Response generation device, method, and program

Abstract

When the same question is input to multiple conversational AIs to engage in multi-turn dialogue, a natural dialogue is continued with each of the multiple conversational AIs. [Solution] The response generation device displays multiple interactive screens for each type of response generation module, each of which includes a question input by a user and answers to the question provided by multiple types of response generation modules using different large-scale language models.When a next question in response to a previous answer is input to a first interactive screen, the device updates the previous answer displayed in the second interactive screen with the previous answer in the first interactive screen, displays in the first interactive screen an answer to the next question provided by the response generation module corresponding to the first interactive screen, and displays in the second interactive screen an answer to the next question reflecting the updated previous answer provided by the response generation module corresponding to the second interactive screen.

Description

[Technical field]

[0001] The present disclosure relates to a response generation device, a response generation method, and a response generation program. [Background technology]

[0002] In recent years, high-performance large-scale language models (LLMs) have been released one after another, and various types of conversational AI (Artificial Intelligence) using LLMs are being offered. In addition, there are services that allow multiple conversational AIs to input the same text and output answers simultaneously.

[0003] As a technology related to multiple dialogue-type AIs, for example, a dialogue device has been proposed that has a task-specific dialogue control unit that controls task-specific dialogue based on a user's input sentence and past dialogue history to output a task-specific response, a chat-type dialogue control unit that controls chat-type dialogue based on the input sentence and dialogue history to output a chat-type response, and a response appropriateness evaluation unit that calculates a response appropriateness evaluation value that evaluates whether the chat-type response is an appropriate response based on the dialogue history and the purpose of the task, and decides whether to adopt a chat-type response or a task-specific response based on the response appropriateness evaluation value (see Patent Document 1).

[0004] Also, for example, a screen dialogue processing device has been proposed which performs processing by interacting with a screen and is characterized by having a dialogue screen generating means which generates multiple dialogue screens based on a single request (see Patent Document 2).

[0005] Also, for example, a method has been proposed that includes the steps of: determining a natural language text string associated with a request based on at least a request received at an endpoint of a system of two or more communicatively coupled computing devices; determining one or more guidance mechanisms associated with the request based at least on the natural language text string and a task associated with the endpoint; processing the natural language text string using a language model and at least based on the one or more guidance mechanisms to generate a text result corresponding to the request; and generating a response to the request using the endpoint, the response including at least the text result (see Patent Document 3). [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2020-118842 A [Patent Document 2] Japanese Patent Application Publication No. 11-3203 [Patent Document 3] JP 2024-43563 A Summary of the Invention [Problem to be solved by the invention]

[0007] Consider inputting the same question to multiple conversational AIs and having them output answers at the same time to conduct a multi-turn dialogue. In this case, each conversational AI may output a different answer to the same question, so if the same question is input to each conversational AI even after the second turn, the dialogue may not be established. For example, if a question is asked in the next turn about a topic that is included only in the answer of one conversational AI, the answer of the other conversational AI may be inconsistent. Even if the other conversational AI is able to output an answer that barely establishes a dialogue in the next turn, the possibility of the dialogue breaking down increases as the turns proceed.

[0008] The present disclosure has been made in consideration of the above points, and aims to provide a response generation device, method, and program that can continue a natural dialogue with each of multiple dialogue AIs when the same question is input to multiple dialogue AIs to engage in multi-turn dialogue. [Means for solving the problem]

[0009] A response generation device according to a first aspect of the present disclosure includes a processor that displays a plurality of interactive screens for each type of response generation module, each of which includes a question input by a user and an answer to the question, the answer being provided by a plurality of types of response generation modules each using a different large-scale language model, and when a next question in response to a previous answer is input to a first interactive screen, the processor updates the previous answer displayed in a second interactive screen with the previous answer in the first interactive screen, displays an answer to the next question provided by the response generation module corresponding to the first interactive screen in the first interactive screen, and displays an answer to the next question reflecting the updated previous answer provided by the response generation module corresponding to the second interactive screen in the second interactive screen.

[0010] A response generation method according to a second aspect of the present disclosure includes a computer that displays a plurality of interactive screens for each type of response generation module, each of which includes a question input by a user and answers to the question provided by a plurality of types of response generation modules each using a different large-scale language model, and when a next question in response to a previous answer is input to a first interactive screen, the computer updates the previous answer displayed in a second interactive screen with the previous answer in the first interactive screen, displays an answer to the next question provided by the response generation module corresponding to the first interactive screen in the first interactive screen, and displays an answer to the next question provided by the response generation module corresponding to the second interactive screen that reflects the updated previous answer in the second interactive screen.

[0011] A response generation device according to a third aspect of the present disclosure causes a computer to perform the following operations: displaying a plurality of interactive screens for each type of response generation module, each of which includes a question input by a user and answers to the question, the answers being provided by a plurality of types of response generation modules each using a different large-scale language model; when a next question in response to a previous answer is input to a first interactive screen, updating the previous answer displayed in a second interactive screen with the previous answer in the first interactive screen, displaying an answer to the next question provided by the response generation module corresponding to the first interactive screen in the first interactive screen, and displaying an answer to the next question provided by the response generation module corresponding to the second interactive screen that reflects the updated previous answer in the second interactive screen. Effect of the Invention

[0012] According to the response generation device, method, and program disclosed herein, when the same question is input to multiple dialogue AIs to engage in multi-turn dialogue, it is possible to continue a natural dialogue with each of the multiple dialogue AIs. [Brief description of the drawings]

[0013] [Figure 1]FIG. 2 is a block diagram showing a hardware configuration of the response generating device. [Diagram 2] 2 is a block diagram showing an example of a functional configuration of a response generating device. FIG. [Diagram 3] FIG. 13 is a diagram showing an example of a dialogue screen. [Figure 4] FIG. 13 is a diagram showing an example of a dialogue screen. [Diagram 5] FIG. 13 is a diagram showing an example of a dialogue screen. [Figure 6] FIG. 13 is a diagram showing an example of a dialogue screen. [Figure 7] FIG. 1 is a diagram illustrating an example of a method for managing a dialogue history using a tree structure. [Figure 8] 13 is a flowchart showing the flow of a response generation process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Hereinafter, an example of an embodiment of the present disclosure will be described with reference to the drawings. In each drawing, the same or equivalent components and parts are given the same reference numerals. In addition, the dimensions and ratios of the drawings are exaggerated for the convenience of explanation and may differ from the actual ratios.

[0015] Fig. 1 is a block diagram showing the hardware configuration of a response generation device 10 according to this embodiment. As shown in Fig. 1, the response generation device 10 has a CPU (Central Processing Unit) 12, a memory 14, a storage device 16, an input device 18, an output device 20, a storage medium reading device 22, and a communication I / F (Interface) 24. Each component is connected to each other via a bus 26 so as to be able to communicate with each other.

[0016] The storage device 16 stores a response generation program for executing a response generation process described below. The CPU 12 is a central processing unit, and executes various programs and controls each component. That is, the CPU 12 reads the program from the storage device 16 and executes the program using the memory 14 as a working area. The CPU 12 controls each component and performs various arithmetic processes according to the program stored in the storage device 16.

[0017] The memory 14 is made up of a RAM (Random Access Memory) and serves as a working area to temporarily store programs and data. The storage device 16 is made up of a ROM (Read Only Memory), a HDD (Hard Disk Drive), a SSD (Solid State Drive), etc., and stores various programs including an operating system and various data.

[0018] The input device 18 is a device for performing various inputs, such as a keyboard, a mouse, etc. The output device 20 is a device for outputting various information, such as a display, a printer, etc. A touch panel display may be used as the output device 20 to function as the input device 18.

[0019] The storage medium reader 22 reads data stored in various storage media such as CD (Compact Disc)-ROM, DVD (Digital Versatile Disc)-ROM, Blu-ray disc, USB (Universal Serial Bus) memory, etc., and writes data to the storage media. The communication I / F 24 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, or Wi-Fi (registered trademark).

[0020] Next, the functional configuration of the response generating device 10 according to this embodiment will be described.

[0021] Fig. 2 is a block diagram showing an example of a functional configuration of the response generation device 10. As shown in Fig. 2, the response generation device 10 includes, as its functional configuration, a reception unit 32, a display control unit 34, and a management unit 36. A response generation DB 40 and a dialogue history DB 42 are stored in a predetermined storage area of ​​the response generation device 10. Each functional configuration is realized by the CPU 12 reading out a response generation program stored in the storage device 16, expanding it in the memory 14, and executing it.

[0022] The response generation DB 40 stores a plurality of types of response generation modules using different LLMs. The response generation module is a computation unit of an interactive AI that realizes a multi-turn dialogue that outputs an answer to an input question and outputs an answer to the next question based on past questions and answers. Examples of the response generation module include ChatGPT by OpenAI, Claude3 by Anthropic, Llama3 and Fugaku by Meta, and OpenCalm by CyberAgent.

[0023] The reception unit 32 receives a question input by a user as text data. Although details will be described later, in this embodiment, a question is input to one of multiple dialogue AIs, and this question is simultaneously reflected in the other dialogue AIs. The reception unit 32 also receives information specifying which dialogue AI the input question was input to, for example, identification information of the dialogue AI. The question may be input by voice. In this case, the reception unit 32 converts the input voice into text data by voice recognition.

[0024] The reception unit 32 also receives various selection information input by the user. The selection information is, for example, the type of interactive AI to be used, history selection information (details will be described later), etc. The reception unit 32 passes the received information to the display control unit 34.

[0025] The display control unit 34 arranges multiple interactive screens for each type of interactive AI, each screen including a question input by a user and an answer to that question output by each interactive AI, and displays them on a display, which is an example of the output device 20.

[0026] FIG. 3 shows an example of the dialogue screen 50. Although FIG. 3 shows an example in which two dialogue AIs are used, three or more dialogue AIs may be used. As shown in FIG. 3, the dialogue screen 50 includes a chat UI (User Interface) 52A for a dialogue AI of type A (hereinafter also referred to as "chat A") and a chat UI 52B for a dialogue AI of type B (hereinafter also referred to as "chat B"). In the following description, when the types of dialogue AIs are not differentiated, the suffixes A and B are omitted.

[0027] The chat UI 52 includes a selection component 54 for selecting the type of dialogue AI. The selection component 54 may be, for example, a pull-down menu, and the options are identification information such as the name of the dialogue AI whose response generation module is stored in the response generation DB 40. The type of dialogue AI selected by the selection component 54 is received by the reception unit 32 as one of the above-mentioned selection information. The chat UI 52 also includes a dialogue display area 56 in which a question input by a user and an answer output by the dialogue AI, that is, text data of the dialogue, are displayed. The chat UI 52 also includes a question input area 58 in which a question from a user is input. The question input in the question input area 58 and the type of dialogue AI selected by the selection component 54 are received by the above-mentioned reception unit 32.

[0028] The display control unit 34 also reflects the question input to the first chat UI 52 in the second chat UI. Hereinafter, the chat UI 52 to which the question is input is also referred to as a "selected chat", and the other chat UIs 52 are also referred to as "non-selected chats". For example, it is assumed that a question associated with the identification information of chat A is transferred from the reception unit 32 to the display control unit 34. That is, as shown in FIG. 4, the question is input in the question input area 58A of the chat UI 52A. In this case, as shown in S of FIG. 4, the display control unit 34 copies the question input in the question input area 58A to the question input area 58B of the chat UI 52B.

[0029] The display control unit 34 inputs the question input in the question input area 58 as a prompt to the response generation module of the dialogue AI selected by the selection component 54, and obtains the response output from the response generation module. That is, responses to the same question are simultaneously generated by each of the dialogue AIs displaying the chat UI 52 on the dialogue screen 50. The display control unit 34 displays the input question and the responses obtained from each of the response generation modules in the dialogue display area 56 of the corresponding chat UI 52.

[0030] The example of FIG. 4 shows examples of responses from chat A and chat B to the question "Please suggest some recommended sightseeing spots for the weekend." In this example, assume that a question about "Hakone" that exists only in the response of chat A (for example, "Hakone sounds nice. What kind of places should I go to?") is input in the question input area 58A of the chat UI 52A as the next question in the multi-turn. In this case, a question about "Hakone" is also asked to chat B, which did not mention "Hakone" in its previous response, so the responses of chat B in the next and subsequent turns may be inconsistent.

[0031] Therefore, in this embodiment, when a next question to a previous answer is input to a selected chat, the display control unit 34 updates the display of the dialogue display area 56 by overriding (overwriting) the previous answer displayed in the non-selected chat with the previous answer in the selected chat. Then, the display control unit 34 copies the question input to the selected chat to the non-selected chat. As a result, when an answer to a newly input question reflecting the answers up to the previous time is output from the response generation module in each dialogue-based AI, an answer reflecting the updated answers up to the previous time is output in the non-selected chat.

[0032] For example, as shown in FIG. 5, assume that a question such as "Hakone sounds nice. Where should I go?" is input in the question input area 58A of the chat UI 52A. In this case, the display control unit 34 overrides the previous answer portion in the dialogue display area 56A with the previous answer portion in the dialogue display area 56B, as shown in T of FIG. 5. As a result, the previous answer in chat B also includes content related to "Hakone" (the dashed line portion in FIG. 5). Therefore, even if the question input in chat A is reflected in chat B as shown in S of FIG. 5, the question related to "Hakone" can also be naturally answered in chat B.

[0033] In addition, although details will be described later, in this embodiment, the history of questions and answers is managed in the dialogue history DB 42, and when an answer is updated, both the original answer and the updated answer are stored in the dialogue history DB 42. Therefore, as shown in FIG. 6, the display control unit 34 displays a selection component 60 that allows the selection of the original answer and the updated answer in association with each answer displayed in the dialogue display area 56. In the example of FIG. 6, the selection component 60 includes a notation "n / N" indicating that the nth answer is displayed among N answers stored as answers to the same question, and a button for increasing or decreasing n. Information on the nth answer selected by the selection component 60 is received by the above-mentioned receiving unit 32 as selection information of the history. The display control unit 34 deletes questions and answers after the selected nth answer from the dialogue display area 56, reads the nth answer from the dialogue history DB 42, and updates the display in the dialogue display area 56.

[0034] The management unit 36 ​​manages the history of questions and answers for each type of response generation module, i.e., for each type of interactive AI, in the dialogue history DB 42 as a tree-structured history in which the original answers output by the response generation module and the answers updated based on the selected chat are branched for the corresponding questions.

[0035] 7 shows an example of a method for managing dialogue history using a tree structure. Qi is a question input in the i-th turn, and Ai is the dialogue AI's answer to Qi. When an answer is updated, the management unit 36 ​​assigns a branch number 1 to the original answer, and assigns an incremented branch number to the updated answer each time the answer is updated.

[0036] The example of FIG. 7 is a tree structure showing the history of the dialogue of chat A in FIG. 4 to FIG. 6. Q1 is "Suggest some recommended sightseeing spots for the weekend", and A1 is an answer to Q1 that includes the content related to "Hakone". In response to this answer, the next question Q2 "Hakone sounds nice. Where should I go?" is input to chat A, so Q2 is stored in association with A1. Assume that A2 does not include the content related to "Hakone Shrine" as an answer to Q1, and the answer of chat B includes the content related to "Hakone Shrine". In response to this answer, if the next question Q3 "Oh, Hakone Shrine? That's nice. What kind of power spot is it?" is input in chat B, the answer A2 is overridden by the answer of chat B. Therefore, the management unit 36 ​​sets the original answer A2 as A2-1, and stores the updated answer A2-2 under the question Q2, branching off from A2-1. Then, the management unit 36 ​​stores the question Q3 under the answer A2-2 in association with it. In this case, a selection component 60 such as "← 2 / 2 →" is displayed in association with the answer A2-2 in the dialogue display area 56A. Note that "←" and "→" represent buttons in the example of FIG.

[0037] Next, the operation of the response generating device 10 according to this embodiment will be described.

[0038] Fig. 8 is a flowchart showing the flow of the response generation process executed by the CPU 12 of the response generation device 10. The CPU 12 reads the response generation program from the storage device 16, expands it in the memory 14, and executes it, whereby the CPU 12 functions as each functional component of the response generation device 10, and the response generation process shown in Fig. 8 is executed.

[0039] First, in step S10, the display control unit 34 displays the dialogue screen 50 in an initial state, for example as shown in Fig. 3. At this time, the reception unit 32 receives from the user the number of types of dialogue AI to be used, and the display control unit 34 displays the dialogue screen 50 including the chat UI 52 for the received number.

[0040] Next, in step S12, the reception unit 32 receives the type of interactive AI to be used through the selection component 54 of each chat UI 52 included in the dialogue screen 50. Next, in step S14, the reception unit 32 receives input of a first question into the question input area 58 in any of the chat UIs 52. Then, the reception unit 32 associates the question input into the question input area 58 with the type of interactive AI selected in the selection component 54 included in the same chat UI as the question input area 58 into which the question was input, and passes them to the display control unit 34.

[0041] Next, in step S16, the display control unit 34 copies the question input in the question input area 58 of the selected chat to the question input area 58 of the non-selected chat. Next, in step S18, the display control unit 34 inputs the question input in the question input area 58 as a prompt to the response generation module of the dialogue AI selected by the selection component 54, and obtains the answer output from the response generation module. That is, answers to the same question are simultaneously generated by each of the dialogue AIs displaying the chat UI 52 on the dialogue screen 50. Then, the display control unit 34 updates the dialogue screen 50 by displaying the input question and the answers obtained from each of the response generation modules in the dialogue display area 56 of the corresponding chat UI 52.

[0042] Next, in step S20, the management unit 36 ​​stores the history of questions and answers in a tree structure for each type of dialogue AI in the dialogue history DB 42. Next, in step S22, the reception unit 32 determines whether or not a question for the next turn has been received. If a question for the next turn has been received, the process proceeds to step S24, and if not, the process proceeds to step S28.

[0043] In step S24, the display control unit 34 updates the display in the dialogue display area 56 by overriding the previous reply in the selected chat with the previous reply displayed in the non-selected chat. Next, in step S26, the management unit 36 ​​adds the overridden reply, branching off from the original reply, below the previous question in the tree structure of the dialogue history, and returns to step S16.

[0044] On the other hand, in step S28, the reception unit 32 determines whether or not selection information from the history has been received by operating any of the selection components 60 in the chat UI 52. If the selection information from the history has been received, the process proceeds to step S30, and if the selection information from the history has not been received, the process proceeds to step S32. In step S30, the display control unit 34 deletes questions and answers following the answer indicated by the selection information from the history from the dialogue display area 56. Then, the display control unit 34 reads out the selected answer from the tree structure managed in the dialogue history DB 42 and displays it in the dialogue display area 56, thereby updating the dialogue screen 50, and the process returns to step S22.

[0045] In step S32, the reception unit 32 judges whether or not to end the ongoing multi-turn dialogue. For example, when an instruction to end the dialogue screen 50 is given, when a different theme from the ongoing dialogue is selected, etc., it is judged that the dialogue is to be ended. If the dialogue is not to be ended, the process returns to step S22, and if the dialogue is to be ended, the response generation process is ended.

[0046] As described above, the response generation device according to the present embodiment displays a plurality of chat UIs for each type of dialogue AI, including a question input by a user and an answer to the question, which is an answer by a plurality of types of dialogue AIs using different LLMs. In addition, when a next question for a previous answer is input to a first chat UI (selected chat), the response generation device updates the previous answer displayed in a second chat UI (non-selected chat) with the previous answer in the selected chat. Then, the response generation device displays an answer to the next question by a dialogue AI corresponding to the selected chat in the selected chat, and displays an answer to the next question by a dialogue AI corresponding to a non-selected chat that reflects the updated previous answer in the non-selected chat. This allows a natural dialogue to be continued with each of the multiple dialogue AIs when the same question is input to multiple dialogue AIs to perform a multi-turn dialogue.

[0047] In the above embodiment, the case where two types of dialogue AIs are used has been described, but as described above, three or more types of dialogue AIs may be used. In this case, the chat UI into which the question is input is the selected chat, and the other two or more types of chat UIs are non-selected chats.

[0048] In addition, the type of dialogue AI to be used may be added, deleted, or changed during a multi-turn dialogue. In this case, for the dialogue AI to be newly used due to the addition or change, the chat UI 52 may be displayed in a state that reflects the questions and dialogue so far by referring to the dialogue history.

[0049] In addition, the response generation process executed by the CPU by reading the software (program) in the above embodiment may be executed by various processors other than the CPU. In this case, examples of the processor include a PLD (Programmable Logic Device) such as an FPGA (Field-Programmable Gate Array) whose circuit configuration can be changed after manufacture, and a dedicated electric circuit such as an ASIC (Application Specific Integrated Circuit) which is a processor having a circuit configuration designed exclusively for executing a specific process. The response generation process may be executed by one of these various processors, or may be executed by a combination of two or more processors of the same or different types (for example, a plurality of FPGAs, a combination of a CPU and an FPGA, etc.). A GPU (Graphics Processing Unit) may be used as the processor. More specifically, the hardware structure of these various processors is an electric circuit that combines circuit elements such as semiconductor elements.

[0050] In the above embodiment, the response generation program is pre-stored (installed) in the storage device, but the present invention is not limited to this. The program may be provided in a form recorded on a recording medium such as a CD-ROM, a DVD-ROM (Digital Versatile Disc Read Only Memory), or a USB (Universal Serial Bus) memory. The program may also be downloaded from an external device via a network. [Explanation of symbols]

[0051] 10 Response Generation Device 12 CPU 14. Memory 16 Storage device 18 Input Devices 20 Output Devices 22 Storage media reader 24 Communication I / F 26 Bus 32 Reception 34 Display control section 36 Management Department 40 Response Generation DB 42 Dialogue History DB 50 Dialogue Screen 52 Chat UI 54 Selected Parts 56 Interactive display area 58 Question input area 60 Selected Parts

Claims

1. A processor is included. The processor, displaying, side by side, a plurality of dialogue screens for each type of response generation module, the dialogue screens including a question input by a user and answers to the question, the answers being provided by a plurality of types of response generation modules each using a different large-scale language model; when a next question for a previous answer is input to the first interactive screen, the previous answer displayed on the second interactive screen is updated with the previous answer on the first interactive screen; displaying, on the first interactive screen, an answer to the next question generated by a response generation module corresponding to the first interactive screen, and displaying, on the second interactive screen, an answer to the next question that reflects the updated previous answer generated by a response generation module corresponding to the second interactive screen. Response generation device.

2. 2. The response generation device according to claim 1, wherein the history of questions and answers for each type of response generation module is managed as a tree-structured history in which the original answer output by the response generation module and the answer updated based on the first interactive screen are branched for the corresponding question.

3. For each answer displayed on the interactive screen, the original answer and the updated answer can be selected; updating the currently displayed answer to the selected answer based on the history of the tree structure; 3. A response generating device according to claim 2.

4. The computer displaying, side by side, a plurality of dialogue screens for each type of response generation module, the dialogue screens including a question input by a user and answers to the question, the answers being provided by a plurality of types of response generation modules each using a different large-scale language model; when a next question for a previous answer is input to the first interactive screen, the previous answer displayed on the second interactive screen is updated with the previous answer on the first interactive screen; displaying, on the first interactive screen, an answer to the next question generated by a response generation module corresponding to the first interactive screen, and displaying, on the second interactive screen, an answer to the next question that reflects the updated previous answer generated by a response generation module corresponding to the second interactive screen. The response generation method.

5. On the computer, displaying, side by side, a plurality of dialogue screens for each type of response generation module, the dialogue screens including a question input by a user and answers to the question, the answers being provided by a plurality of types of response generation modules each using a different large-scale language model; when a next question for a previous answer is input to the first interactive screen, the previous answer displayed on the second interactive screen is updated with the previous answer on the first interactive screen; displaying, on the first interactive screen, an answer to the next question generated by a response generation module corresponding to the first interactive screen, and displaying, on the second interactive screen, an answer to the next question that reflects the updated previous answer generated by a response generation module corresponding to the second interactive screen. A response generator to make that happen.

Images