System, Program, and Document Information Manufacturing Method

Abstract

The objective of this invention is to provide a system suitable for generating travel-related texts. [Solution] The system according to the present invention comprises one or more computers. The first document is a document relating to travel and includes a plurality of second documents. The control circuits of the one or more computers acquire definition information indicating the policy of the first document, instruct the first LLM to generate configuration information indicating the policy of the plurality of second documents based on the definition information, and instruct the second LLM to generate a plurality of second document information indicating the plurality of second documents based on the configuration information and definition information generated by the first LLM.

Description

Technical Field

[0001] The present invention relates to a system, a program, and a method for manufacturing text information.

Background Art

[0002] As an invention related to a conventional system, for example, a picture book generation system described in Patent Document 1 is known. This picture book generation system includes a reception unit, an illustration unit, a story unit, and an integration unit. The reception unit receives user input information including at least one of a photograph, text, and a theme. The illustration unit transmits an illustration generation instruction to an illustration generation unit using the user input information. The story unit transmits a story generation instruction to a story generation unit using the user input information. The integration unit integrates the illustration generated by the illustration generation unit and the story generated by the story generation unit to generate picture book data. According to such a picture book generation system, a picture book having an illustration and a story suitable for the illustration can be generated using the photograph taken by the user and the information input by the user.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, a system suitable for generating text related to travel is required.

[0005] Therefore, an object of the present invention is to provide a system, a program, and a method for manufacturing text information suitable for generating text related to travel.

Means for Solving the Problems

[0006] The first aspect is The system consists of one or more computers. The first sentence is about travel and includes multiple second sentences. The control circuits of the one or more computers described above are: Obtain definition information that shows the policy of the first document mentioned above, Based on the definition information, instructions are given to the first LLM (Large Language Models) to generate configuration information indicating the policies of the multiple second documents. Based on the configuration information and definition information generated by the first LLM, instructions are given to the second LLM to generate a plurality of second document information representing the plurality of second documents. The plurality of second sentence information generated by the second LLM is obtained. It is a system.

[0007] The second form is, The control circuits of the one or more computers described above are: Obtain theme information indicating the theme of the first document mentioned above, Based on the definition information and the theme information, instructions are given to the first LLM to generate configuration information indicating the policies of the multiple second documents. This is the system described in the first form.

[0008] The third form is, The control circuits of the one or more computers described above are: By combining the aforementioned multiple second sentences, first sentence information representing the first sentence is generated. The system is as described in the first or second form.

[0009] The fourth form is, The control circuits of the one or more computers described above are: Retrieve the project proposal information entered by the user. The aforementioned proposal information is converted into a data exchange format by the third LLM to obtain the aforementioned definition information. This is the system described in the second form.

[0010] The fifth form is The first text is a travel article, The theme information includes the area input by the user. It is the system described in the second form.

[0011] The sixth form is The first text is a travel article, The theme information includes the area input by the user, The plan information includes a plan for generating a travel article related to the area included in the theme information. It is the system described in the fifth form.

[0012] The seventh form is The first text includes a plurality of items, The plurality of second texts correspond to the plurality of items. It is the system described in the fourth form or the sixth form.

[0013] The eighth form is The plan information includes a plurality of paragraphs, The control circuit of the one or more computers Based on the plurality of paragraphs included in the plan information, causes the third LLM to generate the definition information including the plurality of items. It is the system described in the seventh form.

[0014] The ninth form is The control circuit of the one or more computers Based on the definition information and the theme information, collects the first collection information using the fourth LLM, Based on the definition information, the theme information, and the first collection information, gives an instruction to cause the first LLM to generate the configuration information. It is the system described in any one of the second form, the fourth form to the eighth form.

[0015] The tenth form is The first article above is a travel article. The definition information includes instructions for generating text about the area identified by the theme information and instructions for creating text about multiple spots included in the area identified by the theme information. The control circuits of the one or more computers mentioned above: First collection information relating to the area identified by the thematic information and the multiple spots is collected using the fourth LLM. This is the system described in Form 9.

[0016] The 11th form is, The control circuits of the one or more computers described above are: Based on the definition information, theme information, and configuration information, the second collection information is collected using the 5th LLM. Based on the configuration information and the second collected information, instructions are given to the second LLM to generate the plurality of second sentence information. This is the system described in any of the second, fourth, or tenth forms.

[0017] The 12th form is, The second type of collected information includes external information existing on the internet and internal information stored on servers whose external access is restricted. The aforementioned internal information includes the first sentence that was generated in the past. This is the system described in the 11th form.

[0018] The 13th form is, The control circuits of the one or more computers described above are: When collecting the second collection information using the 5th LLM, information that does not meet the predetermined conditions will not be collected as the second collection information. This is the system described in the 12th form.

[0019] The 14th form is, The control circuits of the one or more computers described above are: The sixth LLM is used to perform fact-checking of the multiple second documents. The system is one of the first to thirteenth forms.

[0020] The 15th form is, The first article above is a travel article. At least a portion of the aforementioned multiple second pieces of information include links to information about travel products. The system is one of the first to fourteenth forms.

[0021] The 16th form is, The configuration information includes instructions for causing the second LLM to output one or more image pieces of information related to the plurality of second sentences. At least a portion of the plurality of second textual information includes the one or more image information. The system is one of the first to fifteenth forms.

[0022] The 17th form is, The configuration information includes instructions to cause the second LLM to output the URLs referenced during the generation of the multiple second sentences. At least a portion of the plurality of second document information includes URLs referenced by the second LLM when generating the plurality of second documents. The system is one of the first to sixteenth forms.

[0023] The 18th form is, It is a program, The first sentence is about travel and includes multiple second sentences. The program is installed in the control circuits of one or more computers. Obtain definition information that shows the policy of the first document mentioned above, Based on the definition information, instructions are given to the first LLM (Large Language Models) to generate configuration information indicating the policies of the multiple second documents. Based on the configuration information and definition information generated by the first LLM, instructions are given to the second LLM to generate a plurality of second document information representing the plurality of second documents. The second LLM generates the plurality of second sentence information, It is a program.

[0024] The 19th form is, A method for producing written information, The first sentence is about travel and includes multiple second sentences. The above-mentioned method for producing text information involves a control circuit of one or more computers, Obtain definition information that shows the policy of the first document mentioned above, Based on the definition information, instructions are given to the first LLM (Large Language Models) to generate configuration information indicating the policies of the multiple second documents. Based on the configuration information and definition information generated by the first LLM, instructions are given to the second LLM to generate a plurality of second document information representing the plurality of second documents. The second LLM generates the plurality of second sentence information, This is a method for producing written information. [Effects of the Invention]

[0025] According to the present invention, a system, program, and method for producing textual information suitable for generating travel-related text can be provided. [Brief explanation of the drawing]

[0026] [Figure 1] Figure 1 is a block diagram of System 1. [Figure 2] Figure 2 is a block diagram of the first terminal 10. [Figure 3] Figure 3 is a block diagram of the first server 210. [Figure 4] Figure 4 is a block diagram of the second server 310. [Figure 5]Figure 5 shows the first text information I1. [Figure 6] Figure 6 is a flowchart showing the operation of System 1. [Figure 7] Figure 7 shows the IK (Inverse Kinematics) information in the project proposal. [Figure 8] Figure 8 shows the IK (Inverse Kinematics) information in the project proposal. [Figure 9] Figure 9 shows the information IS1 from the first definition document. [Figure 10] Figure 10 shows the second definition document information IS2. [Figure 11] Figure 11 shows the first prompt P1. [Figure 12] Figure 12 shows the configuration information IST. [Figure 13] Figure 13 shows the second prompt P2. [Figure 14] Figure 14 shows the second set of textual information I2-1 and I2-2. [Modes for carrying out the invention]

[0027] (Embodiment) A system 1, a program PG, and a document information manufacturing method according to one embodiment of the present invention will be described with reference to the drawings.

[0028] [System Structure] First, the overall configuration of System 1 will be explained with reference to the diagrams. Figure 1 is a block diagram of System 1. Figure 2 is a block diagram of the first terminal 10. Figure 3 is a block diagram of the first server 210. Figure 4 is a block diagram of the second server 310.

[0029] System 1, shown in Figure 1, comprises a first terminal 10, a first server 210, and a second server 310. The first terminal 10, the first server 210, and the second server 310 can communicate with each other via a communication network. The network can be the internet, an intranet, or the like.

[0030] The first terminal 10 is a computer used by a user. The user is, for example, a staff member of a travel agency who wants to generate travel product titles. The first terminal 10 is, for example, a personal computer, a smartphone, or a tablet device. In this embodiment, the first terminal 10 is a personal computer located in a travel agency. As shown in Figure 2, the first terminal 10 includes a control circuit 12, a memory circuit 14, a network interface 16, a graphics processing unit 18, a display 20, an operation unit 26, and a touch panel 28.

[0031] The memory circuit 14 stores programs and data. The memory circuit 14 is, for example, a combination of ROM (Read Only Memory), RAM (Random Access Memory), and storage (for example, flash memory or hard disk).

[0032] The program includes, for example, the following: • Programs for the OS (Operating System) • Programs for applications that perform information processing (e.g., web browsers, or target applications described later)

[0033] The data includes, for example, the following: • Databases referenced in information processing • Data obtained by performing information processing (i.e., the results of information processing)

[0034] The control circuit 12 realizes the functions of the first terminal 10 by executing the program stored in the memory circuit 14. The control circuit 12 is, for example, at least one of the following: ·CPU(Central Processing Unit) ·GPU(Graphic Processing Unit) ·ASIC(Application Specific Integrated Circuit) ·FPGA(Field Programmable Array)

[0035] The network interface 16 controls communication between the first terminal 10 and an external device. The external device is the first server 210.

[0036] The graphics processing unit 18 displays an image on the display 20 based on the image data generated by the control circuit 12. The display 20 is either a liquid crystal display or an organic EL (Electro-Luminescence) display.

[0037] The operation unit 26 generates an operation signal based on user input via the touch panel 28 and outputs the operation signal to the control circuit 12.

[0038] The first server 210 is a computer. The first server 210 is an application server that operates according to predefined logic based on requests from users. The first server 210 receives instructions from the first terminal 10 and outputs processing commands to the second server 310. The first server 210 is, for example, a server for Dify®. As shown in Figure 3, the first server 210 includes a control circuit 212, a memory circuit 214, and a network interface 216.

[0039] The memory circuit 214 stores programs and data. The memory circuit 214 is, for example, a combination of ROM (Read Only Memory), RAM (Random Access Memory), and storage (for example, flash memory or hard disk).

[0040] The control circuit 212 implements the functions of the first server 210 by executing the program stored in the memory circuit 214. The control circuit 212 is, for example, at least one of the following: ·CPU(Central Processing Unit) ·GPU(Graphic Processing Unit) ·ASIC(Application Specific Integrated Circuit) ·FPGA(Field Programmable Array)

[0041] The control circuit 212 includes, as functional blocks, a definition information acquisition means 220, a configuration information generation instruction means 222, a second document information generation instruction means 224, a theme information acquisition means 226, a fact-checking means 228, a second document information acquisition means 230, a project proposal information acquisition means 232, a first collected information collection means 234, a first document information generation means 236, and a second collected information collection means 238.

[0042] The network interface 216 controls communication between the first server 210 and external devices. The external devices are the first terminal 10 and the second server 310.

[0043] The second server 310 is an inference server equipped with an LLM (Large Language Model). As shown in Figure 4, the second server 310 includes a control circuit 312, a memory circuit 314, and a network interface 316.

[0044] The memory circuit 314 stores programs and data. The memory circuit 314 is, for example, a combination of ROM (Read Only Memory), RAM (Random Access Memory), and storage (for example, flash memory or hard disk).

[0045] The control circuit 312 implements the functions of the second server 310 by executing the program stored in the memory circuit 314. The control circuit 312 is, for example, at least one of the following: ·CPU(Central Processing Unit) ·GPU(Graphic Processing Unit) ·ASIC(Application Specific Integrated Circuit) ·FPGA(Field Programmable Array)

[0046] The network interface 316 controls communication between the second server 310 and an external device. The external device is the first server 210.

[0047] Furthermore, the second server 310 is equipped with the first LLM 318, the second LLM 320, the third LLM 322, the fourth LLM 324, the fifth LLM 326, and the sixth LLM 328. LLMs are natural language processing models trained using a large amount of text data. LLMs can perform natural language processing such as text generation by, for example, fine-tuning a pre-trained model.

[0048] [System 1 Operation] The operation of System 1 will be explained below with reference to the diagrams. Figure 5 is a diagram showing the first document information I1. Figure 6 is a flowchart showing the operation of System 1. Figures 7 and 8 are diagrams showing the project proposal information IK. Figure 9 is a diagram showing the first definition document information IS1. Figure 10 is a diagram showing the second definition document information IS2. Figure 11 is a diagram showing the first prompt P1. Figure 12 is a diagram showing the configuration information IST. Figure 13 is a diagram showing the second prompt P2. Figure 14 is a diagram showing the second document information I2-1 and I2-2.

[0049] The control circuit 212 of the first server 210 reads the program PG stored in the memory circuit 214, and these program PGs cause the control circuit 212 of the first server 210 to execute the operations described below. The program PGs cause the control circuit 212 to execute the following: definition information acquisition means 220, configuration information generation instruction means 222, second document information generation instruction means 224, theme information acquisition means 226, fact checking means 228, second document information acquisition means 230, project proposal information acquisition means 232, first collected information collection means 234, first document information generation means 236, and second collected information collection means 238.

[0050] Furthermore, the memory circuit 214 of the first server 210 stores the first prompt P1, the second prompt P2, and the third prompt P3. The first prompt P1, the second prompt P2, and the third prompt P3 may be generated by the user or by the LLM.

[0051] System 1 aims to generate the first document information I1 shown in Figure 5. The first document information I1 is a digital document. In this embodiment, the first document information I1 shown in Figure 5 is a web page displayed on a travel agency's website. The first document information I1 contains the first document Tx1 related to travel. Specifically, the first document Tx1 related to travel is a travel article introducing 10 recommended hotels in Australia. The first document Tx1 contains multiple items S1 to SN. In Figure 5, only items S1 and S2 are shown. N is an integer greater than or equal to 2. Items S1 to SN are the titles of the topics of the first document Tx1. That is, items S1 to SN are titles for segmenting and explaining the first document Tx1.

[0052] Furthermore, the first document Tx1 contains the second documents Tx2-1 to Tx2-N. The second documents Tx2-1 to Tx2-N correspond to items S1 to SN. The second documents Tx2-1 to Tx2-N are written under each of the items S1 to SN. Each of the second documents Tx2-1 to Tx2-N is a document that explains the details of items S1 to SN.

[0053] System 1 generates the first document Tx1 according to the following procedure. First, System 1 causes the first LLM318 to generate the structure of the first document Tx1. The structure of the first document Tx1 is the generation of items S1 to SN. Then, System 1 causes the second LLM320 to generate the second documents Tx2-1 to Tx2-N, which correspond to each of the items S1 to SN that make up the first document Tx1. Finally, System 1 generates the first document Tx1 by combining the second documents Tx2-1 to Tx2-N.

[0054] The operation of System 1 will be described below with reference to Figure 6. First, the user inputs the project proposal information IK shown in Figures 7 and 8 by operating the operation unit 26 of the first terminal 10. As a result, the control circuit 12 of the first terminal 10 acquires the project proposal information IK (step S1).

[0055] The project information IK contains text created by the user. The project information IK contains a project proposal for generating travel articles related to an area included in the theme information ITE described later. In this embodiment, the project information IK contains a project proposal for creating a first article Tx1 (travel article) titled "Top 10 Recommended Hotels in XX (Country Name)". The theme information ITE described later includes an area corresponding to XX (Country Name). The project information IK contains the writing procedure A1, a published article A2, and an outline of the first article Tx1 (travel article) A3.

[0056] The writing procedure A1 shows the procedure for creating the first document Tx1 (travel article), and in this embodiment, it includes three steps. Specifically, in step 1, keyword research is conducted; in step 2, detailed hotel information is researched using the keywords; and in step 3, a rough structure of the first document Tx1 is created.

[0057] The published article A2 is the URL of a published travel article. The published article A2 is the result of a user's research.

[0058] The summary A3 of the first document Tx1 is used to generate the first document Tx1. The summary of the first document Tx1 includes multiple paragraphs s1~sM such as "Eye-catching image" and "Description sheet". M is an integer of 2 or greater. Each of the multiple paragraphs s1~sM contains the conditions required for the <Body> which will be part of the travel article contained in the first document Tx1. Each of the multiple paragraphs s1~sM also contains a <Body example> corresponding to the <Body>. The <Body example> may be a document created by the user or a document created by LLM. Thus, the proposal information IK contains information about multiple items S1~SN contained in the first document Tx1. Note that the multiple paragraphs s1~sM contained in the proposal information IK may or may not match multiple items S1~SN. In Figure 8, the multiple paragraphs s1~sM contained in the proposal information IK do not match multiple items S1~SN.

[0059] Next, the user inputs theme information ITE by operating the operation unit 26 of the first terminal 10. As a result, the control circuit 12 of the first terminal 10 acquires the theme information ITE (step S2). The theme information ITE indicates the theme of the first document Tx1. In this embodiment, the theme information ITE includes an area entered by the user. An area means a region, and in this embodiment, it means a country. In this embodiment, the user enters "Australia" as the area for the theme information ITE. Thus, the project proposal information IK contains information for generating a travel article without specifying an area. On the other hand, the theme information ITE includes an area. Therefore, by combining the project proposal information IK and the theme information ITE, the input information for generating a travel article for a specific area is obtained.

[0060] Next, the control circuit 12 of the first terminal 10 transmits the project proposal information IK and theme information ITE to the first server 210 via the network interface 16. In response, the network interface 216 of the first server 210 receives the project proposal information IK and theme information ITE and outputs the project proposal information IK and theme information ITE to the control circuit 212. As a result, the control circuit 212 of the first server 210 (control circuits of one or more computers, project proposal information acquisition means 232, theme information acquisition means 226) acquires the project proposal information IK and theme information ITE entered by the user (step S11).

[0061] Next, the control circuit 212 of the first server 210 generates first input information II1 which is input to the third LLM 322 (step S12). The first input information II1 includes a third prompt P3, project proposal information IK, and theme information ITE. The third prompt P3 is stored by the memory circuit 214. The third prompt P3 is an instruction to cause the third LLM 322 to generate first definition document information IS1 and second definition document information IS2 based on the project proposal information IK and theme information ITE.

[0062] The first definition document information IS1 and the second definition document information IS2 (definition document information) indicate the policy for the first document Tx1. Specifically, as shown in Figure 9, the first definition document information IS1 includes instructions for generating multiple paragraphs s1~sM contained in the project proposal information IK, and example sentences for the text contained in the multiple paragraphs s1~sM. These instructions are generated by the third LLM322 based on the writing procedure A1 of the project proposal information IK, the published article A2, and the summary A3 of the first document Tx1 (travel article). In addition, the second definition document information IS2 includes multiple items S1~SN, as shown in Figure 10. The multiple items S1~SN are generated by the third LLM322 based on the writing procedure A1 of the project proposal information IK, the published article A2, and the summary A3 of the first document Tx1 (travel article).

[0063] The first definition information IS1 and the second definition information IS2 described above are written in a data conversion format. In this embodiment, the first definition information IS1 and the second definition information IS2 are written in XML (eXtensible Markup Language) format. Therefore, the third prompt P3 includes the following three instructions, as shown in Figure 11. • The first definition document information IS1 is generated by converting the project proposal information IK into XML format. By extracting multiple paragraphs s1~sM from the project proposal information IK, a second definition document information IS2 containing multiple items S1~SN is generated. • The second definition document information IS2 is described in XML format.

[0064] The control circuit 212 of the first server 210 transmits the first input information II1 to the second server 310 via the network interface 216 (step S13). In response, the network interface 316 of the second server 310 receives the first input information II1 and outputs the first input information II1 to the control circuit 312. As a result, the control circuit 312 of the second server 310 acquires the first input information II1.

[0065] The third LLM322 reads the project proposal information IK and theme information ITE according to the instructions of the third prompt P3. The third LLM322 generates the first definition document information IS1 and the second definition document information IS2 based on the project proposal information IK and theme information ITE according to the instructions of the third prompt P3.

[0066] Next, the control circuit 312 of the second server 320 transmits the first definition document information IS1 and the second definition document information IS2 to the first server 210 via the network interface 316. Accordingly, the network interface 216 of the first server 210 receives the first definition document information IS1 and the second definition document information IS2 and outputs the first definition document information IS1 and the second definition document information IS2 to the control circuit 212. As a result, the control circuit 212 of the first server 210 acquires the first definition document information IS1 and the second definition document information IS2 (step S14). As described above, in steps S12 to S14, the control circuit 212 of the first server 210 (control circuit of one or more computers / definition document information acquisition means 220) acquires the first definition document information IS1 and the second definition document information IS2 (definition document information) by converting the project proposal information IK into XML format (data exchange format) using the third LLM 322. In this process, the control circuit 212 of the first server 210 (control circuit of one or more computers / definition document information acquisition means 220) causes the third LLM 322 to generate a second definition document information IS2 (definition document information) containing multiple items S1 to SN based on multiple paragraphs s1 to sM contained in the project proposal information IK.

[0067] Next, the control circuit 212 of the first server 210 (control circuit of one or more computers, first information collection means 234) collects first collection information IR1 using the fourth LLM 324 based on the first definition information IS1, the second definition information IS2 (definition information), and theme information ITE (step S15). The control circuit 212 of the first server 210 transmits the first definition information IS1, the second definition information IS2, and theme information ITE to the second server 310 via the network interface 216. The network interface 316 of the second server 310 receives the first definition information IS1, the second definition information IS2, and theme information ITE and outputs the first definition information IS1, the second definition information IS2, and theme information ITE to the control circuit 312. As a result, the control circuit 312 of the second server 310 acquires the first definition information IS1, the second definition information IS2, and theme information ITE.

[0068] Here, the project information IK includes instructions for generating text about the area identified by the theme information ITE and instructions for creating text about multiple spots included in the area identified by the theme information ITE. Spots include hotels and surrounding facilities of hotels. Specifically, the project information IK includes writing procedure A1. Writing procedure A1 includes three steps. Specifically, in step 1, a keyword search is conducted; in step 2, a detailed search of the hotel is conducted using the keywords; and in step 3, a rough structure of the first text Tx1 is created. The first definition information IS1 is information obtained by converting the project information IK into XML format, and therefore includes information equivalent to the above writing procedure A1. Accordingly, the first definition information IS1 also includes instructions for generating text about the area identified by the theme information ITE and instructions for creating text about multiple spots included in the area identified by the theme information ITE. Then, the fourth LLM324 generates the first collection information IR1 by executing steps 1 and 2. The first collected information IR1 is information related to the first document Tx1 and is used to generate the first document Tx1. The first collected information IR1 includes, for example, the URL of the hotel and the URLs of facilities around the hotel.

[0069] The control circuit 312 of the second server 310 transmits the first collection information IR1 to the first server 210 via the network interface 316. In response, the network interface 216 of the first server 210 receives the first collection information IR1 and outputs the first collection information IR1 to the control circuit 212. As a result, the control circuit 212 of the first server 210 acquires the first collection information IR1. As described above, in step S15, the control circuit 212 of the first server 210 (control circuits of one or more computers, first collection information collection means 234) collects the first collection information IR1 relating to the area and multiple spots identified by the theme information ITE using the fourth LLM 324.

[0070] Next, the control circuit 212 of the first server 210 generates second input information II2 which is input to the first LLM 318 (step S16). The second input information II2 includes the first prompt P1, theme information ITE, first definition document information IS1, second definition document information IS2, and first collection information IR1 shown in Figure 11. The first prompt P1 is an instruction to cause the first LLM 318 to generate configuration information IST shown in Figure 12 based on the theme information ITE, first definition document information IS1, second definition document information IS2, and first collection information IR1. More specifically, the configuration information IST indicates the policies for multiple second documents Tx2-1 to Tx2-N, each contained in each of the multiple items S1 to SN. In this embodiment, the configuration information IST includes titles assigned to the multiple items S1 to SN and instructions for creating the body text of each of the multiple items S1 to SN. Furthermore, the configuration information IST includes instructions for the second LLM320 to output one or more image information IPs related to multiple second documents Tx2-1 to Tx2-N. If the first document Tx1 is a web page of a travel company's website, the one or more image information IPs include images contained in the web page. In addition, the configuration information IST includes instructions for the second LLM320 to output URLs referenced during the generation of multiple second documents Tx2-1 to Tx2-N.

[0071] The first prompt P1 includes a task, research findings, and instructions so that the configuration information IST described above is generated. The task contains instructions to generate an outline (structure) of the travel article (first document Tx1). The research findings contain instructions to analyze the first collected information IR1. The instructions also contain instructions to generate an outline (structure) of the travel article (first document Tx1) using the first collected information IR1. Furthermore, there are instructions to output one or more image information IPs related to multiple second documents Tx2-1 to Tx2-N. The instructions also contain instructions to output the URLs referenced when generating multiple second documents Tx2-1 to Tx2-N. In addition, the instructions include instructions to refer to the second definition information IS2, as well as an example of the configuration information IST.

[0072] Next, the control circuit 212 of the first server 210 transmits the second input information II2 to the second server 310 via the network interface 216 (step S17). Accordingly, the network interface 316 of the second server 310 receives the second input information II2 and outputs the second input information II2 to the control circuit 312. As a result, the control circuit 312 of the second server 310 acquires the second input information II2.

[0073] The first LLM318 analyzes the first collected information IR1 in accordance with the instructions of the first prompt P1. Based on the analysis results of the first definition information IS1, the second definition information IS2, the theme information ITE, and the first collected information IR1, the first LLM318 generates configuration information IST. At this time, as shown in Figure 12, the first LLM318 includes instructions in the configuration information IST to cause the second LLM320 to output the URLs referenced when generating multiple second documents Tx2-1 to Tx2-N, and one or more image information IPs related to the multiple second documents Tx2-1 to Tx2-N. As described above, in steps S16 and S17, the control circuit 212 of the first server 210 (control circuit / configuration information generation instruction means 222 of one or more computers) instructs the first LLM 318 to generate configuration information IST, which indicates the policies of multiple second documents Tx2-1 to Tx2-N contained in each of the multiple items S1 to SN, based on the theme information ITE, the first collected information IR1, the first definition document information IS1, and the second definition document information IS2 (definition document information).

[0074] Next, the control circuit 312 of the second server 310 transmits the configuration information IST to the first server 210 via the network interface 316. In response, the network interface 216 of the first server 210 receives the configuration information IST and outputs the configuration information IST to the control circuit 212. As a result, the control circuit 212 of the first server 210 acquires the configuration information IST (step S18).

[0075] Next, the control circuit 212 of the first server 210 generates third input information II3 which is input to the second LLM 320 (step S19). The third input information II3 includes the second prompt P2, theme information ITE, first definition document information IS1, second definition document information IS2, first collection information IR1, and configuration information IST shown in Figure 13. The second prompt P2 is an instruction to the second LLM 320 to generate multiple second document information I2-1 to I2-N, which represent multiple second documents Tx2-1 to Tx2-N contained in each of the multiple items S1 to SN, based on the configuration information IST, first definition document information IS1, and second definition document information IS2 (definition document information) generated by the first LLM 318. In this embodiment, as shown in Figure 13, the second prompt P2 includes an instruction to cause the second LLM320 to read the configuration information IST, and an instruction to cause the second LLM320 to generate a plurality of second sentences Tx2-1 to Tx2-N with reference to the first collected information IR1.

[0076] Next, the control circuit 212 of the first server 210 transmits the third input information II3 to the second server 310 via the network interface 216 (step S20). In response, the network interface 316 of the second server 310 receives the third input information II3 and outputs the second input information II2 to the control circuit 312. As a result, the control circuit 312 of the second server 310 acquires the third input information II3.

[0077] The second LLM320 generates multiple second documents Tx2-1 to Tx2-N as shown in Figure 14, following the instructions of the second prompt P2. Specifically, as shown in Figure 14, the second LLM320 creates the body of the lead statement for second document Tx2-1 according to the instructions for the lead statement in the configuration information IST. Similarly, the second LLM320 creates the body of the hotel summary for second document Tx2-1 according to the instructions for the hotel summary in the configuration information IST. However, the configuration information IST includes instructions to cause the second LLM320 to output URLs referenced during the generation of multiple second documents Tx2-1 to Tx2-N, and one or more image information IPs related to the multiple second documents Tx2-1 to Tx2-N. Therefore, the second LLM320 searches for one or more image information IPs and attaches the images indicated by the image information IPs near the body of the lead statement for second document Tx2-1. Furthermore, the second LLM320, for example, includes the hotel's URL in the body of the hotel summary in the second document Tx2-1. In this way, at least a portion of the multiple second document information I2-1 to I2-N contains one or more image information IPs and URLs referenced by the second LLM320 during the generation of the multiple second documents Tx2-1 to Tx2-N.

[0078] As described above, in steps S19 and S20, the control circuit 212 of the first server 210 (control circuit of one or more computers, second document information generation instruction means 224) instructs the second LLM 320 to generate a plurality of second document information I2-1 to I2-N, which represent a plurality of second documents Tx2-1 to Tx2-N contained in each of the plurality of items S1 to SN, based on the configuration information IST, first definition document information IS1, and second definition document information IS2 (definition document information) generated by the first LLM 318.

[0079] Next, the control circuit 312 of the second server 310 transmits multiple second document information I2-1 to I2-N to the first server 210 via the network interface 316. Accordingly, the network interface 216 of the first server 210 receives the multiple second document information I2-1 to I2-N and outputs the multiple second document information I2-1 to I2-N to the control circuit 212. As a result, the control circuit 212 of the first server 210 (control circuit of one or more computers, second document information acquisition means 230) acquires the multiple second document information I2-1 to I2-N generated by the second LLM 320 (step S21).

[0080] Next, the control circuit 212 of the first server 210 (control circuit of one or more computers / fact-checking means 228) performs fact-checking of multiple second documents Tx2-1 to Tx2-N using the sixth LLM 328 (step S22). More specifically, the memory circuit 214 of the first server 210 stores the prompt for fact-checking. The control circuit 212 of the first server 210 transmits the prompt for fact-checking and the multiple second document information I2-1 to I2-N to the second server 310 via the network interface 216. Accordingly, the network interface 316 of the second server 310 receives the prompt for fact-checking and the multiple second document information I2-1 to I2-N and outputs the prompt for fact-checking and the multiple second document information I2-1 to I2-N to the control circuit 312. As a result, the control circuit 312 of the second server 310 acquires the prompt for fact-checking and the multiple second document information I2-1 to I2-N.

[0081] The 6th LLM328 performs fact-checking on multiple second documents Tx2-1 to Tx2-N, following the instructions of the fact-checking prompts. Specifically, the 6th LLM328 collects information about the multiple second documents Tx2-1 to Tx2-N via a network and compares the collected information with the multiple second documents Tx2-1 to Tx2-N. Then, the 6th LLM328 determines whether the multiple second documents Tx2-1 to Tx2-N contain any errors. For example, the 6th LLM328 may determine whether numerical values ​​such as the population of Australia are incorrect, or whether a hotel address is incorrect. The 6th LLM328 generates fact-checking information IFC, which shows the results of the fact-checking. Therefore, the fact-checking information IFC contains information that points out the errors contained in the multiple second documents Tx2-1 to Tx2-N. However, fact-checking information IFC may indicate that multiple second sentences Tx2-1 to Tx2-N are error-free if they are error-free.

[0082] Next, the control circuit 312 of the second server 310 transmits fact-checking information IFC to the first server 210 via the network interface 316. In response, the network interface 216 of the first server 210 receives the fact-checking information IFC and outputs it to the control circuit 212. As a result, the control circuit 212 of the first server 210 acquires the fact-checking information IFC.

[0083] Next, the control circuit 212 of the first server 210 (control circuit of one or more computers, first document information generation means 326) generates first document information I1 representing the first document Tx1 by combining a plurality of second documents Tx2-1 to Tx2-N (step S23). As shown in Figure 5, the first document Tx1 is a web page included in the travel agency's website.

[0084] Next, the control circuit 212 of the first server 210 transmits the first document information I1 and fact-checking information IFC to the first terminal 10 via the network interface 216. Accordingly, the network interface 16 of the first terminal 10 receives the first document information I1 and fact-checking information IFC and outputs the first document information I1 and fact-checking information IFC to the control circuit 12. As a result, the control circuit 12 of the first terminal 10 acquires the first document information I1 and fact-checking information IFC (step S4).

[0085] Finally, the control circuit 12 of the first terminal 10 displays the first document information I1 and fact-checking information IFC on the display 20 (step S5). This allows the user to visually inspect the first document Tx1 and perform fact-checking based on the fact-checking information IFC.

[0086] [effect] According to System 1, a system suitable for generating travel-related texts can be provided. More specifically, the first text Tx1 contains multiple items S1~SN related to travel. Therefore, the information contained in the first text Tx1 spans a wide range of fields. Consequently, it is difficult for the LLM to generate the first text Tx1 with a single instruction.

[0087] Therefore, in System 1, the following steps are performed: the first LLM 318 determines the policies for multiple second documents Tx2-1 to Tx2-N, and the second LLM 320 generates the second documents Tx2-1 to Tx2-N according to the policies for the multiple second documents Tx2-1 to Tx2-N. Specifically, the control circuit 212 of the first server 210 instructs the first LLM 318 to generate configuration information IST, which indicates the policies for multiple second documents Tx2-1 to Tx2-N contained in each of the multiple items S1 to SN, based on the first definition document information IS1 and the second definition document information IS2. Furthermore, the control circuit 212 of the first server 210 instructs the second LLM 320 to generate multiple second document information I2-1 to I2-N, which indicate the multiple second documents Tx2-1 to Tx2-N, based on the configuration information IST, the first definition document information IS1, and the second definition document information IS2. As a result, the first LLM318 can generate configuration information IST with high accuracy. Then, the second LLM320 can generate multiple second document information I2-1 to I2-N representing multiple second documents Tx2-1 to Tx2-N based on the highly accurate configuration information IST. Thus, System 1 can provide a system suitable for generating travel-related documents. In this specification, high accuracy means that the information is highly accurate and / or that the content of the information meets the user's expectations.

[0088] According to System 1, the first LLM 318 can generate configuration information IST that takes into account the theme of the first document Tx1. More specifically, the control circuit 212 of the first server 210 instructs the first LLM 318 to generate configuration information IST that indicates the policies of multiple second documents Tx2-1 to Tx2-N contained in each of the multiple items S1 to SN, based on the first definition information IS1, the second definition information IS2, and the theme information ITE. Thus, the instruction for generating the configuration information IST includes theme information ITE that indicates the theme of the first document Tx1. As a result, the first LLM 318 can generate configuration information IST that takes into account the theme of the first document Tx1.

[0089] Furthermore, in System 1, the theme information ITE includes areas entered by the user. Therefore, the first LLM318 can generate configuration information IST that takes these areas into consideration. Area plays a very important role in travel articles. Therefore, the first LLM318 can generate configuration information IST with high accuracy.

[0090] According to System 1, the first LLM318 can generate configuration information IST with high accuracy. More specifically, the control circuit 212 of the first server 210 obtains the first definition document information IS1 and the second definition document information IS2 by having the third LLM322 convert the project proposal information IK into a data exchange format. In this way, when the first definition document information IS1 and the second definition document information IS2 are described in the data exchange format, the first LLM318 can accurately understand the contents of the first definition document information IS1 and the second definition document information IS2. As a result, the first LLM318 can generate configuration information IST with high accuracy based on the first definition document information IS1 and the second definition document information IS2.

[0091] In System 1, the project proposal information IK contains information about multiple items S1 to SN included in the first document Tx1. The first definition document information IS1 and the second definition document information IS2 are generated based on the project proposal information IK. Therefore, the first definition document information IS1 and the second definition document information IS2 contain information about multiple items S1 to SN. As a result, the first LLM can generate configuration information IST that reflects information about multiple items S1 to SN based on the first definition document information IS1 and the second definition document information IS2.

[0092] In System 1, the control circuit 212 of the first server 210 instructs the first LLM 318 to generate configuration information IST based on the first definition information IS1, the second definition information IS2, theme information ITE, and first collection information IR1. The first collection information IR1 is information collected by the fourth LLM based on the first definition information IS1, the second definition information IS2, and theme information ITE. Therefore, the first LLM 318 can reflect the first collection information IR1, collected via the network, in the configuration information IST. In addition, the first collection information IR1 is information about the area and multiple spots identified by the theme information ITE. Therefore, the first LLM 318 can reflect the area and multiple spots identified by the theme information ITE in the configuration information IST.

[0093] In System 1, the control circuit 212 of the first server 210 uses the 6th LLM328 to perform fact-checking on multiple second document information I2-1 to I2-N. This reduces the probability of generating multiple second document information I2-1 to I2-N that contain incorrect information.

[0094] (Other embodiments) Furthermore, the system according to the present invention is not limited to System 1 and can be modified within the scope of its gist.

[0095] Furthermore, the control circuit 212 of the first server 210 (control circuit of one or more computers, second information collection means 238) may collect second information IR2 using the fifth LLM 326 based on the first definition information IS1, the second definition information IS2 (definition information), theme information ITE, and configuration information IST.Then, the control circuit 212 of the first server 210 (control circuit of one or more computers, second document information generation instruction means 224) may instruct the second LLM 320 to generate a plurality of second document information I2-1 to I2-N based on the configuration information IST and the second information collection IR2.As a result, the second LLM 320 can generate a plurality of second document information I2-1 to I2-N based on the second information collection IR2 collected based on the configuration information IST.

[0096] Furthermore, the second collected information IR2 may include external information existing on the internet and internal information stored on servers whose external access is restricted. Such internal information may include the first document Tx1 that was generated in the past. A server whose external access is restricted is, for example, a data server owned by a travel company. As a result, the second LLM320 can generate multiple second document information I2-1 to I2-N based on the first document Tx1 that was generated in the past. In addition, the first document Tx1 that was generated in the past may be a document that has been viewed many times.

[0097] Furthermore, when the control circuit 212 of the first server 210 (control circuit of one or more computers, second information collection means 238) collects the second information collection IR2 using the fifth LLM 326, it does not have to collect information that does not conform to predetermined conditions as the second information collection IR2. Information that does not conform to predetermined conditions is, for example, information that is generated around the same time as the first document information I1 is generated and contains articles with similar content to the first document Tx1. Content similar to the first document Tx1 means, for example, that the area is the same.

[0098] Furthermore, if the first document Tx1 is a travel article, at least some of the information in the second documents I2-1 to I2-N may include links to information about travel products. Travel products include, for example, tours, hotels, and options.

[0099] The second server 310 is equipped with the first LLM318, the second LLM320, the third LLM322, the fourth LLM324, the fifth LLM326, and the sixth LLM328. However, the second server 310 is equipped with the first LLM318 but does not have to be equipped with the second LLM320, the third LLM322, the fourth LLM324, the fifth LLM326, and the sixth LLM328. In this case, system 1 further includes the fourth to the eighth server. Each of the fourth to the eighth server is equipped with the second LLM320, the third LLM322, the fourth LLM324, the fifth LLM326, and the sixth LLM328.

[0100] Note that the first LLM318, the third LLM322, the fourth LLM324, the fifth LLM326, and the sixth LLM328 may be a single common LLM.

[0101] The flowchart shown in Figure 6 illustrates the operation of the control circuit 12 of the first terminal 10 and the control circuit 212 of the first server 210. However, the flowchart in Figure 6 may be executed by the control circuits of one computer, or by the control circuits of three or more computers.

[0102] Furthermore, the memory circuit 214 of the first server 210 may pre-store the first definition document information IS1 and the second definition document information IS2. In this case, the project proposal information IK is not a mandatory configuration requirement. Also, the first definition document information IS1 and the second definition document information IS2 may be generated by the LLM or entered by the user.

[0103] Note that digital documents are not limited to web pages. Digital documents can also be PDF or Word files.

[0104] Furthermore, the first definition document information IS1 and the second definition document information IS2 are described using a data exchange format. However, the first definition document information IS1 and the second definition document information IS2 may also be textual information such as the project proposal information IK.

[0105] Furthermore, the first terminal 10 may perform the processing in steps S22 and S23 on behalf of the first server 210.

[0106] Furthermore, LLM 6328 may perform fact-checking of multiple second-document information items I2-1 to I2-N, as well as evaluation of multiple second-document information items I2-1 to I2-N. In this case, a prompt for evaluation is prepared. The prompt for evaluation is input to LLM 6328. The prompt for evaluation determines, for example, whether multiple second-document information items I2-1 to I2-N contain appropriate information. Appropriate information includes, for example, whether multiple second-document information items Tx2-1 to Tx2-N are related to areas included in theme information ITE, or whether they are in line with the content of proposal information IK. In addition, LLM 6 may evaluate multiple second-document information items I2-1 to I2-N by scoring them. [Explanation of symbols]

[0107] 1: System 10: Terminal 1 12: Control circuits 14:Memory circuit 16: Network Interface 18: Graphics Processing Unit 20: Display 26:Operation section 28: Touch panel 210: Server 1 212: Control circuit 214:Memory circuit 216: Network Interface 220: Definition document information acquisition means 222: Configuration information generation instruction means 224: Second text information generation instruction means 226: Means of obtaining theme information 228: Fact-checking methods 230: Second text information acquisition means 232: Means of obtaining project proposal information 234: First Information Gathering Method 236: First text information generation means 238: Second Information Gathering Method 310: Second Server 312: Control circuit 314:Memory circuit 316: Network Interface 320: Second Server I1: 1st text information I2-1~I2-N: 2nd text information IFC: Fact-checking information II1: First Input Information II2: Second Input Information II3: Third Input Information IK: Project Proposal Information IP: Image information IR1: First Collection Information IR2: Second Collection of Information IS1: First definition document information IS2:Second definition information IST: Configuration Information ITE: Theme Information P1: First prompt P2: Second prompt P3: Third Prompt PG: Program Tx1: 1st sentence Tx2-1~Tx-N: 2nd sentence

Claims

1. The system consists of one or more computers. The first sentence is about travel and includes multiple second sentences. The control circuits of the one or more computers described above are: Obtain definition information that shows the policy of the first document mentioned above, Based on the definition document information, instructions are given to the first LLM (Large Language Models) to generate configuration information indicating the policies of the multiple second documents. Based on the configuration information and definition information generated by the first LLM, instructions are given to the second LLM to generate a plurality of second document information representing the plurality of second documents. The plurality of second sentence information generated by the second LLM is acquired. system.

2. The control circuits of the one or more computers described above are: Obtain theme information indicating the theme of the first document, Based on the definition information and the theme information, instructions are given to the first LLM to generate configuration information indicating the policies of the multiple second documents. The system according to claim 1.

3. The control circuits of the one or more computers described above are: By combining the aforementioned plurality of second sentences, first sentence information representing the first sentence is generated. The system according to claim 1.

4. The control circuits of the one or more computers described above are: Retrieve the project proposal information entered by the user. The aforementioned proposal information is converted into a data exchange format using the 3rd LLM to obtain the aforementioned definition information. The system according to claim 2.

5. The first article above is a travel article. The aforementioned theme information includes an area to be entered by the user. The system according to claim 2.

6. The first article above is a travel article. The aforementioned theme information includes an area to be entered by the user, The aforementioned project proposal information includes a project proposal for generating travel articles related to the area included in the aforementioned theme information. The system according to claim 4.

7. The first sentence above contains multiple items, The aforementioned multiple second sentences correspond to the aforementioned multiple items, The system according to claim 4.

8. The aforementioned project proposal information includes multiple paragraphs, The control circuits of the one or more computers described above are: Based on the multiple paragraphs included in the project proposal information, the 3LLM generates the definition document information including the multiple items. The system according to claim 7.

9. The control circuits of the one or more computers described above are: Based on the definition information and theme information, the first collection information is collected using the fourth LLM. Based on the definition information, the theme information, and the first collected information, instructions are given to the first LLM to generate the configuration information. The system according to claim 2.

10. The first article above is a travel article. The definition information includes instructions for generating text about the area identified by the theme information and instructions for creating text about multiple spots included in the area identified by the theme information. The control circuits of the one or more computers mentioned above: First collection information relating to the area identified by the thematic information and the multiple spots is collected using the 4th LLM. The system according to claim 9.

11. The control circuits of the one or more computers described above are: Based on the definition information, theme information, and configuration information, the second collection information is collected using the fifth LLM. Based on the configuration information and the second collected information, instructions are given to the second LLM to generate the plurality of second sentence information. The system according to claim 2.

12. The second type of collected information includes external information existing on the internet and internal information stored on servers whose external access is restricted. The aforementioned internal information includes the first sentence that was generated in the past. The system according to claim 11.

13. The control circuits of the one or more computers described above are: When collecting the second collection information using the 5th LLM, information that does not conform to predetermined conditions will not be collected as the second collection information. The system according to claim 12.

14. The control circuits of the one or more computers described above are: The sixth LLM is used to perform fact-checking of the multiple second documents. The system according to any one of claims 1 to 3.

15. The first article above is a travel article. At least a portion of the aforementioned multiple second pieces of information includes links to information about travel products. The system according to any one of claims 1 to 3.

16. The configuration information includes instructions for causing the second LLM to output one or more image pieces of information related to the plurality of second sentences. At least a portion of the plurality of second textual information includes one or more image information. The system according to any one of claims 1 to 3.

17. The configuration information includes instructions to cause the second LLM to output the URLs referenced during the generation of the multiple second sentences. At least a portion of the plurality of second document information includes URLs referenced by the second LLM when generating the plurality of second documents. The system according to any one of claims 1 to 3.

18. It is a program, The first sentence is about travel and includes multiple second sentences. The program is installed in the control circuits of one or more computers. Obtain definition information that shows the policy of the first document mentioned above, Based on the definition document information, instructions are given to the first LLM (Large Language Models) to generate configuration information indicating the policies of the multiple second documents. Based on the configuration information and definition information generated by the first LLM, instructions are given to the second LLM to generate a plurality of second document information representing the plurality of second documents. The plurality of second sentence information generated by the second LLM is to be obtained. program.

19. A method for producing textual information, The first sentence is about travel and includes multiple second sentences. The above-mentioned method for producing text information involves a control circuit of one or more computers, Obtain definition information that shows the policy of the first document mentioned above, Based on the definition document information, instructions are given to the first LLM (Large Language Models) to generate configuration information indicating the policies of the multiple second documents. Based on the configuration information and definition information generated by the first LLM, instructions are given to the second LLM to generate a plurality of second document information representing the plurality of second documents. The plurality of second sentence information generated by the second LLM is to be obtained. Text information production method.

Images