Information processing device, non-transitory storage medium in which information processing program is stored, information processing method, and information processing system

Abstract

[Problem] To provide an information processing device that can carry out interaction by using information obtained by compressing interaction history. [Solution] An information processing device according to the present invention is connected to: an external device that uses a first graph document in which a document structure is graphed; and a large-scale language model that generates the first graph document. The information processing device is characterized by having a control means that creates a first prompt on the basis of information including the first graph document, transmits the first prompt to the large-scale language model, acquires, from the large-scale language model, the first graph document that was updated on the basis of the first prompt, and transmits the acquired first graph document to the external device.

Description

Information processing device, non-transitory storage medium storing information processing program, information processing method, and information processing system

[0001] The present invention relates to an information processing device, a non-transitory storage medium storing an information processing program, an information processing method, and an information processing system, and can be applied to, for example, an interactive system using a graph document.

[0002] In recent years, with the development of language models, the performance of dialogue systems has improved, and various techniques have been developed.

[0003] For example, Patent Document 1 discloses a device that uses a language model to interactively search for documents to be searched for, and Patent Document 2 discloses a device that interactively issues instructions to edit a document and modifies the document.

[0004] Incidentally, Non-Patent Document 1, a public document from the Industrial Japanese Language Research Association, states that there are various advantages to using graph documents instead of text documents. A graph document represents a document structure as a graph, where a sentence of about one sentence size is represented as one node, relationships between sentences are connected by edges, and labels that indicate the meaning of the relationships are assigned to the edges, thereby making it possible to represent the structure and meaning of a document.

[0005] JP 2023-076413 A JP 2023-017938 A

[0006] "Industrial Japanese Language Research Association, Document Creation Support Subcommittee Public Materials" [Retrieved April 17, 2024], [Online], INTERNET,<URL:https: / / tech-jpn.jp / wp-content / uploads / symposium / 13 / 13-2-2.pdf>

[0007] By using a language model such as that described in the above-mentioned Patent Document 1, a dialogue system can create a response sentence that is appropriate to a certain extent in response to an utterance from a user.

[0008] However, because there is a limit to the length of prompts used to create response sentences to be sent to a language model, if the dialogue history to be sent becomes long, it is necessary to delete old dialogue history before sending it. In this case, the content of the deleted dialogue history is not passed to the language model, which results in a dialogue that does not take into account the information in the deleted dialogue history.

[0009] Therefore, there is a demand for an information processing device, a non-transitory storage medium storing an information processing program, an information processing method, and an information processing system that are capable of conducting interactions using graph documents.

[0010] The first invention is an information processing device comprising a display control unit that displays the content of a dialogue with a large-scale language model on a display unit as a graph document that graphs a document structure. The second invention is an information processing device connected to an external device that uses a first graph document that graphs a document structure and a large-scale language model that generates the first graph document, comprising control means that creates a first prompt based on information including the first graph document, transmits the first prompt to the large-scale language model, obtains the first graph document updated based on the first prompt from the large-scale language model, and transmits the obtained first graph document to the external device.

[0011] A third non-transitory storage medium storing an information processing program of the present invention is characterized in that it stores a program that causes a computer mounted on an information processing device connected to an external device that uses a first graph document that graphs a document structure and a large-scale language model that generates the first graph document to function as a control means that creates a first prompt based on information including the first graph document, sends the first prompt to the large-scale language model, obtains the first graph document updated based on the first prompt from the large-scale language model, and sends the obtained first graph document to the external device.

[0012] A fourth aspect of the present invention is an information processing method used in an information processing device connected to an external device that uses a first graph document in which a document structure is graphed, and a large-scale language model that generates the first graph document, wherein a control means creates a first prompt based on information including the first graph document, sends the first prompt to the large-scale language model, obtains the first graph document updated based on the first prompt from the large-scale language model, and sends the obtained first graph document to the external device.

[0013] The fifth aspect of the present invention is an information processing system having one or more external devices that use a first graph document in which a document structure is graphed, a large-scale language model that generates the first graph document, and an information processing device, characterized in that the information processing device of the second aspect of the present invention is applied to the information processing device.

[0014] According to the present invention, interactions can be carried out using graph documents.

[0015] FIG. 1 is a block diagram showing a configuration of a graph document dialogue system according to a first embodiment. FIG. 2 is an explanatory diagram showing an example of the screen configuration of a graph document dialogue input / output unit according to the first embodiment. FIG. 3 is an explanatory diagram showing an example of the configuration of a prompt template according to the first embodiment. FIG. 4 is an explanatory diagram showing an example of a dialogue history according to the first embodiment. FIG. 5 is an explanatory diagram showing an example of a graph document according to the first embodiment. FIG. 6 is a flowchart showing a characteristic operation of a graph document dialogue device (dialogue control unit) according to the first embodiment. FIG. 7 is a flowchart showing a characteristic operation of a graph document dialogue input / output unit according to the first embodiment. FIG. 8 is a block diagram showing a configuration of a graph document dialogue system according to a second embodiment. FIG. 9 is an explanatory diagram showing an example of the configuration of a prompt template for search according to the second embodiment. FIG. 10 is an explanatory diagram showing an example of the configuration of a prompt template for dialogue according to the second embodiment. FIG. 11 is a flowchart showing a characteristic operation of a graph document dialogue device (dialogue control unit) according to the second embodiment. FIG. 12 is a flowchart showing a characteristic operation of a graph document storage and search unit according to the second embodiment. FIG. 13 is an explanatory diagram (part 1) showing an example of a search result searched by the graph document storage and search unit according to the second embodiment. FIG. 14 is an explanatory diagram (part 2) showing an example of a search result searched by the graph document storage and search unit according to the second embodiment. FIG. 15 is a block diagram showing a configuration of a graph document dialogue system according to a third embodiment. FIG. 16 is a flowchart showing a characteristic operation of a graph document dialogue input / output unit according to the third embodiment.

[0016] (A) First Embodiment Hereinafter, a first embodiment of an information processing device, a non-transitory storage medium storing an information processing program, an information processing method, and an information processing system according to the present invention will be described in detail with reference to the drawings. In the following, an example in which the information processing system of the present invention is applied to a graph document processing system will be shown.

[0017] (A-1) Configuration of the First Embodiment FIG. 1 is a block diagram showing the configuration of a graph document dialogue system according to the first embodiment.

[0018] In FIG. 1, a graph document dialogue system 1 includes a graph document dialogue device 10, a graph document dialogue input / output unit 20 as an external device, and a large-scale language model 30.

[0019] In addition, in the graph document dialogue system 1, the graph document dialogue device 10, the graph document dialogue input / output unit 20, and the large-scale language model 30 can be connected via a network using various communication methods, whether wired or wireless. As a variant, the graph document dialogue device 10 may store the large-scale language model 30 internally.

[0020] The graph document dialogue input / output unit 20 is a functional unit that can input user utterances, display user utterances and system utterances, and display and edit graph documents. The graph document dialogue input / output unit 20 can be realized, for example, by a PC, a tablet, a smartphone, etc.

[0021] FIG. 2 is an explanatory diagram showing an example of the screen configuration of the graph document dialogue input / output unit according to the first embodiment.

[0022] In FIG. 2, a graph document dialogue input / output screen 200 has a graph document display / edit field 201 and a dialogue history display / input field 202 .

[0023] The graph document display / edit field 201 allows the display and editing of the graph document. It is assumed that the graph document can be edited by adding nodes and edges on the screen, but it is also possible to only display the graph document without editing it. The edited graph document can be sent to the graph document interaction device 10 by, for example, pressing the send button 205, or it can be sent to the graph document interaction device 10 at any time after editing is complete, but the trigger for sending the graph document is not particularly limited.

[0024] The dialogue display field 203 in the dialogue history display / input field 202 displays the dialogue between the user and the system. The dialogue display format is not particularly limited, but for example, in Fig. 2, a speech bubble from the right indicates a sentence spoken by the user, and a speech bubble from the left indicates a sentence spoken by the system. It is assumed that the dialogue can be input by the user entering any text in the text input field 204 at the bottom and pressing the send button 205.

[0025] The graph document interaction device 10 includes an interaction control unit 11 and a storage unit 12 .

[0026] The graph document interaction device according to the embodiment may be configured as hardware, such as a dedicated IC chip equipped with the components shown in Fig. 1, or may be configured as software, centered around a CPU and a program executed by the CPU, and functionally can be represented as shown in Fig. 1. Also, a non-transitory storage medium storing the program may be provided.

[0027] The dialogue control unit 11 is a functional unit that creates a response sentence (system utterance sentence) in cooperation with the large-scale language model 30 described later in response to a user utterance input via the graph document dialogue input / output unit 20, and controls the dialogue with the user.

[0028] The storage unit 12 is a functional unit that stores data used by the dialogue control unit 11 , and includes a prompt template 13 , a dialogue history 14 , and a current graph document 15 .

[0029] The prompt template 13 is template data for constructing an instruction sentence (prompt) for the large-scale language model 30 to generate a system utterance.

[0030] FIG. 3 is an explanatory diagram illustrating an example of the configuration of a prompt template according to the first embodiment.

[0031] In FIG. 3, the prompt template 13 includes an overall prompt 13-1, a variable portion for each domain 13-2, a detailed prompt 13-3, a current graph document 13-4, and a dialogue history 13-5.

[0032] The general prompt 13-1 and the detailed prompt 13-3 are fixed and preset independently of the domain.

[0033] The general prompt 13-1 contains instructions to have a dialogue while creating a graph document, such as "Please have a dialogue with the user while summarizing the contents of the dialogue with the user in a graph document."

[0034] Detailed prompt 13-3 describes the specific format of the graph document and notes to be taken into consideration. For example, the instruction might be something like, "A graph document is a set of nodes and labeled directed edges. Select edge labels from the following: 'Result,' 'Cause,' 'Background,' 'Example,' 'Conclusion,' 'Constraint,' 'Purpose,' and 'Adversity.' The node content should be a long sentence."

[0035] The domain-specific variable part 13-2 has instructions that change depending on the domain and must be determined before execution. For example, the instructions might be something like, "You (large-scale language model 30) are a paper writer. Based on the information about the paper the user wants to write, please elicit from the user the information necessary for writing the paper. If necessary, please also provide appropriate information."

[0036] The current graph document 13-4 and the dialogue history 13-5 are filled with the values ​​of the current graph document 15 and the dialogue history 14 by the dialogue control unit 11 at the time of execution.

[0037] FIG. 4 is an explanatory diagram showing an example of a dialogue history according to the first embodiment. The dialogue history 14 shown in FIG. 4 shows the content displayed in the dialogue display field 203 of FIG. 2 in a predetermined data format. Note that while FIG. 4 shows the dialogue history 14 in JSON format, the data format used is not particularly limited. As the dialogue between the user and the system progresses, the dialogue history 14 becomes longer as dialogue content (user utterances and system utterances) is embedded. In other words, the dialogue history 14 stored in the storage unit 12 is updated as needed by instructions from the dialogue control unit 11.

[0038] Fig. 5 is an explanatory diagram showing an example of a current graph document according to the first embodiment. The current graph document 15 shown in Fig. 5 shows the content displayed in the graph document display / edit field 201 in Fig. 2 in a predetermined data format (JSON format as above).

[0039] As shown in Fig. 5, the node contents and edge contents are listed in "nodes" and "edges," respectively. The format does not have to be exactly like this example, as long as similar contents are listed.

[0040] The large-scale language model 30 is assumed to be a generally distributed language model or a language model that can be used via an API (Application Programming Interface), such as generative AI such as ChatGPT. When a prompt is input, the large-scale language model 30 generates and outputs text data corresponding to the prompt (a system utterance in response to a user utterance, or a graph document in which a dialogue history is structured). Note that, although the first embodiment illustrates an example in which the large-scale language model 30 handles only text data, as a modified example, the large-scale language model 30 may handle information other than text, such as images and sounds.

[0041] (A-2) Operation of the First Embodiment Next, the operation of the graph document dialogue system 1 according to the first embodiment having the above-described configuration will be described.

[0042] (A-2-1) Processing of the Graph Document Interactive Device 10 FIG. 6 is a flowchart showing the characteristic operations of the graph document interactive device (interactive control unit) according to the first embodiment.

[0043] <S101> The dialogue control unit 11 acquires a user utterance and a graph document from the graph document dialogue input / output unit 20. Initially, the user utterance sentence and the graph document may be empty. If the user utterance sentence is empty, the system starts with an utterance.

[0044] <S102, S103> The dialogue control unit 11 adds the acquired user utterance to the end of the dialogue history 14. Similarly, the dialogue control unit 11 stores the acquired graph document in the current graph document 15.

[0045] <S104> The dialogue control unit 11 creates a prompt by embedding the current graph document 15 in the place of the current graph document 13-4 in the prompt template 13 and embedding the dialogue history 14 in the place of the dialogue history 13-5.

[0046] <S105> The dialogue control unit 11 transmits the prompt created in step S104 to the large-scale language model 30. The large-scale language model 30 generates a graph document and a system utterance based on the prompt.

[0047] <S106> The dialogue control unit 11 acquires the graph document and the system utterance sentence from the large-scale language model 30.

[0048] <S107, S108> The dialogue control unit 11 stores the graph document acquired in step S106 in the current graph document 15. Similarly, the dialogue control unit 11 adds the acquired system utterance to the end of the dialogue history 14.

[0049] <S109> Then, the dialogue control unit 11 transmits the current graph document 15 and the dialogue history 14 to the graph document dialogue input / output unit 20.

[0050] After the process of step S109, the process returns to step S101 and repeats the series of processes. After that, the process may be ended at any point due to the system being shut down, etc.

[0051] (A-2-2) Processing of the Graph Document Dialogue Input / Output Unit 20 FIG. 7 is a flowchart showing the characteristic operation of the graph document dialogue input / output unit according to the first embodiment.

[0052] <S201> The graph document dialogue input / output unit 20 creates a graph document display / edit column 201 on the left side of the screen and a dialogue history display / input column 202 on the right side, and displays information, as shown in the example screen of Fig. 2. Initially, both the graph document and the dialogue history are empty.

[0053] <S202> When the user presses the send button 205, the graph document dialogue input / output unit 20 transmits the user utterance (the text entered in the text input field 204) and the graph document to the dialogue control unit 11. In addition, the graph document dialogue input / output unit 20 empties the text input field 204 at the same time as transmitting the data.

[0054] <S203> After that, the graph document dialogue input / output unit 20 acquires the dialogue history and the graph document from the dialogue control unit 11.

[0055] <S204> The graph document dialogue input / output unit 20 reflects the acquired dialogue history and graph document in the graph document display / edit field 201 and the dialogue history display / input field 202, and makes it possible to input a user utterance.

[0056] After the process of step S204, the process returns to step S202 and repeats the series of processes. After that, the process may be ended at any point by shutting down the system, etc.

[0057] (A-3) Effects of the First Embodiment The graph document dialogue system 1 allows dialogues to be conducted using the large-scale language model 30 while displaying and storing dialogue content in the form of a graph document. In other words, at any point in time, the content of the dialogue up to that point is displayed and stored in the form of a graph document, allowing the user to check the outcome of the dialogue. Furthermore, by specifying any node displayed in the graph document and engaging in dialogue with that node, a dialogue can be conducted for that node. In this case, when sending a user utterance to the large-scale language model 30, the ID (or IDs) of the specified node are also assigned, thereby informing the large-scale language model 30 of the node to be dialogued and clearly indicating that the dialogue is for that node, and appropriate system utterances and graph documents can be obtained.

[0058] For example, suppose a user is having a conversation about a new idea and comes up with a technology necessary to realize that idea. The user can specify the node that describes the technology and have the large-scale language model 30 provide other ideas for that node. In this case, the large-scale language model 30 will provide multiple new ideas based on the other nodes connected to the specified node and their relationships, giving the user the effect of brainstorming alone. Similarly, by having the large-scale language model 30 provide the next step for the specified node, the user can achieve the effect of having an expert ask the appropriate probing questions.

[0059] Furthermore, since there is generally a limit to the length of prompts sent to the large-scale language model 30, if the dialogue history to be sent becomes long, it is necessary to delete old dialogue history before sending it, in which case the content of the deleted dialogue history is not passed to the large-scale language model 30, resulting in a dialogue that does not take into account the information in the deleted dialogue history.In this embodiment, the content of the dialogue is compactly represented in a graph document, so the dialogue takes into account the information in the deleted dialogue history, making it possible to obtain more accurate system utterances.

[0060] In this case, as described in Non-Patent Document 1, a graph document can express the content of a dialogue in a structured form, and is also very well organized. Therefore, the large-scale language model 30 refers not only to the dialogue history but also to the graph document in which the dialogue content is properly organized, and the system makes utterances that are in line with the graph document, thereby achieving the effect of having a more accurate dialogue that is more focused on the dialogue content.

[0061] Furthermore, the graphical document containing the dialogue content can be converted as needed into a form of expression required by the user, such as presentation materials, documents, specifications, proposals, patent documents, papers, procedure manuals, programs, etc. This allows these to be created by utilizing the knowledge held by the large-scale language model 30 in the dialogue.

[0062] (B) Second Embodiment Hereinafter, a second embodiment of an information processing device, a non-transitory storage medium storing an information processing program, an information processing method, and an information processing system according to the present invention will be described in detail with reference to the drawings. In the following, an example in which the information processing system of the present invention is applied to a graph document processing system will be shown.

[0063] (B-1) Configuration of the Second Embodiment FIG. 8 is a block diagram showing the configuration of a graph document dialogue system according to the second embodiment.

[0064] 8, the graph document dialogue system 1A includes a graph document dialogue device 10A, the above-mentioned graph document dialogue input / output unit 20, and the above-mentioned large-scale language model 30. The following description of the graph document dialogue device 10A will focus on the differences from the first embodiment.

[0065] The graph document interaction device 10A includes a graph document storage and search unit 41 in addition to the configuration of the interaction control unit 11 and memory unit 12 described above. The memory unit 12 also stores a graph document group 42 and a search prompt template 43 in addition to a prompt template 13A for interaction, a dialogue history 14, and a current graph document 15.

[0066] The dialogue control unit 11 uses the large-scale language model 30 to create a search query statement for searching for a reference graph document from the graph document group 42 managed by the graph document storage and search unit 41 based on the dialogue content.

[0067] FIG. 9 is an explanatory diagram showing an example of the configuration of a prompt template for search according to the second embodiment.

[0068] In FIG. 9, a prompt template 43 for search has an overall prompt 43-1, a variable part for each domain 43-2, a current graph document 43-3, and a dialogue history 43-4.

[0069] The overall prompt 43-1 contains instructions to create a search query statement necessary to search for a reference graph document from the dialogue content. The domain-specific variable part 43-2 contains instructions such as the field and purpose of the dialogue, the contents of which change depending on the domain. The current graph document 43-3 and dialogue history 43-4 are filled with the values ​​of the current graph document 15 and dialogue history 14 by the dialogue control unit 11 at runtime.

[0070] FIG. 10 is an explanatory diagram showing an example of the configuration of a prompt template for a dialogue according to the second embodiment.

[0071] 10, the prompt template 13A has a search result 13-6 in addition to the components of the above-described prompt template 13. The search result 13-6 is filled with the search results obtained by the graph document storage and search unit 41, which will be described later.

[0072] The graph document group 42 is a plurality of graph documents managed (stored) by the graph document storage and search unit 41 .

[0073] The graph document storage and search unit 41 vectorizes the contents of the graph fragments (such as one node, one node and one edge connected to that node, or two connected nodes and one edge) of each graph document in the graph document group 42 it holds, making them searchable.

[0074] The graph document storage and search unit 41 performs a search in two stages. First, it creates a list of graph documents or graph document fragments with content similar to the search query statement (see FIG. 12, which will be described later). Specifically, the graph document storage and search unit 41 uses a vector representation of a character string to perform a vector search using a technique such as approximate nearest neighbor search. In addition to converting the text representation of an entire graph document into a vector representation, it also stores vectors of graph document fragments (such as the content of one node only, the label of the edge connecting the node content, or the content of two connecting nodes and the label of their edges) so that searches can be performed using graph document fragments. This results in the vector search results shown in FIG. 13.

[0075] Furthermore, based on the vector search results, the graph document storage and search unit 41 also includes in the final search results the nodes connected by specific labels from the nodes contained therein. For example, a final search result such as that shown in FIG. 14 is created, including the nodes connected by labels such as "reference," "overview," and "function." By using this as the final search result, not only the vector search results but also information about the nodes, such as their reference destinations, can be included in the prompt, thereby improving the results returned by the large-scale language model 30. Generally, the large-scale language model 30 tends to generate incorrect content when the information necessary for answering is not included in the prompt. Therefore, by providing sufficient necessary information in the prompt in this way, the results returned by the large-scale language model 30 can be made more accurate.

[0076] (B-2) Operation of the Second Embodiment Next, the operation of the graph document dialogue system 1A according to the second embodiment having the above-described configuration will be described.

[0077] (B-2-1) Processing of the Graph Document Interactive Device 10A Figure 11 is a flowchart showing the characteristic operations of the graph document interactive device (interactive control unit) according to the second embodiment. Note that the processes of steps S101 to S103 and steps S105 to S109 in Figure 11 are the same as those explained in Figure 6 above, and therefore detailed explanations will be omitted. The following explanation will focus on the processes of steps S301 to S307, which are unique to the second embodiment.

[0078] <S301> After step S103 described above, the dialogue control unit 11 creates a search prompt by embedding the current graph document 15 in the current graph document 43-3 position of the search prompt template 43 shown in FIG. 9 and embedding the dialogue history 14 in the dialogue history 43-4 position.

[0079] <S302> The dialogue control unit 11 transmits the search prompt created in step S301 to the large-scale language model 30. The large-scale language model 30 creates a search query sentence based on the search prompt.

[0080] <S303> The dialogue control unit 11 acquires a search query sentence from the large-scale language model 30.

[0081] <S304> The dialogue control unit 11 sends the acquired search query statement to the graph document storage and search unit 41, causing it to search the graph document group 42.

[0082] <S305> The dialogue control unit 11 acquires the search results shown in FIG.

[0083] <S306> The dialogue control unit 11 creates a dialogue prompt by embedding the current graph document 15 in the current graph document 13-4 position of the dialogue prompt template 13A shown in FIG. 10, embedding the dialogue history 14 in the dialogue history 13-5 position, and embedding the search results obtained in step S305 above in the search result 13-6 position.

[0084] <S307> After the above steps S105 to S108, the dialogue control unit 11 transmits the current graph document 15 to the graph document storage and search unit 41, and stores it in the graph document group 42 in the storage unit 12.

[0085] The subsequent processing is the same as that in the first embodiment, and therefore a description thereof will be omitted.Similarly, the processing of the graph document dialogue input / output unit 20 is the same as that in the first embodiment, and therefore a description thereof will be omitted.

[0086] (B-2-2) Processing of the Graph Document Storage and Search Unit 41 FIG. 12 is a flowchart showing the characteristic operation of the graph document storage and search unit according to the second embodiment.

[0087] <S401> The graph document storage and search unit 41 acquires, via the dialogue control unit 11, the search query statement generated by the large-scale language model 30 in step S302 described above.

[0088] <S402> The graph document storage and search unit 41 converts the search query statement into a vector, searches for graph fragments having vectors close to the vector of the search query statement from the graph document group 42 in the storage unit 12, and creates a list of vector search results as shown in FIG. 13.

[0089] <S403> The graph document storage and search unit 41 adds to the list the graph fragments connected to the nodes included in the list created in step S103 above by specific labels that should be included in the search results, such as “reference,” “overview,” and “function,” to create a final list of search results as shown in FIG. 14, and sends this list to the dialogue control unit 11.

[0090] (B-3) Advantages of the Second Embodiment In addition to the advantages of the first embodiment, the second embodiment provides the following advantages.

[0091] In the graphical document dialogue system 1A, dialogue can be carried out by appropriately using information on other accumulated graphical documents.

[0092] For example, if the accumulated graph documents include terms and conditions for a certain service or organizational rules, a user can have a dialogue about those terms and conditions. The large-scale language model 30 can generate an answer using graph fragments with content similar to the content of the question posed by the user. If the terms and conditions, rules, etc. are stored as graph nodes, a vector search can be used to search for terms with content similar to the question. Furthermore, by including nodes linked with labels such as "reference," "overview," and "function" in the search results, the information necessary for the large-scale language model 30 to generate text can be included in the prompt, thereby preventing the large-scale language model from outputting plausible lies (hallucinations).

[0093] For example, if you are having a conversation to deepen your ideas, you can find out whether anyone else has a similar idea, whether there are other ideas based on similar technology, or whether there are attempts to apply the same technology in a different field.

[0094] (C) Third Embodiment Hereinafter, a third embodiment of an information processing device, a non-transitory storage medium storing an information processing program, an information processing method, and an information processing system according to the present invention will be described in detail with reference to the drawings. Hereinafter, an example in which the information processing system of the present invention is applied to a graph document processing system will be shown.

[0095] (C-1) Configuration of the Third Embodiment FIG. 15 is a block diagram showing the configuration of a graph document dialogue system according to the third embodiment.

[0096] 15, the graph document dialogue system 1B includes the above-mentioned graph document dialogue device 10A, a plurality of graph document dialogue input / output units 20A (20A-1 to 20A-n), and the above-mentioned large-scale language model 30. The following description will focus on the differences between the graph document dialogue input / output unit 20A and the first and second embodiments.

[0097] The third embodiment differs from the first and second embodiments in that it includes a plurality of graph document dialogue input / output units 20 A. Each graph document dialogue input / output unit 20 A is connected to the graph document dialogue device 10 A (dialogue control unit 11).

[0098] Since each graph document interactive input / output unit 20A can operate simultaneously, multiple users using these units can share a single graph document and interact with it, and multiple users can cooperate to interact and create graph documents.

[0099] (C-2) Operation of the Third Embodiment Next, we will explain the operation of the graph document dialogue system 1B according to the third embodiment having the above configuration. Since the third embodiment is characterized by the operation of the graph document dialogue input / output unit 20A, the following explanation will focus on the operation of the graph document dialogue input / output unit 20A.

[0100] When a user utterance and a graph document are transmitted from a connected graph document dialogue input / output unit 20A, the graph document dialogue device 10A (dialogue control unit 11) executes the series of processes shown in Fig. 11. When a user utterance and a graph document are transmitted from any of the graph document dialogue input / output units 20A, the same processes as above are executed. In other words, the process shown in Fig. 11 is executed repeatedly every time a user utterance and a graph document are transmitted from the graph document dialogue input / output unit 20A.

[0101] FIG. 16 is a flowchart showing the characteristic operation of the graph document dialogue input / output unit according to the third embodiment.

[0102] <S501> The graph document dialogue input / output unit 20A creates a graph document display / edit field 201 on the left side of the screen and a dialogue history display / input field 202 on the right side, and displays information, as shown in the example screen of Fig. 2. Initially, both the graph document and the dialogue history are empty.

[0103] <S502> The graph document dialogue input / output unit 20A waits for the graph document displayed in the graph document display / editing column 201 to be modified, or for the graph document and dialogue history to be sent from the dialogue control unit 11, or for the send button 205 on the screen to be pressed.

[0104] <S503> The graph document dialogue input / output unit 20A determines whether the graph document has been modified. If the graph document has been modified, the graph document dialogue input / output unit 20A proceeds to step S504. On the other hand, if the graph document has not been modified, the graph document dialogue input / output unit 20A proceeds to step S505.

[0105] <S504> If the graph document dialogue input / output unit 20 determines in the above step S503 that the graph document has been modified, it transmits the modified graph document to the dialogue control unit 11, and returns to the above step S502.

[0106] <S505> If the graph document dialogue input / output unit 20A determines in step S503 that the graph document has not been modified, it determines whether or not the graph document and dialogue history have been transmitted from the dialogue control unit 11. If the graph document and dialogue history have been transmitted, the graph document dialogue input / output unit 20A proceeds to step S506, whereas if the graph document and dialogue history have not been transmitted, it proceeds to step S507.

[0107] <S506> If the graph document dialogue input / output unit 20A determines in the above-mentioned step S505 that the graph document and dialogue history have been transmitted, it reflects the acquired dialogue history and graph document in the graph document display / editing field 201 and the dialogue history display / input field 202, and returns to the above-mentioned step S502.

[0108] <S507> The graph document dialogue input / output unit 20A determines whether or not the send button 205 on the screen has been pressed. If the send button 205 has been pressed, the graph document dialogue input / output unit 20A proceeds to step S508, but if the send button 205 has not been pressed, the graph document dialogue input / output unit 20A returns to the above-mentioned step S502.

[0109] <S508> The graph document dialogue input / output unit 20A transmits the user utterance on the screen and the graph document to the dialogue control unit 11, and empties the text input field 204.

[0110] <S509> The graph document dialogue input / output unit 20A acquires the dialogue history and the graph document from the dialogue control unit 11.

[0111] <S510> The graph document dialogue input / output unit 20A reflects the acquired dialogue history and graph document in the graph document display / edit field 201 and the dialogue history display / input field 202, makes the text input field 204 ready for input, and returns to the above-mentioned step S502.

[0112] (C-3) Advantages of the Third Embodiment As described above, the third embodiment provides the following advantages in addition to the advantages of the first and second embodiments.

[0113] Multiple users can share one graph document / dialogue via the graph document dialogue input / output unit 20A. This allows multiple users to share an idea, plan, proposal, way of thinking, etc. in the form of a graph document / dialogue, and multiple users can cooperate to create new dialogues / graph documents, enabling them to cooperate in accomplishing things that they could not do alone.

[0114] By sharing graph documents and dialogues, even users who join midway through can view the dialogue history and understand the dialogues and graph document modifications that have been made up to that point, enabling more appropriate dialogue and graph document editing.

[0115] Furthermore, by checking the dialogue and graph document corrections made by the large-scale language model 30 with multiple participating users, it is possible to monitor whether the large-scale language model 30 is outputting anything strange.

[0116] (D) Other Embodiments Although various modified embodiments have been mentioned in the above-described embodiments, the following modified embodiments can also be applied.

[0117] (D-1) In the first to third embodiments described above, the input / output devices (graph document dialogue input / output units 20 (20A)) may communicate with each other to share the graph contents and dialogue history. Also, when transmitting the graph contents and dialogue history, only the differences may be transmitted.

[0118] (D-2) In the first to third embodiments described above, the screen for displaying and inputting dialogues may be omitted. A user's utterance in a dialogue may be generated by directly pointing to a portion of the graph with a mouse and specifying a corresponding request using a pull-down menu or the like. System utterances may also be displayed by adding nodes directly to the graph display. This allows a screen to be configured with only a graph display, without a dialogue display portion.

[0119] Conversely, it is also possible to configure the screen so that it only displays and inputs dialogues, without displaying graphs. In this case, the graph is not displayed, but is retained internally but not displayed, and the screen is configured so that only the display and inputs dialogues are displayed. Even in this case, the current graph content is updated internally and the dialogue is based on that. Also, instead of displaying the graph all the time, it can be displayed as needed. It can be displayed when the dialogue is finished, or at the user's request.

[0120] (D-3) In the third embodiment described above, conditions (additional information such as user or organization name, project, access rights, etc.) that further narrow down the graph documents may be added to the search query statement. For example, it may be possible to narrow down the search to only those created by a specific user, to only those belonging to a specific department, to only those belonging to a specific project, or to only those that a specific user may access.

[0121] 1, 1A, 1B...Graph document dialogue system, 10, 10A...Graph document dialogue device, 11...Dialogue control unit, 12...Memory unit, 13...Prompt template, 13-1...Overall prompt, 13-2...Variable part, 13-3...Detailed prompt, 13-4...Graph document, 13-5...Dialogue history, 13-6...Search results, 13A...Prompt template, 14...Dialogue history, 15...Graph document, 20, 20A...Graph document dialogue input / output unit, 30...Large-scale language model, 41...Search unit, 42...Graph document group, 43...Prompt template, 43-1...Overall prompt, 43-2...Variable part, 43-3...Graph document, 43-4...Dialogue history, 200...Graph document dialogue input / output screen, 201...Edit field, 202...Input field, 203...Dialogue display field, 204...Text input field, 205...Send button.

Claims

1. An information processing device characterized by comprising a display control unit that displays the content of a dialogue with a large-scale language model on a display unit as a graph document that graphs the document structure.

2. The information processing device according to claim 1, characterized in that the graph document is editable, and further comprising an interaction control unit that inputs the edited graph document into the large-scale language model.

3. The information processing device according to claim 2, characterized in that the graph document includes a plurality of selectable nodes, and the dialogue control unit inputs the graph document including the selected nodes into the large-scale language model.

4. The information processing device according to claim 3, wherein the dialogue control unit acquires the dialogue result corresponding to the selected node from the large-scale language model.

5. The information processing device according to claim 1, wherein the display control unit causes the display unit to simultaneously display the dialogue history relating to the dialogue content and the graph document.

6. An information processing device connected to an external device that uses a first graph document in which a document structure is graphed, and a large-scale language model that generates the first graph document, characterized in that the information processing device has control means for creating a first prompt based on information including the first graph document, sending the first prompt to the large-scale language model, obtaining the first graph document updated based on the first prompt from the large-scale language model, and sending the obtained first graph document to the external device.

7. The information processing device according to claim 6, wherein the large-scale language model also generates system utterances in response to user utterances input via the external device, the first prompt includes, in addition to the first graph document, dialogue history information of the user utterances and the system utterances, and the control means acquires from the large-scale language model the first graph document and the system utterances updated based on the first prompt, and transmits the acquired first graph document and the system utterances to the external device.

8. The information processing device according to claim 7, further comprising: a graph document search unit that searches for the second graph document based on the search query sentence from a group of graph documents stored in the large-scale language model; the large-scale language model also generates a search query sentence required to search for a reference second graph document from the dialogue content; creates a second prompt based on information including the first graph document and the dialogue history information; sends the second prompt to the large-scale language model; and retrieves the search query sentence generated based on the second prompt from the large-scale language model; and the first prompt includes information about the second graph document found by the graph document search unit.

9. The information processing device according to claim 8, characterized in that the graph document search unit can obtain graph fragments whose content is similar to the search query statement by vector search, and further obtains graph fragments linked to the graph fragments by specific labels, and returns these as search results.

10. The information processing device described in claim 8, characterized in that the graph document search unit can add additional information to the first graph document and add conditions based on the additional information to the search query statement so as to narrow down the second graph document to be searched.

11. A non-transitory storage medium storing an information processing program characterized by causing a computer mounted on an information processing device connected to an external device that uses a first graph document in which a document structure is graphed and a large-scale language model that generates the first graph document to function as a control means that creates a first prompt based on information including the first graph document, sends the first prompt to the large-scale language model, obtains the first graph document updated based on the first prompt from the large-scale language model, and transmits the obtained first graph document to the external device.

12. An information processing method used in an information processing device connected to an external device that uses a first graph document in which a document structure is graphed, and a large-scale language model that generates the first graph document, wherein a control means creates a first prompt based on information including the first graph document, sends the first prompt to the large-scale language model, obtains the first graph document updated based on the first prompt from the large-scale language model, and sends the obtained first graph document to the external device.

13. An information processing system comprising one or more external devices that utilize a first graph document in which a document structure is graphed, a large-scale language model that generates the first graph document, and an information processing device, wherein the information processing device according to claim 6 is applied to the information processing device.

Images