the user; estimating the funnel to which the user belongs based on the acquired survey results; The information processing method according to claim 1 .

3. The questionnaire includes questions regarding the user's subjective information. The information processing method according to claim 2 .

4. the history information includes time-series information indicating whether or not the user acknowledged the predetermined notification means; When the time series information is input as an observation value, the funnel to which the user belongs is estimated using an estimation model that outputs the funnel to which the user belongs in time series as a latent variable. The information processing method according to claim 1 .

5. specifying notification content to be notified to the user by the next notification means according to the estimated funnel; The information processing method according to any one of claims 1 to 4.

6. A funnel to which the user belongs, a survey result answered by the user, and user information of the user are treated as random variables, and a user information distribution representing a probability density of the random variables is estimated; estimating an item information distribution representing a probability density of the random variable; estimating a probability distribution of the user's recognition rate for candidate notification contents to be notified to the user by the next notification means based on the estimated user information distribution and item information distribution; specifying notification content to be notified to the user by a next notification means according to the probability distribution of the estimated recognition rate; The information processing method according to any one of claims 1 to 4.

7. acquiring event history information including at least one of the user's website browsing history and the user's app log history; At least one of estimating a funnel to which the user belongs and specifying notification content to be notified to the user by a next notification means based on the event history information. The information processing method according to any one of claims 1 to 4.

8. A control unit is provided, The control unit Acquire history information regarding a predetermined notification means used to notify the user; estimating a funnel to which the user belongs based on the acquired history information; Information processing device.

9. Acquire history information regarding a predetermined notification means used to notify the user; estimating a funnel to which the user belongs based on the acquired history information; A computer program that causes a computer to perform a process.

Images