Device and method
Patent Information
- Application Number
- PCT/JP2025/014673
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-04-14
- Publication Date
- 2026-10-22
Smart Images
Figure JP2025014673_22102026_PF_FP_ABST
Abstract
Description
Apparatus and Method
[0001] The present invention relates to an apparatus and a method for generating advertisement text to be distributed to users targeted for distribution.
[0002] Patent Document 1 describes an apparatus that distributes advertisements to users by adding the advertisements to web pages for which a browsing request has been received from a user.
[0003] Japanese Unexamined Patent Application Publication No. 2010-026792
[0004] From the perspective of enhancing advertising effectiveness, it is preferable to generate advertisement text for each individual user. However, generating advertisement text for each user poses a problem of inefficiency, as the checking of the advertisement text becomes enormous. Accordingly, it is conceivable to classify users into classes and generate one piece of advertisement text for each class. However, if the classification of users is not appropriate, there arises a problem that expected advertising effectiveness cannot be achieved.
[0005] An objective of the present disclosure is to provide an apparatus and a method capable of generating advertisements that are efficient and have high advertising effectiveness.
[0006] The apparatus of the present disclosure comprises: an extraction unit that extracts candidate words serving as candidates for appeal elements for each service from a plurality of distributed advertisement texts; an appeal element determination unit that determines the appeal elements of the service based on the candidate words and the service feature quantity of the service corresponding to the candidate words; and an advertisement text generation model learning unit that learns an advertisement text generation model for generating advertisement text from user feature quantities and appeal elements, based on the appeal elements of the service, the user feature quantities of reaction users who have reacted to the distributed advertisement texts, and the distributed advertisement texts to which the reaction users have reacted.
[0007] According to the present disclosure, it is possible to generate advertisements that are efficient and have high advertising effectiveness.
[0008] Figure 1 is a diagram showing the system configuration of an advertising generation system including an advertising copy generation device according to an embodiment of the present disclosure. Figure 2 is a block diagram showing the functional configuration of the advertising copy generation device. Figure 3 is a diagram showing an example of advertising copy history stored in a knowledge database. Figure 4 is a diagram showing an example of a list of service features stored in a knowledge database. Figure 5 is a diagram showing an example of a list of user features stored in a knowledge database. Figure 6 is a diagram showing an example of prompts for training an advertising copy generation model. Figure 7 is a diagram showing an example of prompts for generating advertising copy candidates corresponding to a sample user. Figure 8 is a diagram showing an example of multiple generated advertising copy candidates. Figure 9 is a diagram showing an example of a list of selected representative advertising copy. Figure 10 is a diagram showing an example of a list of sample users associated with a class. Figure 11 is a flowchart showing the processing operation of the advertising copy generation device. Figure 12 is a diagram showing an example of the hardware configuration of the advertising copy generation device according to an embodiment of the present disclosure.
[0009] Embodiments of this disclosure will be described with reference to the attached drawings. Where possible, the same parts will be denoted by the same reference numerals, and redundant descriptions will be omitted.
[0010] Figure 1 shows the system configuration of an advertising generation system including an advertising text generation device 100 according to an embodiment of the present disclosure. As shown in the figure, the system comprises an advertising text generation device 100, a large-scale language model 200, and a knowledge database 300. These servers and databases are configured to communicate with each other via a network.
[0011] The ad copy generation device 100 is a device for generating ad copy to be distributed. The users to whom the ad copy generation device 100 distributes ad copy are called target users. Target users include not only users who have received ads in the past, but also new users who have not received ads in the past. Ad copy that has been distributed in the past is called distributed ad copy.
[0012] The ad copy generator 100 retrieves multiple previously delivered ad copy from the knowledge database 300. From the multiple previously delivered ad copy, the ad copy generator 100 extracts candidate words that can serve as appealing elements for each service.
[0013] Next, the ad copy generator 100 determines the appeal elements of a service based on the extracted candidate words for each service and the service features of the services corresponding to the candidate words. For example, the ad copy generator 100 vectorizes the extracted candidate words for each service and the service features of the services corresponding to the candidate words, and determines that the similarity of the vectorized candidate words for each service and the service features of the services corresponding to the candidate words is above a threshold as the appeal elements of the service. For example, the ad copy generator 100 has a large-scale language model 200 perform this vectorization.
[0014] Next, the ad copy generation device 100 obtains from the knowledge database 300 the user characteristics, which are the characteristics of users who responded to delivered ad copy, and the delivered ad copy to which the users responded. Based on the appeal elements of the determined service, the user characteristics of users who responded to delivered ad copy, and the delivered ad copy to which the users responded, the ad copy generation device 100 learns an ad copy generation model that generates ad copy from the user characteristics and appeal elements. For example, the ad copy generation device 100 may train the ad copy generation model using a large-scale language model 200 or a Transformer model.
[0015] Next, the ad copy generation device 100 generates ad copy candidates corresponding to each sample user, based on the learned ad copy generation model, using the user features of multiple sample users and the target appeal elements that the new ad wants to appeal to. For example, the ad copy generation device 100 causes the large-scale language model 200 to generate multiple ad copy candidates corresponding to multiple sample users.
[0016] Next, the ad copy generation device 100 clusters multiple ad copy candidates and multiple sample users into multiple classes (groups). For example, the ad copy generation device 100 vectorizes the multiple ad copy candidates and clusters the multiple ad copy candidates into multiple classes based on the Euclidean distance between the vectors. Then, the ad copy generation device 100 clusters the multiple sample users into multiple classes by associating them with the multiple clustered classes. For example, the ad copy generation device 100 has a large-scale language model 200 perform this clustering.
[0017] Next, the ad copy generation device 100 selects a representative ad copy from multiple ad copy candidates for each of the multiple classes. For example, the ad copy generation device 100 vectorizes the multiple ad copy candidates and, for each of the multiple classes, selects the ad copy candidate that is closest to the average vector as the representative ad copy. For example, the ad copy generation device 100 has the large-scale language model 200 perform the vectorization of the multiple ad copy candidates and the selection of the representative ad copy.
[0018] Next, the ad copy generator 100 learns a classification model that classifies user features into multiple classes. For example, the ad copy generator 100 trains the classification model on a large-scale language model 200.
[0019] Next, the ad copy generation device 100 estimates the class to which the target user belongs based on the user characteristics of the target user, using a classification model.
[0020] Next, the ad copy generation device 100 delivers the estimated representative ad copy for the class to the user terminal 400 of the target user.
[0021] The large-scale language model 200 is a generative AI model that, in response to input from a prompt containing input information, generates content according to the instructions, context, questions, and output format indicated by the prompt, and returns that content as response information. In this disclosure, a prompt is information indicating instructions or questions entered by a user in an interactive system such as a dialogue with a generative AI model or a command-line interface (CLI). The prompt may also include input information, in which case the generative AI model generates response information targeting the input information. The generative AI model may be an interactive AI model that includes, for example, a large-scale language model (LLM) and a user interface (UI) for dialogue with the user, enabling text chat or voice chat with the user. Examples of such generative AI models include ChatGPT, GPT®-3.5, GPT-4V, GPT-4o, PaLM2, and tsuzumi.
[0022] Furthermore, all or some of the large-scale language model 200 and the knowledge database 300 may be placed in the advertising copy generation device 100 or the user terminal 400. The user terminal 400 may also have the functionality of the advertising copy generation device 100 and function as the advertising copy generation device 100. Some generation AI models, like Tsuzumi, have the generation AI model located within the user terminal 400. In this type, the RAG (Retrieval-Augmented Generation) application is also provided on the user terminal 400. However, the information (knowledge database) accessed by the RAG application may be located within the user terminal 400 or on the network. There are also types, like ChatGPT, where the generation AI model is located on the network. In this type, the RAG application is provided on the user terminal 400. However, the information (knowledge database) accessed by the RAG is on the network.
[0023] Figure 2 is a block diagram showing the functional configuration of the advertising copy generation device 100. As shown in Figure 2, the advertising copy generation device 100 is composed of an extraction unit 101, a selling element determination unit 102, an advertising copy generation model learning unit 103, an advertising copy candidate generation unit 104, a clustering unit 105, a selection unit 106, a classification model learning unit 107, a class estimation unit 108, and a distribution unit 109.
[0024] The extraction unit 101 extracts candidate words that are potential selling points for each service from multiple previously delivered ad copy. Previously delivered ad copy refers to ad copy that has been delivered in the past. Examples of services include food and beverage services and clothing services. Multiple candidate words are extracted, for example, by extracting important words using TF-IDE.
[0025] The extraction unit 101 retrieves, for example, multiple previously delivered ad copy from the knowledge database 300. Then, the extraction unit 101 extracts candidate words that are potential appeal elements for each service from the multiple previously delivered ad copy retrieved from the knowledge database 300.
[0026] Figure 3 shows an example of an advertising copy history stored in the knowledge database 300. In the advertising copy history shown in Figure 3, the user ID of the user who delivered the advertising copy, the service targeted by the delivered advertising copy, the delivered advertising copy, and the user's response to the delivered advertising copy are stored in association with each other. The first line of the advertising copy history stores "AAA" as the user ID, "Service A" as the service, "Seasonal menu items available! Why not eat at the cafe?" as the advertising copy, and "0" as the user response. The second line of the advertising copy history stores "BBB" as the user ID, "Service A" as the service, "Coupons available. One free drink!" as the advertising copy, and "1" as the user response. The third line of the advertising copy history stores "BBB" as the user ID, "Service B" as the service, "Trendy clothes have arrived! Why not go to the clothing store?" as the advertising copy, and "1" as the user response. The fourth line of the ad copy history stores "CCC" as the user ID, "Service B" as the service, "Great sale underway! Why not come to the clothing store?" as the ad copy, and "0" as the user response. The user response is represented as "0" if the user did not respond and "1" if the user responded.
[0027] The extraction unit 101, for example, acquires the advertising copy history "We have seasonal menu items! Why not eat at our cafe?" and "Coupons are being distributed. One free drink!" stored in the knowledge database 300 as delivered advertising copy for service A. The extraction unit 101 also acquires the advertising copy history "Trendy clothes have arrived! Why not come to the clothing store?" and "Great sale underway! Why not come to the clothing store?" stored in the knowledge database 300 as delivered advertising copy for service B. Then, from the acquired delivered advertising copy, the extraction unit 101 uses TF-IDF to acquire "seasonal," "menu," "cafe," and "eat" as candidate words for service A, and "trendy" and "sale" as candidate words for service B.
[0028] The appeal element determination unit 102 determines the appeal elements of a service based on the candidate words for each service extracted by the extraction unit 101, and the service features of the services corresponding to the candidate words. Service features are information that indicates the type and content of the service. The appeal element determination unit 102 obtains service features from, for example, the knowledge database 300. Then, the appeal element determination unit 102 determines the appeal elements of a service based on the candidate words for each service extracted by the extraction unit 101, and the service features of the services corresponding to the candidate words obtained from the knowledge database 300.
[0029] Figure 4 shows an example of a list of service features stored in the knowledge database 300. In the list of service features shown in Figure 4, services, service categories, and service subcategories are stored in relation to each other. The first row of the list of service features stores "Service A" as the service, "Restaurant" as the category, and "Café" as the subcategory. The second row of the list of service features stores "Service B" as the service, "Clothing Store" as the category, and "Women's Clothing" as the subcategory.
[0030] The appeal element determination unit 102, for example, acquires "restaurants" and "cafes" from the list of service features stored in the knowledge database 300 as service features for service A, and acquires "clothing stores" and "women's clothing" from the list of service features stored in the knowledge database 300 as service features for service B.
[0031] The appeal element determination unit 102 generates prompts for vectorizing the candidate words for each service extracted by the extraction unit 101 and the service features obtained from the list of service features for each service, and sends the generated prompts to the large-scale language model 200. The appeal element determination unit 102 then compares both vectors obtained from the large-scale language model 200 and determines the candidate words whose similarity is above a threshold as the appeal elements for that service. For example, the appeal element determination unit 102 determines "cafe" and "eat" as the appeal elements for service A.
[0032] The ad copy generation model learning unit 103 learns an ad copy generation model that generates ad copy from user features and appeal elements based on the appeal elements of the service determined by the appeal element determination unit 102, the user features of users who responded to delivered ad copy, and the delivered ad copy to which the responding users responded.
[0033] The ad copy generation model learning unit 103, for example, obtains from the knowledge database 300 the responding users who responded to delivered ad copy, the user features of the responding users, and the delivered ad copy to which the responding users responded.
[0034] Figure 5 shows an example of a list of user features stored in the knowledge database 300. In the list of user features shown in Figure 5, user ID, age, gender, occupation, etc., are stored in association with each other. The first row of the list of user features stores "XXX" as the user ID, "30" as the age, "female" as the gender, and "housewife" as the occupation. The second row of the list of user features stores "YYY" as the user ID, "50" as the age, "male" as the gender, and "company employee" as the occupation.
[0035] The ad copy generation model learning unit 103, for example, retrieves responding users who reacted to delivered ad copy and delivered ad copy to which responding users reacted from the ad copy history (see Figure 3) stored in the knowledge database 300, and retrieves the features of responding users from the list of user features (see Figure 5) stored in the knowledge database 300.
[0036] The ad copy generation model learning unit 103 generates a prompt for learning an ad copy generation model that generates ad copy from user features and appeal elements, based on, for example, the appeal elements of the service determined by the appeal element determination unit 102, the user features of responding users who responded to delivered ad copy, and the delivered ad copy to which the responding users responded, and sends the generated prompt to the large-scale language model 200. This prompt includes, for example, a task, appeal elements, and user features. The task is, for example, a task that indicates to generate ad copy that appeals to users with these user features using these appeal elements. As a result, the large-scale language model 200 learns an ad copy generation model that generates ad copy from user features and appeal elements.
[0037] Figure 6 shows an example of prompts for training an ad copy generation model. The prompts shown in Figure 6 include "Consider ad copy that appeals to the appeal elements [cafe] [eat] for a user with the following features," and "#User features Age: xxx, Gender: xxx, Occupation: xxx."
[0038] The ad copy candidate generation unit 104 generates ad copy candidates corresponding to each sample user, based on the ad copy generation model learned by the ad copy generation model learning unit 103, using the user features of multiple sample users and the target appeal elements that the new ad wants to appeal to. In other words, it generates multiple ad copy candidates corresponding to multiple sample users. The sample users may be any user stored in the knowledge database 300, or they may be newly created virtual users. Furthermore, if the target users to whom the new ad copy will be delivered are known, these target users may be used as sample users.
[0039] The ad copy candidate generation unit 104 generates prompts for generating ad copy candidates corresponding to each sample user, for example, from the user features and target appeal elements of multiple sample users, and transmits the generated prompts to the large-scale language model 200 that has trained the ad copy generation model. These prompts include, for example, a task, an appeal element, and user features. The task is, for example, a task that indicates the generation of ad copy that appeals to the user with these user features using these appeal elements. When the sample users are those stored in the knowledge database 300, the ad copy candidate generation unit 104 obtains the user features of the sample users from the knowledge database 300, for example, and uses the obtained user features as the user features in the prompt. As a result, the large-scale language model 200 generates multiple ad copy candidates corresponding to multiple sample users.
[0040] Figure 7 shows an example of a prompt for generating multiple ad copy candidates corresponding to multiple sample users. The prompt shown in Figure 7 includes the phrase, "Consider ad copy that appeals to the appeal element [xxx] [xxx] for users with the following features," and "#User features Age: xxx, Gender: xxx, Occupation: xxx." This appeal element is, for example, "water conservation."
[0041] Figure 8 shows an example of a list of generated ad copy candidates. In the ad copy candidate list shown in Figure 8, the user ID and the ad copy candidate are stored in association with each other. The first line of the ad copy candidate list stores "XXX" as the user ID and "Let's conserve water and be environmentally conscious!" as the ad copy candidate. The second line of the ad copy candidate list stores "ZZZZ" as the user ID and "Let's save on water bills by turning off the shower when washing our bodies and hair!" The third line of the ad copy candidate list stores "yyy" as the user ID and "Many people are taking environmentally conscious actions. Why don't we all be mindful of conserving water together?" The second line of the ad copy candidate list stores "Vvv" as the user ID and "Our water conservation is necessary to protect the planet!" as the ad copy candidate.
[0042] The clustering unit 105 clusters the plurality of advertisement text candidates generated by the advertisement text candidate generation unit 104 and the plurality of sample users into a plurality of classes (groups).
[0043] The clustering unit 105, for example, generates a prompt for vectorizing the plurality of advertisement text candidates, and transmits the generated prompt to the large language model 200. Then, the clustering unit 105 clusters the plurality of advertisement text candidates into a plurality of classes using k-means clustering or the like, based on the Euclidean distance between the two vectors acquired from the large language model 200. Then, the clustering unit 105 clusters the plurality of sample users into a plurality of classes by associating the plurality of sample users with the plurality of clustered classes.
[0044] The selection unit 106 selects a representative advertisement text from the plurality of advertisement text candidates in each of the plurality of classes clustered by the clustering unit 105. The representative advertisement text of each class is the advertisement text that serves as the representative of the corresponding class.
[0045] Figure 9 is a diagram illustrating an example of a list of selected representative advertisement texts. In the list of representative advertisement texts shown in Figure 9, classes and representative advertisement texts are stored in association with each other. The first line of the list of representative advertisement texts stores "A" as the class, and "Let's save water bills by turning off the shower while washing your body and hair!" as the representative text. The second line of the list of representative advertisement texts stores "B" as the class, and "Many people are taking environmentally conscious actions. Why not be mindful of saving water together?" as the representative advertisement text. The third line of the list of advertisement text candidates stores "C" as the class, and "Our water saving is necessary to protect the Earth!" as the representative advertisement text.
[0046] FIG. 10 is a diagram illustrating an example of a list of sample users associated with classes. In the list of sample users shown in FIG. 10, user IDs and classes are stored in association with each other. In the first line of the list of sample users, "XXX" as a user ID and "A" as a class are stored. In the second line of the list of sample users, "ZZZZ" as a user ID and "A" as a class are stored. In the third line of the list of sample users, "yyy" as a user ID and "B" as a class are stored. In the fourth line of the list of sample users, "Vvv" as a user ID and "B" as a class are stored.
[0047] A classification model learning unit 107 learns a classification model that classifies user feature values into a plurality of classes clustered by a clustering unit 105.
[0048] The classification model learning unit 107 generates a prompt for learning a classification model that classifies user feature values into a plurality of classes, for example, from the respective user feature values of a plurality of sample users and the list of sample users associated with the classes (see FIG. 10), and transmits the generated prompt to a large language model 200. This prompt includes, for example, a task, the user feature values, and the classes. The task is, for example, a task indicating learning a classification model that classifies the user feature values into the plurality of classes. Accordingly, the large language model 200 learns the classification model that classifies the user feature values into the plurality of classes. When users stored in a knowledge database 300 are set as the sample users, the classification model learning unit 107 acquires, for example, the user feature values of the users serving as the sample users from the knowledge database 300, and sets the acquired user feature values as the user feature values in the prompt.
[0049] A class estimation unit 108 estimates a class into which a distribution target user is classified, from the user feature values of the distribution target user, based on the classification model learned by the classification model learning unit 107.
[0050] If the target user is a user stored in the knowledge database 300, the class estimation unit 108 obtains the user's features from the knowledge database 300, for example. If the target user is not a user stored in the knowledge database 300, the class estimation unit 108 obtains the user's features from input from an input device, for example.
[0051] The class estimation unit 108 generates a prompt, for example, from the user features of the target user to estimate the class to which the target user belongs, and sends the generated prompt to the large-scale language model 200 that has trained the classification model. This prompt includes, for example, a task and user features. The task is, for example, a task that indicates the class to which these user features belong.
[0052] The distribution unit 109 distributes representative ad copy for the class estimated by the class estimation unit 108 to the target users. However, if the target users are sample users, the class of the target users is known, so the class estimation unit 108 may distribute representative ad copy for the class to which the target users belong to without performing an estimation of the class to which the target users belong.
[0053] Next, the operation of the advertising copy generation device 100 configured in this way will be explained. Figure 11 is a flowchart showing the processing operation of the advertising copy generation device.
[0054] As shown in Figure 11, the extraction unit 101 extracts candidate words that are potential appeal elements for each service from multiple delivered ad copy (S101).
[0055] Next, the appeal element determination unit 102 determines the appeal elements of a service based on candidate words for each service and the service features of the service corresponding to the candidate words (S102). For example, the appeal element determination unit vectorizes the candidate words for each service and the service features of the service corresponding to the candidate words, and determines candidate words whose similarity to the service features of the service is above a threshold as the appeal elements of the service.
[0056] Next, the ad copy generation model learning unit 103 learns an ad copy generation model that generates ad copy from user features and appeal elements based on the determined service appeal elements, the user features of users who responded to the delivered ad copy, and the delivered ad copy to which the responding users responded (S103). If the knowledge database 300 stores delivered ad copy, user features of users to whom the delivered ad copy was delivered, and users who responded to the delivered ad copy, the ad copy generation model learning unit 103 learns the ad copy generation model based, for example, on the information stored in the knowledge database 300.
[0057] Next, the ad copy candidate generation unit 104 generates multiple ad copy candidates corresponding to multiple sample users based on the ad copy generation model, using the user characteristics of multiple sample users and the target appeal elements which are the appeal elements to be emphasized in the new ad (S104).
[0058] Next, the clustering unit 105 clusters the multiple ad copy candidates into multiple classes (S105). For example, the clustering unit 105 vectorizes the multiple ad copy candidates and clusters the multiple ad copy candidates based on the Euclidean distance between the vectors.
[0059] Next, the selection unit 106 selects a representative advertising slogan from multiple advertising slogan candidates for each of the multiple classes (S106).
[0060] Next, the classification model learning unit 107 learns a classification model that classifies user features into multiple classes (S107).
[0061] Next, the class estimation unit 108 estimates the class to which the target user belongs based on the user characteristics of the target user, using the classification model (S108).
[0062] Next, the distribution unit 109 distributes the representative advertising text for the class estimated by the class estimation unit 108 to the target users (S109).
[0063] Next, the effects and advantages of the embodiments of this disclosure will be described.
[0064] The appeal elements in advertising copy, which depend on user characteristics such as expression methods, significantly influence advertising effectiveness. However, since appeal elements often differ greatly between different services, extracting candidate words for appeal elements from advertising copy of different services makes it difficult to generate advertising copy that depends on user characteristics such as expression methods.
[0065] In contrast, the ad copy generation device 100 of this disclosure extracts candidate words that are candidates for appeal elements for each service from multiple delivered ad copy. The ad copy generation device 100 then determines the appeal elements for a service based on the candidate words for each service and the service features of the service corresponding to the candidate words. Based on the determined service appeal elements, the user features of users who responded to the delivered ad copy, and the delivered ad copy to which the responding users responded, the ad copy generation model learns an ad copy generation model that generates ad copy from user features and appeal elements. As a result, by generating ad copy based on the learned ad copy generation model, it is possible to generate ad copy with appeal elements that depend on user features such as expression methods. Therefore, even when generating ad copy on a class (group) basis, it is possible to generate ad copy with high advertising effectiveness. This makes it possible to generate efficient and highly effective advertisements.
[0066] Furthermore, the advertising copy generation device 100 vectorizes candidate words for each service and the service features of the service corresponding to the candidate words, and determines candidate words whose similarity to the service features is above a threshold as the appeal elements of the service. This makes it easy to determine the appeal elements of the service.
[0067] Furthermore, the advertising copy generation device 100 learns an advertising copy generation model based on the information stored in the knowledge database 300. Therefore, the advertising copy generation model can be easily learned.
[0068] Furthermore, the ad copy generation device 100 generates multiple ad copy candidates corresponding to multiple sample users based on the ad copy generation model, using user characteristics of multiple sample users and target appeal elements that the new ad wants to highlight. Therefore, by selecting a class of ad copy from the generated multiple ad copy candidates, it is possible to generate ad copy with appeal elements that depend on user characteristics such as expression methods.
[0069] Furthermore, by using sample users as the target users to whom the ad copy is delivered, the ad copy generation device 100 can deliver ad copy tailored to the target user's class to the target users without having to separately estimate the target user's class.
[0070] Furthermore, the ad copy generation device 100 clusters multiple ad copy candidates into multiple classes, and in each of the multiple classes, selects a representative ad copy from the multiple ad copy candidates. This enables the efficient generation of highly effective advertisements.
[0071] Furthermore, the ad copy generation device 100 can easily cluster multiple ad copy candidates by vectorizing them and clustering them based on the Euclidean distance between the vectors.
[0072] Furthermore, the ad copy generation device 100 learns a classification model that classifies user characteristics into multiple classes, and based on the classification model, it estimates the class to which a target user belongs based on their user characteristics. This allows the target user to be classified into a class corresponding to their user characteristics.
[0073] Furthermore, the ad copy generation device 100 can enhance advertising effectiveness by delivering representative ad copy for each class to target users.
[0074] The apparatus and method of this disclosure have the following configurations.
[0075] [1] An apparatus comprising: an extraction unit that extracts candidate words that are candidates for appeal elements for each service from multiple delivered ad copy; an appeal element determination unit that determines the appeal elements of the service based on the candidate words and the service features of the service corresponding to the candidate words; and an ad copy generation model learning unit that learns an ad copy generation model that generates ad copy from user features and appeal elements based on the determined appeal elements of the service, the user features of users who responded to the delivered ad copy, and the delivered ad copy to which the responding users responded.
[0076] [2] The apparatus according to [1], wherein the appeal element determination unit vectorizes the candidate words for each service and the service features of the service corresponding to the candidate words, and determines the candidate words whose similarity to the service features of the service is equal to or greater than a threshold as the appeal elements of the service.
[0077] [3] The apparatus according to [1] or [2], further comprising a knowledge database storing the delivered ad copy, the user characteristics of the users to whom the delivered ad copy was delivered, and the users who responded to the delivered ad copy, wherein the ad copy generation model learning unit learns an ad copy generation model based on the information stored in the knowledge database.
[0078] [4] The apparatus according to any one of [1] to [3], further comprising an ad copy candidate generation unit that generates multiple ad copy candidates corresponding to the multiple sample users from the user characteristics of the multiple sample users and the target appeal elements which are appeal elements to be appealed in the new advertisement, based on the ad copy generation model.
[0079] [5] The apparatus described in [4], wherein the sample user is a target user to whom the advertisement text is delivered.
[0080] [6] The apparatus according to [4] or [5], further comprising: a clustering unit that clusters the plurality of advertising copy candidates into a plurality of classes; and a selection unit that selects a representative advertising copy from the plurality of advertising copy candidates in each of the plurality of classes.
[0081] [7] The apparatus according to [6], wherein the clustering unit vectorizes the plurality of advertising copy candidates and clusters the plurality of advertising copy candidates based on the Euclidean distance between the vectors.
[0082] [8] The apparatus according to [6] or [7], comprising: a classification model learning unit that learns a classification model for classifying user characteristics into the plurality of classes; and a class estimation unit that estimates the class into which a target user is classified from the user characteristics of the target user based on the classification model.
[0083] [9] The apparatus according to [8], further comprising a distribution unit that distributes the representative advertising text of the class estimated by the class estimation unit to the target users.
[0084]
[10] A method comprising: an extraction step of extracting candidate words that are candidates for appeal elements for each service from a plurality of delivered ad copy; an appeal element determination step of determining the appeal elements of the service based on the candidate words and the service features of the service corresponding to the candidate words; and an ad copy generation model learning step of learning an ad copy generation model that generates ad copy from user features and appeal elements based on the determined appeal elements of the service, the user features of responding users who responded to the delivered ad copy, and the delivered ad copy to which the responding users responded.
[0085] The block diagram used in the description of the above embodiment shows functional units. These functional blocks (components) are realized by any combination of at least one of hardware and software. Furthermore, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one device that is physically or logically coupled, or it may be realized using two or more physically or logically separated devices that are directly or indirectly connected (for example, using wired or wireless connections). A functional block may be realized by combining the one or more devices with software.
[0086] Functions include, but are not limited to, judgment, decision, determination, calculation, calculation, processing, derivation, investigation, exploration, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, assumption, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating (mapping), and assigning. For example, a functional block (configuration part) that enables transmission is called a transmitting unit or transmitter. In all cases, as mentioned above, the method of implementation is not particularly limited.
[0087] For example, the advertising text generation device 100 in one embodiment of the present disclosure may function as a computer that processes the prompt generation method of the present disclosure. Figure 12 is a diagram showing an example of the hardware configuration of the advertising text generation device 100 according to one embodiment of the present disclosure. The advertising text generation device 100 described above may be physically configured as a computer device including a processor 1001, memory 1002, storage 1003, communication device 1004, input device 1005, output device 1006, bus 1007, etc.
[0088] In the following explanation, the term "device" can be replaced with "circuit," "device," "unit," etc. The hardware configuration of the advertising copy generation device 100 may include one or more of the devices shown in the figure, or it may be configured to omit some of the devices.
[0089] Each function in the advertising copy generation device 100 is realized by loading predetermined software (programs) onto hardware such as the processor 1001 and memory 1002, which causes the processor 1001 to perform calculations, control communication by the communication device 1004, and control at least one of the reading and writing of data in the memory 1002 and storage 1003.
[0090] The processor 1001 controls the entire computer, for example, by running the operating system. The processor 1001 may be composed of a central processing unit (CPU) that includes interfaces with peripheral devices, control devices, arithmetic units, registers, etc. For example, the extraction unit 101, appeal element determination unit 102, advertising copy generation model learning unit 103, advertising copy candidate generation unit 104, clustering unit 105, selection unit 106, classification model learning unit 107, and class estimation unit 108 described above may be implemented by the processor 1001.
[0091] Furthermore, the processor 1001 reads programs (program code), software modules, data, etc., from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes accordingly. The program used is one that causes the computer to execute at least a part of the operations described in the above embodiment. For example, the extraction unit 101, the appeal element determination unit 102, the advertising copy generation model learning unit 103, the advertising copy candidate generation unit 104, the clustering unit 105, the selection unit 106, the classification model learning unit 107, and the class estimation unit 108 may be stored in the memory 1002 and implemented by a control program that operates on the processor 1001, and other functional blocks may be implemented similarly. The above-described various processes have been explained as being executed by one processor 1001, but they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may also be transmitted from a network via a telecommunications line.
[0092] The memory 1002 is a computer-readable recording medium and may consist of at least one of the following: ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), etc. The memory 1002 may also be called a register, cache, main memory, etc. The memory 1002 can store executable programs (program code), software modules, etc., for carrying out the prompt generation method according to one embodiment of the present disclosure.
[0093] The storage 1003 is a computer-readable recording medium and may consist of at least one of the following: an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disc, a digital multipurpose disc, a Blu-ray® disc), a smart card, flash memory (e.g., a card, a stick, a key drive), a floppy® disk, a magnetic strip, etc. The storage 1003 may also be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, server, or other suitable medium including at least one of memory 1002 and storage 1003.
[0094] The communication device 1004 is hardware (transmitting / receiving device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as a network device, network controller, network card, communication module, etc. The communication device 1004 may be configured to include high-frequency switches, duplexers, filters, frequency synthesizers, etc., in order to implement at least one of frequency division duplex (FDD) and time division duplex (TDD). For example, the distribution unit 109 described above may be implemented by the communication device 1004. The communication device 1004 may be implemented with physically or logically separated transmitting and receiving units.
[0095] The input device 1005 is an input device that accepts input from an external source (e.g., a keyboard, mouse, microphone, switch, button, sensor, etc.). The output device 1006 is an output device that outputs to an external source (e.g., a display, speaker, LED lamp, etc.). The input device 1005 and the output device 1006 may be configured as an integrated unit (e.g., a touch panel).
[0096] Furthermore, each device, such as the processor 1001 and memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or different buses may be configured for each device.
[0097] Furthermore, the advertising copy generation device 100 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array), and some or all of each functional block may be realized by such hardware. For example, the processor 1001 may be implemented using at least one of these hardware components.
[0098] Information notification is not limited to the embodiments described herein and may be carried out by other means. For example, information notification may be carried out by physical layer signaling (e.g., DCI (Downlink Control Information), UCI (Uplink Control Information)), upper layer signaling (e.g., RRC (Radio Resource Control) signaling, MAC (Medium Access Control) signaling, broadcast information (MIB (Master Information Block), SIB (System Information Block))), other signals, or combinations thereof. RRC signaling may also be called RRC messages, and may be, for example, RRC Connection Setup messages, RRC Connection Reconfiguration messages, etc.
[0099] The processing procedures, sequences, flowcharts, etc., of each aspect / embodiment described in this disclosure may be reordered, provided they do not contradict each other. For example, the methods described in this disclosure present various step elements using exemplary order and are not limited to the specific order presented.
[0100] Input and output information may be stored in a specific location (e.g., memory) or managed using a management table. Input and output information may be overwritten, updated, or appended to. Output information may be deleted. Input information may be transmitted to other devices.
[0101] The determination may be made by a value represented by one bit (0 or 1), by a boolean value (true or false), or by a numerical comparison (for example, a comparison with a predetermined value).
[0102] Each aspect / embodiment described in this disclosure may be used individually, in combination, or switched between as needed during implementation. Furthermore, notification of specific information (e.g., notification that "X is") is not limited to explicit notification, but may also be implicit (e.g., by not providing such notification).
[0103] Although the present disclosure has been described in detail above, it will be clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the intent and scope of the present disclosure as defined by the claims. Accordingly, the descriptions in the present disclosure are illustrative and not intended to be restrictive in any way.
[0104] Software should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, and so on, whether they are called software, firmware, middleware, microcode, hardware description languages, or by any other name.
[0105] Furthermore, software, instructions, information, etc., may be transmitted and received via a transmission medium. For example, if software is transmitted from a website, server, or other remote source using at least one of wired technologies (such as coaxial cable, fiber optic cable, twisted pair, or digital subscriber line (DSL)) and wireless technologies (such as infrared or microwave), then at least one of these wired and wireless technologies is included in the definition of a transmission medium.
[0106] The information, signals, etc. described in this disclosure may be represented using any of the various different techniques. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.
[0107] In addition, terms used in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings. For example, at least one of the channel and symbol may be a signal (signaling). Also, a signal may be a message. Furthermore, a component carrier (CC) may be called a carrier frequency, cell, frequency carrier, etc.
[0108] Furthermore, the information, parameters, etc., described in this disclosure may be expressed using absolute values, relative values from a given value, or other corresponding information. For example, wireless resources may be indicated by an index.
[0109] The names used for the parameters described above are not restrictive in any way. Furthermore, the formulas and other expressions using these parameters may differ from those expressly disclosed in this disclosure. Various channels (e.g., PUCCH, PDCCH, etc.) and information elements can be identified by any suitable name, and therefore, the various names assigned to these various channels and information elements are not restrictive in any way.
[0110] In this disclosure, terms such as "Mobile Station (MS)," "user terminal," "User Equipment (UE)," and "terminal" may be used interchangeably.
[0111] A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other appropriate term.
[0112] As used in this disclosure, the terms “determining” and “determining” may encompass a wide variety of actions. “Determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, or inquiring (e.g., searching in a table, database, or other data structure), or ascertaining. “Determining” may also include, for example, receiving (e.g., receiving information), transmitting (e.g., sending information), inputting, outputting, or accessing (e.g., accessing data in memory). Furthermore, "judgment" and "decision" can include considering something as having been "judged" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. In other words, "judgment" and "decision" can include considering something as having been "judged" or "decided" after some action. Also, "judgment (decision)" can be reinterpreted as "assuming," "expecting," or "considering."
[0113] The terms “connected,” “coupled,” or any variation thereof, mean any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” with each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, “connection” may be reinterpreted as “access.” As used in this disclosure, two elements may be considered to be “connected” or “coupled” with each other using at least one of one or more wires, cables, and printed electrical connections, and, in some non-limiting and non-exclusive examples, electromagnetic energy having wavelengths in the radio frequency domain, microwave domain, and optical (both visible and invisible) domain.
[0114] In this disclosure, the phrase "based on" does not mean "based solely on" unless otherwise specified. In other words, the phrase "based on" means both "based solely on" and "based at least on."
[0115] Any reference to elements using designations such as “first,” “second,” etc., as used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Accordingly, references to first and second elements do not imply that only two elements may be employed, or that the first element must precede the second element in any way.
[0116] Where the terms “include,” “including,” and their variations are used in this disclosure, these terms are intended to be inclusive, as is the term “comprising.” Furthermore, the term “or” as used in this disclosure is not intended to be exclusive OR.
[0117] In this disclosure, if articles are added by translation, such as a, an, and the in English, this disclosure may include the fact that the noun following these articles is plural.
[0118] In this disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "combine" may be interpreted similarly to "different."
[0119] 100...Advertising text generation device, 101...Extraction unit, 102...Appeal element determination unit, 103...Advertising text generation model learning unit, 104...Advertising text candidate generation unit, 105...Clustering unit, 106...Selection unit, 107...Classification model learning unit, 108...Class estimation unit, 109...Distribution unit, 200...Large-scale language model, 300...Knowledge database, 400...User terminal, 1001...Processor, 1002...Memory, 1003...Storage, 1004...Communication device, 1005...Input device, 1006...Output device, 1007...Bus.
Claims
1. An apparatus comprising: an extraction unit that extracts candidate words that are candidates for appeal elements for each service from multiple delivered ad copy; an appeal element determination unit that determines the appeal elements of the service based on the candidate words and the service features of the service corresponding to the candidate words; and an ad copy generation model learning unit that learns an ad copy generation model that generates ad copy from user features and appeal elements based on the determined appeal elements of the service, the user features of users who responded to the delivered ad copy, and the delivered ad copy to which the responding users responded.
2. The apparatus according to claim 1, wherein the appeal element determination unit vectorizes the candidate words for each service and the service features of the service corresponding to the candidate words, and determines the candidate words whose similarity to the service features of the service is above a threshold as the appeal elements of the service.
3. The apparatus according to claim 1, further comprising a knowledge database storing the delivered ad copy, the user characteristics of the users to whom the delivered ad copy was delivered, and the users who responded to the delivered ad copy, wherein the ad copy generation model learning unit learns an ad copy generation model based on the information stored in the knowledge database.
4. The apparatus according to claim 1, further comprising an ad copy candidate generation unit that generates a plurality of ad copy candidates corresponding to the plurality of sample users from the user characteristics of the plurality of sample users and the target appeal elements which are appeal elements to be appealed in the new advertisement, based on the ad copy generation model.
5. The apparatus according to claim 4, wherein the sample user is a target user to whom the advertisement text is delivered.
6. The apparatus according to claim 4, further comprising: a clustering unit that clusters the plurality of advertising copy candidates into a plurality of classes; and a selection unit that selects a representative advertising copy from the plurality of advertising copy candidates in each of the plurality of classes.
7. The apparatus according to claim 6, wherein the clustering unit vectorizes the plurality of advertising copy candidates and clusters the plurality of advertising copy candidates based on the Euclidean distance between the vectors.
8. The apparatus according to claim 6, comprising: a classification model learning unit that learns a classification model for classifying user characteristics into the plurality of classes; and a class estimation unit that estimates the class into which a target user is classified from the user characteristics of the target user based on the classification model.
9. The apparatus according to claim 8, further comprising a distribution unit that distributes the representative advertising text of the class estimated by the class estimation unit to the target users.
10. A method comprising: an extraction step of extracting candidate words that are candidates for appeal elements for each service from multiple delivered ad copy; an appeal element determination step of determining the appeal elements of the service based on the candidate words and the service features of the service corresponding to the candidate words; and an ad copy generation model learning step of learning an ad copy generation model that generates ad copy from user features and appeal elements based on the determined appeal elements of the service, the user features of users who responded to the delivered ad copy, and the delivered ad copy to which the responding users responded.