Automatic response control system, method, and program

Abstract

To provide an automatic response control system for providing an inquiry window by a chat bot in various Web pages.SOLUTION: The automatic answering control server includes avatar image display control means for displaying an avatar image on the website, chatbot control means for causing a chatbot to function on the website, response sentence generation means for acquiring language information of the website as external information, generating a text by inputting a question sentence combined with the external information to a large-scale language model, and generating a response sentence, and synchronization control means for synchronizing the chatbot with display of the avatar image, the web server includes, as a part of the website construction means, activation control means for receiving an activation input from the user, starting communication with the automatic answering control server, and activating and operating the avatar image display control means, the answer sentence generation means, the chatbot control means, and the synchronization control means.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an automatic response control system, method, and program. [Background technology]

[0002] Conventionally, automated response systems using chatbot technology have been known, and these automated response systems are used to respond to inquiries from service users. In addition, a method is known in which such automated response systems output an image of a character called an avatar along with text and voice, making it appear as if the avatar is speaking.

[0003] The applicant is aware of the following Patent Document 1 as a prior art for such an automatic response system. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7367870

[0005] The above-mentioned Patent Document 1 relates to a "conversation control program" and includes the following description:

[0006]

[0011] (Embodiment 1) First, a system configuration example of a conversation control system 100 including an information processing device 101 according to the first embodiment will be described. The conversation control system 100 is applied to, for example, a computer system that provides a service using two-way communication in cooperation with a chatbot.

[0012] Fig. 1 is an explanatory diagram showing an example of a system configuration of a conversation control system 100 according to a first embodiment. In Fig. 1, the conversation control system 100 includes an information processing device 101, a plurality of information processing terminals 102 (three in the example of Fig. 1), and an administrator terminal 103. In the conversation control system 100, the information processing device 101, the information processing terminals 102, and the administrator terminal 103 are connected via a wired or wireless network 110. The network 110 is, for example, a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, or the like.

[0013] Here, the information processing device 101 is a computer that controls conversations of a chatbot. The information processing device 101 is, for example, a server. The information processing device 101 has, for example, an FAQ master 120, a script table 130, and a chat log DB (Database) 140. The FAQ master 120 stores FAQs (Frequently Asked Questions).

[0017] The information processing terminal 102 is a computer that a user uses when having a conversation with a chatbot. The information processing terminal 102 is, for example, a smartphone, a tablet PC (Personal Computer), or a PC. The information processing terminal 102 may also be a digital board installed in stations, airports, stores, commercial facilities, etc.

[0019] The administrator terminal 103 is a computer used by an administrator of the conversation control system 100. The administrator of the conversation control system 100 is, for example, a designer of a script that defines the conversation flow of a chatbot. The administrator terminal 103 is, for example, a PC. On the administrator terminal 103, the administrator can, for example, design and change scripts and set various thresholds.

[0057] Fig. 7 is an explanatory diagram (part 1) showing an example of a chat screen. In Fig. 7, a chat screen 700 is an example of an operation screen that displays a conversation with a chatbot on the information processing terminal 102. In Fig. 7, an avatar av is a character that represents the chatbot.

[0180] The determination unit 1901 may also use a machine learning model M to determine whether or not a specific word is included in input information received in a conversation with a chatbot in the information processing terminal 102. Here, the specific word is, for example, a password, a credit card number, etc. The machine learning model M is a machine learning model for classifying whether or not an input character string is a specific word.

[0007] The above-mentioned Patent Document 1 discloses that an information processing device 101 controls a conversation with a chatbot, a user uses an information processing terminal 102 to converse with the chatbot, and an avatar av is displayed on an operation screen that displays the conversation with the chatbot. It also discloses that a machine learning model M is used to determine whether or not a specific word is included in the conversation with the chatbot. DISCLOSURE OF THE INVENTION [Problem to be solved by the invention]

[0008] In the technology of Patent Document 1, businesses wishing to set up a chatbot must prepare an information processing device 101, which is a computer that controls the chatbot's conversations. This makes setting up a chatbot extremely expensive, making it difficult to set up a new one. Meanwhile, with the labor shortage caused by the declining birthrate and aging population, the need for automated inquiry desks is increasing, and there is a growing demand for chatbot systems that can be set up cheaply and easily.

[0009] The present invention has been made to solve the above problems and has the following object. An automatic response control system, method, and program are provided that can easily set up a chatbot inquiry desk on various web pages. [Means for solving the problem]

[0010] In order to achieve the above object, the automatic response control system according to claim 1 employs the following configuration. The system is configured to include a web server on which a website is stored and an automatic response control server, The automatic response control server is avatar image display control means for reading the avatar image data from the avatar image data storage unit and displaying the avatar image on the website; a chatbot control means for controlling the reception of input questions and the output of responses on the website, receiving the questions input by users, and outputting responses to the questions, thereby causing the chatbot to function; a response generation means for acquiring language information recorded on the website as external information, inputting the question in combination with the external information into a large-scale language model trained by machine learning to generate text, and generating the response to be output by the chatbot control means; a synchronization control means for synchronizing reception of an input of a user's question and output of a response by the chatbot control means with display of an avatar image by the avatar image display control means, The web server is As part of the website construction means for constructing the website, there is a startup control means that starts communication with the automatic response control server by accepting startup input from a user, and starts and operates at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means.

[0011] The automatic response control system according to claim 2 employs the following configuration in addition to the configuration according to claim 1. When the activation of the activation control means is the first time, The response sentence generating means acquires language information recorded on the website as the external information to be combined in text generation using a large-scale language model.

[0012] The automatic response control system according to claim 3 employs the following configuration in addition to the configuration according to claim 2. When the activation of the activation control means is the second or subsequent time, The response sentence generating means again acquires language information recorded on the website as the external information to be combined in text generation using a large-scale language model.

[0013] The automatic response control system described in claim 4 employs the following configuration in addition to the configuration described in claim 1. When the start-up control means is started, The avatar image display control means is operated to display an avatar image on the website, the chatbot control means is operated to make the chatbot function, and the synchronization control means is operated to synchronize the operation of the chatbot with the display of the avatar image.

[0014] The automatic response control system according to claim 5 employs the following configuration in addition to the configuration according to any one of claims 1 to 4. The response sentence generation means acquires language information recorded on the website as well as language information from websites other than the website, as the external information to be combined in text generation using a large-scale language model.

[0015] The automatic response control system according to claim 6 employs the following configuration in addition to the configuration according to any one of claims 1 to 3. The chatbot control means includes a voice control unit that accepts the input of the question sentence and outputs the response sentence by voice.

[0016] In order to achieve the above object, the automatic response control method according to claim 7 employs the following configuration. Using a web server on which a website is stored and an automatic response control server, The automatic response control server is an avatar image display control step by avatar image display control means for reading avatar image data from an avatar image data storage unit and displaying the avatar image on the website; a chatbot control step by a chatbot control means for controlling the reception of input of a question and the output of a response on the website, receiving the question input by a user, and outputting a response to the question, thereby causing the chatbot to function; a response sentence generation step by a response sentence generation means that acquires language information recorded on the website as external information, inputs the question sentence into a large-scale language model that has been machine-learned in combination with the external information to generate text, and generates the response sentence to be output in the chatbot control step; a synchronization control step by a synchronization control means for synchronizing the reception of an input of a question sentence from a user and the output of a response sentence by the chatbot control means with the display of an avatar image by the avatar image display control means; The web server is In the website construction step of constructing the website, a startup control step is executed in which communication with the automatic response control server is initiated by accepting startup input from the user, and at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means is started and operated.

[0017] In order to achieve the above object, the automatic response control program according to claim 8 employs the following configuration. A program to be executed by a web server storing a website and an automatic response control server, The automatic response control server is an avatar image display control step by avatar image display control means for reading avatar image data from an avatar image data storage unit and displaying the avatar image on the website; a chatbot control step by a chatbot control means for controlling the reception of input of a question and the output of a response on the website, receiving the question input by a user, and outputting a response to the question, thereby causing the chatbot to function; a response generation step by a response generation means for acquiring language information recorded on the website as external information, inputting the question into a large-scale language model trained by machine learning in combination with the external information to generate text, and generating the response to be output by the chatbot control means; a synchronization control step by a synchronization control means for synchronizing the reception of an input of a question sentence from a user and the output of a response sentence by the chatbot control means with the display of an avatar image by the avatar image display control means; The web server is In a part of the website construction step of constructing the website, a start-up control step is executed in which communication with the automatic response control server is initiated by receiving a start-up input from a user, and at least one of the avatar image display control means, the response sentence generation means, the chatbot means, and the synchronization control means is started and operated. The above-mentioned web server and automatic response control server are made to execute the above. [Effects of the Invention]

[0018] The automatic response control system of claim 1 includes a web server storing a website and an automatic response control server. The automatic response control server includes avatar image display control means, chatbot control means, response sentence generation means, and synchronization control means. The avatar image display control means reads avatar image data from an avatar image data storage unit and displays an avatar image on the website. The chatbot control means controls the reception of questions and the output of responses on the website, receiving the questions entered by users and outputting responses to the questions, thereby causing the chatbot to function. The response sentence generation means acquires linguistic information recorded on the website as external information, combines the questions with the external information, and inputs the combined information into a large-scale language model trained by machine learning to generate text, thereby generating the responses to be output by the chatbot control means. The synchronization control means synchronizes the reception of questions from users and the output of responses by the chatbot control means with the display of avatar images by the avatar image display control means. The web server includes a launch control means as part of a website construction means for constructing the website. The launch control means starts communication with the automatic response control server upon receiving a launch input from a user, and launches and operates at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means. Therefore, when a user connects their user terminal to the web server, displays the website on the user terminal, and performs an activation input for the activation control means on the website, communication between the web server and the automatic response control server begins. Then, at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means is activated and operates. The avatar image display control means reads avatar image data from an avatar image data storage unit and displays an avatar image on the website. The response sentence generation means acquires linguistic information recorded on the website as external information, combines the question with the external information, and inputs the combined information into a large-scale language model trained by machine learning to generate text, thereby generating the response sentence by the chatbot control means. The chatbot control means controls the input acceptance of a question sentence and the output of a response sentence on the website, receiving the question sentence entered by the user and outputting a response sentence to the question sentence, thereby functioning as a chatbot. Furthermore, the synchronization control means synchronizes the reception of the user's question input by the chatbot control means and the output of the response by the chatbot control means with the display of the avatar image by the avatar image display control means. Through the above operations, a chatbot synchronized with an avatar image functions on the website on the web server, making it possible to receive and respond to inquiries from the user, and making it easy to set up a chatbot inquiry desk on various web pages.

[0019] In the automatic response control system described in claim 2, when the startup control means is started for the first time, the response sentence generation means acquires language information recorded on the website as the external information to be combined with text generation using a large-scale language model. By the response sentence generation means acquiring the language information recorded on the website as the external information to be combined with text generation using a large-scale language model, the chatbot can output an appropriate response sentence that matches the information recorded on the website in response to a question from a user regarding the content of the website.

[0020] In the automatic response control system described in claim 3, when the activation control means is activated for the second time or later, the response sentence generation means again acquires language information recorded on the website as the external information to be combined with text generation using a large-scale language model. For example, if a certain amount of time has passed since the last response was generated, there is a possibility that the information posted on the website has been updated. By re-obtaining the updated language information, the chatbot can output an appropriate response to the user's question that matches the updated information recorded on the website.

[0021] In the automatic response control system described in claim 4, when the startup control means is started, the avatar image display control means is operated to display an avatar image on the website, the chatbot control means is operated to make the chatbot function, and the synchronization control means is operated to synchronize the operation of the chatbot with the display of the avatar image. Each time the activation control means is activated, the behavior of the chatbot and the display of the avatar image are synchronized. Also, for example, since the response sentence generation means has already obtained the linguistic information recorded on the website when the activation control means is activated for the first time, the chatbot outputs an appropriate response sentence that matches the information recorded on the website in response to a user's question when activated for the second time or later.

[0022] In the automatic response control system described in claim 5, the response sentence generation means acquires language information recorded on the website as well as language information from websites other than the website as the external information to be combined in text generation using a large-scale language model. Because the chatbot acquires not only the language information posted on the website but also the language information of other websites, it outputs an appropriate response to a user's question that takes into account information recorded on the website as well as information from other websites.

[0023] The automatic response control system according to claim 6 includes a voice unit means in the chatbot control means for receiving the input of the question sentence and outputting the response sentence by voice. In chatbots, users can input questions by voice rather than by text, and the output responses can be obtained by listening to the voice rather than reading the text, making them more convenient.

[0024] The automatic response control method of claim 7 uses a web server storing a website and an automatic response control server. The automatic response control server executes an avatar image display control step, a chatbot control step, a response sentence generation step, and a synchronization control step. In the avatar image display control step, an avatar image display control means reads avatar image data from an avatar image data storage unit and displays an avatar image on the website. In the chatbot control step, a chatbot control means controls the input acceptance of a question sentence and the output of a response sentence on the website, receives the question sentence entered by a user, and outputs a response sentence to the question, thereby causing the chatbot to function. In the response sentence generation step, a response sentence generation means acquires linguistic information recorded on the website as external information, combines the question sentence with the external information, and inputs it into a large-scale language model trained by machine learning to generate text, thereby generating the response sentence to be output in the chatbot control step. The synchronization control step is performed by the synchronization control means in such a way that the reception of a user's question input by the chatbot control means and the output of a response sentence are synchronized with the display of an avatar image by the avatar image display control means. Then, in the website construction step of constructing the website, the web server executes a start-up control step of starting communication with the automatic response control server by receiving a start-up input from the user, and starting and operating at least one of the avatar image display control means, the response sentence generation means, the chatbot means, and the synchronization control means. Therefore, when a user connects their user terminal to the web server, displays the website on the user terminal, and performs an activation input for the activation control means on the website, communication between the web server and the automatic response control server begins. Then, at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means is activated and operates. The avatar image display control means reads avatar image data from an avatar image data storage unit and displays an avatar image on the website. The response sentence generation means acquires linguistic information recorded on the website as external information, combines the question with the external information, and inputs the combined information into a large-scale language model trained by machine learning to generate text, thereby generating the response sentence by the chatbot control means. The chatbot control means controls the input acceptance of a question sentence and the output of a response sentence on the website, receiving the question sentence entered by the user and outputting a response sentence to the question sentence, thereby functioning as a chatbot. Furthermore, the synchronization control means synchronizes the reception of the user's question input by the chatbot control means and the output of the response by the chatbot control means with the display of the avatar image by the avatar image display control means. Through the above operations, a chatbot synchronized with an avatar image functions on the website on the web server, making it possible to receive and respond to inquiries from the user, and making it easy to set up a chatbot inquiry desk on various web pages.

[0025] The automatic response control program according to claim 8 is a program executed by a web server storing a website and the automatic response control server. The automatic response control server executes an avatar image display control step, a chatbot control step, a response sentence generation step, and a synchronization control step. The avatar image display control step reads avatar image data from an avatar image data storage unit and displays an avatar image on the website using an avatar image display control means. The chatbot control step controls, using a chatbot control means, the reception of a question sentence and the output of a response sentence on the website, receiving the question sentence entered by a user and outputting a response sentence to the question, thereby causing a chatbot to function. The response sentence generation step uses a response sentence generation means to acquire linguistic information recorded on the website as external information, combine the question sentence with the external information, and input the resultant text to a large-scale language model trained by machine learning, thereby generating the response sentence to be output in the chatbot control step. The synchronization control step is performed by the synchronization control means in such a way that the chatbot control means synchronizes the reception of a user's question input and the output of a response with the display of an avatar image by the avatar image display control means.The web server then executes a launch control step as part of the website construction step of constructing the website.The launch control step starts communication with the automatic response control server by receiving a launch input from the user, and launches and operates at least one of the avatar image display control means, the response generation means, the chatbot means, and the synchronization control means. Therefore, when a user connects their user terminal to the web server, displays the website on the user terminal, and performs an activation input for the activation control means on the website, communication between the web server and the automatic response control server begins. Then, at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means is activated and operates. The avatar image display control means reads avatar image data from an avatar image data storage unit and displays an avatar image on the website. The response sentence generation means acquires linguistic information recorded on the website as external information, combines the question with the external information, and inputs the combined information into a large-scale language model trained by machine learning to generate text, thereby generating the response sentence by the chatbot control means. The chatbot control means controls the input acceptance of a question sentence and the output of a response sentence on the website, receiving the question sentence entered by the user and outputting a response sentence to the question sentence, thereby functioning as a chatbot. Furthermore, the synchronization control means synchronizes the reception of the user's question input by the chatbot control means and the output of the response by the chatbot control means with the display of the avatar image by the avatar image display control means. Through the above operations, a chatbot synchronized with an avatar image functions on the website on the web server, making it possible to receive and respond to inquiries from the user, and making it easy to set up a chatbot inquiry desk on various web pages. [Brief explanation of the drawings]

[0026] [Figure 1] 1 is a system configuration diagram illustrating an embodiment of an automatic response control system of the present invention. [Figure 2] FIG. 2 is a system configuration diagram illustrating an automatic response control server used in the above embodiment. [Figure 3] FIG. 2 is a system configuration diagram illustrating a web server used in the embodiment. [Figure 4]FIG. 3 is a sequence diagram illustrating a first example of an automatic response control method according to the present embodiment. [Figure 5] FIG. 10 is a diagram showing an example of a web page and a chatbot screen. [Figure 6] FIG. 10 is a sequence diagram illustrating a second example of the automatic response control method according to the present embodiment. [Figure 7] FIG. 10 is a system configuration diagram illustrating a second example of an automatic response control server. [Figure 8] FIG. 10 is a diagram showing a second display example of a web page and a chatbot screen. [Figure 9] FIG. 10 is a system configuration diagram illustrating a third example of an automatic response control server. DETAILED DESCRIPTION OF THE INVENTION

[0027] Next, an embodiment of the present invention will be described.

[0028] 1 to 6 are diagrams illustrating one embodiment of the present invention.

[0029] [Basic configuration] FIG. 1 is a system configuration diagram illustrating an automatic response control system according to this embodiment. The automatic response control system of this embodiment includes a web server 300 storing a website, and an automatic response control server 200. The web server 300 and the automatic response control server 200 are both connected to the Internet 100 and are interconnected so as to be able to communicate with each other via the Internet 100. In this embodiment, a large-scale language model 400, a user terminal 500 operated by a user, and a web server 600 other than the web server 300 are also connected to the Internet 100 and are interconnected so as to be able to communicate with each other via the Internet 100.

[0030] In other words, the web server 300, the automatic response control server 200, the large-scale language model 400, the user terminal 500, and the other web server 600 are each systems equipped with a communication interface, a central processing unit, a large-capacity data storage device, etc., and the central processing unit reads out a program recorded on an external storage device such as a hard disk drive, executes processing in accordance with the program, and also communicates data with the other system that has accessed via the Internet 100 in accordance with a protocol such as HTTP (HyperText Transfer Protocol).

[0031] The automatic response control server 200 includes an avatar image display control means 22, a chatbot control means 24, a response sentence generation means 26, and a synchronization control means 27 (details of which will be described later).

[0032] The avatar image display control means 22 reads avatar image data from the avatar image data storage unit 23 and displays the avatar image on the website. The chatbot control means 24 controls the reception of a question input and the output of a response on the website, receiving the question input by a user and generating and outputting a response to the question, thereby causing the chatbot to function. The response generation means 26 acquires linguistic information recorded on the website as external information, combines the question with the external information, and inputs it into a machine-learned large-scale language model 400 to generate text, thereby generating the response to be output by the chatbot control means 24. The synchronization control means 27 synchronizes the reception of a question input from a user and the output of a response by the chatbot control means 24 with the display of an avatar image by the avatar image display control means 22.

[0033] The web server 300 has a startup control command 35 as part of a web program 33 that constructs the website. The startup control command 35 starts communication with the automatic response control server 200 by accepting a startup input from a user, and starts and operates at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27.

[0034] [Basic functions] In the automatic response control system of this embodiment, a user connects his / her user terminal 500 to the web server 300, displays the website on the user terminal 500, and performs a startup input on the website, which is realized by the startup control command 35, thereby starting communication between the web server 300 and the automatic response control server 200. Then, at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27 is started and operates.

[0035] As a result, the avatar image display control means 22 reads avatar image data from the avatar image data storage unit 23 and displays the avatar image on the website. The response sentence generation means 26 acquires linguistic information recorded on the website as external information, combines the question with the external information, and inputs the combined information into a machine-learned large-scale language model 400 to generate text, thereby generating the response sentence to be output by the chatbot control means 24. The chatbot control means 24 controls the acceptance of input of a question sentence and the output of a response sentence on the website, receiving the question sentence entered by a user and outputting a response sentence to the question sentence, thereby functioning as a chatbot. Furthermore, the synchronization control means 27 synchronizes the acceptance of input of a question sentence from a user and the output of a response sentence by the chatbot control means 24 with the display of an avatar image by the avatar image display control means 22.

[0036] The automatic response control system of this embodiment executes the automatic response control program of this embodiment, thereby realizing the automatic response control method of this embodiment.

[0037] [Automatic response control server 200] FIG. 2 is a system configuration diagram for explaining the automatic response control server 200. As shown in FIG. The automatic response control server 200 has a transmission / reception control unit 21, an avatar image display control means 22, an avatar image data storage unit 23, a chatbot control means 24, a voice control unit 25, a response sentence generation means 26, and a synchronization control means 27.

[0038] ▽Transmission and reception control unit 21 The transmission / reception control unit 21 performs data communication with a remote system that has accessed via the Internet 100, for example, in accordance with the Internet Protocol (IP).

[0039] ▽Avatar image display control means 22 The avatar image display control means 22 reads out avatar image data from the avatar image data storage unit 23 and displays the avatar image on the website of the web server 300. The avatar image displayed on the website may be a still image, or, when synchronized with a chatbot function as described below, may be a moving image that changes lip-synching, gestures, facial angle, facial expression, etc. in response to the input of questions and the output of responses in the chatbot.

[0040] The avatar image can be displayed over the entire web page that makes up the website, hiding the background web page, or by making the background web page slightly transparent. Alternatively, the avatar image can be displayed as a small image in the lower right corner of the screen displaying the web page.

[0041] ▽ Chatbot control method 24 The chatbot control means 24 controls the input acceptance of questions and the output of responses on the website of the web server 300, and causes the chatbot to function by receiving the questions input by the user and outputting responses to the questions.

[0042] The chatbot control means 24 displays a question field for accepting input of a question on the website of the web server 300, and receives the question input therein. The chatbot control means 24 also displays a response field for displaying a response on the website of the web server 300, and outputs and displays a response obtained by a response generation means 26 (described later) here.

[0043] When the chatbot control means 24 receives a text question, it generates an appropriate response through inference using a machine learning model obtained by the response generation means 26, which will be described later. To infer the response, an AI (artificial intelligence) technique using a large-scale language model 400 trained by machine learning is used. That is, to infer the response, the large-scale language model 400 trained by machine learning is used for the relationship between the question and the response. The response generation means 25 applies the large-scale language model 400 trained by machine learning to the text question, thereby generating a response that is text as an inference result, and the chatbot control means 24 outputs it.

[0044] ▽Audio control unit 25 The voice control unit 25 receives the question and outputs the response by voice in the chatbot control means 24. That is, when a question is input by voice, the voice control unit 25 recognizes the voice and converts it into text data. Based on this text data, the text data of the response obtained by the chatbot control is converted into voice data by voice synthesis, and the voice response is output. The voice input of the question is performed at the user terminal 500, and is received by the automatic response server 200 via the web server 300 and used as input data for the chatbot. The voice output of the response is received by the user terminal 500 via the web server 300 from the automatic response server 200 and used as output data for the chatbot. The voice recognition of the voice question can be realized by applying, for example, ASR (Automatic Speech Recognition) or the like. Furthermore, the technology for synthesizing the text data of the response and converting it into voice data can be realized, for example, by existing TTS (Text to Speech) voice synthesis.

[0045] ▽ Response sentence generation means 26 The response sentence generation means 26 acquires language information recorded on the website of the web server 300 as external information, and generates text by inputting the question sentence in combination with the external information into, for example, a large-scale machine-learned language model 400 existing on the web, and generates the response sentence to be output by the chatbot control means 24.

[0046] The large-scale language model 400 is used by the chatbot control means 24 as a model for generating a response sentence by inference based on an input question sentence. The large-scale language model 400 can use, for example, ChatGPT or other LLMs (large-scale language models) available on the web. The large-scale language model 400 can also use an LLM that is not published on the web.

[0047] When the launch control means (described later) in the web server 300 is launched for the first time, the response sentence generation means 26 communicates with the web server 300 and acquires language information recorded on the website of the web server 300 as the external information to be combined with text generation by the large-scale language model 400. By acquiring the external information in this manner, the chatbot can output an appropriate response sentence that matches the information recorded on the website in response to a question from a user about the content of the website.

[0048] The external information acquired in the above manner is stored in a memory unit within the system and is used when generating a response to a question in chatbot control using the response generation means 26.

[0049] As described above, the response sentence generation means 26 acquires language information recorded on the website as external information, combines the question sentence with the external information and inputs it into the machine-learned large-scale language model 400 to generate text, and generates the response sentence to be output by the chatbot control means 24. This function can be realized, for example, by applying a technology called RAG (Retrieval Augmented Generation). RAG is also called "retrieval augmented generation," and is a technology that combines the external information with text generation by the large-scale language model 400.

[0050] In the RAG, when a question is first input, the RAG retrieves external information as needed using a search algorithm. Once relevant external information is retrieved, preprocessing such as tokenization (a technique that replaces confidential data with random strings or numbers while maintaining the data format) and stemming (a text preprocessing method in natural language processing that converts word inflections into stems) is performed, and the preprocessed external information is incorporated into a large-scale language model 400 that has been machine-learned in advance. The large-scale language model 400 generates a more accurate response based on the context enhanced by combining the external information.

[0051] In the response sentence generation means 26 using such a RAG, when a user inputs a question about the content of the website to the chatbot, the RAG searches the website using a search algorithm or the like and acquires the linguistic information recorded there as external information. Once the external information is acquired in this manner, preprocessing such as the tokenization and stemming described above is performed, and the preprocessed external information is incorporated into the large-scale language model 400. The large-scale language model 400 receives a response sentence obtained based on the context enhanced by combining the external information. The response sentence generated in this manner is output by the chatbot control means 24.

[0052] In this way, by linking the linguistic information recorded on the website, it becomes possible to generate a response sentence that includes external information in a range that cannot be covered by the large-scale language model 400 alone. It also reduces hallucination that tends to occur in the large-scale language model 400. Furthermore, highly accurate and context-appropriate response sentences that utilize external information are output, making it possible to obtain effective response sentences for questions related to specific fields or specialized areas.

[0053] When the activation control means in the web server 300 is activated for the second or subsequent time, for example, if a predetermined time has passed since the previous external information acquisition, the response sentence generation means 26 communicates with the web server 300 and again acquires, as the external information, the language information recorded on the website of the web server 300. In other words, if a predetermined time has passed since the previous external information acquisition, the information posted on the website may have been updated, so the updated language information is acquired again as the external information. This allows the chatbot to output an appropriate response to the user's question that is compatible with the updated information recorded on the website.

[0054] The timing for re-obtaining the language information recorded on the website of the web server 300 as the external information can be, in addition to "the passage of a predetermined time" as described above, "when the administrator manually presses the update button" or "when the file size or update date and time of the web page has previously been obtained and has changed."

[0055] The response sentence generation means 26 acquires language information recorded on the website of the web server 300 as the external information to be combined with text generation by the large-scale language model 400, and can also acquire language information from other websites of other web servers 600 other than the website of the web server 300. Because it acquires not only language information posted on the website of the web server 300 but also language information from other websites, the chatbot outputs an appropriate response to a user's question that takes into account information recorded on the website as well as information from other websites.

[0056] In acquiring the external information by the response sentence generation means 26, if the web server 300 is configured with multiple servers and the multiple web pages that make up the website are distributed across the multiple servers, all of the web pages can be the targets for acquiring external information. If the website of the web server 300 has multiple web pages, all of the web pages can be the targets for acquiring external information. Similarly, if another website of another web server 600 has multiple web pages, all of the web pages can be the targets for acquiring external information.

[0057] ▽ Synchronous control means 27 The synchronization control means 27 synchronizes the reception of a user's question input by the chatbot control means 24 and the output of a response with the display of an avatar image by the avatar image display control means 22. For example, if the avatar image displayed on the website is a still image, the response is output in synchronization with the display of the still image. The response may be output as text or as audio. Both text and audio may be output. If the avatar image displayed on the website is a moving image, the response is output in synchronization with the display of the moving image. The moving image may be, for example, a moving image that changes lip-synching, gestures, facial angle, facial expression, etc., in response to the reception of a question input by the chatbot and the output of a response. When the response is output as text, the text may be displayed in a speech bubble displayed for the avatar image.

[0058] When displaying in a chatbot, it is necessary to display the avatar image and audio at roughly the same timing as the question or response (because, for example, it would sound strange if all the text was displayed, then the audio was played, and then the audio was lip-synced). Therefore, the output (display) of the text (response or question) and the output (display) of the avatar image and audio by the synchronization control means 27 are controlled so that the output of the text and the output of the avatar image and audio start to be displayed (played) almost simultaneously. In addition, depending on the length of the response text, it is possible to stop playback of the video or play back the audio while appropriately delimiting the content of the response text with punctuation marks, etc. In this way, the output of the text and the output of the avatar image and audio are synchronized.

[0059] [Web server 300] FIG. 3 is a system configuration diagram for explaining the web server 300. As shown in FIG. The web server 300 comprises a transmission / reception control section 31, a web program execution section 32, a web program 33, a data storage section 34, and a start control command 35.

[0060] ▽Transmission and reception control unit 31 The transmission / reception control unit 31 performs data communication with the other system that has accessed via the Internet 100 in accordance with a protocol such as HTTP (HyperText Transfer Protocol).

[0061] A unique network address (URL) is assigned to the web server 300. A user of a user terminal 500 can access the web server 300 by specifying the URL on the user terminal 500.

[0062] ▽Web Program 33 The web program 33 is a program for constructing a website written in a predetermined language such as HTML, PHP, XML, etc. In response to a request from the user terminal 500, the web program 33 publishes the website, which is data including documents and images written in a digital document description language such as HTML or PHP, on the Internet 100 and displays it on the user terminal 500.

[0063] ▽Data storage unit 34 The data storage unit 34 stores language information and image data for constructing the website.

[0064] ▽Web Program Execution Department 32 The web program execution unit 32 reads out the language information and image data stored in the data storage unit 34, executes the web program 33, constructs a predetermined website, and publishes it on the Internet 100. In other words, the web program and the web program execution unit 32 constitute the website construction means of the present invention, and the website construction means executes the website construction step of the present invention.

[0065] ▽Start control means A launch control command 35 exists as part of the web program 33 that constructs the website. The launch control command 35 is executed by the web program execution unit 32. Therefore, the launch control command 35 and the web program execution unit 32 constitute the launch control means of the present invention, and the launch control step of the present invention is executed by the launch control means. If the web program 33 is written in HTML, for example, the launch control command 35 can be a command using HTML tags.

[0066] The startup control command 35 exists as part of the web program 33 that constructs the website, and upon receiving startup input from the user, starts communication with the automatic response control server 200, and starts and operates at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27.

[0067] That is, by including the launch control command 35 in a part of the web program 33 that constructs the website, a function is realized to launch and operate at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27. In this way, the function of the present invention can be utilized by simply adding, for example, a few lines of launch control command 35, without making any major changes to the originally operated web program 33, and therefore the hurdle for adopting this technology is low for the operator of the website.

[0068] When the startup control means is started for the first time, as described above, in the automatic response control server 200, the response sentence generation means 26 acquires language information recorded on the website of the web server 300 as the external information to be combined with text generation by the large-scale language model 400.

[0069] When the startup control means is started for the second time or later, as described above, in the automatic response control server 200, if a predetermined time has elapsed since the previous machine learning, the response sentence generation means 26 again acquires language information recorded on the website of the web server 300 as the external information to be combined with text generation by the large-scale language model 400.

[0070] The external information acquired in the above manner is stored in a memory unit within the system and is used when generating a response to a question in chatbot control using the response generation means 26.

[0071] When the activation control means is activated, as described above, in the automatic response control server 200, the avatar image display control means 22 is operated to display an avatar image on the website of the web server 300, the chatbot control means 24 is operated to make the chatbot function, and the synchronization control means 27 is operated to synchronize the operation of the chatbot with the display of the avatar image.

[0072] That is, a website is displayed on the display of the user terminal 500 connected to the web server 300, and the website displays an avatar image, a question field for the chatbot, and a response field. At this time, the display of the chatbot and the display of the avatar image are synchronized. For example, in response to the input of a question or the output of a response by the chatbot, a moving image can be displayed as the avatar image, with lip-synching, gestures, and other movements, and with changes in facial angle and expression.

[0073] [Another web server 600] The other web server 600 has a different management entity from the web server 300, for example, and publishes a different website on the Internet 100 that stores different information than the website of the web server 300. The basic system configuration is the same as that of the web server 300 (although the launch control command 35 may not exist). The other web server 600 hosts a different website that contains documents and images written in a predetermined language such as HTML, PHP, or XML. By communicating with the automatic response control server 200, the response sentence generation means 26 acquires the language information of the other website as the external information to be combined with text generation by the large-scale language model 400.

[0074] The external information obtained in this way from other websites is also stored in the memory unit within the system and is used when generating a response to a question in chatbot control using the response generation means 26.

[0075] [Large-scale language model 400] The large-scale language model 400 is a deep learning model that uses machine learning to pre-train a language model, which models natural language based on its occurrence probability, from a huge amount of data. In other words, the large-scale language model 400 is a natural language processing model trained using a large amount of text data, and when a sentence in natural language is input as question information, a sentence in natural language is output as answer information. For example, in a chatbot, when a question sentence is input as question information, a response sentence is output as answer information. As the large-scale language model 400, for example, ChatGPT based on GPT-4 provided by OpenAI, Inc. in the United States, or other LLMs (large-scale language models) can also be used.

[0076] [User terminal 500]

[0077] The user terminal 500 may be, for example, a smartphone, tablet, or computer. The user terminal 500 is connected to the web server 300 via the Internet 100, allowing the user to browse websites using a browser, including documents and images written in a digital document description language such as HTML. The user terminal 500 also communicates with the web server 300 via the Internet 100 and executes processes. The user terminal 500 communicates with the web server 300 via the Internet 100, inputs and outputs data such as documents, images, and audio, as well as authentication information, as needed, and executes predetermined processes on the website. This allows the functions of the chatbot described above to be realized.

[0078] [Sequence Example 1] FIG. 4 is a sequence diagram illustrating a first example of the automatic response control method of this embodiment. FIG. 5 is a diagram showing an example of a web page and a chatbot screen.

[0079] In this embodiment, automatic response control is performed by a system that uses the above-mentioned web server 300, automatic response control server 200, user terminal 500, and large-scale language model 400.

[0080] First, the web server 300 acquires the activation control command 35 issued by the automatic response control server 200 and incorporates it into the web program 33 (S10). The activation control command 35 causes the activation control means of the present invention to function as described above. The activation control command 35 may be a command using HTML tags, for example, and functions as a trigger to accept the activation input from the user, start communication with the automatic response control server 200, and activate and operate at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27.

[0081] In this way, the functions of the present invention can be utilized by simply adding, for example, a few lines of startup control command 35 to the originally operating web program 33 without making any major changes to it, and therefore the barrier to adoption of this technology is low for operators of the above-mentioned websites.

[0082] In this state, the user terminal 500 connects to the web server 300, displays the web page on the display of the user terminal 500, and starts browsing the website.

[0083] As shown in FIG. 5(A), in the above state, a chatbot start button 41 is displayed on the website in accordance with the above-mentioned activation control command 35 (for example, a command using an HTML tag).

[0084] A start control step is executed by clicking the start button 41. As will be described later, in the start control step, the web program 33 that constructs the website starts communication with the automatic response control server 200 by accepting a start input from the user, and starts and operates at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot means 24, and the synchronization control means 27.

[0085] That is, when the start button 41 is clicked on the display of the user terminal 500, the start control means is started and starts communication with the web server 300. In this example, the start control means is started for the first time (S20). At this time, a question is also entered into the question display unit 44, and this question is accepted by the automatic response control server 200 (S25).

[0086] When the activation control means is activated for the first time, the response sentence generation means 26 of the automatic response control server 200 acquires language information recorded on the website of the web server 300 as external information, combines the question sentence with the external information and inputs it into the machine-learned large-scale language model 400 to generate text, and generates a response sentence to be output by chatbot control (S30).Then, the response sentence is output by the chatbot control means 24 of the automatic response control server 200 to the user terminal 500 via the web server 300 and displayed on its display (S35).

[0087] When the activation control means is activated, the avatar image display control means 22 is activated to display an avatar image on the website. The chatbot control means 24 is activated to activate the chatbot. The synchronization control means 27 is activated to synchronize the operation of the chatbot with the display of the avatar image.

[0088] The external information acquired in the manner described above is stored in a memory unit within the system and is used when generating responses to questions from the second time onwards in chatbot control using the response generation means 26.

[0089] Next, the second activation of the activation control means will be explained. When the user terminal 500 is connected to the web server 300 and a web page is displayed on the display of the user terminal 500, and the start button 41 is clicked, the activation control means is activated and starts communication with the web server 300 (S40).

[0090] When the activation control means is activated, avatar image display control means 22 is operated to display an avatar image on the website. At this time, an avatar image display control step is executed by avatar image display control means 22, which reads avatar image data from avatar image data storage unit 23 and displays avatar image 42 on the website.

[0091] The chatbot is made to function by operating the chatbot control means 24. At this time, a chatbot control step is executed by the chatbot control means 24, which controls the input acceptance of a question and the output of a response on the website, receives the question input by the user, and makes the chatbot function by outputting a response to the question.

[0092] Then, the synchronization control means 27 is operated to synchronize the operation of the chatbot with the display of the avatar image 42. That is, a synchronization control step is executed by the synchronization control means 27, in which the chatbot control means 24 accepts an input of a question from a user and outputs a response in synchronization with the display of the avatar image 42 by the avatar image display control means 22.

[0093] That is, as shown in FIG. 5(B), the website of the web server 300 displayed on the display of the user terminal 500 displays an avatar image 42 and a question box 43 for the chatbot.

[0094] When a question is entered as text in the question field 43 displayed on the display of the user terminal 500 and sent, the question is sent to the automatic response control server 200 via the web server 300 (S50). In the automatic response control server 200, the chatbot control means 24 generates text by inference based on the entered question using the response generation means 26 and the large-scale language model 400, and a response is generated (S55). At this time, the text is generated by combining external information stored in a memory unit within the system with the large-scale language model 400. The response generated by the chatbot control means 24 is output to the user terminal 500 via the web server 300 and displayed on its display (S60).

[0095] 5(C), the website of web server 300 displayed on the display of user terminal 500 displays avatar image 42 and a question box 43 for the chatbot, as well as a question display section 44 in which the question is displayed and a response display section 45 in which a response to the question is displayed. In the illustrated example, response display section 45 is displayed as a speech bubble that looks like a statement made by avatar image 42.

[0096] At this time, the synchronization control means 27 operates to synchronize the behavior of the chatbot with the display of the avatar image 42. For example, in response to the output of a response sentence from the chatbot, it is possible to display a moving image of the avatar image 42, which changes lip-synching, gestures, and other movements, as well as facial angles and expressions.

[0097] The response sentence can be output not only as text as described above, but also as voice. In this case, the response sentence generated by the chatbot control means 24 is voice-synthesized by the voice control unit 25 to generate voice data, which is output from the speaker of the user terminal 500.

[0098] The above-mentioned input of a question sentence (S50) and output of a response sentence (S60) can be repeated multiple times as necessary.

[0099] Next, the third activation of the activation control means will be explained. When the user terminal 500 is connected to the web server 300 and a web page is displayed on the display of the user terminal 500, and the start button 41 is clicked, the activation control means is activated and communication with the web server 300 begins. In this case, the activation of the activation control means is the third time, so it is the second or subsequent time (S70). At this time, a question is also entered into the question display unit 44, and this question is accepted by the automatic response control server 200 (S75).

[0100] Then, since this is the second or subsequent activation of the activation control means, in the automatic response control server 200, if a predetermined time has passed since the previous machine learning, the response sentence generation means 26 of the automatic response control server 200 again acquires language information recorded on the website of the web server 300 as the external information, inputs the question sentence in combination with the external information into the large-scale language model 400 to generate text, and receives it as a response sentence for chatbot control (S80).Then, the response sentence is output to the user terminal 500 via the web server 300 by the chatbot control means 24 of the automatic response control server 200 (S85).

[0101] In this way, if a predetermined amount of time has passed since the last machine learning, there is a possibility that the information posted on the website has been updated. Therefore, by performing machine learning on the updated language information again to obtain a machine learning model, the chatbot can output an appropriate response to the user's question that is compatible with the updated information recorded on the website.

[0102] Furthermore, the above-described input of a question and output of a response can be repeated as necessary. That is, when a question is input as text in the question field 43 displayed on the display of the user terminal 500 and sent, the question is transmitted to the automatic response control server 200 via the web server 300 (S90). In the automatic response control server 200, the chatbot control means 24 generates text by inference based on the input question using the response generation means 26 and the large-scale language model 400, and generates a response (S95). At this time, the text is generated by combining external information stored in a memory unit within the system with the large-scale language model 400. The response generated by the chatbot control means 24 is output to the user terminal 500 via the web server 300 and displayed on its display (S100).

[0103] [Sequence Example 2] FIG. 6 is a sequence diagram illustrating a second example of the automatic response control method according to the present embodiment. In this example, the response sentence generation means 26 acquires, as external information, language information recorded on the website of the web server 300, and also acquires, as external information, language information from other websites of other web servers 600 other than the web server 300. Other than that, the example is the same as the first example in Fig. 4, and similar explanations may be omitted.

[0104] First, the web server 300 acquires the start control command 35 issued by the automatic response control server 200 and incorporates it into the web program 33 (S10).

[0105] In this state, the user terminal 500 connects to the web server 300, displays the web page on the display of the user terminal 500, and starts browsing the website.

[0106] Next, when the start button 41 is clicked on the display of the user terminal 500, the start control means is started and starts communication with the web server 300. In this example, the start control means is started for the first time (S20). At this time, a question is also entered into the question display unit 44, and this question is accepted by the automatic response control server 200 (S25).

[0107] When the activation control means is activated for the first time, the response sentence generation means 26 of the automatic response control server 200 acquires, as external information, linguistic information recorded on the website of the web server 300. At this time, the response sentence generation means 26 acquires, as external information, linguistic information from other websites of web servers 600 other than the web server 300 (S30). Then, the question sentence is combined with both pieces of external information and input to the machine-learned large-scale language model 400 to generate text, and the large-scale language model 400 generates a response sentence for chatbot control (S35). Then, the response sentence is output by the chatbot control means 24 of the automatic response control server 200 to the user terminal 500 via the web server 300 and displayed on its display (S36). As a result, the chatbot outputs an appropriate response to the user's question, taking into account information recorded on the website as well as information from other websites.

[0108] When the activation control means is activated, the avatar image display control means 22 is activated to display an avatar image on the website. The chatbot control means 24 is activated to activate the chatbot. The synchronization control means 27 is activated to synchronize the operation of the chatbot with the display of the avatar image.

[0109] The external information acquired in the manner described above is stored in a memory unit within the system and is used when generating responses to questions from the second time onwards in chatbot control using the response generation means 26.

[0110] Next, the second activation of the activation control means will be explained. When the user terminal 500 is connected to the web server 300 and a web page is displayed on the display of the user terminal 500, and the start button 41 is clicked, the activation control means is activated and starts communication with the web server 300. In this case, since this is the second activation of the activation control means, this is the second or subsequent time (S40).

[0111] When the activation control means is activated, the avatar image display control means 22 is activated to display an avatar image on the website, the chatbot control means 24 is activated to activate the chatbot, and the synchronization control means 27 is activated to synchronize the operation of the chatbot with the display of the avatar image.

[0112] When a question is entered as text in the question field 43 displayed on the display of the user terminal 500 and sent, the question is sent to the automatic response control server 200 via the web server 300 (S50). In the automatic response control server 200, the chatbot control means 24 generates text by inference based on the entered question using the learning model obtained by the response generation means 26, and generates a response (S55). The response generated by the chatbot control means 24 is output to the user terminal 500 via the web server 300 and displayed on its display (S60).

[0113] At this time, the synchronization control means 27 operates to synchronize the behavior of the chatbot with the display of the avatar image 42. For example, in response to the output of a response sentence from the chatbot, it is possible to display a moving image of the avatar image 42, which changes lip-synching, gestures, and other movements, as well as facial angles and expressions.

[0114] If necessary, the above-mentioned input of a question sentence (S50) and output of a response sentence (S60) can be repeated multiple times.

[0115] Next, the third activation of the activation control means will be explained. When the user terminal 500 is connected to the web server 300 and a web page is displayed on the display of the user terminal 500, and the start button 41 is clicked, the activation control means is activated and communication with the web server 300 begins. In this case, the activation of the activation control means is the third time, so it is the second or subsequent time (S70). At this time, a question is also entered into the question display unit 44, and this question is accepted by the automatic response control server 200 (S75).

[0116] Since this is the second or subsequent activation of the activation control means, in the automatic response control server 200, if a predetermined time has elapsed since the previous machine learning, the response sentence generation means 26 of the automatic response control server 200 again acquires, as external information, language information recorded on the website of the web server 300. At this time, the response sentence generation means 26 acquires, as external information, language information from other websites of web servers 600 other than the web server 300 (S80). Then, the question sentence is combined with the external information and input to the machine-learned large-scale language model 400 to generate text, thereby generating a response sentence for chatbot control (S85). The response sentence is then output by the chatbot control means 24 of the automatic response control server 200 to the user terminal 500 via the web server 300 and displayed on its display (S86). This allows the chatbot to output an appropriate response to a user's question that takes into account information recorded on the website as well as information from other websites.

[0117] FIG. 7 is a system configuration diagram illustrating a second example of the automatic response control server 200 of this embodiment. This example includes an image generation AI 28. The image generation AI 28 generates an image from an input question. The generated image is displayed on the website when the chatbot is activated by the chatbot control means 24. The image generation AI 28 is an AI that generates an image from text, and can use, for example, Stable Diffusion, Midjourney, etc., but is not limited to these. The rest is the same as the embodiment described in Figure 2, and similar parts are given the same reference numerals and will not be described again.

[0118] FIG. 8 is a diagram showing a second display example of the web page and chatbot screen. This example shows a web page and chatbot screen according to the second example of the automatic response control server 200. In this second display example, the website of the web server 300 displayed on the display of the user terminal 500 displays a question display section 44 in which the question is displayed, a response display section 45 in which a response to the question is displayed, and an image 46 is displayed in the response display section 45. In the illustrated example, in response to the question "Please tell me how to get to work," the image generation AI 28 generates a map image 46 and displays it in the response display section 45. The manner in which the image is displayed is not limited to the manner described above.

[0119] FIG. 9 is a system configuration diagram illustrating a third example of the automatic response control server 200 of this embodiment. This example includes image recognition means 29. The image recognition means 29 receives image data input to the user terminal 500 via the web server 300 and performs image recognition on the received image data. The response sentence generation means 26 generates the response sentence by referring to the recognition results. That is, the image recognition means 29 analyzes the pixels of the received image data, analyzes the image based on past learning data, and converts it into data (i.e., verbalizes it). For example, the "meaning and characteristics of the image," such as what is depicted in the image, its positional relationship, and what text, if any, is written, are recognized and converted into verbalized data. The chatbot control means 24 creates a response sentence based on the verbalized data. Other than that, the embodiment is similar to that described in FIG. 2, and similar parts are designated by the same reference numerals, and their description will be omitted.

[0120] [Effects of each embodiment] As described above, the above-described embodiments provide the following advantages.

[0121] This embodiment includes a web server 300 storing a website and an automatic response control server 200. The automatic response control server 200 includes an avatar image display control unit 22, a chatbot control unit 24, a response sentence generation unit 26, and a synchronization control unit 27. The avatar image display control unit 22 reads avatar image data from an avatar image data storage unit 23 and displays an avatar image 42 on the website. The chatbot control unit 24 controls the input acceptance of a question sentence and the output of a response sentence on the website, receiving the question sentence entered by a user and outputting a response sentence to the question sentence, thereby causing the chatbot to function. The response sentence generation unit 26 acquires linguistic information recorded on the website as external information, combines the question sentence with the external information, and inputs it into a machine-learned large-scale language model 400 to generate text, thereby generating the response sentence to be output by the chatbot control unit. The synchronization control means 27 synchronizes the reception of a user's question input by the chatbot control means 24 and the output of a response sentence with the display of the avatar image 42 by the avatar image display control means 22. The web server 300 has a launch control means as part of the website construction means that constructs the website. Upon receiving a launch input from a user, the launch control means starts communication with the automatic response control server 200 and launches and operates at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27. Therefore, when a user connects his / her user terminal 500 to the web server 300, displays the website on the user terminal 500, and performs an activation input for the activation control means on the website, communication between the web server 300 and the automatic response control server 200 begins. Then, at least one of the avatar image display control means 22, the response sentence generation means 26, the chatbot control means 24, and the synchronization control means 27 is activated and operates. As a result, the avatar image display control means 22 reads avatar image data from the avatar image data storage unit 23 and displays the avatar image 42 on the website. The response sentence generation means 26 acquires language information recorded on the website as external information, and generates the response sentence by the chatbot control means 22 in combination with a large-scale language model 400 trained by machine learning. The chatbot control means 24 controls the input acceptance of a question and the output of a response on the website, receives the question input by a user, and outputs a response to the question, thereby functioning as a chatbot. Furthermore, the synchronization control means 27 synchronizes the input acceptance of a question from a user and the output of a response by the chatbot control means 24 with the display of an avatar image 42 by the avatar image display control means 22. By the above operation, a chatbot synchronized with the avatar image 42 functions on the website on the web server 300, and is able to receive and respond to inquiries from the user, making it easy to set up a chatbot inquiry desk on various web pages.

[0122] In this embodiment, when the startup control means is started for the first time, the response sentence generation means 26 acquires language information recorded on the website as the external information to be combined with text generation by the large-scale language model 400. The response sentence generation means 26 acquires the language information recorded on the website as the external information to be combined with text generation by the large-scale language model 400, so that the chatbot can output an appropriate response sentence that matches the information recorded on the website in response to a question from a user about the content of the website.

[0123] In this embodiment, when the activation control means is activated for the second time or later, the response sentence generation means 26 again acquires the language information recorded on the website as the external information to be combined with text generation by the large-scale language model 400. For example, if a predetermined amount of time has passed since the last response was generated, there is a possibility that the information posted on the website has been updated. By re-obtaining the updated language information, the chatbot will output an appropriate response to the user's question that matches the updated information recorded on the website.

[0124] In this embodiment, when the activation control means is activated, the avatar image display control means 22 is operated to display the avatar image 42 on the website, the chatbot control means 24 is operated to make the chatbot function, and the synchronization control means 27 is operated to synchronize the operation of the chatbot with the display of the avatar image 42. Every time the activation control means is activated, the behavior of the chatbot is synchronized with the display of the avatar image 42. Furthermore, for example, since the response sentence generation means 26 has already obtained the language information recorded on the website when the activation control means is activated for the first time, the chatbot outputs an appropriate response sentence that matches the information recorded on the website in response to a user's question when activated for the second time or later.

[0125] In this embodiment, the response sentence generation means 26 acquires language information recorded on the website as well as language information from websites other than the website itself as the external information to be combined in text generation by the large-scale language model 400. Because the chatbot acquires not only the language information posted on the website but also the language information of other websites, it outputs an appropriate response to a user's question that takes into account information recorded on the website as well as information from other websites.

[0126] In this embodiment, the chatbot control means 24 includes a voice unit means 25 that receives the input of the question sentence and outputs the response sentence by voice. In chatbots, users can input questions by voice rather than by text, and the output responses can be obtained by listening to the voice rather than reading the text, making them more convenient.

[0127] [Other Modifications] The above describes a particularly preferred embodiment of the present invention, but the present invention is not intended to be limited to the illustrated embodiment, and can be modified and implemented in various ways, and the present invention is intended to encompass various modified examples.

[0128] For example, the present invention also includes a mode in which a question sentence is input and a response sentence is output only by voice. Furthermore, the above-mentioned web server 300 and other web servers 600 may each be composed of a single server or multiple servers. Furthermore, the devices for performing input and output processing are not limited to the above-mentioned devices that input or output text data, devices that input or output images, and devices that input or output voice, but may also include devices that perform various types of input or output. [Explanation of symbols]

[0129] 21: Transmission and reception control section 22: Avatar image display control means 23: Avatar image data storage section 24: Chatbot control means 25: Audio control unit 26: Response sentence generation means 27: Synchronous control means 28: Image generation AI 29: Image recognition method 31: Transmission and reception control section 32: Web Program Execution Department 33: Web Program 34: Data storage section 35: Startup control command 41: Start button 42:Avatar image 43: Questions 44:Question display section 45: Response display unit 46: Image 100: Internet 200: Automatic response control server 300: Web server 400: Large-scale language model 500: User terminal 600: Other web servers

Claims

1. The system is configured to include a web server on which a website is stored and an automatic response control server, The automatic response control server is avatar image display control means for reading the avatar image data from the avatar image data storage unit and displaying the avatar image on the website; a chatbot control means for controlling the reception of input questions and the output of responses on the website, receiving the questions input by users, and outputting responses to the questions, thereby causing the chatbot to function; a response generation means for acquiring language information recorded on the website as external information, inputting the question in combination with the external information into a large-scale language model trained by machine learning to generate text, and generating the response to be output by the chatbot control means; a synchronization control means for synchronizing reception of an input of a user's question and output of a response by the chatbot control means with display of an avatar image by the avatar image display control means, The web server is As part of the website construction means for constructing the website, there is provided a start-up control means for starting communication with the automatic response control server by accepting a start-up input from a user, and for starting and operating at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means. An automatic response control system characterized by:

2. When the activation of the activation control means is the first time, The response sentence generation means acquires language information recorded on the website as the external information to be combined with text generation using a large-scale language model.

2. The automatic response control system according to claim 1.

3. When the activation of the activation control means is the second or subsequent time, The response sentence generation means acquires language information recorded on the website as the external information to be combined again in text generation using a large-scale language model.

3. The automatic response control system according to claim 2.

4. When the start-up control means is started, The avatar image display control means is operated to display an avatar image on the website, the chatbot control means is operated to cause the chatbot to function, and the synchronization control means is operated to synchronize the operation of the chatbot with the display of the avatar image.

2. The automatic response control system according to claim 1.

5. The response sentence generation means acquires language information recorded on the website as well as language information from websites other than the website as the external information to be combined in text generation using a large-scale language model. The automatic response control system according to any one of claims 1 to 4.

6. The chatbot control means includes a voice control unit that receives the question and outputs the response by voice. The automatic response control system according to any one of claims 1 to 3.

7. Using a web server on which a website is stored and an automatic response control server, The automatic response control server is an avatar image display control step by avatar image display control means for reading avatar image data from an avatar image data storage unit and displaying the avatar image on the website; a chatbot control step by a chatbot control means for controlling the reception of input of a question and the output of a response on the website, receiving the question input by a user, and outputting a response to the question, thereby causing the chatbot to function; a response sentence generation step by a response sentence generation means that acquires language information recorded on the website as external information, inputs the question sentence into a large-scale language model that has been machine-learned in combination with the external information to generate text, and generates the response sentence to be output in the chatbot control step; a synchronization control step by a synchronization control means for synchronizing the reception of an input of a question sentence from a user and the output of a response sentence by the chatbot control means with the display of an avatar image by the avatar image display control means; The web server is In the website construction step of constructing the website, a start-up control step is executed in which communication with the automatic response control server is initiated by receiving a start-up input from a user, and at least one of the avatar image display control means, the response sentence generation means, the chatbot control means, and the synchronization control means is started and operated. An automatic response control method comprising:

8. A program to be executed by a web server storing a website and an automatic response control server, The automatic response control server is an avatar image display control step by avatar image display control means for reading avatar image data from an avatar image data storage unit and displaying the avatar image on the website; a chatbot control step by a chatbot control means for controlling the reception of input of a question and the output of a response on the website, receiving the question input by a user, and outputting a response to the question, thereby causing the chatbot to function; a response generation step by a response generation means for acquiring language information recorded on the website as external information, inputting the question into a large-scale language model trained by machine learning in combination with the external information to generate text, and generating the response to be output by the chatbot control means; a synchronization control step by a synchronization control means for synchronizing the reception of an input of a question sentence from a user and the output of a response sentence by the chatbot control means with the display of an avatar image by the avatar image display control means; The web server is In a part of the website construction step of constructing the website, a start-up control step is executed in which communication with the automatic response control server is initiated by receiving a start-up input from a user, and at least one of the avatar image display control means, the response sentence generation means, the chatbot means, and the synchronization control means is started and operated. The automatic response control program is characterized by causing the web server and the automatic response control server to execute the above.

Images