Estimation device and estimation method

Abstract

This estimation device comprises: an acquisition unit that acquires a customer behavior sequence, which is a sequence of the past behavior of a customer, and a seller behavior sequence, which is a sequence of the past behavior of a seller toward the customer; and an estimation unit that uses a trained behavior effect estimation model to which the customer behavior sequence and the seller behavior sequence are input and which estimates behavior effects relating to the past behavior of the seller and including the state of the seller, the behavior of the customer, or the behavior of another seller, and that estimates the behavior effects resulting when the acquired customer behavior sequence and the seller behavior sequence are used as input.

Description

Estimation device and estimation method

[0001] The technology of the present disclosure relates to an estimation device and an estimation method.

[0002] Because customer behavior is directly related to a company's profits, research into learning customer behavior patterns from large amounts of customer behavior history and predicting future behavior is accelerating.

[0003] BERT4Rec (Non-Patent Document 1) is known as a method for predicting future behavior by learning behavioral history using only an encoder.

[0004] Fei Sun, et al. BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer. CIKM, 2019. https: / / arxiv.org / abs / 1904.06690

[0005] Conventional technology predicts future behavior based on a customer's past behavior, but it is not possible to predict behavior with high accuracy because it cannot take into account seller-side actions such as campaign activities and new product launches.

[0006] The disclosed technology has been made in consideration of the above points, and aims to provide an estimation device and estimation method that can accurately estimate the effects of seller behavior by taking into account both customer behavior and seller behavior.

[0007] A first aspect of the present disclosure is an estimation device including: an acquisition unit that acquires a customer behavior series, which is a past behavior series of a customer, and a seller behavior series, which is a past behavior series of a seller toward the customer; and an estimation unit that receives the customer behavior series and the seller behavior series as input, and estimates a behavioral effect when the acquired customer behavior series and the seller behavior series are input, using a trained behavioral effect estimation model that estimates a behavioral effect related to the seller's past behavior and including the seller's situation, the behavior of the customer, or the behavior of other sellers.

[0008] A second aspect of the present disclosure is an estimation method in which a computer acquires a customer behavior series, which is a series of past behaviors of a customer, and a seller behavior series, which is a series of past behaviors of a seller toward the customer, and uses the customer behavior series and the seller behavior series as inputs to estimate the behavioral effect when the acquired customer behavior series and the seller behavior series are inputs using a trained behavioral effect estimation model that estimates a behavioral effect related to the seller's past behavior and that includes the seller's situation, the behavior of the customer, or the behavior of other sellers.

[0009] According to the disclosed technology, it is possible to take into account both customer behavior and seller behavior and accurately estimate the effect of seller behavior.

[0010] 1 is a schematic block diagram of an example of a computer that functions as a learning device and an estimation device of this embodiment. FIG. 2 is a diagram showing an example of a customer behavior sequence. FIG. 3 is a diagram showing an example of a seller behavior sequence. FIG. 4 is a diagram showing an example of a behavior effect. FIG. 5 is a diagram showing an example of a behavior effect. FIG. 6 is a block diagram showing the functional configuration of the learning device of this embodiment. FIG. 7 is a diagram showing the configuration of a behavior effect estimation model. FIG. 8 is a diagram showing an example of layer design of a behavior effect estimation model. FIG. 9 is a block diagram showing the functional configuration of the estimation device of this embodiment. FIG. 10 is a diagram showing how a part of a seller behavior sequence is changed and input. FIG. 11 is a diagram showing how a part of a customer behavior sequence is changed and input. A flowchart showing the flow of the learning process of this embodiment. A flowchart showing the flow of the estimation process of this embodiment. A diagram showing the configuration of a behavior effect estimation model in a modified example.

[0011] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. Note that the same or equivalent components and parts in each drawing are given the same reference numerals. Also, the dimensional proportions in the drawings are exaggerated for the convenience of explanation and may differ from the actual proportions.

[0012] <Outline of this Embodiment> Because customer behavior is directly related to a company's profits, research into learning customer behavior patterns from a large amount of customer behavior history and predicting future behavior is accelerating.

[0013] The developers of the technology disclosed herein have developed a technology for predicting future behavior based on a contextual understanding of continuous customer behavior by introducing a function for processing time information and data types other than character strings, which are important for behavioral analysis but are not taken into account when processing natural language, based on an existing natural language processing model.

[0014] However, even if you can predict the behavior of each customer, you cannot formulate an effective improvement policy unless you visualize the impact on the company's profits.Furthermore, a company's profits are influenced not only by customer behavior but also by factors such as the behavior of sellers, both the company itself and its competitors.

[0015] For example, even if it is predicted that a certain customer is highly likely to purchase green juice within the next week, there is a possibility that the customer will be lost to a competitor due to the influence of a competitor's campaign activities, etc.

[0016] Also, even if it is predicted that a certain customer is unlikely to purchase green juice within the next week, there is a possibility that they will become a customer of the company's green juice product through improvement activities such as the release of a new product by the company.

[0017] In this way, the actions of your company or your competitors affect the actions of your customers, and ultimately your company's profits.

[0018] In order to express this chain reaction of influence in a model, in this embodiment, a large number of customer behavior sequences and seller behavior sequences are input, the correct labels of the content that is to be predicted as the effect of seller behavior are compiled, and a model that estimates the effect of seller behavior is trained using supervised learning.

[0019] Using a model that estimates the effects of seller behavior using customer behavior patterns and seller behavior patterns as inputs, customer behavior, the status of the seller itself, or the behavior of other sellers is estimated. Furthermore, a portion of the customer behavior patterns or seller behavior patterns is changed, and the estimated results are simulated. Based on the simulated estimation results, it is possible to adjust target customers or business strategies such as products, advertising, distribution, and prices, as needed.

[0020] <Configuration of Learning Device According to This Embodiment> FIG. 1 is a block diagram showing the hardware configuration of a learning device 10 according to this embodiment.

[0021] 1, the learning device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.

[0022] The CPU 11 is a central processing unit that executes various programs and controls each component. That is, the CPU 11 reads programs from the ROM 12 or the storage 14 and executes the programs using the RAM 13 as a work area. The CPU 11 controls the above components and performs various arithmetic processing in accordance with the programs stored in the ROM 12 or the storage 14. In this embodiment, a learning program is stored in the ROM 12 or the storage 14. The learning program may be a single program, or a group of programs consisting of multiple programs or modules.

[0023] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including the operating system and various data.

[0024] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used to input various types of data including a customer behavior series, which is a series of past behaviors for each customer, a seller behavior series, which is a series of past behaviors of a seller toward a customer, and learning data including behavioral effects related to the seller's past behavior, including the seller's situation, customer behavior, or behavior of other sellers.

[0025] For example, as shown in Fig. 2, a customer behavior sequence, which is a sequence of past behaviors manually prepared for each customer, is input to the input unit 15. Here, the customer behavior sequence refers to table data in which customer behavior data that may contribute to estimating behavioral effects is collected and organized in chronological order.

[0026] FIG. 2 shows an example in which a customer behavior sequence is made up of records including customer, date and time, gender, advertisement viewing, product purchase, and amount.

[0027] In this example, "Customer" is a column that represents customer identification, and "Date and Time" is a column that represents time. Representing customer identification requires a column that represents time. At least one event-related column, such as "Ad View," must be selected. Attribute-related columns, such as "Gender," are not required, but are preferable for gaining a deeper understanding of customer behavior. If the time granularity of the collected customer behavior data is not constant, the customer behavior data may be aggregated at predetermined time intervals. In this case, if the data elements are a set of labels, the data can be aggregated by summing the sets (combining the arrays). If the data elements are numerical, a representative value can be calculated using functions such as the average, sum, variance, maximum, and minimum.

[0028] Furthermore, the data type of each column is not limited to this example, and may be other data types.

[0029] In this example, the first record indicates that customer "1" whose gender is "male" viewed advertisements for "green juice" and "potatoes" on the date and time "2024-01-01 9:10" and purchased the product "green juice" for a total amount of 520. The second record indicates that customer "1" whose gender is "male" viewed an advertisement for "green juice" and purchased the product "potatoes" for a total amount of 190 on the date and time "2024-01-01 9:20".

[0030] The customer and date and time are required items, the gender is an item indicating the customer's attributes, and the advertisement viewing, product purchase, and amount are items indicating behavioral events.

[0031] Further, as shown in FIG. 3, the input unit 15 receives a seller behavior sequence, which is a past behavior sequence manually prepared for each seller.

[0032] Here, the seller behavior sequence refers to tabular data in which behavioral data of the user's company or competitors that may contribute to customer behavior is collected and organized in chronological order.

[0033] FIG. 3 shows an example in which a seller's behavior sequence is made up of records including company, date and time, stock price, product name, packaging, advertising music, material, market share, unit price, and event.

[0034] In this example, "Company" is a column that represents the company identification, and "Date and Time" is a column that represents the time. The columns that represent the company identification and the time are required.

[0035] You must select one or more attribute columns that represent the attributes of the subject of the action (the company or a specific product), such as "Stock price" or "Product name." You must also select one or more event columns that represent specific actions, such as "Event."

[0036] If there are events with closely spaced dates and times, the behavior data may be aggregated at predetermined time intervals, similar to customer behavior data.

[0037] Furthermore, the data type of each column is not limited to this example, and may be other data types.

[0038] In this example, the first record indicates that company "1" advertised a newly released product on date and time "2024-01-01," with a stock price of "4000.5," product name "True Milk," packaging and advertising song "Do-do-do," ingredients "Milk," market share of "25%," and unit price of "110."

[0039] The company and date are required items, and the stock price, product name, packaging, advertising music, material, market share, and unit price are items that indicate the attributes of the seller.

[0040] Furthermore, as shown in FIG. 4A, customer behaviors are manually prepared for each customer and input to the input unit 15 as behavioral effects related to the seller's past behaviors.

[0041] Here, customer behavior as a behavioral effect includes, for example, whether or not to purchase a product within a certain period of time in the future, which product or combination of products to purchase within a certain period of time in the future, when to purchase a certain product or combination of products, or how much total product to purchase within a certain period of time in the future.

[0042] 4A shows an example where customer behavior indicates which combination of products will be purchased within a certain period of time, and is composed of records including customers and product purchases. In this example, the first record indicates that customer "1" purchased the products "green juice" and "potatoes."

[0043] Alternatively, as shown in FIG. 4B, the input unit 15 receives input of the status of the seller, which is the user himself, as a manually prepared behavioral effect related to the seller's past behavior.

[0044] The seller's status as a behavioral effect includes, for example, sales of the entire company or the division in question within a certain period of time in the future, profits of the entire company or the division in question within a certain period of time in the future, or the time when a target sales index can be achieved.

[0045] 4B shows an example in which the seller's status represents the sales and profits of the seller for a certain period of time in the future, and is composed of records including date and time, sales, and profits. In this example, the first record indicates that the sales for the date and time "first week of 2024" are "5000" and the profit is "1000".

[0046] Alternatively, as shown in FIG. 4C, the input unit 15 receives input of actions of other sellers as action effects related to the seller's past actions, which are manually prepared for each seller.

[0047] Behavioral effects of other sellers include, for example, whether a competitor will perform an event within a certain period of time in the future, which event or combination of events a competitor will perform within a certain period of time in the future, and when a competitor will perform a certain event or combination of events.

[0048] 4C shows an example in which the behavior of other sellers indicates whether they will release new products within a certain period of time, and is composed of records including the company and the new product release. In this example, the first record indicates that company "1" has released a new product.

[0049] The display unit 16 is, for example, a liquid crystal display, and displays various information including the processing results. The display unit 16 may be a touch panel type and function as the input unit 15.

[0050] The communication interface 17 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, and Wi-Fi (registered trademark).

[0051] Next, a description will be given of the functional configuration of the learning device 10. Fig. 5 is a block diagram showing an example of the functional configuration of the learning device 10.

[0052] As shown in FIG. 5, the learning device 10 functionally includes an acquisition unit 20, a learning unit 22, and a model storage unit 24.

[0053] The acquisition unit 20 acquires the input learning data.

[0054] The learning unit 22 uses a plurality of learning data sets, and inputs customer behavior sequences and seller behavior sequences, to learn a behavior effect estimation model that estimates behavior effects.

[0055] Specifically, as shown in FIG. 6, the behavioral effect estimation model 100 has a Transformer encoder 100A and a decoder 100B. The encoder 100A receives a customer behavior sequence for a given period as input, and the decoder 100B receives the output of the encoder 100A and a seller behavior sequence for that period as input, and outputs a behavioral effect.

[0056] In Figure 6 above, an example is shown in which a customer behavior sequence consisting of records including an item for identifying the customer, a date and time item, an item for customer attributes, and an item for an event indicating the behavior is input to encoder 100A, and a seller behavior sequence consisting of records including an item for identifying the seller company, a date and time item, an item for seller attributes, and an item for an event indicating the behavior is input to decoder 100B, and the behavioral effect including the date and time, the customer's behavior, the seller's situation, or the behavior of other sellers is estimated from the output of decoder 100B.

[0057] Note that the behavioral effect estimation model 100 is not limited to a Transformer, but may be any model that can accept past customer behavioral sequences and seller behavioral sequences and output future behavioral effects, such as a multi-layer neural network or gradient boosting. The behavioral effect estimation model 100 can handle both regression problems and classification problems.

[0058] 7 shows an example of the layer design of the action effect estimation model 100. The encoder 100A includes a Self-Attention layer, a Layer Norm layer, a Feed Forward layer, and a Layer Norm layer. The decoder 100B includes a Self-Attention layer, a Layer Norm layer, a Cross-Attention layer, a Layer Norm layer, a Feed Forward layer, and a Layer Norm layer.

[0059] The customer behavior sequence is converted into a multidimensional vector and input to the encoder 100A. Specifically, the conversion is performed using a vector conversion technique suitable for the type of data included in the customer behavior sequence. For example, an embedding layer is used to vectorize numeric values ​​such as attributes, labels representing behavior content, and date and time, character strings are vectorized using a language model (e.g., GPT), and images are vectorized using an image model (e.g., Stable Diffusion). The seller behavior sequence is similarly converted into a multidimensional vector and input to the decoder 100B.

[0060] The output layer of the decoder 100B is configured to convert the data into the type of data desired to be obtained as an estimation result, and is configured to output, for example, a numerical value or a label.

[0061] The learning unit 22 learns parameters relating to the part of the behavioral effect estimation model that receives customer behavior sequences as inputs through self-supervised learning using the customer behavior sequences.

[0062] For example, in self-supervised learning for the encoder 100A, a customer behavior sequence input to the encoder 100A is masked, and the parameters of the encoder 100A are repeatedly updated so that the masked portions can be output from the encoder 100A, thereby further improving the accuracy of behavioral effect estimation.

[0063] Furthermore, in the self-supervised learning of the encoder 100A, learning is performed so that the result obtained through the Output layer for the output of the encoder 100A when a masked behavior sequence is input matches the output when the same behavior sequence is input without being masked.

[0064] The learning unit 22 learns the behavioral effect estimation model 100 in two stages: in the first stage, self-supervised learning is performed on the encoder 100A using only the customer behavior series, and in the second stage, learning is performed on the entire behavioral effect estimation model 100 using the customer behavior series, seller behavior series, and behavioral effects. In the second stage, learning may be performed on only the decoder 100B.

[0065] The learning unit 22 stores the learned action effect estimation model 100 in the model storage unit 24 .

[0066] <Configuration of Estimation Device According to This Embodiment> FIG. 1 is a block diagram showing the hardware configuration of an estimation device 50 according to this embodiment.

[0067] 1 , the estimation device 50, like the learning device 10, has a CPU 11, a ROM 12, a RAM 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other so as to be able to communicate with each other via a bus 19. An estimation program is stored in the ROM 12 or the storage 14.

[0068] The input unit 15 is used to input various data including past customer behavior patterns and seller behavior patterns.

[0069] Next, a description will be given of the functional configuration of the estimation device 50. Fig. 8 is a block diagram showing an example of the functional configuration of the estimation device 50.

[0070] As shown in FIG. 8, the estimation device 50 functionally includes a model storage unit 52, an acquisition unit 54, an estimation unit 56, and a change unit 58.

[0071] The model storage unit 52 stores a trained action effect estimation model 100 similar to that of the model storage unit 24 of the learning device 10 .

[0072] The acquisition unit 54 acquires the input past customer behavior sequences and seller behavior sequences.

[0073] The estimation unit 56 uses the trained behavioral effect estimation model 100 to estimate the seller's situation, the customer's behavior, or the behavior of other sellers as the behavioral effect when the acquired customer behavior series and seller behavior series are input.

[0074] For example, the estimation unit 56 inputs the customer behavior sequence to the encoder 100A and the seller behavior sequence to the decoder 100B, and uses the output of the decoder 100B to estimate the seller's situation, the customer's behavior, or the behavior of other sellers as a behavioral effect.

[0075] The change unit 58 changes a part of either the acquired customer behavior series or the acquired seller behavior series. Specifically, the change unit 58 accepts an instruction from the user to change a part of either the acquired customer behavior series or the acquired seller behavior series, and changes the part of either the acquired customer behavior series or the acquired seller behavior series.

[0076] The estimation unit 56 further uses a behavioral effect estimation model to estimate the behavioral effect when inputting either one of the partially changed customer behavior series or seller behavior series and the other of the acquired customer behavior series or seller behavior series.

[0077] For example, as shown in Figure 9, a part of the acquired seller behavior sequence is changed, and the behavioral effect is estimated again using the behavioral effect estimation model. Figure 9 shows an example in which the sequence [water, honey, shiso] included in the ingredients field of the record for company "2" in the seller behavior sequence is changed to [water, honey, shiso, celery]. It also shows an example in which the estimated results of sales and profits, which indicate the seller's situation and are included in the behavioral effect, have changed compared to before the change.

[0078] As another example, as shown in Figure 10, a part of the acquired customer behavior sequence is changed, and the behavioral effect is estimated again using the behavioral effect estimation model. Figure 10 shows an example in which the customer attribute "gender" field of each record in the customer behavior sequence is changed from "male" to "female." The example also shows how the estimated results of sales and profits, which indicate the seller's situation and are included in the behavioral effect, have changed compared to before the change.

[0079] <Operation of the Learning Device According to the Present Embodiment> Next, the operation of the learning device 10 according to the present embodiment will be described.

[0080] 11 is a flowchart showing the flow of the learning process by the learning device 10. The learning process is performed by the CPU 11 reading out a learning program from the ROM 12 or storage 14, expanding it into the RAM 13, and executing it. It is also assumed that learning data including past customer behavior patterns, seller behavior patterns, and behavioral effects related to seller behavior have been input to the learning device 10.

[0081] In step S100 , the CPU 11 functions as the acquisition unit 20 to acquire a plurality of pieces of learning data received by the input unit 15 .

[0082] In step S102, the CPU 11, functioning as the learning unit 22, performs self-supervised learning on the encoder 100A of the behavioral effect estimation model 100 using customer behavior sequences included in the multiple pieces of learning data.

[0083] In step S104, the CPU 11, as the learning unit 22, performs supervised learning on the entire action effect estimation model 100 using multiple pieces of learning data, stores the learned action effect estimation model 100 in the model storage unit 24, and terminates the learning process.

[0084] <Operation of the Estimation Device According to the Present Embodiment> Next, the operation of the estimation device 50 according to the present embodiment will be described.

[0085] 12 is a flowchart showing the flow of estimation processing by the estimation device 50. The estimation processing is performed by the CPU 11 reading out an estimation program from the ROM 12 or storage 14, deploying it in the RAM 13, and executing it. The model storage unit 52 of the estimation device 50 stores a trained behavioral effect estimation model 100 trained by the learning device 10. It is also assumed that past customer behavior sequences and seller behavior sequences have been input to the estimation device 50.

[0086] In step S110, the CPU 11 functions as the acquisition unit 54 to acquire past customer behavior patterns and seller behavior patterns.

[0087] In step S112, the CPU 11, as the estimation unit 56, uses the trained behavioral effect estimation model 100 to estimate the seller's situation, the customer's behavior, or the behavior of other sellers as the behavioral effect when the acquired customer behavior series and seller behavior series are input.

[0088] In step S114, the CPU 11 functions as the change unit 58 to change a part of either the acquired customer behavior sequence or seller behavior sequence.

[0089] In step S116, the CPU 11, as the estimation unit 56, uses the behavioral effect estimation model to estimate the seller's situation, the customer's behavior, or the behavior of other sellers as the behavioral effect when either the partially changed customer behavior series or the seller behavior series and the other of the acquired customer behavior series or the seller behavior series are input, and then ends the estimation process.

[0090] As described above, the estimation device according to this embodiment receives as input a customer behavior sequence, which is a sequence of past customer behavior, and a seller behavior sequence, which is a sequence of past seller behavior toward the customer, and estimates behavioral effects using a trained behavioral effect estimation model that estimates behavioral effects related to the seller's past behavior and that include the seller's situation, customer behavior, or the behavior of other sellers. This allows for accurate estimation of the effects of seller behavior, taking both customer behavior and seller behavior into consideration.

[0091] In marketing, there is a method for analyzing the optimal mix of product, advertising, distribution, and price as components of marketing activities in order to effectively approach customers from one's own company. However, manually implementing the conclusions of desk studies is risky. The estimation device of this embodiment makes it possible to simulate the strategy that a seller plans to implement by modifying part of the seller's behavioral sequence and estimating the behavioral effects. In the example of Figure 9 above, in order to compete with other companies, celery was added to the ingredients of the company's products. When the estimated behavioral effects were confirmed, it was found that there was a high possibility of a decrease in profits, and therefore the strategy needed to be reconsidered.

[0092] Furthermore, by changing part of the customer behavior sequence and observing the change in behavioral effect, it is possible to consider the effectiveness of management strategies such as products, advertising, distribution, and prices. Specifically, it is possible to simulate behavioral effects by changing the customer attributes or events included in the customer behavior sequence. In the example of Figure 10 above, when the target customer was changed to female and the estimated behavioral effect was confirmed, it was found that there was a high possibility that this would lead to increased profits. In this way, it is possible to consider the rationality of target customers by changing part of the customer behavior sequence and observing the change in behavioral effect.

[0093] Furthermore, the present invention is not limited to the device configuration and operation of the above-described embodiment, and various modifications and applications are possible within the scope of the gist of the present invention.

[0094] For example, as shown in FIG. 13, a seller behavior sequence consisting of records including an item identifying the seller company, a date and time item, an item of seller attributes, and an item of an event indicating behavior can be input to encoder 100A, and a customer behavior sequence consisting of records including an item identifying the customer, a date and time item, an item of customer attributes, and an item of an event indicating behavior can be input to decoder 100B, and the behavioral effect including the date and time, customer behavior, seller status, or other seller behavior can be estimated from the output of decoder 100B.

[0095] Although the above description has been given with reference to an example in which the estimation device and the learning device are configured separately, the present invention is not limited to this and the estimation device and the learning device may be configured as a single device.

[0096] In addition, the various processes executed by the CPU after reading the software (program) in the above embodiment may be executed by various processors other than the CPU. Examples of processors in this case include dedicated electrical circuits, such as programmable logic devices (PLDs) whose circuit configuration can be changed after manufacture, such as field-programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs), which are processors having a circuit configuration designed specifically for executing specific processes. Furthermore, the learning process and estimation process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, a combination of a CPU and an FPGA, etc.). The hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor elements.

[0097] In addition, in each of the above embodiments, the learning program and the estimation program are described as being pre-stored (installed) in the storage 14, but the present invention is not limited to this. The programs may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The programs may also be downloaded from an external device via a network.

[0098] The following additional notes are provided regarding the above-described embodiments.

[0099] (Supplementary Item 1) An estimation device including: a memory; and at least one processor connected to the memory, wherein the processor is configured to: acquire a customer behavior sequence, which is a past behavior sequence of a customer, and a seller behavior sequence, which is a past behavior sequence of a seller toward the customer; and estimate a behavioral effect when the acquired customer behavior sequence and seller behavior sequence are input, using a trained behavioral effect estimation model that estimates a behavioral effect related to the seller's past behavior and that includes the seller's situation, the behavior of the customer, or the behavior of other sellers.

[0100] (Supplementary Item 2) A non-transitory storage medium storing a program executable by a computer to execute an estimation process, wherein the estimation process is configured to: acquire a customer behavior series, which is a series of past behaviors of a customer, and a seller behavior series, which is a series of past behaviors of a seller toward the customer; and estimate the behavioral effect when the acquired customer behavior series and seller behavior series are input using a trained behavioral effect estimation model that estimates behavioral effects related to the seller's past behavior and that include the seller's situation, the behavior of the customer, or the behavior of other sellers.

[0101] REFERENCE SIGNS LIST 10 Learning device 11 CPU 13 RAM 14 Storage 15 Input unit 16 Display unit 20, 54 Acquisition unit 22 Learning unit 24, 52 Model storage unit 50 Estimation device 56 Estimation unit 58 Change unit 100 Action effect estimation model 100A Encoder 100B Decoder

Claims

1. An estimation device including: an acquisition unit that acquires a customer behavior sequence, which is a sequence of a customer's past behavior, and a seller behavior sequence, which is a sequence of a seller's past behavior toward the customer; and an estimation unit that receives the customer behavior sequence and the seller behavior sequence as input and estimates the behavioral effect when the acquired customer behavior sequence and the seller behavior sequence are input, using a trained behavioral effect estimation model that estimates the behavioral effect related to the seller's past behavior and including the seller's situation, the behavior of the customer, or the behavior of other sellers.

2. The estimation device according to claim 1, wherein the behavioral effect estimation model has a Transformer encoder and decoder, wherein the encoder receives either the customer behavior series or the seller behavior series as input, and the decoder receives the encoder output and the other of the customer behavior series and the seller behavior series as input, and outputs the behavioral effect.

3. The estimation device according to claim 1, further comprising a modification unit that modifies a part of either the acquired customer behavior series or the acquired seller behavior series, wherein the estimation unit further uses the behavior effect estimation model to estimate the behavior effect when the modified either the customer behavior series or the seller behavior series and the other either the customer behavior series or the seller behavior series are input.

4. An estimation method executed by a computer, which includes acquiring a customer behavior sequence, which is a sequence of a customer's past behavior, and a seller behavior sequence, which is a sequence of a seller's past behavior toward said customer, and using said customer behavior sequence and said seller behavior sequence as inputs, and using a trained behavior effect estimation model that estimates behavior effects related to the seller's past behavior and that include the seller's situation, the behavior of the customer, or the behavior of other sellers, to estimate the behavioral effects when said acquired customer behavior sequence and said seller behavior sequence are inputs.

Images