Notification device, notification method, and notification program

Abstract

This enables personalized push notifications for each user. [Solution] The notification device 100 generates a push notification based on information about a user using the target service and a predetermined model configured to generate a push notification, which is a message output from the target service to the user to encourage a predetermined action by the user using the target service. The notification device 100 outputs the generated push notification to the user.

Description

Technical Field

[0004] , , , , ,

[0006] , , , , , ,

[0005] , , , , , ,

[0001] The present invention relates to a notification device, a notification method, and a notification program. [[ID=­7]]

Background Art

[0002] In recent years, a technique for generating conversation content for users using generative AI (Artificial Intelligence) such as large language models has been known. For example, a conventional technique for generating response content of a dialogue agent based on user language information and user non-verbal information using a large language model is known (see, for example, Patent Document 1).

[0003] And, by automatically generating appropriate conversation content in response to an input of a conversation from a user based on the above-described conventional technique, for example, an automatic response to an inquiry regarding a service from a user can be realized.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the conventional technique has a problem in performing push notifications according to users. For example, the conventional technique can generate conversation content according to input information from a user using a large language model. However, the conventional technique only generates conversation content according to an input from a user, and there is a problem in spontaneously transmitting (notifying) a message (push notification) including a hearing for the user and a proposal related to a service without waiting for an input from the user.

Means for Solving the Problems

[0006] Therefore, in order to solve the above-mentioned problems and achieve the objective, the notification device of the present invention is characterized by having a generation unit that generates a push notification based on information about a user using the target service and a predetermined model configured to generate a push notification, which is a message output from the target service to the user in order to prompt the user to take a predetermined action using the target service, and an output unit that outputs the generated push notification to the user. [Effects of the Invention]

[0007] The present invention has the effect of enabling push notifications tailored to each user. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1 is a diagram illustrating the overall processing of the notification device according to this embodiment. [Figure 2] Figure 2 shows the configuration of the notification device according to the embodiment. [Figure 3] Figure 3 is a table diagram showing an example of user information according to the embodiment. [Figure 4] Figure 4 is a table diagram showing an example of classification conditions according to the embodiment. [Figure 5] Figure 5 is a diagram illustrating a first example according to the embodiment. [Figure 6] Figure 6 shows an example of a prompt used in the first example according to the embodiment. [Figure 7] Figure 7 illustrates a second example according to the embodiment. [Figure 8] Figure 8 shows an example of a prompt used in a second example according to the embodiment. [Figure 9] Figure 9 illustrates a third example according to the embodiment. [Figure 10] Figure 10 shows an example of a prompt used in a third example according to the embodiment. [Figure 11]Figure 11 is a flowchart showing the processing performed by the notification device according to the embodiment. [Figure 12] Figure 12 shows an example of a computer that implements a notification device according to the embodiment. [Modes for carrying out the invention]

[0009] Hereinafter, embodiments for carrying out the present invention (hereinafter referred to as "embodiments") will be described with reference to the drawings. However, each embodiment is not limited to those described below.

[0010] <Overview> (background) In recent years, numerous solutions and communication services (hereinafter sometimes referred to as "Target Services") have been provided for use in sales activities, organizational management, etc. Users of the Target Services (hereinafter sometimes simply referred to as "Users") may have various opinions, complaints, requests, and questions (hereinafter sometimes simply referred to as "Requests") regarding the Target Services, and it is the responsibility of the administrators of the Target Services to provide support to such users.

[0011] Therefore, in some cases, natural language-based user support is provided using reference technologies (see, for example, Reference 1 listed below), such as generating the content of a dialogue agent's response based on the user's linguistic and non-linguistic information using large-scale language models.

[0012] (Reference 1) Japanese Patent Publication No. 2024-112283

[0013] By using the aforementioned reference technologies to automatically generate appropriate conversation content in response to user input, it becomes possible to provide human-free responses to user inquiries, including opinions, complaints, requests, and questions regarding the target service.

[0014] However, since the above-described related art only generates conversation content in response to an input from the user, there is a problem in actively outputting a message including a hearing for the user and a proposal related to the service without waiting for an input from the user.

[0015] (Processing by Notification Device 100) Therefore, the notification device 100 according to the present embodiment realizes the generation and output of push notifications based on information regarding a user who uses a target service and a predetermined model such as a large language model.

[0016] The above-mentioned "information regarding a user who uses a target service" is information including attribute information related to the user, information related to the conversation history, information related to the usage status of the target service, etc., and may hereinafter be simply referred to as "user information". Also, the above-mentioned "information related to the conversation history" may hereinafter be referred to as "conversation history information". Also, the above-mentioned "information related to the usage status of the target service" may hereinafter be referred to as "service usage information".

[0017] The above-mentioned "push notification" is a message output from the target service side to the user in order to prompt a predetermined action by the user who uses the target service (for example, a response to a hearing or proposal from an administrator of the target service, etc., a change in the user's behavior, etc.). Also, the above-mentioned "administrator" includes a person who manages the target service provided to users such as customers and company employees based on the notification device 100, and for example, a person in charge of the customer, a person who supports the end user, etc.

[0018] Here, the overall picture of the processing by the notification device 100 will be described. FIG. 1 is a diagram for explaining the overall picture of the processing of the notification device 100 according to the embodiment. The notification device 100 shown in FIG. 1 is an example of a computer that provides a technology for realizing the information processing described below.

[0019] The notification device 100 receives prompts in natural language that instruct it to generate a push notification from a large-scale language model 10, which is provided with prior knowledge such as user information, and generates a push notification (Figure 1 (1-1) to (1-3)).

[0020] Next, the notification device 100 outputs a push notification generated based on user information and the large-scale language model 10 to user 1 at a predetermined timing based on a predetermined format such as email, chat, audio, or video (Figure 1(2)).

[0021] The notification device 100 can then output information to administrator 2, such as a response to user 1 and information regarding escalation to the administrator of the target service, based on user 1's response to the push notification (Figure 1 (3-1)). (Figure 1 (3-2)) The "escalation information" mentioned above is information that communicates the user's situation to the administrator so that the administrator can take appropriate action, and may be referred to as "escalation information" hereafter.

[0022] Through the process described above, the notification device 100 proactively outputs the generated push notification to the target user from the target service side, enabling administrators to appropriately grasp user trends such as opinions, requests, and potential dissatisfactions, propose new services to users, and follow up with users. In other words, the notification device 100 has the effect of shifting from a "defensive" stance, which was previously to send messages in response to user input, to an "offensive" stance, which is to proactively send push notifications tailored to the user.

[0023] <Description of notification device 100> The configuration of the notification device 100 according to this embodiment will now be described. Figure 2 is a diagram showing the configuration of the notification device 100 according to this embodiment. As shown in Figure 2, the notification device 100 has a communication unit 110, a storage unit 120, and a control unit 130.

[0024] Although not shown in Figure 2, the notification device 100 may be equipped with an input unit such as a keyboard or mouse to receive input such as operations from an administrator. The notification device 100 may also be equipped with a display or the like to show user information, conversation history information, the content of a push notification, or the user's response to the push notification to an administrator.

[0025] (Communications Department 110) The communication unit 110 performs data communication related to the input of user information, conversation history information, and statistical information on the response time for each user. The communication unit 110 also performs data communication related to the output of push notifications generated by the generation unit 132, which will be described later.

[0026] The communication unit 110 is implemented using a NIC (Network Interface Card) or the like, and controls communication via telecommunication lines such as a LAN (Local Area Network) or the Internet. The communication unit 110 can be connected to the network via wired or wireless connections as needed, and can send and receive information bidirectionally with other server devices or user-operated terminal devices.

[0027] (Storage unit 120) The storage unit 120 stores data and programs used for various processes performed by the control unit 130, as well as various data acquired through the operation of the control unit 130. The storage unit 120 is implemented using semiconductor memory elements such as RAM (Random Access Memory) and flash memory, or storage devices such as hard disks and optical discs. As shown in Figure 2, the storage unit 120 also includes a user information DB 121, a statistical information DB 122, a classification condition DB 123, and a model DB 124.

[0028] (User Information DB121) User Information DB121 is a database that stores user information such as attribute information, past conversation history information, service usage status, and recommended notification times for push notifications to users of the target service. Here, an example of user information stored in User Information DB121 will be explained using Figure 3. Figure 3 is a table diagram showing an example of user information according to the embodiment.

[0029] Specifically, the user information DB121 associates information related to each of the following items in the user information: "No," which identifies individual data contained in the user information; "User Identification Information"; "Attribute Information"; "Conversation History Information"; "Service Usage Status"; and "Recommended Notification Time." This information is stored, for example, in a table format as shown in Figure 3. The letters "A through E" listed for each item in the table diagram shown in Figure 3 are legends for the information contained in each item.

[0030] The "user identification information" mentioned above refers to information that identifies a user of the service in question, and includes, for example, information such as the user's name, nickname, telephone number, email address, and user-specific information that is an arbitrary combination of text, numbers, symbols, etc.

[0031] "Attribute information" includes demographic data such as age, gender, affiliation, occupation, income, and residential area of ​​the user, as well as psychographic data such as hobbies, preferences, values, and lifestyle. Furthermore, attribute information may also include the priority of support for each user.

[0032] "Conversation history information" refers to information about the history of conversations between users of the target service and the administrators or chatbots of the target service, and includes, for example, information in which the conversation exchange is expressed in natural language. "Service usage status" refers to the status of the user's use of the target service, and includes, for example, information such as the name of the contracted service, contract details, usage time, frequency of use, and amount of data used.

[0033] The "recommended notification time" is information about the time period when the user is most likely to respond when a push notification is sent to them, and includes, for example, the "recommended notification time for push notifications" calculated by the calculation unit 133 described later.

[0034] (Statistics information DB122) The statistical information DB122 is a database that stores statistical information, such as conversation history information from which personally identifiable information has been removed, and information such as the reply time ranges for different attributes of multiple users, calculated by the calculation unit 133 described later, using conversation history information stored in the user information DB121.

[0035] Specifically, the statistical information DB122 uses information about conversations and interactions between users of the target service and the administrators or chatbots of the target service to store conversation history information as statistical information, after any personally identifiable information (personal information) has been removed.

[0036] For example, the conversation history information from which the above-mentioned personal information has been deleted may include information such as "attribute information of multiple users," "text of the conversation," "date and time the conversation took place," and "information other than the conversation text that was provided to the user."

[0037] Furthermore, the statistical information DB122 stores, as statistical information, information regarding the response time, such as the time of day when replies (responses) are made to push notifications and other messages, for each user group classified by predetermined attributes, as calculated by the calculation unit 133 described later.

[0038] (Classification condition DB123) The classification condition DB123 is a database that stores classification conditions, which are information about the conditions used when executing processing in response to a user's response to a push notification. Here, an example of classification conditions stored in the classification condition DB123 will be explained using Figure 4. Figure 4 is a table diagram showing an example of classification conditions according to the embodiment.

[0039] Specifically, the classification condition DB123 stores information related to each item, such as the "classification" which identifies individual classification conditions, and the "conditions" and "processing details" corresponding to each classification, in a table format, for example, as shown in Figure 4. From here, we will explain each of the conditions shown in the table diagram in Figure 4.

[0040] Conditions identified by classification "0" include "the conversation history does not contain keywords indicating an intention or indication of cancellation" and "it is evaluated as not containing negative emotions." If the conditions identified by classification "0" are met, the user is in a state such as "there are no potential complaints, etc. (neutral)." When classified as classification "0," the generation unit 132 performs, for example, "generating new proposals tailored to the user."

[0041] Conditions identified by classification "1" include "the conversation history contains keywords indicating a potential cancellation" and "it is evaluated as a negative emotion." For example, if the conditions identified by classification "1" are met, the user is in a state where "there are potential dissatisfactions, opinions, requests, etc., and if left unaddressed, it may lead to cancellation of the service."

[0042] If the case is classified as category "1", the generation unit 132 generates a response and escalation information for the user. For example, the escalation information related to category "1" described above includes information such as, "There are potential complaints, opinions, requests, etc., and if left unattended, this may lead to the user canceling the service, so action from the administrator is required."

[0043] Conditions identified by classification "2" include "the conversation history contains keywords indicating an intention to cancel" and "it is evaluated as a negative emotion." If the conditions identified by classification "2" are met, the user is in a state where "there are explicit dissatisfactions, opinions, requests, etc., which will lead to the cancellation of the service."

[0044] If the case is classified as category "2", the generation unit 132 generates a response and escalation information for the user. For example, the escalation information related to category "2" described above includes information such as, "There are explicit complaints, opinions, requests, etc., which are likely to lead to service cancellation, and therefore urgent action from the administrator is required."

[0045] The conditions identified by classification "3" include "the inclusion of keywords related to requests for service improvements," etc. If the conditions identified by classification "3" are met, the user is in a state where "there are no overt complaints, but there are requests for service improvements, etc."

[0046] If the case is classified as category "3", the generation unit 132 generates escalation information that includes the response to the user and information such as requests for improvement extracted from the response from the user. For example, the escalation information related to category "3" described above includes information such as, "There are no overt complaints, but there are requests for service improvements, so it is necessary to consider improving the service."

[0047] The conditions identified by classification "4" include "includes keywords related to opinions about the service," etc. If the conditions identified by classification "4" are met, the user is in a state where "there are no overt complaints, but there are opinions about the service or push notifications," etc.

[0048] If the case is classified as category "4", the generation unit 132 generates escalation information that includes the response to the user and information such as opinions extracted from the response from the user. For example, the escalation information related to category "4" described above includes information such as, "There are no overt complaints, but there are opinions about the service, so it is necessary to consider improving the service and the content of push notifications based on these opinions."

[0049] Furthermore, if the user is classified as category "4", the learning unit 136, described later, uses the opinions extracted from the user's responses to train a predetermined model, such as a large-scale language model.

[0050] The conditions identified by classification "5" include "keywords related to the user's efforts and challenges in health management." If the conditions identified by classification "5" are met, the user is in a state where "challenges regarding the user's health management have become apparent."

[0051] If the user is classified as category "5", the generation unit 132 generates escalation information that includes information such as issues extracted from the user's response. For example, the escalation information related to category "5" described above includes information such as, "The user has the following issue regarding health management, so follow-up is necessary for this user."

[0052] The conditions identified by classification "6" include conditions such as "the content of the user inquiry is included." If the conditions identified by classification "6" are met, the user is in a state such as "an inquiry has been made by the user."

[0053] If the inquiry is classified as category "6", the generation unit 132 generates a response to the user's inquiry. For example, the escalation information related to category "6" mentioned above includes information such as, "The user may have questions about the service in question regarding XX, therefore, follow-up is necessary for the user."

[0054] (Model DB124) Model DB124 is a database that stores predetermined models used for generating push notifications by the generation unit 132 (described later) and for calculating the recommended notification time for push notifications by the calculation unit 133 (described later).

[0055] Specifically, the model DB 124 stores a large-scale language model as a predetermined model used by the generation unit 132. For example, the model DB 124 can store at least one of the following as a large-scale language model: "ChatGPT®", which is a large-scale language model with general knowledge, and "tsuzumi®", which is a predetermined large-scale language model on which adapter tuning is performed (see, for example, references 2 and 3).

[0056] (Reference 2):ChatGPT(OpenAI),<URL:https: / / openai.com / chatgpt> ,<Searched on January 20, 2020> (Reference 3): NTT version of large-scale language model "tsuzumi",<URL:https: / / www.rd.ntt / research / LLM_tsuzumi.html> ,<Searched on January 20, 2020>

[0057] Furthermore, the model DB124 stores, as a predetermined model, a machine learning model that has been trained to generate push notifications and responses corresponding to user information, etc., using user information as input, which is used by the generation unit 132.

[0058] Furthermore, the model DB124 stores a machine learning model, which is used by the calculation unit 133 and has been trained to calculate the recommended notification time for push notifications related to a user, as a predetermined model.

[0059] The machine learning models mentioned above are not particularly limited and may include, for example, any known machine learning models trained based on supervised learning, unsupervised learning, reinforcement learning, etc.

[0060] (Control unit 130) Now, let's return to Figure 2 and continue the explanation. The control unit 130 has an internal memory for temporarily storing programs and processing data that define various processing procedures of the notification device 100, and is realized by electronic circuits such as a CPU (Central Processing Unit) and an MPU (Micro Processing Unit), and integrated circuits such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array). As shown in Figure 2, the control unit 130 has an acquisition unit 131, a generation unit 132, a calculation unit 133, a classification unit 134, an output unit 135, and a learning unit 136.

[0061] (Acquisition part 131) The acquisition unit 131 can acquire user information and conversation history information from an external information processing device, etc., via the communication unit 110 described above, and store it in the storage unit 120. In addition, the acquisition unit 131 can acquire user information entered by the administrator of the target service, etc., via the input unit described above, and store it in the storage unit 120.

[0062] (Generation unit 132) The generation unit 132 generates a push notification based on user information and a predetermined model configured to generate push notifications. For example, the generation unit 132 generates a push notification that includes at least one of the following: a conversational text for hearing the user's requests, and a conversational text for making a predetermined proposal to the user regarding services tailored to the user.

[0063] In terms of the specific processing flow, the generation unit 132 inputs prompts, expressed in natural language, which include a command to use user information relating to the user of the target service as prior knowledge, and a command to generate a push notification tailored to the user based on the prior knowledge, into a predetermined large-scale language model to generate a push notification. The prior knowledge (user information) mentioned above includes, for example, at least one of the following: user attribute information, conversation history information, and service usage information.

[0064] Here, we will illustrate with several examples the instructions for the large-scale language model included in the prompts mentioned above.

[0065] (Prompt example 1) The prompt described above may include a command that provides a large-scale language model with prior knowledge regarding the user's usage of the target service, and a command that generates conversational text for the target service to proactively propose a usage plan for the target service to the user in accordance with the user's usage.

[0066] (Prompt example 2) The prompts described above may include instructions that provide a large-scale language model with prior knowledge regarding the user's usage of the target service, and instructions that generate conversational text for the target service to proactively communicate with the user, such as thanks or alerts, depending on the user's usage of the service.

[0067] (Prompt example 3) The prompts described above may include instructions for generating conversational text that proactively proposes to the user the service provider's policies and user interaction methods set by the service provider's administrator, as well as usage plans tailored to the user's service contract status and usage.

[0068] The conversational text used by the service provider to proactively make suggestions to the user in question includes "conversational text that proactively suggests plans that the user is likely to subscribe to, plans that the administrator particularly wants to suggest, and plans that are more suitable for the user's usage (for example, suggesting a higher bandwidth plan to a user with high data usage), etc., in accordance with the user's usage trends of the service provider."

[0069] For example, information selected by the administrator from pre-registered information on events, campaigns, recommended products, etc., related to the target service may be included in the conversation text of the push notification. As a specific example, if "Service A" is registered as a recommended product for the target service, and the administrator's request is to hear about the user's problems, then conversation text including "Do you have any problems? Security awareness has been increasing recently, and we have been receiving more inquiries about our Service A" may be generated.

[0070] (Prompt example 4) The prompts described above may include instructions to generate conversational messages that proactively initiate thank-you messages, inquiries, and suggestions from the service provider in response to changes in the user's usage of the service. Examples of such conversational messages that proactively initiate thank-you messages, inquiries, and suggestions from the service provider in response to changes include conversational messages to the user expressing gratitude when it is detected that a user has added a location to their network, and conversational messages suggesting measures to improve bandwidth congestion.

[0071] Furthermore, the generation unit 132 can also generate push notifications based on a machine learning model that has been trained to generate push notifications and responses to the user based on pre-set conversation branches.

[0072] For example, the generation unit 132 can generate push notifications based on a machine learning model that has been trained to output questions such as "Are you interested in the XX service?", options such as "Yes or No", and information about the destination from the options, such as an application link if yes, or information about the archive if no.

[0073] (Calculation section 133) The calculation unit 133 calculates the time period when the response rate from users is highest when push notifications or other messages are sent to users of the target service as the recommended notification time for push notifications.

[0074] Specifically, the calculation unit 133 inputs statistical information on the response times of individual target users into a pre-trained machine learning model and calculates the recommended notification time for push notifications to those individual target users. Here, the calculation unit 133 can use a machine learning model that has been trained to output the recommended notification time for push notifications to individual users in response to the input of statistical information on the response times of individual users, using statistical information on the response times of multiple users by attribute, which is calculated using the attribute information of each user and the past response times of each user.

[0075] Furthermore, the calculation unit 133 can calculate statistics on response times by attribute using past conversation history information for multiple users and information on the response times of users. For example, the calculation unit 133 calculates the recommended notification time for push notifications based on the time of day when the response rate is maximized for multiple users and the time of day when the response rate is maximized for the target user.

[0076] Furthermore, the calculation unit 133 can also input a prompt to the large-scale language model that includes a command to calculate the recommended notification time for push notifications, thereby calculating the recommended notification time for push notifications tailored to the user.

[0077] For example, the calculation unit 133 can input a prompt to the large-scale language model that includes a command to provide the target user's conversation history information as prior knowledge to the large-scale language model, and a command to calculate the time at which the target user's response rate and read rate to push notifications are maximized based on the prior knowledge, thereby calculating the recommended notification time for push notifications related to the target user.

[0078] (Classification section 134) The classification unit 134 classifies responses from users of the service based on predetermined keywords included in the user's response to the push notification, or based on the results of sentiment analysis.

[0079] Specifically, the classification unit 134 determines whether or not pre-registered keywords such as "cancellation, dissatisfaction, slow, frequently interrupted, difficult to connect, unable to connect, disconnected, difficult to understand, don't understand, expensive, no contact received" are included in the user's response.

[0080] Furthermore, the classification unit 134 performs sentiment analysis on the user using the user's responses. For example, the classification unit 134 inputs the user's conversation history over a certain period of time and prompts containing instructions to summarize the user's statements based on that conversation history into a large-scale language model to summarize the user's responses. Next, the classification unit 134 performs sentiment analysis on the summarized responses based on known techniques.

[0081] As described above, the classification unit 134 classifies the user's response based on the results of the evaluation of the user's response using pre-registered keywords and the results of the sentiment analysis. For example, if the user's response contains keywords indicating a potential cancellation and the numerical value related to negative emotions exceeds a pre-set threshold, the classification unit 134 classifies it as "Classification: 1".

[0082] (Output section 135) The output unit 135 outputs the generated push notification to the user. For example, the output unit 135 outputs a push notification to the target individual user based on the calculated recommended notification time for the push notification to that individual user. An example of the output by the output unit 135 will be explained in detail in the sections related to the first to third examples described below.

[0083] (Learning Section 136) When the learning unit 136 classifies a user's response as containing an opinion about the service, it learns a predetermined model using the user's opinion extracted from the conversation history of users who use the service.

[0084] For example, if the classification unit 134 classifies the user's response as "Classification: 4", the learning unit 136 extracts the user's opinion from that response. The learning unit 136 then learns a predetermined model to generate a push notification corresponding to the extracted opinion.

[0085] In this embodiment, "learning" includes providing prior knowledge to a large-scale language model and training a machine learning model. For example, the learning unit 136 can learn the large-scale language model by inputting prompts that include commands to reflect the extracted opinions in the conversation text included in the push notification.

[0086] (An example of processing) From here, we will explain an example of processing by the notification device 100 using Figures 5 to 10. Figures 5 and 6 explain an example of a "sales support use case" as the first example. Figures 7 and 8 explain an example of a "health management support use case" as the second example. Figures 9 and 10 explain an example of an "inquiry response use case" as the third example.

[0087] (Example 1) First, the processing flow of "An Example of a Sales Support Use Case" will be explained using Figure 5. Figure 5 is a diagram illustrating the first example according to the embodiment. The example shown in Figure 5 is a use case for support (sales support) in order to address issues related to contact with existing customers, such as hearing about potential issues that customers have and making proposals to customers. In the first example, the user is customer 1a who uses the target service.

[0088] Traditionally, when conducting sales activities with customers, each salesperson needs to take appropriate actions according to the attributes of the target customer. However, when the target customers have attributes such as "medium to low average customer spending" and "a large number of customers," it can be difficult to properly support each individual customer. As a result, it becomes difficult to grasp customer trends due to insufficient support, which can lead to missed additional business opportunities or service cancellations without any complaints.

[0089] Therefore, in the first example, the notification device 100 is provided with user information such as attribute information, conversation history information, and service usage status related to customer 1a as prior knowledge, and generates push notifications with content tailored to customer 1a using a large-scale language model that has been configured with a personality and role that is tailored to customer 1a. The notification device 100 in the first example then actively outputs the generated push notifications to customer 1a.

[0090] From here, we will explain the sequence of operations of the notification device 100 in the first example. The notification device 100 (generation unit) in the first example inputs a prompt (P1 in Figure 5) into the large-scale language model 11 that describes the acquisition of prior knowledge, the setting of the role of the large-scale language model 11, and the content of the push notification to be generated, and generates a push notification ((1-1) in Figure 5).

[0091] Specifically, the notification device 100 (generation unit) inputs a prompt P1, which includes a command to generate a conversational text for hearing customer 1a's requests regarding the target service or making a predetermined proposal regarding the service, based on customer 1a's attribute information and conversational history information, to a large-scale language model 11 that has been configured to set the words to be used for push notifications notified from the target service and a predetermined conversational development method based on prior knowledge such as conversational history information of operators, etc., in a contact center. The notification device 100 (generation unit) generates the conversational text for hearing customer 1a's requests regarding the service or making a predetermined proposal regarding the service as a push notification.

[0092] Here, an example of a prompt (P1 in Figure 5) relating to the first example will be explained using Figure 6. Figure 6 is a diagram illustrating an example of a prompt used in the first example according to the embodiment. The prompt P1 shown in Figure 6 includes "<definition of role> (Figure 6 (1))" and "description for causing the large-scale language model to execute <task> according to <constraints> (Figure 6 (2) to (4))" which are set for the large-scale language model.

[0093] In the "<Role Definition>" shown in (1) of Figure 6, for example, a command is described that sets the role of the large-scale language model as "a person who proposes services to users (customers) or a person who listens to customer complaints and opinions."

[0094] The “<Processing Instruction>” shown in Figure 6 (2) contains instructions to execute the “<Task>” described in the prompt according to the “<Constraints>”.

[0095] The "<Task>" shown in (3) of Figure 6 contains instructions such as "a command to generate a push notification tailored to the customer" and "a command to generate predetermined output information (such as a response or escalation information) corresponding to the customer's response if a response is received."

[0096] The "<Constraints>" shown in (4) of Figure 6 describes the following conditions that the large-scale language model must strictly adhere to when performing the <Task>. The "<Constraints>" described in area (4) of Figure 6 allows the notification device 100 (generation unit) to cause the large-scale language model to generate push notifications that the administrator of the target service or others desire to output to customers.

[0097] For example, the constraints described in area (4-1) of Figure 6 define the following operator profile (person profile) for the large-scale language model. (1) Carefully listen to customer requests and concerns to contribute to improving customer satisfaction. (2) Respond using polite yet friendly language. (3) When a customer's problem or request is related to the service in question, we will investigate the matter further and respond appropriately. (4) Messages such as push notifications to customers are not verbose and are concise.

[0098] Furthermore, due to the constraints described in area (4-2) of Figure 6, the notification device 100 (generation unit) can cause the large-scale language model to generate push notifications for conducting interviews and making proposals tailored to the target customer.

[0099] Specifically, the notification device 100 (generation unit) can generate conversational text that prompts a response from the customer (user) based on the customer's attribute information stored in the user information DB 121, according to the assumed customer profile. For example, if the notification device 100 (generation unit) assumes the customer's profile is "quick-tempered" based on the customer's attribute information, it can generate a conversational text as a push notification that takes into consideration not to provoke anger in the customer.

[0100] Furthermore, the constraints described in area (4-3) of Figure 6 allow the notification device 100 (generation unit) to generate push notifications or output information corresponding to the classification using a large-scale language model. For example, in the case of "Classification: 0", the notification device 100 (generation unit) generates conversational text ("proactive" content) for making a new proposal to the customer regarding the target service.

[0101] In the case of "Category 1," the notification device 100 (generation unit) generates a conversational message ("defense" content) for conducting an interview with the customer to prevent the customer from canceling the service. In the case of "Category 2," the notification device 100 (generation unit) generates a conversational message ("defense" content) for apologizing to the customer and conducting an interview to prevent the customer from canceling the service, as well as escalation information. In addition, in the case of "Category 3" or "Category 4," the notification device 100 (generation unit) extracts "requests" and "opinions" included in the customer's response and generates escalation information.

[0102] Now, let's return to Figure 5 and continue the explanation. The notification device 100 (output unit) outputs a push notification (Figure 5 (1-2)) containing conversational text such as "suggestions regarding the target service" and "hearings regarding improvement requests, opinions, complaints, etc." to customer 1a at the recommended notification time (Figure 5 (2)). As a result, the notification device 100 can reduce customer 1a's aversion by outputting the push notification during the time when customer 1a's response rate is high.

[0103] Next, the notification device 100 (classification unit) performs classification on customer 1a's response to the push notification (Figure 5 (3-1)). For example, based on the content of customer 1a's response, such as "I am dissatisfied with the quality and would like to cancel the contract" (Figure 5 (3-2)), the notification device 100 (classification unit) classifies customer 1a's response as containing keywords or negative emotions related to service cancellation (classification "2"). Then, the notification device 100 (generation unit) causes the large-scale language model 11 to generate the response to customer 1a corresponding to classification "2" (Figure 5 (3-3)) and escalation information to be output to customer representative 2a (Figure 5 (3-4)) as output information.

[0104] Specifically, the notification device 100 (generation unit) generates a response to customer 1a's statement, "I am dissatisfied with the quality and would like to cancel the contract," which includes conversational phrases such as, "We are very sorry. We would like to offer some suggestions for improvement, but could you please tell us more about your dissatisfaction?" (Figure 5 (3-5)).

[0105] Furthermore, the notification device 100 (generation unit) generates conversational text such as, "The customer is dissatisfied with the service and may cancel the service. Please follow up immediately" as escalation information (output information) for the service administrator (customer representative 2a) (Figure 5 (3-6)).

[0106] As described above, the notification device 100 enables customer representative 2a to follow up with customer 1a (by phone) through appropriate escalation according to customer 1a's situation (Figure 5 (4)). Furthermore, the notification device 100 can improve the accuracy of subsequent push notifications by learning a large-scale language model using opinions and other information included in customer 1a's responses.

[0107] (Second example) Next, the processing flow of "An Example of a Health Management Support Use Case" will be explained using Figure 7. Figure 7 is a diagram illustrating a second example according to the embodiment. The example shown in Figure 7 is a use case for supporting behavioral changes of managers (users) within a company based on the vital data of its employees, in order to address challenges faced by companies engaged in health management.

[0108] In the second example, the user refers to a manager 1b (an employee engaged in management, etc.), who is a senior employee who manages and supervises the organization and employees of a company that uses the health management support service. In the second example, "vital data" includes biometric data such as hourly heart rate data acquired from sensor devices worn by employees, and is used to analyze autonomic nervous system balance, the degree of fatigue accumulation, concentration / stress levels during work, etc.

[0109] The aforementioned "health management" is a management approach that emphasizes strategically implementing employee health management from a business perspective. It involves understanding the health status of employees based on their vital data and implementing various measures. However, while various surveys, including the acquisition of vital data, are conducted to understand the current situation, it can be difficult to identify issues from the results obtained and to encourage behavioral changes among managers to resolve those issues.

[0110] Therefore, in the second example, the notification device 100 is provided with prior knowledge of conversation history information of experts related to health management, which is a management method that emphasizes employee health management (conversation history information of health management experts), and a large-scale language model is set up to perform consulting to encourage behavioral change in manager 1b. Based on the employee's vital data, the notification device 100 generates push notifications with content tailored to manager 1b. The notification device 100 in the second example then proactively outputs the generated push notifications to manager 1b.

[0111] From here, we will explain the sequence of operations of the notification device 100 in the second example. The notification device 100 (generation unit) in the second example inputs a prompt (P2 in Figure 7) into the large-scale language model 12 that describes the acquisition of prior knowledge, the setting of the role of the large-scale language model 12, and the content of the push notification to be generated, and generates a push notification ((1-1) in Figure 7).

[0112] Specifically, the notification device 100 (generation unit) inputs a prompt to a large-scale language model 12, which is provided with conversation history information of health management experts as prior knowledge. This prompt includes a command to generate information that encourages behavioral change by manager 1b in accordance with the employee's vital data, which is service usage information. The notification device then generates this information as a push notification.

[0113] Here, an example of a prompt (P2 in Figure 7) relating to the second example will be explained using Figure 8. Figure 8 is a diagram illustrating an example of a prompt used in the second example according to the embodiment. The prompt P2 shown in Figure 8 includes "<definition of role> (Figure 8 (1))" and "description for causing the large-scale language model to execute <task> according to <constraints> (Figure 8 (2) to (4))" which are set for the large-scale language model.

[0114] The “<Role Definition>” shown in Figure 8 (1) contains, for example, a directive that sets the role of the large-scale language model as “an expert who supports the health management of an organization.”

[0115] The “<Processing Instruction>” shown in Figure 8 (2) contains instructions to execute the “<Task>” described in the prompt according to the “<Constraints>”.

[0116] The "<Task>" shown in (3) of Figure 8 includes instructions such as "an instruction to analyze employee vital data and generate a push notification that includes an explanation of the organization's state and suggestions for actions to encourage behavioral change in managers" and "an instruction to generate predetermined output information (such as responses and escalation information) corresponding to the response received from managers."

[0117] The "<Constraints>" shown in (4) of Figure 8 describes the following conditions that the large-scale language model must strictly adhere to when performing the <Task>. The "<Constraints>" described in area (4) of Figure 8 allows the notification device 100 (generation unit) to cause the large-scale language model to generate push notifications to be output to managers.

[0118] For example, the constraints described in the area of ​​Figure 8(4-1) define the following profile of a health management expert in the large-scale language model. (1) Provide expert advice from the perspective of a health management specialist. (2) Briefly describe the state of the organization from the perspective of health management. (3) Propose specific actions that managers should take. (4) Provide support to encourage behavioral change among managers.

[0119] Furthermore, the constraints described in area (4-2) of Figure 8 allow the notification device 100 (generation unit) to generate information in a large-scale language model that encourages behavioral change in a manner appropriate to the target manager.

[0120] Specifically, the notification device 100 (generation unit) can generate conversational text that prompts a response from the manager (user) based on the attribute information of the target manager (user) stored in the user information DB 121, according to the assumed personality of the manager. For example, if the notification device 100 (generation unit) assumes that the manager's personality is "conservative" based on the manager's attribute information, it can generate conversational text that takes into consideration not to cause distrust from the manager.

[0121] Furthermore, the constraints described in area (4-3) of Figure 8 allow the notification device 100 (generation unit) to generate output information corresponding to the classification using a large-scale language model. For example, in the case of "Classification: 5", the notification device 100 (generation unit) extracts the "status and challenges of health management initiatives" included in the manager's response and generates escalation information.

[0122] Now, let's return to Figure 7 and continue the explanation. The notification device 100 (output unit) outputs a push notification (Figure 7 (1-2)) containing the generated conversational text such as "an explanation of the organization's status and suggestions for actions to encourage behavioral change in manager 1b" to manager 1b at the recommended notification time (Figure 7 (2)).

[0123] Next, the notification device 100 (classification unit) performs classification on the manager 1b's response to the push notification (Figure 7 (3-1)). For example, based on the content of the response from manager 1b, such as "I'm having trouble with XX (Figure 7 (3-2))", the notification device 100 (classification unit) classifies it as containing keywords related to the challenges in manager 1b's (user) efforts toward health management (classification "5"). Then, based on classification "5", the notification device 100 (generation unit) extracts the challenges from manager 1b's response and causes the large-scale language model 12 to generate a response to manager 1b corresponding to the challenges (Figure 7 (3-3)) and escalation information to administrator 2b (Figure 7 (3-4)) as output information.

[0124] Specifically, the notification device 100 (generation unit) generates a response to manager 1b's statement, "I'm having trouble with 'XX'," which includes conversational phrases such as, "I see you're having trouble with XX. I'd like to offer some suggestions for actions to resolve the issue. Could you please tell me more about your problem?" (Figure 7 (3-5)).

[0125] Furthermore, the notification device 100 (generation unit) generates conversational text such as, "The manager is having trouble with XX, so it may not be possible to proceed with the next action. Please check the situation and follow up as necessary" as escalation information (output information) for the service administrator 2b (Figure 7 (3-6)).

[0126] As described above, the notification device 100 enables manager 2b to understand the status and challenges of health management initiatives and to follow up with manager 1b through appropriate escalation according to the manager 1b's situation (Figure 8 (4)). Furthermore, if manager 1b's response includes an opinion, the notification device 100 can improve the accuracy of subsequent push notifications by training a large-scale language model using that opinion.

[0127] (Third example) Next, the processing flow of "An Example of an Inquiry Response Use Case" will be explained using Figure 9. Figure 9 is a diagram illustrating a third example according to the embodiment. The example shown in Figure 9 is a use case that generates a push notification prompting an end user to make an inquiry in order to address the challenge of generating inquiries from end users who use the target service. In this third example, the user is end user 1c who uses the target service.

[0128] In the third example, we will explain how to generate a response to an inquiry made by an end user in response to a push notification sent to check whether an inquiry has been made. However, this can also be applied to inquiries made when no push notification has been sent.

[0129] Traditionally, when end users have questions or problems with the services they use, they contact a designated contact point using methods such as telephone or email. However, if the end user's literacy level is low or if they have difficulty accurately articulating their questions or problems, they may give up on making an inquiry altogether.

[0130] Therefore, in the third example, the notification device 100 is provided with the usage status of the target service and conversation history information as prior knowledge, and causes a large-scale language model, which is configured to act as a responder responsible for responding to inquiries, to generate a push notification containing a conversational message prompting the user to make the inquiry. Next, the notification device 100 proactively outputs the generated push notification to the end user 1c. Then, the notification device 100 generates a response to the inquiry made based on the push notification and outputs it to the end user 1c.

[0131] From here, we will explain the sequence of operations of the notification device 100 in the third example. The notification device 100 (generation unit) in the third example inputs a prompt (P3 in Figure 9) into the large-scale language model 13 that describes the acquisition of prior knowledge, the setting of the role of the large-scale language model 13, and the content of the push notification to be generated, and generates a push notification (Figure 9 (1-1)).

[0132] Specifically, the notification device 100 (generation unit) inputs a prompt, which includes a command to generate a conversational message prompting the end user 1c to make an inquiry, to a large-scale language model 13 provided with prior knowledge of conversation history information between the end user 1c and the service provider, as well as service usage information. The notification device then generates the conversational message prompting the end user 1c to make an inquiry as a push notification.

[0133] Here, an example of a prompt (P3 in Figure 9) relating to the third example will be explained using Figure 10. Figure 10 is a diagram illustrating an example of a prompt used in the third example according to the embodiment. The prompt P3 shown in Figure 10 includes "<definition of role> (Figure 10 (1))" and "description for causing the large-scale language model to execute <task> according to <constraints> (Figure 10 (2) to (4))" which are set for the large-scale language model.

[0134] In the “<Role Definition>” shown in Figure 10 (1), for example, a command is described that sets the role of the large-scale language model to “a support person who responds to inquiries from end users (users).”

[0135] The “<Processing Instruction>” shown in (2) of Figure 10 contains instructions to execute the “<Task>” described in the prompt according to the “<Constraints>”.

[0136] The "<Task>" shown in (3) of Figure 10 contains instructions such as "a command to generate a push notification that prompts the end user to inquire about any points they are unsure of," "a command to generate predetermined output information (response, escalation, etc.) in response to the answer received from the end user," and "a command to generate proactive answers to anticipated inquiries."

[0137] The "<Constraints>" shown in (4) of Figure 10 describes the following conditions that the large-scale language model must strictly adhere to when executing the <Task>. The "<Constraints>" described in area (4) of Figure 10 enable the notification device 100 (generation unit) to cause the large-scale language model to generate push notifications to be output to the end user.

[0138] For example, the constraints described in the area of ​​Figure 10(4-1) define the profile of the person handling inquiries in the large-scale language model, such as the following: (1) Carefully listen to the end user's questions and concerns. (2) Respond using polite yet friendly language. (3) When an end user's problem or question stems from the service in question, we will investigate the matter further and take appropriate action. (4) Messages such as push notifications to end users are not verbose and are concise.

[0139] Furthermore, the constraints described in area (4-2) of Figure 10 allow the notification device 100 (generation unit) to generate a push notification containing a conversational message prompting the end user to make an inquiry using a large-scale language model.

[0140] Specifically, the notification device 100 (generation unit) can generate conversational text that encourages the end user to make an inquiry, based on the assumed personality of the end user (user) as determined from the attribute information of the target end user (user) stored in the user information DB 121. For example, if the notification device 100 (generation unit) assumes the end user's personality to be "conservative" based on the attribute information of the end user, it can generate conversational text that takes into consideration not to incur the distrust of management.

[0141] Furthermore, the constraints described in area (4-3) of Figure 10 allow the notification device 100 (generation unit) to generate output information corresponding to the classification using a large-scale language model. For example, in the case of "Classification: 3" or "Classification: 4", the notification device 100 (generation unit) extracts "requests" and "opinions" included in the response from the end user and generates escalation information. In the case of "Classification: 6", the notification device 100 (generation unit) extracts "inquiry content" included in the response from the end user and generates escalation information.

[0142] Furthermore, in the third example, the constraints described in the region shown in (4-4) of Figure 10 allow the notification device 100 (generation unit) to generate a "predictive response to the expected inquiry content" in the large-scale language model.

[0143] For example, the notification device 100 (generation unit) can have a large-scale language model generate responses such as "Are you experiencing any problems regarding 'Content A'?" for inquiries that have a high number of occurrences based on past inquiries ("Content A"), and "An outage is scheduled for XX / XX" for inquiries that are based on scheduled outages or maintenance dates.

[0144] Now, let's return to Figure 9 and continue the explanation. The notification device 100 (output unit) outputs a push notification (Figure 9 (1-2)) containing the generated "conversational text prompting the user to make an inquiry" and "proactive notifications for frequently asked questions" to the end user 1c at the recommended notification time for push notifications (Figure 9 (2)).

[0145] Next, the notification device 100 (classification unit) performs classification on the end user 1c's response to the push notification (Figure 9 (3-1)). For example, based on the content of the response from end user 1c, such as "I'm having trouble with '○○' (Figure 9 (3-2))", the notification device 100 (classification unit) classifies it as containing keywords related to the inquiry from end user 1c (user) (classification "6"). Then, based on classification "6", the notification device 100 (generation unit) extracts the issue from the end user 1c's response and causes the large-scale language model 13 to generate output information: a response to the end user 1c corresponding to the issue (Figure 9 (3-3)) and escalation information to administrator 2c (Figure 9 (3-4)).

[0146] Specifically, the notification device 100 (generation unit) generates a response to the end user 1c's statement, "I'm having trouble with XX," which includes conversational phrases such as, "I see you're having trouble with XX. Could you tell me more about your problem?" (Figure 9 (3-5)).

[0147] Furthermore, the notification device 100 (generation unit) generates conversational text such as "An end user is having trouble with XX. Please check the situation and follow up as necessary" as escalation information (output information) for the service administrator 2c (Figure 9 (3-6)).

[0148] As described above, the notification device 100 enables administrator 2c to follow up on inquiries from end-user 1c by appropriately escalating according to the end-user 1c's situation (Figure 10 (4)). Furthermore, if the response from end-user 1c includes an opinion, the notification device 100 can improve the accuracy of subsequent push notifications by training a large-scale language model using that opinion.

[0149] (Procedure for processing by notification device 100) Next, the procedure for processing implemented by the notification device 100 according to this embodiment will be explained using Figure 11. Figure 11 is a flowchart showing the processing performed by the notification device 100 according to this embodiment.

[0150] The notification device 100 waits to process until the pre-calculated recommended time for push notification arrives (No in S101). Then, when the recommended time for push notification arrives, the notification device 100 proceeds with processing (Yes in S101).

[0151] The generation unit 132 inputs a prompt to the large-scale language model (S102). Here, the prompt input by the generation unit 132 may be, for example, one of the prompts shown as an example in the first to third examples.

[0152] The generation unit 132 generates a push notification tailored to the user based on a large-scale language model (S103). Next, the output unit 135 outputs the generated push notification to the user (S104).

[0153] If the user responds to the push notification (Yes in S105), the classification unit 134 classifies the user's response (S106). The notification device 100 then performs processing according to the classification result (S107).

[0154] For example, if escalation is necessary based on the classification result (Yes in S108), the notification device 100 escalates the matter to the administrator (S109). Here, "escalating the matter" means, for example, that the output unit 135 outputs the escalation information generated by the generation unit 132 to the administrator.

[0155] Next, the notification device 100 outputs a response to the user (S110). Then, the notification device 100 terminates the process.

[0156] The phrase "outputting a response to the user" as described above means, for example, that the output unit 135 outputs to the user information related to the response to the user generated by the generation unit 132.

[0157] If escalation is not required (No. in S108), the notification device 100 skips steps S108 and S109. Also, if there is no response from the user to the push notification (No. in S105), the notification device 100 skips steps S106 through S110. Then, the notification device 100 terminates the process.

[0158] (effect) Next, we will explain the effects of the notification device 100 according to this embodiment. Conventionally, since the generation of conversation content based on a large-scale language model is triggered by user input, there has been a challenge in proactively outputting messages to the user, including hearings for the user and service proposals, without waiting for user input.

[0159] Therefore, the generation unit 132 of the notification device 100 according to this embodiment generates a push notification based on the user information of a user using the target service and a predetermined model configured to generate a push notification, which is a message output from the target service to the user to encourage a predetermined action by the user using the target service. The output unit 135 of the notification device 100 outputs the generated push notification to the user.

[0160] Therefore, the notification device 100 according to this embodiment has the effect of enabling push notifications tailored to the user.

[0161] For example, the notification device 100 can detect signs of user behavior based on data obtained by conducting interviews and listening to user concerns via push notifications generated based on a large-scale language model. As a result, the notification device 100 can prevent cancellation by proactively following up, even if, for example, a user is considering canceling the service.

[0162] Furthermore, the notification device 100 proactively outputs push notifications to the user, thereby prompting a response from the user. As a result, the notification device 100 has the effect of enabling the user to easily ask questions or seek advice.

[0163] Furthermore, the notification device 100 can proactively and regularly deliver information and content that users want to know by proactively outputting push notifications to the user. As a result, the notification device 100 can improve the frequency of interaction with the user, thereby enabling the realization of a user-friendly service.

[0164] Furthermore, the notification device 100 according to this embodiment achieves predetermined effects by executing the processes described below.

[0165] The generation unit 132 inputs a prompt, expressed in natural language, into a predetermined large-scale language model, which is a large-scale language model, to generate a push notification. This prompt includes a command to use the user's user information, which includes at least one of the following, as prior knowledge: user attribute information, conversation history information, and service usage information; and a command to generate a push notification tailored to the user based on the prior knowledge.

[0166] Through the process described above, the notification device 100 can generate push notifications using appropriate conversational text that takes into account the user's service contract status, usage status, and the user's gender, age, personality, values, and current situation. Therefore, the notification device 100 enables the flexible generation of conversational text tailored to the user, thereby achieving the output of appropriate push notifications tailored to the user.

[0167] The generation unit 132 generates a push notification that includes at least one of the following: a conversational text for hearing the user's requests, and a conversational text for making a predetermined proposal to the user regarding a service tailored to the user.

[0168] Through the process described above, the notification device 100 can proactively output push notifications that include content for listening to the user's problems, questions, and inquiries, as well as content for proactively making new suggestions to the user from the service side. Therefore, by taking proactive actions towards the user, the notification device 100 can improve service quality and reduce the churn rate.

[0169] The generation unit 132 inputs a prompt to a large-scale language model, which has been configured to set the words to be used in push notifications sent from the target service and a predetermined conversation development method, based on the user's attribute information and conversation history information. This prompt includes a command to generate a conversational text for hearing the user's requests about the service or for carrying out a predetermined proposal regarding the service. The generation unit 132 then generates the conversational text for hearing the user's requests about the service or for carrying out a predetermined proposal regarding the service as a push notification.

[0170] Through the process described above, the notification device 100 proactively outputs push notifications tailored to each customer, thereby enabling support for hearing about potential issues the customer may have and making proposals before the customer even makes an inquiry.

[0171] The generation unit 132 inputs a prompt, which includes a command to generate information that encourages user behavior change in accordance with employee vital data, which is service usage information, to a large-scale language model that is provided with prior knowledge of conversation history information of experts related to health management, which is a management method that emphasizes employee health management. The generation unit 132 generates information that encourages user behavior change in accordance with employee vital data as a push notification.

[0172] Through the process described above, the notification device 100 proactively outputs push notifications to managers, thereby encouraging behavioral changes by managers based on employee vital data. Therefore, the notification device 100 has the effect of solving the challenges of health management and realizing health management.

[0173] The generation unit 132 inputs a prompt, which includes a command to generate a conversational message prompting the end user to make an inquiry, to the large-scale language model 13, which is provided with prior knowledge of conversational history information between the end user and the service provider and service usage information related to the service. The generation unit 132 then generates the conversational message prompting the end user to make an inquiry as a push notification.

[0174] Through the process described above, the notification device 100 proactively outputs push notifications to the end user, thereby promoting inquiries from that end user.

[0175] For example, the above-mentioned facilitation of inquiries from end users can be applied to network-related services. Specifically, the notification device 100 can replace the role of a network service desk, such as answering questions from end users, notifying of faults, and notifying of maintenance dates.

[0176] Furthermore, the notification device 100 can also take over the role of answering questions about contract status and other matters from customers who do not have detailed knowledge of networks. Specifically, the notification device 100 conducts network vulnerability assessments on behalf of end users who do not have detailed knowledge of networks and pushes information about detected vulnerabilities to the end users. In addition, the notification device 100 can take action such as responding to inquiries from end users regarding countermeasures or proposing new products, triggered by the push notifications.

[0177] Furthermore, the notification device 100 enables appropriate support to end users who are unable to investigate the terminal equipment causing the network failure themselves. For example, based on uploaded photographs of terminal equipment, etc., that are considered to be potential causes of the network failure, the notification device 100 can seamlessly determine the status and whether or not the terminal equipment is faulty, suggest how to deal with the failure, and provide information to human support.

[0178] The calculation unit 133 uses attribute information for each user and statistical information on response times for each user attribute, calculated using attribute information for each user and past response times for each user, to input statistical information on the response times of the target individual user into a machine learning model that has been trained to output a recommended notification time for push notifications to individual users in response to the input of statistical information on the response times of individual users, and calculates the recommended notification time for push notifications to the target individual user. The output unit 135 outputs a push notification to the target individual user based on the calculated recommended notification time for push notifications to the target individual user.

[0179] Through the process described above, the notification device 100 can set a delivery schedule based on the time when each user is most likely to respond and output push notifications. Therefore, the notification device 100 can increase the user response rate to push notifications.

[0180] The classification unit 134 classifies responses from users of the service based on predetermined keywords included in the user's response to the push notification or the results of sentiment analysis.

[0181] Through the processing described above, the notification device 100 is able to take appropriate action towards a user based on keywords included in the user's response and the user's emotions. For example, the notification device 100 can detect the timing of additional business opportunities or signs of cancellation when specific keywords are included or when the user has specific emotions. As a result, the notification device 100 can take actions such as responding to the user or escalating to the administrator in real time.

[0182] If the generation unit 132 classifies the user's response as containing keywords related to service cancellation or negative emotions, it generates escalation information for the service administrator.

[0183] Through the process described above, the notification device 100 escalates the matter to the service administrator if the user's response includes cancellation or negative emotions. As a result, the notification device 100 has the effect of appropriately responding to potential signs of cancellation by the user and preventing cancellations from occurring.

[0184] When the learning unit 136 classifies a user's response as containing an opinion about the service, it learns a predetermined model using the user's opinion extracted from the conversation history of users who use the service.

[0185] The above-described process enables the notification device 100 to improve the accuracy of push notification generation based on user feedback. For example, based on user feedback and the user's conversation history, the notification device 100 can generate conversational text that may be of interest to the user in the future.

[0186] Furthermore, the notification device 100 enables interactive responses even if user inquiries contain technical terms, thanks to its individual learning function based on Retrieval-Augmented Generation (RAG) and the like.

[0187] <Variation> The following describes modifications implemented by the notification device 100 according to this embodiment.

[0188] (Data, etc.) The names of the functional parts of the push notification, prompt, notification device 100, steps, processes, and names of steps or processes used in the description of the above embodiment are merely examples and can be changed at will.

[0189] For example, while it was explained that User Information DB121 stores information related to each item, such as "No," which identifies individual data included in user information, "User Identification Information," "Attribute Information," "Conversation History," "Service Usage Status," and "Recommended Notification Time," in a table format such as that shown in Figure 3, the stored items and the information within each item are not limited to those shown in Figure 3. Similarly, while it was explained that Classification Condition DB123 stores information related to each item, such as "Classification," which identifies individual classification conditions, and "Conditions" and "Processing Contents" corresponding to each classification, in a table format such as that shown in Figure 4, the stored items and the information within each item are not limited to those shown in Figure 4.

[0190] (Push notification for first-time users) Although the notification device 100 of this embodiment is described as making push notifications based on the user information of users using the target service, if the user has not yet used the target service (i.e., is a "first-time user"), a push notification can be output to the first-time user based on the following processing.

[0191] The notification device 100 (generation unit) receives a prompt containing a command to generate a push notification tailored to a first-time user from a large-scale language model that is provided with attribute information and conversation content similar to that of a first-time user as prior knowledge, and generates a push notification. The notification device 100 (output unit) then actively outputs the generated push notification to the first-time user.

[0192] Furthermore, the notification device 100 (calculation unit) can calculate the recommended notification time for push notifications related to attribute groups similar to first-time users based on the conversation history accumulated in the past. Then, the notification device 100 (output unit) can output a push notification at the time when first-time users are most likely to reply, according to the calculated recommended notification time for push notifications related to attribute groups similar to first-time users.

[0193] (Regarding the calculation of the recommended notification time for push notifications) Although the notification device 100 of this embodiment is described as calculating the recommended notification time for push notifications based on user attributes and statistical information on past reply times, it is not limited to this. For example, the notification device 100 can calculate the recommended notification time for push notifications using past read rates, push notification times, and subsequent user behavior.

[0194] (Regarding notifications based on user priority) The notification device 100 of this embodiment can output push notifications to target users with priority based on the "user response priority" stored in the user information DB 121. For example, for users whose response priority is set to "high" because they have previously had complaints, the notification device 100 can output push notifications that take priority over other users by changing the notification time, frequency, etc.

[0195] (Regarding adjusting prompts) The prompts used in this embodiment may be adjusted by the administrator. For example, the prompts may be updated based on the actual interaction between the notification device 100 and the user, and based on a persona that has been adapted to behave as desired by the administrator.

[0196] (Regarding the use of the specified model) In this embodiment, the predetermined model used by the notification device 100 is described as being stored in the model DB 124 of the storage unit 120, but this is not limited to this. For example, the notification device 100 can access an external information processing device (such as a server) to use the predetermined model.

[0197] (Flowcharts, etc.) In flowcharts, each step may be rearranged as long as it does not create inconsistencies, and some steps may be omitted. Furthermore, conjunctions such as "next," "continue," "in addition," "at this time," and "on this occasion" in flowchart descriptions do not limit the order or timing of the processes in the flowchart.

[0198] <Hardware Configuration> Each component of the illustrated device is a functional concept and does not necessarily have to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those shown, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads and usage conditions. Furthermore, each processing function performed by each device can be implemented, all or any part of it, by a CPU and the program that is analyzed and executed by that CPU, or by hardware using wired logic.

[0199] Furthermore, among the processes described in this embodiment, all or part of those described as being performed automatically can be performed manually using known methods. In addition, the processing procedures, control procedures, specific names, and information including various data and parameters shown in the drawings can be arbitrarily changed unless otherwise specified.

[0200] <Program> In one embodiment, the various devices constituting the notification device 100 can be implemented by installing a notification program as packaged software or online software on a desired computer. For example, by having the above notification program run on an information processing device, the various devices constituting the notification device 100 can be made to function. The information processing device referred to here includes desktop or notebook personal computers. In addition, the information processing device also includes mobile communication terminals such as smartphones and mobile phones, and slate terminals such as PDAs (Personal Digital Assistants).

[0201] Figure 12 shows an example of a computer that implements the notification device 100 according to the embodiment. The computer 1000 has, for example, memory 1010 and CPU 1020. The computer 1000 also has a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0202] Memory 1010 includes ROM (Read Only Memory) 1011 and RAM 1012. ROM 1011 stores, for example, a boot program such as BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to the hard disk drive 1090. The disk drive interface 1040 is connected to the disk drive 1100. For example, a removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to, for example, a mouse 1110 and a keyboard 1120. The video adapter 1060 is connected to, for example, a display 1130.

[0203] The hard disk drive 1090 stores, for example, an OS (Operating System) 1091, an application program 1092, a program module 1093, and program data 1094. That is, the programs that define the various processes of the various devices constituting the notification device 100 are implemented as program modules 1093 in which executable code for a computer is written. The program modules 1093 are stored, for example, in the hard disk drive 1090. For example, a program module 1093 for performing processes similar to the functional configurations of the various devices constituting the notification device 100 is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD (Solid State Drive).

[0204] Furthermore, the configuration data used in the processing of the embodiment described above is stored as program data 1094 in, for example, memory 1010 or hard disk drive 1090. The CPU 1020 then reads the program module 1093 and program data 1094 stored in memory 1010 or hard disk drive 1090 into RAM 1012 as needed and executes the processing of the embodiment described above.

[0205] Furthermore, the program module 1093 and program data 1094 are not limited to being stored in the hard disk drive 1090; for example, they may be stored in a removable storage medium and read by the CPU 1020 via a disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (LAN, WAN (Wide Area Network), etc.). The program module 1093 and program data 1094 may then be read from the other computer by the CPU 1020 via a network interface 1070.

[0206] <Other> Although this embodiment has been described above, this embodiment is not limited by the description and drawings that constitute part of the disclosure. That is, all other embodiments, examples, and operational techniques made by those skilled in the art based on this embodiment are included in the scope of this embodiment. [Explanation of Symbols]

[0207] 100 Notification device 110 Communications Department 120 Storage section 121 User Information Database 122 Statistics information DB 123 Classification condition DB 124 Model DB 130 Control Unit 131 Acquisition Department 132 Generation part 133 Calculation Section 134 Classification Department 135 Output section 136 Learning Department

Claims

1. A generation unit that generates push notifications based on information about users of the target service and a predetermined model configured to generate push notifications, which are messages output from the target service to the user to encourage a predetermined action by the user of the target service, An output unit that outputs the generated push notification to the user, A notification device characterized by having the following features.

2. The generating unit is A command that uses as prior knowledge information about a user of the target service, including at least one of the user's attribute information, conversation history information, and usage status information of the target service, A command to generate the push notification corresponding to the user based on the prior knowledge, The system inputs a prompt expressed in natural language into a predetermined large-scale language model to generate the push notification. The notification device according to feature 1.

3. The generating unit is A conversational transcript for gathering the user's requests. And, A conversational text for making a predetermined proposal to the user regarding the service tailored to the user, A push notification is generated that includes at least one of the following: The notification device according to feature 2.

4. The generating unit is The large-scale language model, which has been configured to set the language to be used for the push notifications sent from the target service and to set a predetermined conversation development method, is given a prompt that includes a command to generate conversational text for hearing the user's requests regarding the target service or for making a predetermined proposal regarding the service, based on the user's attribute information and information about the conversation history. The system generates a conversational text as a push notification for hearing the user's requests regarding the service or for carrying out a predetermined proposal regarding the service. The notification device according to feature 2.

5. The generating unit is The large-scale language model, which is provided with prior knowledge of the conversation history of experts related to health management, a management method that emphasizes employee health management, is given a prompt that includes a command to generate information that encourages behavioral change of the user in accordance with the vital data of the employee, which is information regarding the usage status of the service. The system generates information that encourages behavioral changes in the user based on the vital data of the employee, as a push notification. The notification device according to feature 2.

6. The generating unit is The large-scale language model, which is provided with prior knowledge of the conversation history between the end user and the service provider and the usage status of the service, is given a prompt that includes a command to generate a conversational sentence prompting the end user to make an inquiry. A conversational message prompting the end user to make an inquiry is generated as the push notification. The notification device according to feature 2.

7. A machine learning model trained to output a recommended notification time for push notifications to individual users, based on input of statistical information on the response times of individual users, using statistical information on the response times of individual users, calculated using attribute information for each user and the past response times of each user, The system further includes a calculation unit that inputs statistical information on the response times of individual target users and calculates the recommended notification time for the push notification to the said individual target user. The output unit is, Based on the calculated recommended notification time for the push notification to the individual target user, the push notification is output to the individual target user. A notification device according to any one of claims 1 to 6.

8. The system further includes a classification unit that classifies responses from users of the service based on predetermined keywords included in the user's response to the push notification or the results of sentiment analysis. A notification device according to any one of claims 1 to 6.

9. The generating unit is If the user's response is classified as containing keywords or negative sentiments related to the cancellation of the service, information regarding escalation to the service administrator will be generated. The notification device according to feature 8.

10. If the user's response is classified as including an opinion about the service, the system further includes a learning unit that learns the predetermined model using the user's opinion extracted from the conversation history of the user using the service. The notification device according to feature 8.

11. A notification method to be executed by a notification device, A generation step for generating a push notification based on information about a user using the target service and a predetermined model configured to generate a push notification, which is a message output from the target service to the user to encourage a predetermined action by the user using the target service; Output step of outputting the generated push notification to the user, A notification method characterized by including the following.

12. A generation step for generating a push notification based on information about a user using the target service and a predetermined model configured to generate a push notification, which is a message output from the target service to the user to encourage a predetermined action by the user using the target service; Output step of outputting the generated push notification to the user, A notification program that instructs a computer to execute a command.

Images