Learning device, estimation device, learning method, estimation method, and program

Abstract

A training video information acquisition unit (102) of a learning device (10) acquires training video information acquired by analyzing a training video advertisement that is a training video advertisement. A training purchase information acquisition unit (103) acquires training purchase information pertaining to purchase by a viewer who trains a video advertisement, that is, a training viewer. The learning unit (104) learns a machine learning model on the basis of the training video information and the training purchase information. An estimated video information acquisition unit (202) of an estimation device (20) acquires estimated video information acquired by analyzing an estimated video advertisement that is a video advertisement for estimation. A model storage unit (201) stores a machine learning model after learning. The estimation unit (203) estimates the advertisement effect of the estimated video advertisement on the basis of the estimated video information and the machine learning model.

Description

Technical Field

[0001] The present disclosure relates to a learning device, an inference device, a learning method, an inference method, and a program. Background Art

[0002] In the past, there is a known technology for estimating the advertising effect of video advertisements. For example, Patent Document 1 describes a method that processes the action data and physiological data of viewers who watch media content such as video advertisements to obtain data points of emotional states in a time series, and outputs predicted effect data representing the effect predicted for the media content by matching the effect data with prediction parameters of the data points of emotional states in a time series from a classification model. The prediction parameters in Patent Document 1 are quantitative indicators of the relative changes in the viewers' responses to video advertisements.

[0003] [Background Technology Literature]

[0004] [Patent Document]

[0005] Patent Document 1: Japanese Patent Publication No. 2020-501260 Summary of the invention

[0006] [Problems to be solved by the invention]

[0007] However, since the classification model of Patent Document 1 only considers data based on the actions of viewers of video ads and outputs predicted effect data, the content of the actually released video ads is not fully considered in the estimation of the advertising effect of the video ads. Therefore, the technology of Patent Document 1 has a situation where the estimation accuracy of the advertising effect of the video ads is insufficient. This is also true in other background technologies other than the technology of Patent Document 1.

[0008] One of the purposes of the present disclosure is to improve the accuracy of estimating the advertising effectiveness of video advertisements.

[0009] [Technical means to solve the problem]

[0010] The learning device disclosed in the present invention includes: a training video information acquisition unit, which acquires training video information obtained by analyzing a video advertisement for training, i.e., a training video advertisement; a training purchase information acquisition unit, which acquires training purchase information related to purchases made by viewers of the training video advertisement, i.e., training viewers; and a learning unit, which performs learning of a machine learning model for estimating the advertising effect of the estimated video advertisement from estimated video information obtained by analyzing a video advertisement for estimation, i.e., an estimated video advertisement, based on the training video information and the training purchase information.

[0011] The inference device disclosed in the present invention includes: an inferred video information acquisition unit, which acquires inferred video information obtained by analyzing a video advertisement for inference, namely, an inferred video advertisement; a model storage unit, which stores a machine learning model after learning based on training video information obtained by analyzing a video advertisement for training, namely, a training video advertisement, and training purchase information related to purchases of viewers of the training video advertisement, namely, training viewers; and an inference unit, which infers the advertising effect of the inferred video advertisement based on the inferred video information and the machine learning model.

[0012] [Effects of the Invention]

[0013] According to the present disclosure, it is possible to improve the accuracy of estimating the advertising effectiveness of video advertisements. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a diagram showing an example of the hardware configuration of each of the learning device and the estimation device.

[0015] Figure 2 This is a diagram showing an example of a training video advertisement distributed by a live broadcast service.

[0016] Figure 3 This is a diagram showing an example of the overview of each of the learning device and the estimation device.

[0017] Figure 4 This is a diagram showing an example of functions implemented by each of the learning device and the estimation device.

[0018] Figure 5 This is a diagram showing an example of a training database.

[0019] Figure 6 This is a diagram showing an example of an estimation database.

[0020] Figure 7 This is a diagram showing an example of a video advertisement database.

[0021] Figure 8 This is a diagram showing an example of processing executed by the learning device.

[0022] Fig. 9 This is a diagram showing an example of the processing executed in S10.

[0023] Fig.10 This is a diagram showing an example of processing executed by the estimation device.

[0024] Fig.11 This is a diagram showing an example of the processing executed in S20.

[0025] Fig.12 This is a diagram showing an example of functions implemented in a modified example. DETAILED DESCRIPTION

[0026] [1. Hardware Configuration of Each of the Learning Device and the Estimation Device]

[0027] An example of an implementation of the learning device, inference device, learning method, inference method, and program disclosed herein is described. In this implementation, the learning device and the inference device are different devices as an example, but the learning device and the inference device may be the same device. In other words, a certain device may function as both a learning device and an inference device.

[0028] In the present embodiment, the learning device performs learning of a machine learning model for estimating the advertising effect of an advertisement using a video, that is, a video advertisement. The estimating device performs estimation based on the learned machine learning model. Hereinafter, a video advertisement used for training (learning) for learning a machine learning model is referred to as a training video advertisement. The performer of the training video advertisement is referred to as a training performer. The viewer of the training video advertisement is referred to as a training viewer. The video advertisement used for estimation that will become the processing object of the learned machine learning model is referred to as an estimated video advertisement. The performer of the estimated video advertisement is referred to as an estimated performer. The viewer of the estimated video advertisement is referred to as an estimated viewer.

[0029] In addition, in this embodiment, when there is no need to distinguish between training video ads and estimated video ads, sometimes they are simply referred to as video ads. When there is no need to distinguish between training performers and estimated performers, sometimes they are simply referred to as performers. When there is no need to distinguish between training viewers and estimated viewers, sometimes they are simply referred to as viewers. In addition, when there is no need to distinguish between the terms of training ○○ and estimated ○○, sometimes they are simply referred to as ○○.

[0030] Figure 1 1 is a diagram showing an example of the hardware configuration of each of the learning device and the estimation device. In this embodiment, an example of a live broadcast system 1 including the learning device 10 and the estimation device 20 is described. Figure 1 In the example of , the live broadcast system 1 includes, in addition to the learning device 10 and the estimation device 20, a server 30, a performer training device 40, a viewer training device 50, an estimated performer device 60, and an estimated viewer device 70. Each of the learning device 10, the estimation device 20, the server 30, the performer training device 40, the viewer training device 50, the estimated performer device 60, and the estimated viewer device 70 is connected to a network N such as the Internet or a LAN (Local Area Network).

[0031] The learning device 10 is a computer for learning a machine learning model. For example, the learning device 10 is a personal computer, a server computer, a tablet, or a smart phone. For example, the learning device 10 includes a control unit 11, a storage unit 12, a communication unit 13, an operation unit 14, and a display unit 15. The control unit 11 includes at least one processor. The storage unit 12 includes a volatile memory such as a RAM (Random Access Memory) and at least one non-volatile memory such as a flash memory. The communication unit 13 includes at least one communication interface for wired communication and a communication interface for wireless communication. The operation unit 14 is an input device such as a touch panel. The display unit 15 is a liquid crystal or an organic EL (Electroluminescence) display.

[0032] The estimation device 20 is a computer that performs estimation based on the learned machine learning model. For example, the estimation device 20 is a personal computer, a server computer, a tablet, or a smart phone. For example, the estimation device 20 includes a control unit 21, a storage unit 22, a communication unit 23, an operation unit 24, and a display unit 25. The hardware configuration of the control unit 21, the storage unit 22, the communication unit 23, the operation unit 24, and the display unit 25 may be the same as that of the control unit 11, the storage unit 12, the communication unit 13, the operation unit 14, and the display unit 15, respectively.

[0033] The server 30 is a server computer managed by an operator of a live broadcast service that publishes videos in real time to an unspecified number of people. In the present embodiment, the case where the operator of the live broadcast service also manages the learning device 10 and the inference device 20 is taken as an example, but the learning device 10 and the inference device 20 may also be managed by other people. For example, the server 30 includes a control unit 31, a storage unit 32, and a communication unit 33. The hardware configuration of the control unit 31, the storage unit 32, and the communication unit 33 may be the same as that of the control unit 11, the storage unit 12, and the communication unit 13, respectively.

[0034] The performer training device 40 is a computer for training performers. For example, the performer training device 40 is a personal computer, a tablet, or a smart phone. For example, the performer training device 40 includes a control unit 41, a storage unit 42, a communication unit 43, an operation unit 44, and a display unit 45. The hardware configuration of the control unit 41, the storage unit 42, the communication unit 43, the operation unit 44, and the display unit 45 can be the same as that of the control unit 11, the storage unit 12, the communication unit 13, the operation unit 14, and the display unit 15, respectively. The performer training device 40 is connected to a photographing unit 46. The photographing unit 46 includes at least one camera. The photographing unit 46 can be included in the performer training device 40.

[0035] The viewer-listener training device 50 is a computer for training the viewer-listener. For example, the viewer-listener training device 50 is a personal computer, a tablet, or a smart phone. For example, the viewer-listener training device 50 includes a control unit 51, a storage unit 52, a communication unit 53, an operation unit 54, and a display unit 55. The hardware configurations of the control unit 51, the storage unit 52, the communication unit 53, the operation unit 54, and the display unit 55 may be the same as those of the control unit 11, the storage unit 12, the communication unit 13, the operation unit 14, and the display unit 15, respectively.

[0036] The estimated performer device 60 is a computer for estimating a performer. For example, the estimated performer device 60 is a personal computer, a tablet, or a smart phone. For example, the estimated performer device 60 includes a control unit 61, a storage unit 62, a communication unit 63, an operation unit 64, and a display unit 65, and is connected to a camera unit 66. The hardware configurations of the control unit 61, the storage unit 62, the communication unit 63, the operation unit 64, the display unit 65, and the camera unit 66 may be the same as those of the control unit 11, the storage unit 12, the communication unit 13, the operation unit 14, the display unit 15, and the camera unit 46, respectively. The camera unit 66 may be included in the estimated performer device 60.

[0037] The estimated viewer device 70 is a computer for estimating the viewer. For example, the estimated viewer device 70 is a personal computer, a tablet, or a smartphone. For example, the estimated viewer device 70 includes a control unit 71, a storage unit 72, a communication unit 73, an operation unit 74, and a display unit 75. The hardware configuration of the control unit 71, the storage unit 72, the communication unit 73, the operation unit 74, and the display unit 75 may be the same as that of the control unit 11, the storage unit 12, the communication unit 13, the operation unit 14, and the display unit 15, respectively.

[0038] In addition, the program stored in the storage unit 12, 22, 32, 42, 52, 62, 72 can be supplied to the learning device 10, the estimation device 20, the server 30, the performer training device 40, the viewer training device 50, the performer estimation device 60, or the viewer estimation device 70 via the network N. In addition, the program stored in the computer-readable information storage medium can be supplied to the learning device 10, the estimation device 20, the server 30, the performer training device 40, the viewer training device 50, the performer estimation device 60, or the viewer estimation device 70 via a reading unit for reading the information storage medium (for example, an optical disk drive or a memory card slot) or an input / output unit for inputting and outputting data with an external device (for example, a USB (Universal Serial Bus) port).

[0039] In addition, the live broadcast system 1 only needs to include at least one computer, not limited to Figure 1For example, the live broadcast system 1 may include a learning device 10, an inference device 20, and a server 30, but may not include a training performer device 40, a training viewer device 50, an inference performer device 60, and an inference viewer device 70. In the case, the training performer device 40, the training viewer device 50, the inference performer device 60, and the inference viewer device 70 exist outside the live broadcast system 1. In addition, for example, the learning device 10 and the inference device 20 may not be included in the live broadcast system 1, but may exist outside the live broadcast system 1.

[0040] [2. Overview of each of the learning device and the estimation device]

[0041] Figure 2 1 is a diagram showing an example of a training video advertisement released by a live broadcast service. For example, a training performer promotes a product or service to be traded by e-commerce in front of a camera unit 46. The training performer device 40 sends a training video advertisement generated by the camera unit 46 to the server 30. The server 30 releases the training video advertisement received from the training performer device 40 to the training viewer device 50 of an unspecified plurality of training viewers in real time. The display unit 55 of the training viewer device 50 displays a screen SC showing the training video advertisement.

[0042] For example, when the training viewer selects button B1, the training viewer device 50 accesses a page showing details of the product or service introduced by the training video advertisement. The training viewer can purchase the product or service from the page. The details can be displayed on the screen SC. When the training viewer selects button B2, the training viewer device 50 accesses a page showing coupons for the product or service introduced by the training video advertisement. The training viewer can obtain coupons for the product on the page. Coupons can also be obtained on the screen SC.

[0043] For example, the training viewer can input comments on the training video advertisement into the input box F. In the display area A of the screen SC, the comments input by the training viewer and each of the other training viewers are displayed. One of the goals of the live broadcast service as described above is to maximize the advertising effect. The advertising effect can also be called conversion. For example, the advertising effect is the number of sales of goods or services, the turnover, the number of additions to the shopping cart, the number of additions to the bookmarks, the number of users who have collected the store from the page of the video advertisement, or a combination thereof. The advertising effect is believed to be affected by various factors. In the present embodiment, the learning device 10 performs learning of a machine learning model for estimating the advertising effect in the live broadcast service. The estimation device 20 performs estimation using the learned machine learning model.

[0044] Figure 31 is a diagram showing an example of the outline of each of the learning device 10 and the estimation device 20. In the present embodiment, the learning device 10 performs various analyses on the released training video advertisements to obtain training video information indicating the characteristics of the training video advertisements. The training video information indicates the characteristics that are considered to have a causal relationship with the advertising effect of the training video advertisements. The details of the method of obtaining the training video information will be described later. For example, the learning device 10 analyzes the expressions of the training performers, the responses of the training performers, and the explanations of the training performers to obtain a plurality of training video information.

[0045] In this embodiment, the learning device 10 obtains training viewer information related to the purchase of the training viewer by tracking the behavior of the training viewer. The training viewer information is equivalent to an indicator indicating the advertising effect of the training video advertisement. The details of the method of obtaining the training purchase information will be described later. For example, the training viewer information tracks whether the training viewer has purchased, etc., and obtains the training viewer information. Figure 3 , one piece of training viewer information is shown, but the learning device 10 may obtain a plurality of pieces of training viewer information.

[0046] exist Figure 3 In the example, the learning device 10 generates training data for learning the machine learning model M based on multiple training video information and one training viewer information. The details of the training data will be described later. For example, the learning device 10 can perform the same processing on each of multiple training video advertisements to generate training data in sequence. The learning device 10 learns the machine learning model M based on the training data. The details of the learning will be described later. The learning device 10 sends the learned machine learning model M to the inference device 20. The inference device 20 records the learned machine learning model M received from the learning device 10 in the storage unit 22.

[0047] For example, the estimation device 20 performs various analyses on the estimated video advertisements to obtain estimated video information in order to analyze the advertising effect of the estimated video advertisements that have been released. Figure 2 The training video advertisement is released in the same process as described above. For example, the estimated performer operates the estimated performer device 60 and uses the camera unit 66 to shoot himself to introduce the product or service. The server 30 releases the estimated video advertisement to the estimated viewer devices 70 of a plurality of unspecified estimated viewers. The details of the estimated video information will be described later, but the method of obtaining the estimated video information can be the same as the method of obtaining the training video information.

[0048] For example, the estimation device 20 inputs a plurality of estimated video information to the machine learning model M. The machine learning model M calculates a feature quantity (embedded expression) based on the plurality of estimated video information. The machine learning model M estimates the advertising effect of the estimated video advertisement based on the feature quantity. For example, the machine learning model M estimates the probability of the estimated viewer who watches the estimated video advertisement purchasing a product or service.

[0049] For example, the estimation device 20 obtains the estimation result output from the machine learning model M. The estimation device 20 can use the estimation result of the machine learning model M for any purpose. For example, the estimation device 20 can cause the display unit 25 to display the estimation result of the machine learning model M. The estimation device 20 can send the estimation result of the machine learning model M to the estimated performer device 60 to feed back the advertising effect of the estimated video advertisement to the estimated performer.

[0050] As described above, the learning device 10 of this embodiment learns the machine learning model M based on the training video information and the training purchase information, thereby creating a machine learning model M with high estimation accuracy of the advertising effect. The estimation device 20 can estimate the advertising effect of the estimated video advertisement with high accuracy by performing estimation based on the machine learning model M learned by the learning device 10. The details of this embodiment will be described below.

[0051] [3. Functions implemented by each of the learning device and the estimation device]

[0052] Figure 4 1 is a diagram showing an example of the functions implemented by each of the learning device 10 and the estimation device 20. Figure 4 In FIG. 1 , the functions implemented by the server 30 are also shown.

[0053] [3-1. Functions implemented by the learning device]

[0054] For example, the learning device 10 includes a data storage unit 100, a model storage unit 101, a training video information acquisition unit 102, a training purchase information acquisition unit 103, and a learning unit 104. The data storage unit 100 and the model storage unit 101 are implemented by the storage unit 12. The training video information acquisition unit 102, the training purchase information acquisition unit 103, and the learning unit 104 are implemented by the control unit 11.

[0055] [Data storage unit]

[0056] The data storage unit 100 stores data necessary for learning the machine learning model M. For example, the data storage unit 100 stores a training database DB1 .

[0057] Figure 51 is a diagram showing an example of a training database DB1. The training database DB1 is a database storing training data for the machine learning model M to learn. In the present embodiment, a unit of single data used when the machine learning model M learns is referred to as training data. A set of training data storing a plurality of training data is referred to as training database DB1. The machine learning model M learns all or part of the plurality of training data stored in the training database DB1.

[0058] exist Figure 5 In the example, "No" is the record number of the training database DB1. In this embodiment, the case where one training data is generated from one training video advertisement is taken as an example. That is, the training video advertisement and the training data are in a one-to-one relationship. Therefore, Figure 5 The "No" of can also be regarded as information that can identify the training video advertisement. In addition, a plurality of training data can be generated from one training video advertisement, and a single training data can be generated from a plurality of training video advertisements.

[0059] For example, the training data includes an input part that is input to the machine learning model M during learning, and an output part that should be output from the machine learning model M during learning. The input part of the training data is in the same format as the input data input to the machine learning model M during estimation. The input part of the training data is sometimes referred to as an explanatory variable. In the present embodiment, the input part of the training data is a plurality of training video information. The output part of the training data is in the same format as the output data output from the machine learning model M during estimation. The output part of the training data is equivalent to the correct answer during learning. The output part of the training data is sometimes referred to as a target variable. In the present embodiment, the output part of the training data is taken as an example of training purchase information.

[0060] In addition, the data storage unit 100 can store arbitrary data. The data stored in the data storage unit 100 is not limited to the training database DB1. For example, the data storage unit 100 can store a learning program representing a series of processes in the learning of the machine learning model M. The data storage unit 100 can store the video data of the training video advertisement downloaded by the learning device 10 from the server 30. The data storage unit 100 can store data attached to the video data of the training video advertisement (for example, the category of the product or service introduced in the training video advertisement, the description input in advance, the information of the coupon, or a combination thereof).

[0061] For example, the data storage unit 100 may store a natural language processing program described later. The data storage unit 100 may store a dictionary database storing keywords used in natural language processing. The data storage unit 100 may store a machine learning model M for natural language processing. The data storage unit 100 may store an image analysis processing program described later. The data storage unit 100 may store a machine learning model M for image analysis processing.

[0062] [Model storage unit]

[0063] The model storage unit 101 stores a machine learning model M. The machine learning model M includes a program portion that represents a process such as calculation of a feature value (embedded representation) of input data written to the machine learning model M, and a parameter portion (for example, a weighting coefficient and a bias) referenced by the program portion. The parameter portion of the machine learning model M is changed by the learning of the learning unit 104. The model storage unit 101 stores a machine learning model M whose parameter portion is an initial value. The machine learning model M whose parameter portion is an initial value is a machine learning model M before learning. The parameter portion of the initial value is overwritten by the processing of the learning unit 104.

[0064] The machine learning model M is a model using a machine learning method. Machine learning itself can use various well-known methods. For example, the machine learning model M may be a model using any of supervised learning, semi-supervised learning, or unsupervised learning. In the present embodiment, the case where the machine learning model M is a neural network is taken as an example, but the machine learning model M may also be a model using other methods, not limited to neural networks. For example, the machine learning model M may be a model using linear regression, a model using logistic regression, a model using a support vector machine, or a model using a decision tree.

[0065] [Training video information acquisition department]

[0066] The training video information acquisition unit 102 acquires training video information acquired by analyzing a video advertisement for training, that is, a training video advertisement. In this embodiment, a case where a video advertisement that has been released in the past is equivalent to a training video advertisement is taken as an example, but an unreleased video advertisement may also be equivalent to a training video advertisement. In addition, a case where a video of a live broadcast service is equivalent to a training video advertisement is taken as an example, but a training video advertisement is not limited to a video of a live broadcast service as long as it is a video advertisement for learning. For example, a website advertisement, a TV advertisement, an outdoor advertisement, a movie advertisement, or other advertisements may also be equivalent to a training video advertisement.

[0067] Analysis of training video ads is a process of extracting features related to training video ads. For example, sound analysis processing, natural language processing, image analysis processing, or a combination thereof is equivalent to analysis of training video ads. The training video information represents features extracted by analysis of training video ads. Next, an example of features acquired by the training video information acquisition unit 102 as training video information is described. In addition, the training video information acquisition unit 102 only needs to acquire features that the person who created the machine learning model M (for example, the operator of the live broadcast service) believes to have a certain causal relationship with the advertising effect as training video information. The features acquired by the training video information acquisition unit 102 as training video information are not limited to the examples in this embodiment.

[0068] In this embodiment, the training video information acquisition unit 102 acquires training video information by performing natural language processing on the training text obtained by analyzing the sound of the training video advertisement. For example, the sound of the training video advertisement is the sound of the training performer. The sound of the training video advertisement can be any sound included in the training video advertisement, and is not limited to the sound of the training performer. For example, the sound of the training video advertisement can be the sound of a staff member other than the training performer, the audience watching the training video advertisement on site, the third party interviewed by the training performer, or a pre-prepared artificial sound.

[0069] The training text is a text generated from the sound of the training video advertisement. For example, the training video information acquisition unit 102 generates the training text by performing sound analysis processing on the sound of the training video advertisement. The sound analysis processing algorithm itself can be a well-known algorithm. For example, the training information acquisition unit generates the training text from the sound of the training video advertisement based on a method using a hidden Markov model, a method using a preset waveform pattern, a machine learning method such as a neural network, or other methods. The training text can be generated by other computers other than the learning device 10. In the case, the training video information acquisition unit 102 obtains the training text generated by other computers from the other computers, and further other computers, or information storage media.

[0070] For example, the training video information acquisition unit 102 acquires the training video information by performing natural language processing on the training text. Natural language processing is a process used by a computer to recognize the language used by humans. Natural language processing can be performed by other computers other than the learning device 10. In the above case, the training video information acquisition unit 102 acquires the training text generated by the natural language processing performed by the other computer from the other computer, and further another computer, or an information storage medium.

[0071] For example, natural language processing may be at least one of an emotion analysis process for analyzing the emotions of the performer of the training video advertisement, i.e., the training performer; a response detection process for detecting the response of the performer of the training video advertisement, i.e., the training performer; a description detection process for detecting descriptions related to the goods or services introduced in the training video advertisement; and a sales promotion detection process for detecting sales promotion related to the training video advertisement.

[0072] For example, the training video information acquisition unit 102 acquires training video information related to the training performer's emotions by performing sentiment analysis processing on the training text. The sentiment analysis processing method may be a well-known method. For example, the training video information acquisition unit 102 may analyze the training performer's emotions based on a rule method for determining whether a keyword representing emotions is included in the training text. In addition, for example, the training video information acquisition unit 102 may analyze the training performer's emotions based on a machine learning method for calculating a feature amount of the training text and outputting a classification result of the emotions.

[0073] For example, the training video information acquisition unit 102 acquires training video information indicating the analysis result of the training performer's emotion through the emotion analysis process. When analyzing the four emotions of joy, anger, sorrow, and happiness, the training video information acquisition unit 102 acquires training video information indicating the degree of the emotion of "joy" (for example, the number of occurrences or the frequency of occurrence of the keyword indicating "joy"), the degree of the emotion of "anger" (for example, the number of occurrences or the frequency of occurrence of the keyword indicating "anger"), the degree of the emotion of "sorrow" (for example, the number of occurrences or the frequency of occurrence of the keyword indicating "sorrow"), and the degree of the emotion of "happiness" (for example, the number of occurrences or the frequency of occurrence of the keyword indicating "happiness"). The training video information acquisition unit 102 may analyze other emotions besides joy, anger, sorrow, and happiness.

[0074] For example, the training video information acquisition unit 102 acquires the training video information related to the response of the training performer by performing response detection processing on the training text. The method of the response detection processing may be a well-known method. For example, the training video information acquisition unit 102 may detect the response of the training performer based on a rule method of determining whether a keyword representing the response is included in the training text. In addition, for example, the training video information acquisition unit 102 may detect the response of the training performer based on a machine learning method of calculating a feature amount of the training text and outputting a classification result of the response.

[0075] In addition, the response detected by the response detection process can also be called a reaction. The response detected by the response detection process can be the response of other training performers to the speech of a certain training performer. The response in the above case can also be called a conversation between training performers. The training video information acquisition unit 102 acquires training video information indicating the detection result of the response of the training performer through the response detection process. For example, when a speech such as "Is that so" is included in the training text, the training video information acquisition unit 102 detects the response of other training performers to the speech of a certain training performer.

[0076] In addition, for example, the response detected by the response detection process may be the response of the training performer to the comment input by the training viewer. When a speech such as "Thank you for the comment" is included in the training text, the training video information acquisition unit 102 detects the response of the training performer to the comment input by the training viewer. The training video information acquisition unit 102 may acquire training video information indicating the number of responses detected by the response detection process, the frequency of responses, the type of responses (for example, what is the response to, or the specific content of the response, etc.), or a combination thereof.

[0077] For example, the training video information acquisition unit 102 acquires training video information related to the explanation of the training performer by performing explanation detection processing on the training text. The explanation detection processing method may be a well-known method. For example, the training video information acquisition unit 102 may detect the response of the training performer based on a rule method for determining whether a keyword representing the explanation of a product or service is included in the training text. In addition, for example, the training video information acquisition unit 102 may detect the explanation of the training performer based on a machine learning method for calculating a feature amount of the training text and outputting a classification result of the explanation of the product or service.

[0078] For example, the training video information acquisition unit 102 acquires training video information indicating the detection result of the training performer's explanation through the explanation detection process. When the training video advertisement introduces desserts, the training video information acquisition unit 102 counts the number of occurrences or the frequency of occurrences of keywords such as "chocolate" that represent the product itself in the training text. The training video information acquisition unit 102 counts the number of occurrences or the frequency of occurrences of keywords such as "delicious" that represent the quality of the product in the training text. The training video information acquisition unit 102 acquires training video information indicating these counting results.

[0079] For example, the sales promotion detection process may be a keyword detection process for detecting at least one of a keyword related to the price of a product or service introduced in the training video advertisement, a keyword related to the sales status of a product or service introduced in the training video advertisement, and a keyword related to the description associated with the training video advertisement. The sales promotion detection process may also be referred to as a process for detecting words related to the sales promotion of a product or service. The training video information acquisition unit 102 may also detect the response of the training performer based on a rule method for determining whether a keyword indicating the price of a product or service is included in the training text.

[0080] For example, the training video information acquisition unit 102 determines whether a keyword stored in a dictionary database storing keywords related to prices (e.g., "1,000 yen", "affordable", "discount", or "coupon") is included in the training text, and detects the keyword. The training video information acquisition unit 102 determines whether a keyword stored in a dictionary database storing keywords related to sales conditions (e.g., "buy" or "not much left") is included in the training text, and detects the keyword.

[0081] The explanation associated with the training video advertisement is an explanation related to the product or service introduced in the training video advertisement. For example, information that the training viewer can appropriately refer to from the screen SC may correspond to the explanation associated with the training video advertisement. Figure 2 In the example of , the detailed description that the training viewer can refer to by selecting button B1 is an example of description associated with the training video advertisement. The coupon information that the training viewer can refer to by selecting button B2 is an example of description associated with the training video advertisement.

[0082] For example, the training video information acquisition unit 102 determines whether a keyword stored in a dictionary database storing keywords related to descriptions of products or services introduced in the training video advertisement (e.g., "limited time only," "delicious," or "topic," etc.) is included in the training text, and detects the keyword. Based on the above-mentioned keyword detection result, the training video information acquisition unit 102 acquires training video information indicating the number of keyword detections or the detection frequency.

[0083] In addition, the sales promotion detection process is not limited to the above example. For example, the training video information acquisition unit 102 may detect sales promotion based on a machine learning method that calculates the feature amount of the training text and outputs an estimated result of sales promotion, rather than a rule method using keywords. In the model used in the machine learning method, various words related to the sales promotion of goods or services have been learned.

[0084] In addition, the training video information acquisition unit 102 can perform any natural language processing on the training text. The natural language processing performed by the training video information acquisition unit 102 is not limited to the example of this embodiment. For example, the training video information acquisition unit 102 can perform morpheme analysis, syntactic analysis, semantic analysis, feature extraction, or a combination thereof as natural language processing for the training text, and acquire training video information based on the execution result of the natural language processing. The training video information acquisition unit 102 can perform natural language processing using a machine learning method such as Transformer on the training text.

[0085] For example, the training video information acquisition unit 102 may acquire training video information related to the expression of the performer of the training video advertisement, that is, the training performer, acquired by analyzing the video of the training video advertisement. The video of the training video advertisement includes a plurality of frames (static images). The training video information acquisition unit 102 acquires the training video information by performing image analysis processing on each frame. The image analysis processing may be performed by another computer other than the learning device 10. In the case, the training video information acquisition unit 102 acquires the training video information generated by the image analysis processing performed by the other computer from the other computer, and further another computer, or an information storage medium.

[0086] The expression of the training performer can also be called the emotion of the training performer. Therefore, the training video information acquisition unit 102 can identify the same emotion as the emotion analysis process through image analysis processing. The method of analyzing a person's expression through image analysis processing can be a well-known method. For example, the training video information acquisition unit 102 can analyze the expression of the training performer in the training video advertisement based on template matching using a template image that represents a person's basic expression. The training video information acquisition unit 102 can analyze the expression of the training performer in the training video advertisement based on the configuration pattern of feature points detected from the video of the training video advertisement. The training video information acquisition unit 102 can analyze the expression of the training performer in the training video advertisement based on a machine learning model that has learned various human expressions.

[0087] In addition, the training video information acquisition unit 102 can analyze features other than the expression of the training performer through image analysis processing. For example, the training video information acquisition unit 102 can analyze the shooting technique, the brightness of the lighting, the posture of the training performer, the effects in the video, the content of the subtitles, or other features in the training video advertisement through image analysis processing. The training video information acquisition unit 102 can acquire training video information indicating these features.

[0088] In addition, in the present embodiment, the case where the training video information acquisition unit 102 acquires a plurality of training video information is taken as an example, but the training video information acquisition unit 102 only needs to acquire at least one training video information. The training video information acquisition unit 102 may acquire only one training video information. For example, the training video information acquisition unit 102 may only perform any one of the methods described above and acquire only one training video information.

[0089] [Training purchase information acquisition department]

[0090] The training purchase information acquisition unit 103 acquires training purchase information related to purchases by viewers of the training video advertisement, that is, the training viewers. The purchases by the training viewers are the purchases of the goods or services introduced in the training video advertisement. The training viewers sometimes make purchases while watching the training video advertisement, and sometimes make purchases after watching the training video advertisement. The training viewers sometimes make purchases not during the live broadcast of the training video advertisement, but after watching the live broadcast of the training video advertisement.

[0091] in the case of Figure 2 In the example of the training video advertisement, the training viewer sometimes selects button B1 on the screen SC and then makes a purchase from the page displayed by the training viewer device 50, and sometimes makes a purchase by an operation other than selecting button B1. For example, the training viewer sometimes selects a link to jump to a page of a product or service introduced in the training video advertisement after temporarily closing the screen SC and then makes a purchase from the page displayed by the training viewer device 50.

[0092] In this embodiment, the distribution unit 301 of the server 30 tracks the behavior of the training viewer and generates the training purchase information. The training purchase information acquisition unit 103 acquires the training purchase information from the server 30. The training purchase information may be generated by a computer other than the server 30. The training purchase information acquisition unit 103 may acquire the training purchase information from the other computer, another computer, or an information storage medium. The training purchase information acquisition unit 103 may acquire the training purchase information by generating the training purchase information itself.

[0093] In this embodiment, the case where the training purchase information acquisition unit 103 acquires training purchase information indicating the ratio of training viewers who have completed a purchase among the training viewers who have watched the training video advertisement (the total number of training viewers who have completed a purchase / the total number of training viewers) is taken as an example. The training purchase information can be any information related to the purchase of the training viewers, and is not limited to the example of this embodiment. For example, the training purchase information acquisition unit 103 can acquire training purchase information related to at least one of whether the training viewers have purchased and information related to the sale of the goods or services introduced in the training video advertisement.

[0094] The training purchase information related to whether or not the training viewers have purchased is information generated based on whether or not the training viewers have purchased. The ratio is also an example of the training purchase information related to whether or not the training viewers have purchased, but the training purchase information related to whether or not the training viewers have purchased can also be other information. For example, the training purchase information related to whether or not the training viewers have purchased can be information indicating whether or not the individual training viewers have purchased, the total number of training viewers who have completed the purchase, the total number of people per unit time, or a combination thereof. The unit time can be any length of time, for example, 1 minute or 5 minutes. For example, the server 30 generates the training purchase information by collecting this information.

[0095] The information related to the sales of the goods or services introduced in the training video advertisement is information generated based on the sales of the goods or services. For example, the information related to the sales may be the sales number (sold number), sales amount, sales number per unit time, sales amount per unit time, or a combination thereof of the goods or services. For example, the server 30 generates training purchase information by collecting such information.

[0096] [Learning Department]

[0097] The learning unit 104 learns a machine learning model M for estimating the advertising effect of the estimated video advertisement from the estimated video information obtained by analyzing the estimated video advertisement, that is, the estimated video advertisement, based on the training video information and the training purchase information. The learning unit 104 generates training data based on the training video information and the training purchase information. In this embodiment, the learning unit 104 generates training data in such a way that the training video information becomes the input part of the training data and the training purchase information becomes the output part of the training data.

[0098] In addition, the learning unit 104 may not directly use the training video information as the input part of the training data, but may perform some processing such as standardization or collection processing on the training video information and use it as the input part of the training data. Similarly, the learning unit 104 may not directly use the training purchase information as the output part of the training data, but may perform some processing such as standardization or collection processing on the training purchase information and use it as the output part of the training data.

[0099] For example, the learning unit 104 learns the machine learning model M based on the training data. In this embodiment, the learning unit 104 generates training data based on each of the multiple training video advertisements and stores them in the training database DB1. The learning unit 104 learns the machine learning model M based on all or part of the multiple training data stored in the training database DB1. A part of the training data can be used to verify the learned machine learning model M. In addition, the learning unit 104 can generate training data based on one training video advertisement instead of multiple training video advertisements.

[0100] Learning is a process of adjusting the parameter part of the machine learning model M. The learning method may be a well-known method used in well-known machine learning. For example, the learning unit 104 performs learning of the machine learning model M in a manner of outputting the output part of the training data when the input part of the training data is input based on a learning method such as error back propagation or gradient descent. The learning unit 104 calculates the loss based on a well-known loss function and performs learning until the loss is reduced to a certain extent. The learning unit 104 can complete the learning by performing learning based on a certain amount of training data.

[0101] For example, when the training video information is obtained by performing natural language processing on the training text, the same data is input to the machine learning model M during estimation. Therefore, the learning unit 104 performs learning of the machine learning model M for estimating the advertising effect from the estimated video information obtained by performing the same processing as the natural language processing on the training text on the estimated text obtained by analyzing the sound of the estimated video advertisement.

[0102] The estimated text is a text generated from the sound of the estimated video advertisement. In the present embodiment, since the processing during estimation is performed by the estimation device 20, the generation of the estimated text is performed by the estimation device 20. The method for generating the estimated text may be the same as the method for generating the training text. The details of the method for generating the estimated text can be replaced by "training" in the description of the method for generating the training text with "estimation". The natural language processing for the estimated text can also be replaced by "training" in the description of the natural language processing for the training text with "estimation".

[0103] For example, when the training video information is obtained by analyzing the video of the training video advertisement, the same data is input to the machine learning model M during estimation. Therefore, the learning unit 104 performs learning of the machine learning model M for estimating the advertising effect from the estimated video information related to the expression of the performer of the estimated video advertisement, that is, the estimated performer, obtained by analyzing the video of the estimated video advertisement.

[0104] The expression of the estimated performer is an expression identified by analyzing the video of the estimated video advertisement. In the present embodiment, since the processing during estimation is performed by the estimation device 20, the analysis of the expression of the estimated performer is performed by the estimation device 20. The analysis method of the expression of the estimated performer can be the same as the analysis method of the expression of the trained performer. The details of the analysis method of the expression of the estimated performer can be replaced by "estimation" in the description of the analysis method of the expression of the trained performer.

[0105] For example, when at least one of information on whether or not the estimated video advertisement viewer, that is, the estimated viewer has purchased and information related to the sale of the product or service introduced in the estimated video advertisement is obtained as training purchase information, the learning unit 104 performs learning of the machine learning model M that estimates the at least one advertisement effect. The learning unit 104 obtains the output part of the training data based on the at least one. The learning method based on the training data is as described above.

[0106] [3-2. Functions implemented by the estimation device]

[0107] For example, the estimation device 20 includes a data storage unit 200, a model storage unit 201, an estimated video information acquisition unit 202, and an estimation unit 203. The data storage unit 200 and the model storage unit 201 are implemented by the storage unit 22. The estimated video information acquisition unit 202 and the estimation unit 203 are implemented by the control unit 21.

[0108] [Data storage unit]

[0109] The data storage unit 200 stores data necessary for estimation based on the machine learning model M. For example, the data storage unit 200 stores an estimation database DB2.

[0110] Figure 6 : is a diagram showing an example of an estimation database DB2. The estimation database DB2 is a database storing data related to estimated video advertisements for enabling the machine learning model M to estimate the advertising effect. In the present embodiment, the case where the video data of each of a plurality of estimated video advertisements is stored in the estimation database DB2 is taken as an example, but the estimated video information generated by the estimated video information acquisition unit 202 from each of the plurality of estimated video advertisements may also be stored in the estimation database DB2. In the estimation database DB2, data representing the advertising effect estimated for the estimated video advertisement may be stored. In addition, the video data also includes sound information.

[0111] For example, the estimation device 20 downloads the video data of all or part of the video advertisement stored in the video advertisement database DB3 described later from the server 30 as the video data of the estimated video advertisement. The estimation device 20 stores the downloaded video data in the estimation database DB2. The estimation device 20 may obtain the video data of the estimated video advertisement from other computers or information storage media other than the server 30.

[0112] In addition, the data storage unit 200 can store any data. The data stored in the data storage unit 200 is not limited to the estimated database DB2. For example, the data storage unit 200 or the data storage unit 200 may store data attached to the video data of the estimated video advertisement (for example, the category of the product or service introduced in the estimated video advertisement, the description input in advance, the information of the coupon, or a combination thereof).

[0113] For example, the data storage unit 200 may store a program for natural language processing. The data storage unit 200 may store a dictionary database storing keywords for natural language processing. The data storage unit 200 may store a machine learning model M for natural language processing. The data storage unit 200 may store a program for image analysis processing. The data storage unit 200 may store a machine learning model M for image analysis processing.

[0114] [Model storage unit]

[0115] The model storage unit 201 stores a machine learning model M that performs learning based on training video information obtained by analyzing a video advertisement for training, i.e., a training video advertisement, and training purchase information related to purchases by viewers of the training video advertisement, i.e., training viewers. The estimation device 20 acquires the machine learning model M learned by the learning unit 104 from the learning device 10. The estimation device 20 records the machine learning model M acquired from the learning device 10 in the model storage unit 201.

[0116] [Estimated video information acquisition unit]

[0117] The estimated video information acquisition unit 202 acquires estimated video information acquired by analyzing the estimated video advertisement, that is, the estimated video advertisement. For example, the estimated video information acquisition unit 202 acquires all or part of the estimated video information of the estimated video advertisement whose video data is stored in the estimation database DB2. The estimated video advertisement whose estimated video information the estimated video information acquisition unit 202 acquires may be specified by the person operating the estimation device 20 or may be determined by a preset method.

[0118] In this embodiment, a case where a video advertisement that has been released in the past is equivalent to an estimated video advertisement is taken as an example, but an unreleased video advertisement may also be equivalent to an estimated video advertisement. In addition, a case where a video of a live broadcast service is equivalent to an estimated video advertisement is taken as an example, but an estimated video advertisement can be any video advertisement that is the object of estimation of the advertising effect, and is not limited to a video of a live broadcast service. For example, website advertisements, TV advertisements, outdoor advertisements, movie advertisements, or other advertisements may also be equivalent to estimated video advertisements.

[0119] The analysis of the estimated video advertisement is a process of extracting features related to the estimated video advertisement. For example, sound analysis processing, natural language processing, image analysis processing, or a combination thereof is equivalent to the analysis of the estimated video advertisement. The estimated video information represents the features extracted by the analysis of the estimated video advertisement. The estimated video information acquisition unit 202 can acquire the estimated video information based on the same acquisition method as the training video information. Therefore, the details of the process of the estimated video information acquisition unit 202 acquiring the estimated video information can replace the "training" in the description of the training video information acquisition unit 102 with "estimated".

[0120] For example, the estimated video information acquisition unit 202 can acquire estimated video information acquired by performing the same natural language processing as that performed on the training text on the estimated text acquired by analyzing the sound of the estimated video advertisement. The processing may be at least one of the sentiment analysis processing, response detection processing, explanation detection processing, and sales promotion detection processing as described above. The details can also replace "training" in the description of the training video information acquisition unit 102 with "estimation". The point that any natural language processing can be performed on the estimated text and that the natural language processing can be performed by other computers other than the estimation device 20 are the same as the natural language processing on the training text.

[0121] For example, the estimated video information acquisition unit 202 can acquire estimated video information related to the expression of the performer of the estimated video advertisement, that is, the estimated performer, acquired by analyzing the video of the estimated video advertisement. The video of the estimated video advertisement includes a plurality of frames (still images). The estimated video information acquisition unit 202 acquires the estimated video information by performing image analysis processing on each frame. The point that any image analysis processing can be performed on the video of the estimated video advertisement and the point that the image analysis processing can be performed by other computers other than the estimation device 20 are the same as the image analysis processing on the video of the training video advertisement.

[0122] [Estimation Department]

[0123] The estimation unit 203 estimates the advertising effect of the estimated video advertisement based on the estimated video information and the machine learning model M. For example, the estimation unit 203 inputs the estimated video information to the machine learning model M. The estimation unit 203 may not directly input the estimated video information, but may input the estimated video information after performing a certain process such as standardization or collection processing to the machine learning model M. The machine learning model M calculates the feature quantity of the estimated video information input to itself based on the parameter part adjusted by learning, and outputs the estimation result corresponding to the feature quantity. The internal processing of the machine learning model M may be a well-known processing adopted in a well-known machine learning method.

[0124] In this embodiment, since the output part of the training data is training purchase information indicating the proportion of training viewers who completed the purchase after watching the training video advertisement, the estimated result output from the machine learning model M is also the same information. For example, the machine learning model M outputs estimated purchase information indicating the proportion of estimated viewers who completed the purchase (the proportion of estimated viewers who expect to purchase, that is, the probability of purchase) among the estimated viewers who watched the estimated video advertisement as the estimated result. The estimation unit 203 obtains the estimated result output from the machine learning model M. The estimation unit 203 can store the estimated result of the machine learning model M in the estimation database DB2.

[0125] In addition, in the case where the training purchase information is the other information, the inferred purchase information becomes the same information as the other information. For example, the machine learning model M can output information indicating whether the inferred viewer has made a purchase, the total number of inferred viewers who are inferred to have completed the purchase, the total number of people per unit time, or a combination thereof as an inferred result. In addition, for example, the machine learning model M can output information related to sales inferred for the goods or services introduced in the inferred video advertisement as an inferred result. The information related to the sales may be the inferred number of sales (number of sales) of the goods or services, the sales amount, the number of sales per unit time, the sales amount per unit time, or a combination thereof.

[0126] [3-3. Functions implemented by the server]

[0127] For example, the server 30 includes a data storage unit 300 and a distribution unit 301. The data storage unit 300 is implemented by the storage unit 32. The distribution unit 301 is implemented by the control unit 31.

[0128] [Data storage unit]

[0129] The data storage unit 300 stores data required for the live broadcast service. For example, the data storage unit 300 stores a video advertisement database DB3.

[0130] Figure 7 1 is a diagram showing an example of a video advertisement database DB3. The video advertisement database DB3 is a database storing data related to video advertisements in a live broadcast service. In the present embodiment, the video advertisements storing data in the video advertisement database DB3 may be at least one of training video advertisements and estimated video advertisements. The video advertisements storing data in the video advertisement database DB3 may be used as both training video advertisements and estimated video advertisements.

[0131] For example, in the video advertisement database DB3, a video advertisement ID (Identity: identification information) that can identify a video advertisement, performer data related to the performer, product / service data related to the product or service introduced in the video advertisement, coupon data related to the coupon of the product or service, video data of the video advertisement, and comment data related to the viewer's comment are stored. Not only the video advertisements during the release period, but also the archives of video advertisements released in the past can be stored in the video advertisement database DB3. In addition, for example, data related to reaction tags can be stored in the video advertisement database DB3. The reaction tag is a function that allows the viewer to react by selecting at least one of a plurality of tags (for example, not understanding or agreeing, etc.) at any time while watching the video advertisement. The reaction tag is an example of a user's response (reaction).

[0132] For example, when a performer performs an operation for starting the release of a new video advertisement, the publishing unit 301 issues a video advertisement ID of the video advertisement. The publishing unit 301 associates the video advertisement ID with the performer data and stores the data such as the performer data in the video advertisement database DB3. The performer data may represent any information related to the performer, for example, the performer's account, name, gender, age, other characteristics, or a combination thereof. The product / service data may represent any information related to a product or service, for example, the category of the product or service, the title of the product or service, the description of the product or service, a link to the details page of the product or service, or a combination thereof.

[0133] The coupon data may represent any information related to the coupon of the product or service, for example, the discount amount, the discount rate, the total number of coupons, the applicable conditions of the coupon, the conditions for obtaining the coupon, the acquisition status of the coupon, or a combination thereof. The video data is generated by recording the video advertisement by the publishing unit 301. The video data may be in any data format. The video data may only represent the image without including the sound. The video data may include basic information of the video advertisement such as the playing time of the video.

[0134] The comment data is data indicating the content of the comment. For example, the comment data indicates the account of the viewer who posted the comment, the text indicating the content of the comment, the date of posting the comment, or a combination thereof. The viewer may give a reaction other than the comment. In the case described above, the data indicating the viewer's reaction is stored in the video advertisement database DB3. In the video advertisement database DB3, data related to the sales status of the goods or services introduced in the video advertisement may be stored.

[0135] In addition, the data storage unit 300 can store arbitrary data. The data stored in the data storage unit 300 is not limited to the video advertisement database DB3. For example, the data storage unit 300 can store data indicating the purchase status of the goods or services introduced in each video advertisement. The data can indicate the account of the viewer who purchased the goods or services, the sales volume of the goods or services, the sales volume, the sales volume per unit time, the sales volume per unit time, or a combination thereof. These data can be stored in the video advertisement database DB3. In the video advertisement database DB3, data related to the viewers who watched the video advertisement can be stored regardless of whether or not comments are posted. Based on the data, information such as how many viewers have watched the video advertisement and how long each viewer watched the video advertisement can be analyzed.

[0136] [Release Department]

[0137] The publishing unit 301 publishes video advertisements to viewers. The processing performed by the publishing unit 301 may be the same as the processing adopted in the well-known live broadcast service. The publishing unit 301 publishes video advertisements to each of the training viewer device 50 and the estimated viewer device 70. The publishing unit 301 accepts input of comments from each of the training viewer device 50 and the estimated viewer device 70. The publishing unit 301 updates the video advertisement database DB3 based on the communication content with each of the training performer device 40, the training viewer device 50, the estimated performer device 60, and the estimated viewer device 70.

[0138] [4. Processing performed by each of the learning device and the estimation device]

[0139] Figure 8 It is a diagram showing an example of processing executed by the learning device 10 . Figure 8 The processing is performed by the control unit 11 according to the program stored in the storage unit 12. Figure 8 When processing, the learning device 10 pre-downloads the video data of the training video advertisement from the server 30. Figure 8 As shown, the learning device 10 obtains training video information by analyzing the training video advertisement (S10).

[0140] Fig. 91 is a diagram showing an example of the processing performed in S10. The learning device 10 captures the sound of the training video advertisement (S100). The learning device 10 converts the sound into text and obtains the training text (S101). The learning device 10 performs natural language processing on the training text (S102). The learning device 10 analyzes the video of the training video advertisement and analyzes the expression of the training performer (S103). The learning device 10 obtains the training video information based on each analysis result (S104). The details of each of the processing of S100 to S104 are as described as the processing of the training video information acquisition unit 102.

[0141] Return to Figure 8 The learning device 10 obtains the training purchase information (S11). The details of the processing of S11 are as described as the processing of the training purchase information acquisition unit 103. The learning device 10 pre-downloads the data required for obtaining the training purchase information from the server 30. The learning device 10 learns the machine learning model M based on the training video information and the training purchase information (S12). The details of the processing of S12 are as described as the processing of the learning unit 104. The learning device 10 sends the learned machine learning model M to the inference device 20 (S13). Figure 8 The processing is finished.

[0142] Fig.10 It is a diagram showing an example of processing executed by the estimation device 20. Fig.10 The processing is executed by the control unit 21 according to the program stored in the storage unit 22. Fig.10 When processing, the estimation device 20 downloads the video data of the estimated video advertisement from the server 30 in advance. Fig.10 As shown, the estimation device 20 acquires estimated video information by analyzing the estimated video advertisement ( S20 ).

[0143] Fig.11 : is a diagram showing an example of the processing performed in S20. The inference device 20 captures the sound of the inferred video advertisement (S200). The inference device 20 converts the sound into text and obtains the inferred text (S201). The inference device 20 performs natural language processing on the inferred text (S202). The inference device 20 analyzes the video of the inferred video advertisement and analyzes the expression of the inferred performer (S203). The inference device 20 obtains the inferred video information based on each analysis result (S204). The details of each of the processing of S200 to S204 are as described as the processing of the inferred video information acquisition unit 202.

[0144] Return to Fig.10 The estimation device 20 estimates the machine learning model M based on the estimated video information and the machine learning model M (S21). Fig.10The processing of S21 is completed. The details of the processing of S21 are as described as the processing of the estimation unit 203.

[0145] [5. Summary of implementation methods]

[0146] The learning device 10 of the present embodiment learns the machine learning model M based on the training video information and the training purchase information. The learning device 10 allows the machine learning model M to learn the training video information and the purchase of the training audience, so that the machine learning model M can estimate the advertising effect from the characteristics of the estimated video advertisement, and thus can create a machine learning model M with high estimation accuracy of the advertising effect. For example, the learning device 10 can allow the machine learning model M to learn the characteristics of the training video advertisement itself by allowing the machine learning model M to learn the training video information that has a causal relationship with the advertising effect. As a result, the machine learning model M can estimate the advertising effect from the characteristics of the estimated video advertisement itself, so that the estimation accuracy of the advertising effect is improved.

[0147] In addition, the learning device 10 obtains training video information obtained by performing natural language processing on the training text. The learning device 10 performs learning of a machine learning model M that estimates the advertising effect from the estimated video information obtained by performing the same processing as natural language processing on the estimated text. The learning device 10 causes the machine learning model M to learn the training video information analyzed by natural language processing, so that the machine learning model M can estimate the advertising effect from the features of the estimated text that is the text of the sound of the estimated video advertisement. Therefore, the learning device 10 can create a machine learning model M with a high estimation accuracy of the advertising effect. For example, by converting the sound into text by the learning device 10, analysis based on various features becomes easy.

[0148] In addition, the natural language processing is at least one of sentiment analysis processing, response detection processing, explanation detection processing, and sales promotion detection processing. For example, the learning device 10 can enable the machine learning model M to learn the causal relationship between the emotions of the training performers and the purchases of the training viewers through sentiment analysis processing. The learning device 10 can enable the machine learning model M to learn the causal relationship between the responses of the training performers and the purchases of the training viewers through response detection processing. The learning device 10 can enable the machine learning model M to learn the causal relationship between the explanations of the training performers and the purchases of the training viewers through explanation detection processing. The learning device 10 can enable the machine learning model M to learn the sales promotion in the training video advertisement and the causal relationship between the purchases of the training viewers through sales promotion detection processing. As a result, the learning device 10 can create a machine learning model M with a higher estimation accuracy of the advertising effect.

[0149] In addition, the sales promotion detection process is a keyword detection process for detecting at least one of the keywords related to the price of the goods or services introduced in the training video advertisement, the keywords related to the sales status of the goods or services introduced in the training video advertisement, and the keywords related to the description associated with the training video advertisement. The learning device 10 can detect sales promotion that can be effectively used in estimating the advertising effect through the keywords. As a result, the learning device 10 can improve the estimation accuracy of the machine learning model M that estimates the advertising effect from the sales promotion in the video advertisement. For example, the learning device 10 can detect keywords such as the coupon acquisition rate as keywords related to the price, and perform analysis including parameters other than the words of the goods or services themselves introduced in the video advertisement to obtain more extensive information.

[0150] In addition, the learning device 10 obtains training video information related to the expression of the performer of the training video advertisement, that is, the training performer, obtained by analyzing the video of the training video advertisement. The learning device 10 performs learning of a machine learning model M that estimates the advertising effect from the estimated video information related to the expression of the performer of the estimated video advertisement, that is, the estimated performer, obtained by analyzing the video of the estimated video advertisement. The learning device 10 enables the machine learning model M to learn the causal relationship between the expression of the training performer and the purchase of the training audience. As a result, the learning device 10 can create a machine learning model M with a high estimation accuracy of the advertising effect. For example, the learning device 10 can analyze the characteristics of each training performer that cannot be analyzed in natural language processing.

[0151] In addition, the learning device 10 obtains training purchase information related to at least one of the information related to whether the training viewer has purchased and the sales of the goods or services introduced in the training video advertisement. The learning device 10 performs learning of a machine learning model M that estimates the viewer of the estimated video advertisement, that is, the estimated viewer has purchased and at least one of the information related to the sales of the goods or services introduced in the estimated video advertisement as the advertising effect. The learning device 10 enables the machine learning model M to learn the training video information and the causal relationship between the at least one. As a result, the learning device 10 can create a machine learning model M with a higher estimation accuracy of the advertising effect. For example, because the unit price varies depending on the goods or services, if only the parameter of sales is analyzed, it is possible that the goods with a higher unit price will play an advantageous role. In this regard, the learning device 10 can create a more accurate machine learning model M by analyzing various features such as the number of sales.

[0152] The estimation device 20 of this embodiment estimates the advertising effect of the estimated video advertisement based on the estimated video information and the machine learning model M. The estimation device 20 can estimate the advertising effect from the characteristics of the estimated video advertisement itself, thereby improving the estimation accuracy of the advertising effect.

[0153] [6. Variations]

[0154] In addition, the present disclosure is not limited to the above-described embodiment, and the present disclosure can be appropriately modified within the scope of the present disclosure.

[0155] Fig.12 FIG. 1 is a diagram showing an example of the function implemented by the variation. Fig.12 As shown, in the modified example described later, the learning device 10 includes a contribution calculation unit 105 and a training viewer information acquisition unit 106. The contribution calculation unit 105 and the training viewer information acquisition unit 106 are implemented by the control unit 11. The estimation device 20 includes an estimated viewer information acquisition unit 204. The estimated viewer information acquisition unit 204 is implemented by the control unit 21.

[0156] [6-1. Modification 1]

[0157] For example, as described in the embodiment, the training video information acquisition unit 102 may acquire a plurality of training video information. The learning unit 104 may further perform learning of the machine learning model M based on the plurality of training video information. The above processing is as described in the embodiment. When the machine learning model M learns a plurality of training video information, there are both training video information that strongly contributes to the advertising effect and training video information that does not contribute to the advertising effect. Therefore, the contribution of each training video information to the advertising effect can be calculated.

[0158] The learning device 10 of variation 1 includes a contribution calculation unit 105. The contribution calculation unit 105 calculates the contribution of each of the multiple training video information to the advertising effect. The contribution is the degree to which the training video information contributes to the advertising effect. In other words, the contribution is the degree to which the training video information contributes to the estimation of the machine learning model M. The contribution is sometimes referred to as importance, score, influence, or other names. In variation 1, the case where the contribution is expressed as a numerical value between 0 and 100 is taken as an example, but the contribution can be expressed by a numerical value in other ranges, or can be expressed in words or symbols instead of numerical values. In variation 1, a higher numerical value represented by the contribution means that the training video information strongly contributes to the advertising effect.

[0159] The method for calculating the contribution may be a well-known method known in the field of machine learning. For example, the contribution calculation unit 105 calculates the contribution of each of the multiple training video information based on calculation methods such as random forest, gradient boosting, SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), L1 standardization, or L2 standardization. For example, when the value of a certain training video information changes, the contribution calculation unit 105 calculates the contribution of the training video information in a manner that the contribution of the training video information becomes higher when the estimated result of the machine learning model M changes significantly. For example, the contribution calculation unit 105 calculates the contribution of the training video information in a manner that the greater the increase in the numerical value represented by the estimated result of the machine learning model M relative to the increase in the numerical value of the certain training video information, the higher the contribution of the training video information.

[0160] The contribution calculation unit 105 records the contribution of each of the plurality of training video information in the data storage unit 100. The contribution can be used for any purpose. For example, the learning device 10 can cause the display unit 15 to display the contribution of each of the plurality of training video information. The learning device 10 can output data representing the contribution of each of the plurality of training video information to another computer or information storage medium. The learning unit 104 can select at least one training video information with a relatively high contribution among the plurality of training video information, and perform learning of a new machine learning model M based only on the selected at least one training video information.

[0161] The learning device 10 of variation 1 calculates the contribution of each of the plurality of training video information to the advertising effect. The learning device 10 can use the contribution of each of the plurality of training video information to identify the training video information that is helpful to the advertising effect. For example, the learning device 10 can prompt the creator of the machine learning model M, the performer, or other persons which of the plurality of training video information is helpful to the advertising effect. For example, if the performer refers to the contribution when creating future video advertisements, then the video advertisement with higher advertising effect can be created. For example, the learning device 10 can create a machine learning model M that can more accurately estimate the advertising effect by identifying which element in the training video information used in the learning of the machine learning model M effectively contributes to the advertising effect through the contribution.

[0162] [6-2. Modification 2]

[0163] For example, in the machine learning model M, not only the features of the training video advertisement but also the features of the training viewer can be learned. The learning device 10 of the variation 2 includes a training viewer information acquisition unit 106. The training viewer information acquisition unit 106 acquires the training viewer information related to the training viewer. For example, the training viewer information acquisition unit 106 acquires the training viewer information from the server 30. The training viewer information acquisition unit can generate the training viewer information based on the data acquired from the server 30. The training viewer information acquisition unit 106 can generate the training viewer information based on the data acquired from the training viewer device 50. The training viewer information acquisition unit 106 only needs to acquire the training viewer information of at least one training viewer, and can also acquire the training viewer information of each of a plurality of training viewers.

[0164] The training viewer information is information indicating a certain characteristic related to the training viewer. For example, the training viewer information is the action of the training viewer, the basic information of the training viewer, or a combination thereof. For example, the training viewer information acquisition unit 106 may acquire the training viewer information related to at least one of the age of the training viewer, the gender of the training viewer, the purchase history of the training viewer in the store, the training comments entered by the training viewer on the training video advertisement, the coupons related to the training video advertisement obtained by the training viewer, the viewing status of the training viewer on the training video advertisement, the access status of the training viewer, and the search of the training viewer.

[0165] The basic information of the training viewer, such as the age and gender of the training viewer, may be stored in the data storage unit 300 of the server 30, or may be input by the training viewer when watching the training video advertisement. The training viewer information acquisition unit 106 may acquire the training viewer information by acquiring the age and gender pre-stored in the data storage unit 300 of the server 30, or by acquiring the age and gender input by the training viewer.

[0166] The purchase history of the training audience in the store is the online or offline purchase history in the store. The purchase history can also be called the training audience's use history of the store. For example, the purchase history is the information of the store used by the training audience, the information of the goods or services purchased by the training audience, the date when the training audience used the store, or a combination thereof. The purchase history of the training audience in the store can be stored in the data storage unit 300 of the server 30, or in other computers or information storage media. The training audience information acquisition unit 106 can obtain the training audience information by obtaining the purchase history of the training audience in the store from the data storage unit 300, other computers, or information storage media.

[0167] For example, comment data representing training comments is stored in the video advertisement database DB3. The training viewer information acquisition unit 106 acquires the training viewer information by acquiring the training comments from the video advertisement database DB3. The training viewer information acquisition unit 106 may perform natural language processing on the training comments, and acquire the training viewer information based on the execution result of the natural language processing. The natural language processing may be the same as the processing on the training text described in the embodiment. For example, the training viewer information acquisition unit 106 may perform natural language processing such as sentiment analysis processing or response detection processing on the training comments.

[0168] For example, coupon data indicating the acquisition status of coupons related to the training video advertisement by the training viewer is stored in the video advertisement database DB3. The training viewer information acquisition unit 106 acquires the training viewer information by acquiring the acquisition status of coupons from the video advertisement database DB3. The acquisition status of coupons may be the total number of coupons acquired by the training viewer, the total number per unit time, the time series transition of the total number, or a combination thereof.

[0169] For example, the viewing status of the training video advertisement for training viewers is stored in the video advertisement database DB3. The training viewer information acquisition unit 106 acquires the training viewer information by acquiring the viewing status from the video advertisement database DB3. The viewing status of the training video advertisement may be any information related to the viewing of the training viewer, for example, the viewing time of the training viewer (the actual viewing time of the training viewer during the broadcasting time of the video advertisement), the ratio of the viewing time to the broadcasting time, the time period when the training viewer watches the video advertisement, or a combination thereof.

[0170] For example, the access status of the training audience is stored in the video advertisement database DB3. The training audience information acquisition unit 106 acquires the training audience information by acquiring the access status from the video advertisement database DB3. The access status may be the access status to any page, for example, the access status to the page of the live broadcast service, or the access status to the page of the e-commerce. The access status may be whether a certain page is visited, the ranking of the visit, the frequency of the visit, the time period of the visit, or a combination thereof.

[0171] For example, the search results of the training audience are stored in the video advertisement database DB3. The training audience information acquisition unit 106 acquires the training audience information by acquiring the access status from the video advertisement database DB3. The training audience search may be a search in a live broadcast service, e-commerce, or other services. The training audience search may be a query input during the search, a time period during which the search is performed, an action of the user after the search, or a combination thereof.

[0172] In addition, the training viewer information can be any feature related to the training viewer, and is not limited to the above examples. For example, the training viewer information acquisition unit 106 can acquire the time elapsed since the training viewer started using the live broadcast service, the goods or services purchased by the training viewer through the live broadcast service, the number of video advertisements watched by the training viewer in the live broadcast service, the annual income of the training viewer, the hobbies of the training viewer, or other information as the training viewer information. The training viewer information acquisition unit 106 can acquire a plurality of training viewer information.

[0173] The learning unit 104 of the second variation also learns a machine learning model M for estimating the advertising effect using the estimated viewer information related to the estimated viewer of the estimated video advertisement, that is, the estimated viewer, based on the training viewer information. For example, the learning unit 104 learns the machine learning model M for estimating the advertising effect using the estimated viewer information related to at least one of the estimated viewer information of the estimated viewer's age, the estimated viewer's gender, the estimated viewer's purchase history in the store, the estimated comment entered by the estimated viewer for the estimated video advertisement, the estimated viewer's acquisition of coupons related to the estimated video advertisement, the estimated viewer's viewing status of the estimated video advertisement, the estimated viewer's access status, and the estimated viewer's search. The learning unit 104 also performs the same learning of the machine learning model M when acquiring other estimated viewer information.

[0174] For example, the learning unit 104 generates the input part of the training data based on the training video information and the training viewer information. The learning unit 104 may not use the training viewer information directly as the input part of the training data, but may use the training viewer information as the input part of the training data after performing some processing such as standardization or collection processing. For example, the learning unit 104 may collect the training video information of each of a plurality of training viewers and use the collected results as the input part of the training data. The learning unit 104 may collect the ratio of the genders or the number of the training viewers who watch the training video advertisement and use the ratio or the number of the training viewers as the input part of the training data.

[0175] In addition, for example, the learning unit 104 may collect the distribution of the ages of the training viewers who watched the training video advertisement, and use the distribution as the input part of the training data. The learning unit 104 may collect the ratio or number of the training viewers who purchased a certain product, and use the ratio or number as the input part of the training data. The same is true for other training viewer information. The learning unit 104 may use it as the input part of the training data after performing the collection process, etc. The input part of the training data is different from the embodiment, but the learning unit 104 causes the machine learning model M to learn the training data as described in the embodiment.

[0176] The learning device 10 of the second variation also performs learning of a machine learning model M for estimating the advertising effect using the estimated viewer information related to the estimated viewer of the video advertisement, that is, the estimated viewer, based on the training viewer information. The learning device 10 can create a machine learning model M with a high estimation accuracy of the advertising effect by making the machine learning model M learn not only the training video information but also the training viewer information. For example, in the case where there is a causal relationship between the characteristics of the training viewer and the advertising effect (for example, in the case where the advertising effect is for a specific age group), the machine learning model M can estimate the advertising effect corresponding to the characteristics of the training viewer. For example, the learning device 10 can expand the scope of learning by making the machine learning model M learn the characteristics of each training viewer that cannot be learned by training purchase information alone.

[0177] In addition, the learning device 10 obtains training viewer information related to at least one of the age of the training viewer, the gender of the training viewer, the purchase history of the training viewer in the store, the training comments entered by the training viewer for the training video advertisement, the acquisition of coupons related to the training video advertisement by the training viewer, the viewing status of the training video advertisement by the training viewer, the access status of the training viewer, and the search of the training viewer. The learning device 10 also performs learning of the machine learning model M that estimates the advertising effect using the estimated viewer information related to the at least one. The learning device 10 can create a machine learning model M with a high estimation accuracy of the advertising effect by making the machine learning model M learn not only the training video information but also the at least one. For example, when there is a causal relationship between the at least one and the advertising effect, the machine learning model M can estimate the advertising effect corresponding to the characteristics of the training viewer.

[0178] [6-3. Modification 3]

[0179] For example, in variation 2, the training viewer information acquisition unit 106 can obtain the training viewer information related to the training comments by performing at least one of an emotion analysis process for analyzing the emotions of the training viewer, a comment classification process for classifying the training comments, and a keyword detection process for detecting keywords from the training comments.

[0180] The sentiment analysis process can be the same as the sentiment analysis process for the training text described in the embodiment. The training viewer information acquisition unit 106 performs sentiment analysis on the training comments, analyzes the emotions of the training viewer, such as joy, anger, sorrow, and happiness, and acquires the training viewer information indicating the analysis result. It is sufficient to replace the "training text" in the description of the sentiment analysis process for the training text in the embodiment with "training comment" and replace "training performer" with "training viewer". The point that various sentiment analysis processes can be performed on the training comments is the same as the sentiment analysis process for the training text.

[0181] The comment classification process is a process of classifying the content of the training comments. The comment classification process can use the method used when the meaning of words is classified by natural language processing. For example, the training viewer information acquisition unit 106 can classify the training comments based on a rule method for determining whether a keyword corresponding to each of a plurality of categories is included in the training comments. In addition, for example, the training viewer information acquisition unit 106 can classify the training comments based on a machine learning method that calculates the feature amount of the training comments and outputs the classification result. For example, in the comment classification process, the training comments are classified into either positive or negative. The classification of the training comments may be other classifications.

[0182] The keyword detection process is a process of detecting keywords included in the training review. The training viewer information acquisition unit 106 determines whether a keyword stored in a dictionary database storing keywords related to prices (e.g., "cheap" or "expensive") is included in the training review, and detects the keyword. The training viewer information acquisition unit 106 determines whether a keyword stored in a dictionary database storing keywords related to the feeling of a product or service (e.g., "cute" or "cool") is included in the training review, and detects the keyword.

[0183] The learning unit 104 also performs learning of the machine learning model M for estimating the advertising effect using the estimated viewer information related to the estimated comments, and the estimated viewer information is obtained by performing at least one of a sentiment analysis process for analyzing the sentiment of the estimated viewer, a comment classification process for classifying the estimated comments, and a keyword detection process for detecting keywords from the estimated comments. The learning unit 104 can collect the execution results of the sentiment analysis process, the comment classification process, and the keyword detection process, and generate the input part of the training data based on the collected results (for example, 100 people commented positively, 50 people commented negatively, etc.). This is also described in the second variation.

[0184] In addition, the details of the processing of the estimated comments can be replaced by "training" in the description of the processing of the training comments. The natural language processing of the estimated comments can also be replaced by "training" in the description of the natural language processing of the training comments. Since the processing during the estimation is performed by the estimation device 20, the processing of the estimated comments is performed by the estimation device 20.

[0185] The learning device 10 of the third variation acquires training viewer information related to the training comments by performing at least one of sentiment analysis processing, comment classification processing, and keyword detection processing on the training comments. The learning device 10 also learns the machine learning model M for estimating the advertising effect using the estimated viewer information acquired by performing at least one of sentiment analysis processing, comment classification processing, and keyword detection processing on the estimated comments. The learning device 10 can create a machine learning model M with high estimation accuracy of the advertising effect by making the machine learning model M learn the analysis results of at least one of sentiment analysis processing, comment classification processing, and keyword detection processing. For example, the learning device 10 can further improve the estimation accuracy of the machine learning model M by performing analysis based on information related to the reaction of the training viewer.

[0186] [6-4. Modification 4]

[0187] For example, as described in the embodiment, the training video information acquisition unit 102 may acquire a plurality of training video information. As described in Modification 2 and Modification 3, the training viewer information acquisition unit 106 may acquire a plurality of training viewer information. In this case, the learning unit 104 performs learning of the machine learning model M for estimating the advertising effect from a plurality of estimated video information and a plurality of estimated viewer information based on the plurality of training video information and the plurality of training viewer information. The above processing is as described in the embodiment, Modification 2, and Modification 3.

[0188] The learning device 10 of variation 4 includes a contribution calculation unit 105. The contribution calculation unit 105 calculates the contribution of each of a plurality of training video information and each of a plurality of training viewer information to the advertising effect. Among the functions of the contribution calculation unit 105, the configuration of calculating the contribution of each of a plurality of training video information may be the same as that of variation 1. The contribution calculation unit 105 of variation 4 calculates not only the contribution of each of a plurality of training video information but also the contribution of each of a plurality of training viewer information. That is, the contribution calculation unit 105 calculates the contribution of a plurality of training viewer information, such as training viewer emotions or training viewer comments, in order to analyze which training viewer information contributes to the advertising effect.

[0189] The meaning of the contribution of the training viewer-listener information is the same as the contribution of the training video information. The meaning of the contribution of the training viewer-listener information can be obtained by replacing "video" in the description of the contribution of the training video information in Variation 1 with "viewer". The contribution calculation unit 105 can calculate the contribution of the training viewer-listener information by the same calculation method as the calculation method of the contribution of the training video information. The method for calculating the contribution of the training viewer-listener information can be obtained by replacing "video" in the description of the method for calculating the contribution of the training video information in Variation 1 with "viewer". The contribution calculation unit 105 of Variation 4 records the contribution of each of the multiple training video information and each of the multiple training viewer-listener information in the data storage unit 100. The fact that the contribution can be used for any purpose is also the same as in Variation 1.

[0190] The learning device 10 of variation 4 calculates the contribution of each of the plurality of training video information and each of the plurality of training viewer information to the advertising effect. The learning device 10 can use the contribution of each of the plurality of training video information and each of the plurality of training viewer information to identify the training video information and the training viewer information that are helpful to the advertising effect. For example, the learning device 10 can prompt the creator of the machine learning model M, the performer, or other person, which of the plurality of training video information and the plurality of training viewer information is helpful to the advertising effect. For example, if the performer refers to the contribution to create future video advertisements, then the video advertisement with higher advertising effect can be created. For example, the learning device 10 can identify which element of the training video information and the training viewer information used in the learning of the machine learning model M effectively plays a role in the advertising effect through the contribution, and can create a machine learning model M that can more accurately estimate the advertising effect.

[0191] [6-5. Variation 5]

[0192] For example, in variation 4, the learning unit 104 may select at least one of the multiple training video information and the multiple training viewer information based on the contribution of each of the multiple training video information and the contribution of each of the multiple training viewer information, and learn a new machine learning model M based on the selected at least one.

[0193] The learning unit 104 selects at least one training video information with a relatively high contribution from the plurality of training video information. For example, the learning unit 104 selects at least one training video information with a contribution greater than a threshold value from the plurality of training video information. The learning unit 104 may select a predetermined number of training video information from the plurality of training video information in descending order of contribution. The learning unit 104 uses only the selected at least one training video information from the plurality of training video information for learning the new machine learning model M.

[0194] The learning unit 104 selects at least one piece of training viewer and listener information with a relatively high contribution from the plurality of training viewer and listener information. For example, the learning unit 104 selects at least one piece of training viewer and listener information with a contribution greater than a threshold value from the plurality of training viewer and listener information. The learning unit 104 may select a predetermined number of training viewer and listener information from the plurality of training viewer and listener information in descending order of contribution. The learning unit 104 uses only the selected at least one piece of training viewer and listener information from the plurality of training viewer and listener information for learning the new machine learning model M.

[0195] For example, the learning unit 104 generates an input portion of the training data based on the at least one selected training video information and the at least one selected training viewer information. The learning unit 104 learns a new machine learning model M based on the training data. Although the training data is different, the learning method itself can be the same as that of the old machine learning model M (the machine learning model M temporarily created for calculating the contribution). The old machine learning model M here is the machine learning model M created by the method described in the embodiment and the modified examples 1 to 4.

[0196] The learning device 10 of the fifth variation selects at least one of the plurality of training video information and the plurality of training viewer information based on the contribution of each of the plurality of training video information and the contribution of each of the plurality of training viewer information, and performs learning of a new machine learning model M based on the selected at least one. Thus, the learning device 10 can reduce the amount of information input to the machine learning model M during estimation, and thus can generate a machine learning model M that can reduce the processing load during estimation. In addition, since the machine learning model M does not learn the training video information and the training viewer information with relatively low contribution, the accuracy of the machine learning model M is improved.

[0197] [6-6. Variation 6]

[0198] For example, the estimation device 20 may perform estimation based on the machine learning model M generated by the method described in Modifications 2 to 5. The machine learning model M of Modification 6 performs learning based on training viewer information related to the training viewer. The machine learning model M is as described in Modifications 2 to 5.

[0199] The estimation device of the variation 6 includes an estimated viewer information acquisition unit 204. The estimated viewer information acquisition unit 204 acquires estimated viewer information related to the estimated viewer of the video advertisement, that is, the estimated viewer. The estimated viewer information acquisition unit 204 only needs to acquire the estimated viewer information based on the same acquisition method as the training viewer information. Therefore, the details of the process of the estimated viewer information acquisition unit 204 acquiring the estimated viewer information can be replaced by "training" in the description of the training viewer information acquisition unit 106.

[0200] For example, the estimated viewer information is the estimated viewer's behavior, the estimated viewer's basic information, or a combination thereof. Such information may be stored in the data storage unit 300 of the server 30. The estimated viewer information acquisition unit 204 may acquire estimated viewer information related to at least one of the estimated viewer's age, the estimated viewer's gender, the estimated viewer's purchase history in a store, the estimated viewer's input of estimated video ads, the estimated viewer's acquisition of coupons related to estimated video ads, the estimated viewer's viewing status of estimated video ads, the estimated viewer's access status, and the estimated viewer's search.

[0201] For example, the estimated viewer information acquisition unit 204 may acquire estimated viewer information related to the estimated comment by performing at least one of a sentiment analysis process of analyzing the sentiment of the estimated viewer, a comment classification process of classifying the estimated comment, and a keyword detection process of detecting a keyword from the estimated comment. The estimated viewer information acquisition unit 204 may acquire a plurality of estimated viewer information.

[0202] The estimation unit 203 of variation 6 also estimates the advertising effect based on the estimated viewer information. In addition to the estimated video information, the estimation unit 203 also inputs the estimated viewer information into the machine learning model M. The input to the machine learning model M is different from that in the embodiment, but the processing performed by the machine learning model M is the same as that in the embodiment. The machine learning model M calculates the feature quantity based on not only the estimated video information but also the estimated viewer information, and outputs the estimation result corresponding to the feature quantity. After collecting the estimated viewer information, the estimation unit 203 may input the collected result into the machine learning model M.

[0203] The estimation device 20 of the variation 6 also estimates the advertising effect based on the estimated viewer information. In addition to estimating the video information, the estimation device 20 also outputs the estimated viewer information to the machine learning model M, thereby improving the estimation accuracy of the advertising effect. For example, in the case where there is a causal relationship between the characteristics of the estimated viewer and the advertising effect, the machine learning model M can estimate the advertising effect corresponding to the characteristics of the estimated viewer. For example, the estimation device 20 can improve the estimation accuracy by making the machine learning model M learn the characteristics of each estimated viewer that cannot be estimated by estimating purchase information alone.

[0204] [6-7. Other Variations]

[0205] For example, variations of the descriptions may be combined.

[0206] For example, the learning device 10 and the estimation device 20 may be the same device. The functions described as functions implemented by the learning device 10 may be shared by a plurality of computers. The functions described as functions implemented by the estimation device 20 may be shared by a plurality of computers. The functions described as functions implemented by the server 30 may be implemented by the learning device 10 or the estimation device 20.

[0207] [7. Notes]

[0208] For example, the learning device can also be configured as follows. (1)

[0210] A learning device, comprising:

[0211] a training video information acquisition unit that acquires training video information acquired by analyzing a video advertisement for training, that is, a training video advertisement;

[0212] a training purchase information acquisition unit that acquires training purchase information related to a purchase by a viewer of the training video advertisement, that is, a training viewer; and

[0213] The learning unit performs learning of a machine learning model for estimating the advertising effect of the estimated video advertisement from estimated video information obtained by analyzing the estimated video advertisement based on the training video information and the training purchase information. (2)

[0215] A learning device as described in (1), wherein

[0216] The training video information acquisition unit acquires the training video information acquired by performing natural language processing on a training text acquired by analyzing the sound of the training video advertisement;

[0217] The learning unit performs learning of the machine learning model for estimating the advertisement effect from the estimated video information obtained by performing a process similar to the natural language processing on the estimated text obtained by analyzing the sound of the estimated video advertisement. (3)

[0219] A learning device as described in (2), wherein

[0220] The natural language processing is at least one of an emotion analysis process for analyzing the emotions of the performer of the training video advertisement, i.e., the training performer; a response detection process for detecting the response of the performer of the training video advertisement, i.e., the training performer; a description detection process for detecting descriptions related to the goods or services introduced in the training video advertisement; and a sales promotion detection process for detecting sales promotion related to the training video advertisement. (4)

[0222] A learning device as described in (3), wherein

[0223] The sales promotion detection process is a keyword detection process that detects at least one of the keywords related to the price of the product or service introduced in the training video advertisement, the keywords related to the sales status of the product or service introduced in the training video advertisement, and the keywords related to the description associated with the training video advertisement. (5)

[0225] A learning device as described in any one of (1) to (4), wherein

[0226] The training video information acquisition unit acquires the training video information related to the expression of the performer of the training video advertisement, that is, the training performer, acquired by analyzing the video of the training video advertisement;

[0227] The learning unit performs learning of the machine learning model for estimating the advertisement effect from the estimated video information related to an expression of a performer of the estimated video advertisement, that is, an estimated performer, obtained by analyzing a video of the estimated video advertisement. (6)

[0229] A learning device as described in any one of (1) to (5), wherein

[0230] The training purchase information acquisition unit acquires the training purchase information related to at least one of whether the training viewer has purchased or not and information related to sales of the product or service introduced in the training video advertisement;

[0231] The learning unit estimates the viewers of the estimated video advertisement, that is, whether the viewers have made a purchase, and at least one of information related to sales of the product or service introduced in the estimated video advertisement, as learning of the machine learning model of the advertising effect. (7)

[0233] A learning device as described in any one of (1) to (6), wherein

[0234] The training video information acquisition unit acquires a plurality of the training video information;

[0235] The learning unit also performs learning of the machine learning model based on the plurality of training video information;

[0236] The learning device further includes a contribution degree calculation unit that calculates a contribution degree of each of the plurality of training video information to the advertising effect. (8)

[0238] A learning device as described in any one of (1) to (7), wherein

[0239] The learning device further includes a training viewer information acquisition unit that acquires training viewer information related to the training viewer;

[0240] The learning unit also performs learning of the machine learning model for estimating the advertisement effect by using estimated viewer information related to estimated viewers of the estimated video advertisement, that is, estimated viewers, based on the training viewer information. (9)

[0242] A learning device as described in (8), wherein

[0243] The training viewer information acquisition unit acquires the training viewer information related to at least one of the age of the training viewer, the gender of the training viewer, the purchase history of the training viewer in a store, the training comments input by the training viewer on the training video advertisement, the acquisition of coupons related to the training video advertisement by the training viewer, the viewing status of the training video advertisement by the training viewer, the access status of the training viewer, and the search of the training viewer;

[0244] The learning unit also uses the estimated viewer information related to at least one of the estimated viewer's age, the estimated viewer's gender, the estimated viewer's purchase history in a store, the estimated comment entered by the estimated viewer regarding the estimated video advertisement, the estimated viewer's acquisition of coupons related to the estimated video advertisement, the estimated viewer's viewing status of the estimated video advertisement, the estimated viewer's access status, and the estimated viewer's search to learn the machine learning model that estimates the advertising effect. (10)

[0246] A learning device as described in (9), wherein

[0247] The training viewer information acquisition unit acquires the training viewer information related to the training review by executing at least one of an emotion analysis process of analyzing the emotion of the training viewer, a comment classification process of classifying the training review, and a keyword detection process of detecting a keyword from the training review.

[0248] The learning unit also performs learning of the machine learning model for estimating the advertising effect by using the estimated viewer information related to the estimated comments obtained by performing at least one of a sentiment analysis process for analyzing the sentiment of the estimated viewer, a comment classification process for classifying the estimated comments, and a keyword detection process for detecting keywords from the estimated comments. (11)

[0250] A learning device as described in any one of (8) to (10), wherein

[0251] The training video information acquisition unit acquires a plurality of the training video information;

[0252] The training viewer information acquisition unit acquires a plurality of training viewer information;

[0253] The learning unit performs learning of the machine learning model for estimating the advertisement effect from the plurality of estimated video information and the plurality of estimated viewer information based on the plurality of training video information and the plurality of training viewer information;

[0254] The learning device further includes a contribution degree calculation unit that calculates a contribution degree of each of the plurality of training video information and each of the plurality of training viewer information to the advertisement effect. (12)

[0256] A learning device as described in (11), wherein the learning unit selects at least one of the multiple training video information and the multiple training viewer information based on the contribution degree of each of the multiple training video information and the contribution degree of each of the multiple training viewer information, and learns a new machine learning model based on the selected at least one.

Claims

1. A learning device, comprising: A training video information acquisition unit acquires training video information acquired by analyzing a video advertisement for training, that is, a training video advertisement; a training purchase information acquisition unit that acquires training purchase information related to a purchase by a viewer of the training video advertisement, that is, a training viewer; and The learning unit performs learning of a machine learning model for estimating the advertising effect of the estimated video advertisement from estimated video information obtained by analyzing the estimated video advertisement based on the training video information and the training purchase information.

2. The learning device according to claim 1, wherein The training video information acquisition unit acquires the training video information acquired by performing natural language processing on a training text acquired by analyzing the sound of the training video advertisement; The learning unit performs learning of the machine learning model for estimating the advertisement effect from the estimated video information obtained by performing a process similar to the natural language processing on the estimated text obtained by analyzing the sound of the estimated video advertisement.

3. The learning device according to claim 2, wherein The natural language processing is at least one of an emotion analysis process for analyzing the emotions of the actors in the training video advertisement, i.e., the training actors; a response detection process for detecting the responses of the actors in the training video advertisement, i.e., the training actors; a description detection process for detecting descriptions related to the goods or services introduced in the training video advertisement; and a sales promotion detection process for detecting sales promotion related to the training video advertisement.

4. The learning device according to claim 3, wherein The sales promotion detection process is a keyword detection process that detects at least one of the keywords related to the price of the product or service introduced in the training video advertisement, the keywords related to the sales status of the product or service introduced in the training video advertisement, and the keywords related to the description associated with the training video advertisement.

5. The learning device according to any one of claims 1 to 4, wherein The training video information acquisition unit acquires the training video information related to the expression of the actor of the training video advertisement, that is, the training actor, obtained by analyzing the video of the training video advertisement; The learning unit performs learning of the machine learning model for estimating the advertisement effect from the estimated video information related to expressions of actors who appear in the estimated video advertisement, that is, estimated actors, obtained by analyzing a video of the estimated video advertisement.

6. The learning device according to any one of claims 1 to 4, wherein The training purchase information acquisition unit acquires the training purchase information related to at least one of whether the training viewer has purchased or not and information related to sales of the product or service introduced in the training video advertisement; The learning unit estimates the viewers of the estimated video advertisement, that is, whether the viewers have made a purchase, and at least one of information related to sales of the product or service introduced in the estimated video advertisement, as learning of the machine learning model of the advertising effect.

7. The learning device according to any one of claims 1 to 4, wherein The training video information acquisition unit acquires a plurality of the training video information; The learning unit also performs learning of the machine learning model based on the plurality of training video information; The learning device further includes a contribution degree calculation unit that calculates a contribution degree of each of the plurality of training video information to the advertising effect.

8. The learning device according to any one of claims 1 to 4, wherein The learning device further includes a training viewer information acquisition unit that acquires training viewer information related to the training viewer; The learning unit also performs learning of the machine learning model for estimating the advertisement effect by using estimated viewer information related to estimated viewers of the estimated video advertisement, that is, estimated viewers, based on the training viewer information.

9. The learning device according to claim 8, wherein The training viewer information acquisition unit acquires the training viewer information related to at least one of the age of the training viewer, the gender of the training viewer, the purchase history of the training viewer in a store, the training comments input by the training viewer on the training video advertisement, the acquisition of coupons related to the training video advertisement by the training viewer, the viewing status of the training video advertisement by the training viewer, the access status of the training viewer, and the search of the training viewer; The learning unit also uses the estimated viewer information related to at least one of the estimated viewer's age, the estimated viewer's gender, the estimated viewer's purchase history in a store, the estimated comment entered by the estimated viewer regarding the estimated video advertisement, the estimated viewer's acquisition of coupons related to the estimated video advertisement, the estimated viewer's viewing status of the estimated video advertisement, the estimated viewer's access status, and the estimated viewer's search to learn the machine learning model for estimating the advertising effect.

10. The learning device according to claim 9, wherein The training viewer information acquisition unit acquires the training viewer information related to the training review by executing at least one of an emotion analysis process of analyzing the emotion of the training viewer, a comment classification process of classifying the training review, and a keyword detection process of detecting a keyword from the training review. The learning unit also performs learning of the machine learning model for estimating the advertising effect by using the estimated viewer information related to the estimated comments obtained by performing at least one of a sentiment analysis process for analyzing the sentiment of the estimated viewer, a comment classification process for classifying the estimated comments, and a keyword detection process for detecting keywords from the estimated comments.

11. The learning device according to any one of claims 8 to 10, wherein The training video information acquisition unit acquires a plurality of the training video information; The training viewer information acquisition unit acquires a plurality of training viewer information; The learning unit performs learning of the machine learning model for estimating the advertisement effect from the plurality of estimated video information and the plurality of estimated viewer information based on the plurality of training video information and the plurality of training viewer information; The learning device further includes a contribution degree calculation unit that calculates a contribution degree of each of the plurality of training video information and each of the plurality of training viewer information to the advertisement effect.

12. The learning device according to claim 11, wherein The learning unit selects at least one of the multiple training video information and the multiple training viewer information based on the contribution degree of each of the multiple training video information and the contribution degree of each of the multiple training viewer information, and learns a new machine learning model based on the selected at least one.

13. A presumption device, comprising: The estimated video information acquisition unit acquires estimated video information acquired by analyzing the estimated video advertisement, that is, the estimated video advertisement; a model storage unit storing a machine learning model for performing learning based on training video information obtained by analyzing a video advertisement for training, i.e., a training video advertisement, and training purchase information related to purchases by viewers of the training video advertisement, i.e., training viewers; The estimating unit estimates the advertising effect of the estimated video advertising information based on the estimated video information and the machine learning model.

14. The estimation device according to claim 13, wherein The machine learning model performs learning based on training viewer information related to the training viewer; The estimation device further includes an estimated viewer information acquisition unit that acquires estimated viewer information related to a viewer of the estimated video advertisement, that is, an estimated viewer; The estimating unit further estimates the advertising effect based on the estimated viewer information.

15. A learning method comprising: a training video information acquisition step of acquiring training video information acquired by analyzing a video advertisement for training, namely, a training video advertisement; a training purchase information acquisition step of acquiring training purchase information related to purchases by viewers of the training video advertisement, namely, the training viewers; and The learning step is to perform learning of a machine learning model for estimating the advertising effect of the estimated video advertisement from estimated video information obtained by analyzing the estimated video advertisement, i.e., the estimated video advertisement, based on the training video information and the training purchase information.

16. A method of presumption, comprising: an estimated video information acquisition step of acquiring estimated video information acquired by analyzing a video advertisement for estimation, namely, an estimated video advertisement; and The estimation step is to estimate the advertising effect of the estimated video advertisement based on a machine learning model that learns training video information obtained by analyzing training video advertisements, i.e., training video advertisements, and training purchase information related to purchases made by viewers of the training video advertisements, i.e., training viewers, and the estimated video information.

17. A program for causing a computer to function as follows: A training video information acquisition unit acquires training video information acquired by analyzing a video advertisement for training, that is, a training video advertisement; a training purchase information acquisition unit that acquires training purchase information related to a purchase by a viewer of the training video advertisement, that is, a training viewer; and The learning unit performs learning of a machine learning model for estimating the advertising effect of the estimated video advertisement from estimated video information obtained by analyzing the estimated video advertisement based on the training video information and the training purchase information.

18. A program for causing a computer to function as follows: an estimated video information acquisition unit that acquires estimated video information acquired by analyzing an estimated video advertisement, that is, an estimated video advertisement; and The estimation unit estimates the advertising effect of the estimated video advertisement based on a machine learning model that performs learning based on training video information obtained by analyzing training video advertisements, i.e., training video advertisements, and training purchase information related to purchases made by viewers of the training video advertisements, i.e., training viewers, and the estimated video information.

Images