Information processing method, information processing device, and computer program

Abstract

Provided are an information processing method, an information processing device, and a computer program.　A computer executes processing for acquiring a query of a user and outputting, by means of a language model using book data, an answer sentence to the acquired query.

Description

Information processing method, information processing device, and computer program

[0001] The present disclosure relates to an information processing method, an information processing device, and a computer program.

[0002] Conventionally, systems have been provided that perform natural language processing such as question-and-answering using a language model (see, for example, Patent Document 1). The language model is a natural language processing model trained using a large amount of text data, and is configured to receive a sentence as input and output a sentence. When applied to a question-and-answering system, the language model is trained to output an answer sentence in response to an input question sentence.

[0003] Japanese Patent Application Laid-Open No. 2022-180282

[0004] Conventional language models only output general answers to user questions, and are unable to reflect the thoughts of specific individuals in the answers.

[0005] In one aspect, an object is to provide an information processing method, an information processing device, and a computer program that can output a response to a user's question that reflects the thoughts of a specific person.

[0006] The information processing method disclosed herein involves a computer executing a process of acquiring a user's query and outputting a response sentence to the acquired query using a language model that uses book data.

[0007] In one aspect, it is possible to output answers to user questions that reflect the thoughts of a specific person.

[0008] 1 is a schematic diagram showing an example of the configuration of an AI chat system according to an embodiment. FIG. 1 is a block diagram showing the internal configuration of an AI server. FIG. 2 is a conceptual diagram showing an example of a book database. FIG. 3 is a block diagram showing the internal configuration of an LLM server. FIG. 4 is a block diagram showing the internal configuration of a user terminal. FIG. 5 is an explanatory diagram explaining an overview of processing in the AI ​​chat system. FIG. 6 is an explanatory diagram explaining an overview of processing in the AI ​​chat system. FIG. 7 is a flowchart explaining the procedure of processing executed by the AI ​​chat system. FIG. 8 is a schematic diagram showing an example of a user interface screen provided by the AI ​​server. FIG. 9 is a conceptual diagram explaining an example of a prompt generated by the AI ​​server. FIG. 10 is a schematic diagram showing an example of an output response sentence to a user's query. FIG. 11 is a conceptual diagram explaining an example of a prompt taking into account the attributes of the questioner (user) and the answerer.

[0009] The present invention will be described in detail below with reference to the drawings illustrating embodiments thereof. (Embodiment 1) FIG. 1 is a schematic diagram showing an example configuration of an AI chat system according to an embodiment. The AI ​​chat system according to this embodiment includes an AI server 10, an LLM server 20, and a user terminal 30. The AI ​​server 10, the LLM server 20, and the user terminal 30 are communicably connected to each other via a communication network NW such as the Internet. The AI ​​server 10 is a server computer that provides chat services to users. The LLM server 20 is a server computer that provides services using a large language model (LLM, hereinafter referred to as LLM). The user terminal 30 is a client computer used by a user.

[0010] The AI ​​server 10 includes a book database (DB) that stores texts contained in books. In this embodiment, a book may be any text containing texts, and may include not only printed and bound publications but also e-books, magazines, comics, company histories, and unpublished drafts. In this embodiment, text data contained in books is converted into data that is easy to handle in a computer system (e.g., numerical data such as vector data) and registered in the book database (DB). The text data registered in the book database (DB) is hereinafter referred to as book data. Book data is registered in the book database (DB) by book, author, and expert. In this embodiment, an expert refers to a group of authors categorized by appropriate category. Experts include managers, entrepreneurs, investors, politicians, critics, thinkers, artists, researchers, persons with professional or tutoring qualifications (e.g., lawyers, tax accountants, patent attorneys, doctors, nurses, teachers), professional athletes, and sports instructors. In this embodiment, an expert may be a group of one or more authors, and the authors included in the group do not necessarily have a common specialty. Furthermore, the book data registered in the book database DB is not limited to sentences extracted from books, but may also be data on sentences contained in content made public by authors on the communication network NW (such as articles posted on SNS or blogs).

[0011] The AI ​​server 10 provides a chat service to users using book data registered in the book database DB. Specifically, the AI ​​server 10 accepts a user's query through the user terminal 30, extracts book data related to the accepted query from the book database DB, and uses the extracted book data to generate input data for the LLM. The input data for the LLM is also called a prompt. The AI ​​server 10 transmits the generated prompt to the LLM server 20 and receives a response from the LLM server 20. The AI ​​server 10 outputs the response received from the LLM server 20 as a response to the user's query.

[0012] The device configuration of each device that makes up the AI ​​chat system will be described below.

[0013] 2 is a block diagram showing the internal configuration of the AI ​​server 10. The AI ​​server 10 is a dedicated or general-purpose server computer, and includes a control unit 11, a memory unit 12, a communication unit 13, an operation unit 14, and a display unit 15.

[0014] The control unit 11 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The ROM included in the control unit 11 stores control programs and the like that control the operation of each hardware component included in the AI ​​server 10. The CPU in the control unit 11 reads and executes the control programs stored in the ROM and the computer programs (described below) stored in the memory unit 12, and controls the operation of each hardware component, thereby causing the entire device to function as the information processing device of the present disclosure. The RAM included in the control unit 11 temporarily stores data used during the execution of calculations and control.

[0015] In the embodiment, the control unit 11 is configured to include a CPU, a ROM, and a RAM, but the configuration of the control unit 11 is not limited to the above. The control unit 11 may be, for example, one or more control circuits or arithmetic circuits including a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), a quantum processor, volatile or non-volatile memory, etc. The control unit 11 may also have functions such as a clock that outputs date and time information, a timer that measures the elapsed time from when a measurement start instruction is given until when a measurement end instruction is given, and a counter that counts numbers.

[0016] The storage unit 12 includes a storage device such as a hard disk drive (HDD), a solid state drive (SSD), etc. The storage unit 12 stores various computer programs executed by the control unit 11 and various data used by the control unit 11.

[0017] The computer program (program product) stored in the storage unit 12 includes a chat generation program PG for causing a computer to execute a process of acquiring a user's query and outputting a response to the acquired query using a language model that uses book data. The chat generation program PG may be a single computer program or a group of programs consisting of multiple computer programs. The chat generation program PG may be executed by multiple computers in cooperation. The chat generation program PG may also partially use an existing library.

[0018] A computer program including the chat generation program PG is provided by a non-transitory recording medium RM on which the computer program is readably recorded. The recording medium RM is a portable memory such as a CD-ROM, USB memory, a Secure Digital (SD) card, a microSD card, or a CompactFlash (registered trademark). The control unit 11 reads various computer programs from the recording medium RM using a reading device (not shown) and stores the read computer programs in the memory unit 12. The computer program including the chat generation program PG may be provided via communications. In this case, the control unit 11 acquires the computer program including the chat generation program PG through communications via the communication unit 13 and stores the acquired computer program in the memory unit 12.

[0019] The AI ​​server 10 includes the book database DB in the storage unit 12. FIG. 3 is a conceptual diagram illustrating an example of the book database DB. The book database DB stores author IDs, group IDs, book IDs, and book data in association with each other. The author ID is an identifier that identifies the author. The group ID is an identifier that identifies the group (expert) to which the author belongs. Each author is assigned one or more group IDs. Some authors may not be assigned a group ID. The book ID is an identifier that identifies books written by the author. Some authors may have multiple book IDs registered. Book data represents data obtained by quantifying sentences contained in books. Existing methods are used to quantify sentences. For example, sentences are quantified by dividing them into words (morphemes), counting the number of occurrences of each word for each sentence, and generating a vector whose length is the number of vocabulary words. The quantification may be performed on a sentence-by-sentence basis or on a sentence-by-sentence basis. In this embodiment, "book data" is a term that refers to a collection of numerical data obtained from a single book. However, the term "book data" does not distinguish between the above and numerical data corresponding to individual sentences or sentences. The control unit 11 of the AI ​​server 10 can use the book database DB to calculate the similarity between a query received from a user and a sentence contained in a book.

[0020] In this embodiment, a book ID is assigned to each book, but a book ID may also be assigned to each chapter within a book, or a book ID may be assigned to a portion of a sentence in a book.

[0021] Furthermore, in this embodiment, book data extracted from a book is registered in association with an author ID. However, data extracted from text included in content published by an author on a communication network NW (such as articles posted on social media or blogs) may also be registered in association with an author ID. In this case, for example, text (text data) may be acquired from a pre-defined author's social media or blog, and the following may be registered in the book database DB: an author ID as an identifier identifying the author; a book ID as an identifier identifying the author's social media or blog; and numerical data converted from the acquired text as book data. By registering text acquired from social media or blogs in this way in the book database DB, the author's expressions on social media or blogs can be treated in the same way as book data extracted from a book. Note that only a portion of the text included in the content may be registered in the book database DB as book data, or the entire text may be registered in the book database DB as book data. Furthermore, the text included in the content may be summarized and registered in the book database DB, or translated and registered in the book database DB.

[0022] The communication unit 13 includes a communication interface for transmitting and receiving various data to and from an external device. The communication interface of the communication unit 13 may be a communication interface conforming to a communication standard such as Wi-Fi (registered trademark), LAN (Local Area Network), Bluetooth (registered trademark), ZigBee (registered trademark), 3G, 4G, 5G, or LTE (Long Term Evolution). When data to be transmitted is input from the control unit 11, the communication unit 13 transmits the data to the destination external device, and when data transmitted from the external device is received, the communication unit 13 outputs the received data to the control unit 11. In this embodiment, an example of an external device is a user terminal 30.

[0023] The operation unit 14 includes operation devices such as a touch panel, a keyboard, and switches, and receives various operations and settings from an administrator, etc. The control unit 11 performs appropriate control based on various pieces of operation information provided by the operation unit 14, and stores setting information in the storage unit 12 as necessary.

[0024] The display unit 15 includes a display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display, and displays information to be presented in response to an instruction from the control unit 11 .

[0025] In this embodiment, the AI ​​server 10 may be a single computer, or may be a computer system configured with multiple computers and peripheral devices, etc. Furthermore, the AI ​​server 10 may be a virtual machine whose entity is virtualized, or may be a cloud.

[0026] 4 is a block diagram showing the internal configuration of the LLM server 20. The LLM server 20 is a dedicated or general-purpose server computer, and includes a control unit 21, a storage unit 22, a communication unit 23, an operation unit 24, and a display unit 25.

[0027] The control unit 21 includes, for example, a CPU, a ROM, a RAM, etc. The CPU of the control unit 21 controls the operation of the various hardware components described above by loading various programs stored in advance in the ROM or the storage unit 22 into the RAM and executing them.

[0028] In the embodiment, the control unit 21 is configured to include a CPU, a ROM, and a RAM, but the configuration of the control unit 21 is not limited to the above. The control unit 21 may be one or more control circuits or arithmetic circuits including a GPU, an FPGA, a DSP, a quantum processor, a volatile or non-volatile memory, etc.

[0029] The storage unit 22 includes storage devices such as a memory and a hard disk. The storage unit 22 stores various computer programs executed by the control unit 21, data necessary for executing the computer programs, etc. The computer programs stored in the storage unit 22 include a natural language processing program that causes the computer to interpret prompts received from outside and generate sentences using the LLM.

[0030] The storage unit 22 also includes an LLM such as GPT-3 (Generative Pretrained Transformer 3), LaMDA (Language Model for Dialogue Applications), PaLM (Pathways Language Model), LLaMA (Large Language Model Meta AI), and Weblab-10B. The LLM included in the LLM server 20 is not limited to the above, and may be any existing model or a model obtained by tuning an existing model.

[0031] The communication unit 23 includes a communication interface for connecting to the communication network NW. The communication interface included in the communication unit 23 is a communication interface conforming to a communication standard such as Wi-Fi (registered trademark), LAN, Bluetooth (registered trademark), ZigBee (registered trademark), 3G, 4G, 5G, LTE, etc. The communication unit 23 transmits various information to be notified to the outside and receives various information transmitted from the outside to the device itself.

[0032] The operation unit 24 is equipped with input devices such as a touch panel and operation buttons, and receives various types of operation information and setting information. The operation unit 24 outputs the received operation information and setting information to the control unit 21. The control unit 21 performs appropriate control based on the operation information input from the operation unit 24, and stores the setting information in the storage unit 22 as necessary.

[0033] The display unit 25 includes a display device such as a liquid crystal display or an organic EL display, and displays information to be presented to the administrator or the like based on a control signal output from the control unit 21 .

[0034] In this embodiment, the AI ​​server 10 and the LLM server 20 are separate entities, but they may also be integrated. That is, the AI ​​server 10 may be configured to include an LLM. In this case, the AI ​​server 10 generates a prompt based on a query received from the user terminal 30, generates an answer sentence based on the generated prompt within the AI ​​server 10, and outputs the answer sentence to the user terminal 30.

[0035] 5 is a block diagram showing the internal configuration of the user terminal 30. The user terminal 30 is a computer such as a smartphone, a tablet terminal, or a personal computer, and includes a control unit 31, a storage unit 32, a communication unit 33, an operation unit 34, and a display unit 35.

[0036] The control unit 31 includes, for example, a CPU, a ROM, a RAM, etc. The CPU of the control unit 31 controls the operation of the various hardware components described above by loading various programs stored in advance in the ROM or the storage unit 32 into the RAM and executing them, causing the entire device to function as the user terminal 30 of the present disclosure.

[0037] The control unit 31 is not limited to the above configuration, and may be configured as a single piece of hardware (SoC: System On a Chip) that integrates a processor, memory, storage, communication interface, etc. The control unit 31 may also have functions such as a clock that outputs date and time information, a timer that measures the elapsed time from when an instruction to start measurement is given until when an instruction to end measurement is given, and a counter that counts numbers.

[0038] The storage unit 32 includes storage such as a memory and a hard disk. The storage unit 32 stores various computer programs executed by the control unit 31, data necessary for executing the computer programs, etc. The computer programs stored in the storage unit 32 include application programs for using the services provided by the AI ​​server 10.

[0039] The programs stored in the storage unit 32 may be provided by a non-transitory storage medium on which the programs are readably recorded. The storage medium is, for example, a portable memory such as a CD-ROM, a USB memory, or an SD card. The control unit 31 reads the various programs from the storage medium using a reading device (not shown) and installs the read programs in the storage unit 32. The programs stored in the storage unit 32 may also be provided by communication. In this case, the control unit 31 acquires the various programs through the communication unit 33 and installs the acquired programs in the storage unit 32.

[0040] The communication unit 33 includes a communication interface for connecting to the communication network NW. The communication interface included in the communication unit 33 is a communication interface conforming to a communication standard such as Wi-Fi (registered trademark), LAN, Bluetooth (registered trademark), ZigBee (registered trademark), 3G, 4G, 5G, LTE, etc. The communication unit 33 transmits various information to be notified to the outside and receives various information transmitted from the outside to the device itself.

[0041] The operation unit 34 is equipped with input devices such as a touch panel and operation buttons, and receives various types of operation information and setting information. The operation unit 34 outputs the received operation information and setting information to the control unit 31. The control unit 31 performs appropriate control based on the operation information input from the operation unit 34, and stores the setting information in the storage unit 32 as necessary.

[0042] The display unit 35 includes a display device such as a liquid crystal display or an organic EL display, and displays information to be presented to the user based on a control signal output from the control unit 31 .

[0043] An overview of the processing in the AI ​​chat system will be explained below. Figures 6 and 7 are explanatory diagrams illustrating an overview of the processing in the AI ​​chat system. Figure 6 shows an overview of the processing in the preparation stage before the service starts, and Figure 7 shows an overview of the processing after the service starts. In the preparation stage before the service starts, the AI ​​server 10 performs a process of extracting book data from books and registering the extracted book data in the book database DB. This registration process is also executed as needed after the service starts.

[0044] When the AI ​​server 10 acquires a book to be registered, it converts the sentences contained in the book into digital text data using, for example, OCR (Optical Character Recognition) technology, and then converts the text data into book data (numeric data) using an existing digitization method. The AI ​​server 10 associates the book data extracted from one book with an author ID, group ID, and book ID, and registers the book data in the book database DB. The AI ​​server 10 performs this registration process in the book database DB as needed.

[0045] For example, if a book contains the sentence "...the group's philosophy, 'The heart is what matters,' is something my teacher taught me when I was a student...", the AI ​​server 10 calculates numerical data (vector data) such as [12.345, 67.890, ...], [33.444, 55.666, ...], ... from the text data of the sentence, and registers the obtained numerical data (book data) in the book database DB together with the author ID, group ID, and book ID.

[0046] The AI ​​server 10 starts providing services once the registration process in the book database DB has progressed, and accepts user queries from the user terminal 30. The AI ​​server 10 quantifies the sentences included in the accepted query and compares them with the book data registered in the book database DB to extract one or more pieces of book data that have a high similarity to the query from the book database DB. A known method such as cosine similarity is used to calculate the similarity. Cosine similarity is an index that indicates the degree to which vectors are oriented in the same direction, and takes on a real value ranging from -1 to +1 depending on the level of similarity.

[0047] For example, when a query such as "Q: What does 'The heart is what matters' mean?" is input from the user terminal 30, the AI ​​server 10 digitizes a sentence included in the query (e.g., the main part of the query, "The heart is what matters' mean) and searches the book database DB based on the resulting numerical data ([12.333, 67.888, ...] in the example of Figure 7), thereby extracting highly similar book data. At this time, the AI ​​server 10 extracts one or more book data having a similarity higher than a threshold value from the book database DB. Alternatively, the AI ​​server 10 may extract only the book data with the highest similarity, or may extract a predetermined number (e.g., three) of book data in descending order of similarity.

[0048] The AI ​​server 10 generates input data for the LLM based on book data extracted from the book database DB. The input data is a prompt that includes a user query as an instruction to the LLM and embeds book data extracted from the book database DB as reference information to be referenced when generating a response sentence.

[0049] In the example of Figure 7, the instruction to the LLM (#question) includes the sentence, "What does it mean that the heart is the most important thing? Please answer by referring to the reference information." The first half is a sentence corresponding to the user's query, and the second half is a predetermined standard sentence. The reference information (#data) is embedded with a book ID, book data extracted from the book database DB, and the similarity between the query and the book data. In the example of Figure 7, "id" represents the book ID, "text" represents the book data extracted from the book database DB, and "score" represents the similarity between the query and the book data.

[0050] The AI ​​server 10 transmits a prompt generated based on the book data to the LLM server 20 and receives an answer sentence from the LLM in response. The AI ​​server 10 transmits the obtained answer sentence from the LLM to the user terminal 30. The user terminal 30 receives the answer sentence from the AI ​​server 10 and displays the received answer sentence on the display unit 35.

[0051] The operation of the AI ​​chat system will be described below. Figure 8 is a flowchart illustrating the processing steps executed by the AI ​​chat system. The user terminal 30 accesses the AI ​​server 10 via a dedicated app or a web browser (step S101). After user authentication, the user terminal 30 displays a user interface screen (UI screen) provided by the AI ​​server 10 on the display unit 35 (step S102).

[0052] FIG. 9 is a schematic diagram illustrating an example of a user interface screen provided by the AI ​​server 10. The user interface screen 100 illustrated in FIG. 9 includes a chat display field 101 that displays a conversation (chat) between a user and the system, a model selection field 102 that accepts an AI model selection, and a chat input field 103 that accepts a user chat. The chat display field 101 displays a user chat entered in the chat input field 103 via the operation unit 34 and a system chat generated by the AI ​​server 10. The model selection field 102 displays multiple AI model options in a pull-down menu format and accepts the user's selection of a desired AI model from the displayed options. In this embodiment, the AI ​​model (answer model) is configured to allow selection of at least one of an author model that simulates an answer by a specific author, an expert model that simulates an answer by a specific expert, and a book model that simulates an answer by a specific book. That is, in this embodiment, the AI ​​model (answer model) is a term that represents an option for selecting an author, an expert, and a book, and does not represent a machine learning model that outputs some output data when input data is input. Each of the author model, the expert model, and the book model includes a plurality of options. The model selection field 102 and the chat input field 103 accept selection operations and input operations via the operation unit 34 of the user terminal 30.

[0053] For example, when an AI model (one of the author models) in which the author is Mr. A is selected in the model selection field 102, the AI ​​server 10 generates system chat in which Mr. A responds to the user's chat input, and displays it in the chat display field 101. When an AI model (one of the expert models) in which a manager is selected in the model selection field 102, the AI ​​server 10 generates system chat in which the manager responds to the user's chat input, and displays it in the chat display field 101. When an AI model (one of the book models) in which the book is Book α is selected in the model selection field 102, the AI ​​server 10 generates system chat in which Book α (or the author of Book α) responds to the user's chat input, and displays it in the chat display field 101.

[0054] In this embodiment, multiple AI model options are displayed in the model selection field 102 and a user selection is accepted. However, an AI model may be automatically designated depending on the user. For example, AI models of the managers of various companies (users) may be prepared for each company, and the AI ​​server 10 may determine access from a company and automatically designate the AI ​​model of the manager of that company as the AI ​​model to be used by that company. In this case, the model selection field 102 displays only the model name of one AI model designated by the AI ​​server 10. Alternatively, the user terminal 30 may display a user interface screen 100 on the display unit 35 that does not include the model selection field 102.

[0055] In step S102 described above, the user terminal 30 displays a user interface screen 100 such as that shown in FIG. 9 on the display unit 35 and accepts the selection of an AI model from the model selection field 102 (step S103). The user terminal 30 notifies the AI ​​server 10 of the identifier of the selected AI model (step S104). Among the AI ​​models, each author model is linked to an author ID, each expert model is linked to a group ID, and each book model is linked to a book ID. Therefore, the user terminal 30 may notify the AI ​​server 10 of the author ID, group ID, or book ID as the identifier of the selected AI model. In this embodiment, the selection of an AI model is not essential and may be omitted.

[0056] The user terminal 30 accepts a user query through the chat input field 103 on the user interface screen 100 (step S105). The query is input as a sentence such as, for example, "What does it mean that the heart is the most important thing?" Before accepting the user query, the system may inquire about the user's attributes (such as name, gender, and age) and the user's current state (such as emotions and interests) through speech. Some of the user's attributes (such as name, gender, and age) may be registered in advance in the storage unit 12 of the AI ​​server 10. In the latter case, the user's attributes may be read from the storage unit 12 during login authentication to the AI ​​server 10.

[0057] The user terminal 30 transmits the query received in step S105 to the AI ​​server 10 (step S106). The query transmitted from the communication unit 33 of the user terminal 30 reaches the AI ​​server 10 via the communication network NW.

[0058] The AI ​​server 10 receives a query transmitted from the user terminal 30 (step S107). The control unit 11 of the AI ​​server 10 searches the book database DB based on the received query and extracts book data corresponding to the query from the book database DB (step S108). Specifically, the control unit 11 quantifies the query (or a key part of the query) using a predetermined method, calculates the similarity with book data registered in the book database DB, and extracts book data with a relatively high similarity from the book database DB. Furthermore, if text (numerical data) obtained from the author's social media or blog is registered in the book database DB, the control unit 11 may calculate the similarity between the calculated numerical value from the query and the numerical data of the text obtained from the social media or blog, and extract text from the social media or blog with a relatively high similarity from the book database DB.

[0059] Note that when an AI model is selected on the user terminal 30 and its identifiers are notified, the search range of the book database DB is limited by the AI ​​model's identifiers (author ID, group ID, book ID). For example, when an author model is selected, the author is identified by the author model's identifier (author ID), so the search range can be book data related to the author identified by the author ID. Also, when an expert model is specified, one or more authors are identified by the expert model's identifier (group ID), so the search range can be book data related to one or more authors identified by the group ID. Furthermore, when a book model is specified, the book is identified by the book model's identifier (book ID), so the search range can be book data of the book identified by the book ID.

[0060] The control unit 11 generates input data for the LLM based on the book data extracted from the book database DB (step S109). The control unit 11 generates a prompt as input data for the LLM, which includes a sentence in response to the user's query as an instruction to the LLM and includes the book data extracted from the book database DB as reference information. The reference information may include not only the book data with the highest similarity, but also multiple book data with relatively high similarity. The reference information may also include a book ID and information on the similarity between the query and the book data.

[0061] FIG. 10 is a conceptual diagram illustrating an example of a prompt generated by the AI ​​server 10. The prompt includes instructions to the LLM and reference information to be referenced by the LLM. In the prompt shown in FIG. 10, the instructions to the LLM are written in text in the "#Question" section, and the reference information to be referenced by the LLM is written in text in the "#Data" section. The instructions to the LLM include a user query and a standard phrase indicating that the LLM should answer by referring to the reference information. The reference information includes a book ID, book data extracted from the book database (DB), and the similarity between the query and the book data. In this embodiment, since the book data is numerical data, the numerical data is converted to the original text data (text) and then embedded in the reference information. For this reason, it is preferable that the book data registered in the book database (DB) be digitized using a reversible digitization method. Alternatively, a separate database may be provided that associates and stores the book data (numerical data) registered in the book database (DB) with the original text data, and the original text data may be converted by referring to the database. The prompt format shown in FIG. 10 is an example, and the actual format may be designed appropriately depending on the LLM.

[0062] In the example of FIG. 10 , book data extracted from the book database DB is embedded in the reference information of the prompt. However, the control unit 11 may also be configured to store a series of dialogues as a history in the storage unit 12 and add the dialogues to the reference information of the prompt. That is, the control unit 11 may register data of a user's query as book data in the book database DB, and later add the data to the prompt as reference information when generating an answer. The control unit 11 may also register data of a generated answer as book data in the book database DB, and later add the data to the prompt as reference information when generating an answer. Furthermore, the control unit 11 may store data of a user's query in association with a user identifier, and extract information regarding the user's emotions, behavior, or state based on the content of the stored query.

[0063] The AI ​​server 10 transmits the prompt generated in step S109 to the LLM server 20 (step S110). The prompt transmitted from the communication unit 13 of the AI ​​server 10 reaches the LLM server 20 via the communication network NW.

[0064] The LLM server 20 receives the prompt sent from the AI ​​server 10 (step S111). The LLM server 20 generates a response sentence to the user's query by referring to the received prompt (step S112). Specifically, the control unit 21 of the LLM server 20 references the reference information according to the instructions included in the prompt and generates a sentence that is highly compatible with the sentence (book data) included in the reference information using the LLM, thereby generating a response sentence to the user's query. The LLM server 20 returns the generated response sentence to the AI ​​server 10 in response to the prompt received in step S111 (step S113).

[0065] The AI ​​server 10 receives the response text sent back from the LLM server 20 (step S114) and outputs the received response text to the user terminal 30 (step S115).

[0066] The user terminal 30 displays the reply sentence output from the AI ​​server 10 on the display unit 35 (step S116). The user terminal 30 may display the reply sentence output from the AI ​​server 10 in the chat display field 101 of the user interface screen 100 described above. FIG. 11 is a schematic diagram showing an example of an output reply sentence to a user's query. The output example of FIG. 11 shows an example in which a reply sentence such as "The heart is what is important" is output by the AI ​​server 10 in response to a user's query of "What does the heart be important?" and displayed in the chat display field 101 of the user interface screen 100.

[0067] As described above, in the first embodiment, a response sentence to a user's query can be output using a language model that uses book data. Because books contain the thoughts and ideas of their authors, by using book data, a response that feels as if one is having a conversation with the author himself / herself can be obtained. In this embodiment, an author model is prepared for specifying an author, an expert model for specifying a specific expert, and a book model for specifying a specific book are also prepared. When the expert model is selected, a response that feels as if one is having a conversation with an expert is obtained, and when the book model is selected, a response that feels as if it were extracted from a book is obtained.

[0068] In this embodiment, a query is received as text data and a reply is output as text data, but a query may be received as voice data and a reply may be output as voice data.

[0069] In addition, in this embodiment, attributes may be assigned to the author model or the expert model. The attributes assigned to the author model or the expert model include the name, gender, age, job title, etc. of the respondent (author or expert). The attribute information is registered in, for example, a book database (DB). When attributes are assigned to the author model or the expert model, the AI ​​server 10 can generate a prompt including information on the respondent's attributes and instruct the LLM server 20 to generate an answer sentence according to the attributes.

[0070] In addition, in this embodiment, user attributes may be acquired. User attributes include the user's name, gender, age, mood, etc. Information on user attributes may be registered in advance for each user in the storage unit 12, or may be acquired by the AI ​​server 10 making an inquiry to the user terminal 30 after login authentication. When user attributes are assigned, the AI ​​server 10 can generate a prompt including information on the user's attributes and instruct the LLM server 20 to generate a response sentence according to the attributes.

[0071] FIG. 12 is a conceptual diagram illustrating an example of a prompt that takes into account the attributes of the questioner (user) and the answerer. FIG. 12 shows an example of a prompt when the answerer's attribute is a man in his 70s and the questioner (user)'s attribute is a man in his 30s. When the author model or expert model is selected by the user, the answerer's attribute is specified by reading the attributes associated with the selected model from the author database DB. The questioner's attribute is read from the storage unit 12 when the user logs in, or is specified by inquiring of the user after the user logs in. In the example of FIG. 12, by adding the instruction "Please answer as if a man in his 70s were speaking to a man in his 30s," an answer that corresponds to the attributes of the questioner and the answerer can be provided to the user.

[0072] In the example of FIG. 12, the attributes of both the questioner and the answerer are specified, but the attributes of either the questioner or the answerer may be specified.

[0073] In this embodiment, input data (prompts) to the LLM are generated based on book data registered in the book database DB. Alternatively, an AI model may be prepared for each author, expert, or book. In this case, the AI ​​model represents a language model (machine learning model) trained to output a sentence corresponding to the input sentence when a certain sentence is input. For example, an AI model for author A is generated by collecting sentence data from A's books and content published by A on the communication network NW, and learning using the collected sentence data as training data. The same applies to AI models specialized for other authors, experts, or books. Note that the learning algorithm used to generate the AI ​​model is not limited to a specific algorithm, and any appropriate learning algorithm may be used. Furthermore, the AI ​​model may be generated by fine-tuning an existing language model.

[0074] The AI ​​models thus generated, specialized for each author, expert, and book, are registered in the memory unit 12 of the AI ​​server 10 or in a server accessible from the AI ​​server 10. After the service starts, the AI ​​server 10 accepts queries from users and also accepts the selection of an AI model. The AI ​​server 10 inputs the accepted user query into the AI ​​model selected by the user and causes the AI ​​model to generate a response sentence corresponding to the user query. The AI ​​server 10 provides a chat service to users by outputting the response sentence generated by the AI ​​model to the user terminal 30.

[0075] The embodiments disclosed herein are illustrative in all respects and should not be considered limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims.

[0076] REFERENCE SIGNS LIST 10 AI server 11 Control unit 12 Memory unit 13 Communication unit 14 Operation unit 15 Display unit 20 LLM server 21 Control unit 22 Memory unit 23 Communication unit 24 Operation unit 25 Display unit 30 User terminal 31 Control unit 32 Memory unit 33 Communication unit 34 Operation unit 35 Display unit PG Chat generation program DB Book database

Claims

1. An information processing method in which a computer executes a process of acquiring a user's query, and outputting an answer to the acquired query using a language model that uses book data.

2. The information processing method according to claim 1, further comprising the steps of: preparing a plurality of answer models for generating said answer sentences; and associating book data with each of said plurality of answer models.

3. The information processing method according to claim 2, wherein the answer model includes at least one of an author model that simulates answers by a specific author, an expert model that simulates answers by a specific expert, and a book model that simulates answers by a specific book.

4. The information processing method of claim 2, wherein the computer executes the following process: referencing a database that stores an identifier for identifying each answer model in association with book data for one or more books, extracting from the database book data to be used for generating input data for the language model based on the identifier of the selected answer model and the acquired query, and generating the input data based on the extracted book data.

5. The information processing method according to claim 4, wherein the input data is a prompt that includes the query as an instruction to the language model and includes book data extracted from the database as reference information to be referred to when generating the answer sentence.

6. The information processing method according to claim 2, wherein attributes are assigned to each of the plurality of answer models, and the computer executes a process of inputting input data including attributes of a selected answer model into the language model to generate an answer sentence according to the attributes.

7. The information processing method according to claim 2, further comprising the steps of: acquiring attributes of the user; and inputting input data including the acquired user attributes into the language model, thereby generating a response sentence according to the attributes.

8. The information processing method according to claim 4, wherein the database stores data on a part or all of a sentence contained in content on a communication network, data on a summary of the sentence, or data on a translation of the sentence as book data, and the computer executes a process of generating input data for the language model based on the book data including data obtained from the content.

9. The information processing method according to claim 1, further comprising the step of: registering the acquired query data as the book data, the process being executed by the computer.

10. The information processing method according to claim 1, further comprising the step of: registering the answer text data as the book data, the process being executed by the computer.

11. The information processing method according to claim 1, wherein the computer executes the process of: storing the acquired query data in association with a user identifier; and extracting information relating to the user's emotions, behavior, or state based on the stored query data.

12. An information processing device comprising at least one processor, the processor acquiring a user's query, and outputting an answer to the acquired query using a language model that uses book data.

13. A computer program for causing a computer to execute a process of acquiring a user's query and outputting a response to the acquired query using a language model that uses book data.

Images