Information processing device and program

Abstract

To provide an information processing device capable of generating a more appropriate sentence based on a user input sentence, and a program.SOLUTION: In a document generation system in which a user device and a response generation device are respectively connected to a sentence generation device via a network so as to enable mutual communication, the document generation device being an information processing device comprises: an input acquisition unit which acquires a plurality of user input sentences respectively associated with a plurality of items from a user; and a sentence generation unit which generates a first instruction sentence instructing to generate a first sentence of a format including at least a plurality of sentences tailored to the plurality of items on the basis of the plurality of user input sentences, and inputs the first instruction sentence to a trained model to generate the first sentence.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus and a program.

Background Art

[0002] In recent years, dialogue AI services using generative Artificial Intelligence have become available.

[0003] For example, Patent Document 1 discloses a technique of providing an answer while referring to an appropriate document in response to a user's question using such generative AI.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] As described above, it is expected to use generative AI to generate an appropriate answer, that is, appropriate text generation. On the other hand, if the question (text) (also called an instruction (text) or a prompt (text)) input to the generative AI is inappropriate, there is a risk that an incomprehensible answer (text) or the like will be generated.

[0006] In view of the above, an object of the present disclosure is to provide a technique capable of generating a more appropriate text based on a user input text.

Means for Solving the Problems

[0007] An information processing apparatus according to an aspect of the present disclosure includes an input acquisition unit that acquires a plurality of user input sentences respectively associated with a plurality of items from a user, and generates a first instruction sentence for generating a first sentence in a format including at least a plurality of sentences adapted to the plurality of items based on the plurality of user input sentences, and inputs the first instruction sentence to a learned model to generate the first sentence, and a sentence generation unit.

Effect of the Invention

[0008] According to an aspect of the present disclosure, a more appropriate sentence can be generated based on a user input sentence.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11A

Figure 11B

Figure 12

Figure 13

Figure 14

Mode for Carrying Out the Invention

[0010] (Embodiment) <Summary> According to an embodiment of the present disclosure, an information processing apparatus (for example, the user device 10 or the sentence generation apparatus 20 described later) includes an input acquisition unit and a sentence generation unit. The input acquisition unit acquires a plurality of user input sentences respectively associated with a plurality of items from a user. The sentence generation unit generates an instruction sentence for generating a sentence in a format including at least a plurality of sentences adapted to the plurality of items based on the plurality of acquired user input sentences. Further, the sentence generation unit inputs the generated instruction sentence into a learned model (for example, a large - language model) to generate the above - mentioned sentences.

[0011] In this way, the information processing apparatus generates a sentence in a format including at least a plurality of sentences adapted to the plurality of items corresponding to each of the plurality of user input sentences from the user, that is, generates a sentence divided into a plurality of sentences according to the plurality of items. Thereby, a more appropriate sentence based on the user input sentence can be generated.

[0012] <Configuration of Educational Support System> FIG. 1 is a block diagram showing a schematic configuration example of a sentence generation system according to an embodiment of the present disclosure.

[0013] As shown in FIG. 1, the sentence generation system 1 includes a user device 10, a sentence generation device 20, and an answer generation device 30. The user device 10 and the sentence generation device 20 are communicably connected to each other via a network 40, and the sentence generation device 20 and the answer generation device 30 are communicably connected to each other via the network 40. The network 40 may be a LAN (Local Area Network), the Internet, an intranet, a VPN (Virtual Private Network), a mobile communication network, or the like, or a combination thereof.

[0014] Based on an input sentence (referred to as a user input sentence) from the user of the user device 10, the sentence generation system 1 generates a more appropriate draft, performs correction on the draft, generates a more appropriate final sentence, and presents the generated final sentence to the user of the user device 10.

[0015] In this embodiment, the text generation system 1 generates and presents a final text for publicity based on the input text from the user of the user device 10. For example, when creating a press release for a company or the like, the person in charge of a project (for example, a product planner or developer) and the person in charge of publicity may create a draft while negotiating and adjusting through meetings and file exchanges. However, the person in charge of the project may lack the perspective of publicity and may not obtain appropriate information for publicity. As a result, the person in charge of publicity needs to create a text in an appropriate expression for publicity, which imposes a burden. On the other hand, with the advent of generative AI, it is expected to use generative AI to generate such texts. However, when specialized terms, business-specific expressions, industry-specific expressions, etc. are required in the generated text, it is difficult to obtain a text with content that satisfies the user from the generative AI. Also, as conditions for text generation, when the specification of detailed terms and notations for the text increases, it often occurs that the generative AI generates a text that does not satisfy some of the conditions. Also, since text generation by generative AI is generation based on probability, there is no guarantee that the desired notations and terms to be included will always be output. As described below, this embodiment can also handle such a situation.

[0016] The user device 10 has a configuration that can be connected to the text generation device 20 via, for example, a dedicated application installed on the user device 10 or a web browser. For example, the following processing of the user device 10 may be performed via a dedicated application or a web browser.

[0017] The user device 10 obtains the input form by receiving screen data including the input form for generating the final text for publicity from the text generation device 20 via the network 40.

[0018] The user device 10 displays a screen including the acquired input form and receives a user input sentence input by the user of the user device 10 on the screen (input form). The user device 10 transmits the received user input sentence to the text generation device 20 via the network 40. In the present disclosure, "transmission" may be replaced with "output".

[0019] The user device 10 obtains the final text by receiving the final text based on the user input sentence transmitted from the text generation device 20 via the network 40. Then, the user device 10 outputs the final text (for example, the final text is displayed on a display device (display unit) provided in the user device 10).

[0020] Note that although one user device 10 is shown in FIG. 1, a plurality of user devices 10 may exist. The user device 10 may be referred to as an information terminal, a user terminal, a terminal, a client, a client device, etc.

[0021] The text generation device 20 is installed in, for example, a cloud environment. Alternatively, the text generation device 20 may be installed in an on-premises environment.

[0022] In response to a request (referred to as a text generation request) for generating a final text for publicity from the user device 10 (the user thereof), the text generation device 20 transmits screen data including an input form to the user device 10 via the network 40.

[0023] The text generation device 20 obtains the user input sentence by receiving the user input sentence from the user device 10 via the network 40.

[0024] The text generation device 20 generates a text generation instruction by converting the information included in the input form sent to the user device 10 and the user input sentence into a text generation instruction to be sent to the answer generation device 30 based on a predetermined conversion rule (which may be called an instruction generation rule). Note that the instruction may be appropriately rephrased as an instruction sentence, a prompt (sentence), a question (sentence), etc.

[0025] The text generation device 20 sends the generated text generation instruction to the answer generation device 30 via the network 40.

[0026] The text generation device 20 obtains a text by receiving, from the answer generation device 30 via the network 40, an answer (text of a publicity-oriented article) to the text generation instruction. The text generation device 20 generates a text in cooperation with the answer generation device 30 in this way.

[0027] The text generation device 20 generates a final text by performing correction on the obtained text. The correction of the text includes a first-stage correction by sending a text correction instruction to the answer generation device 30 to obtain a corrected text, and a second-stage correction (also called re-correction) by performing rule-based (logic-based) correction on the corrected text. Details of the correction to be performed will be described later.

[0028] The text generation device 20 sends the generated final text to the user device 10 via the network 40.

[0029] The answer generation device 30 is installed, for example, in a cloud environment. Alternatively, the answer generation device 30 may be installed in an on-premises environment.

[0030] The answer generation device 30 obtains a text generation instruction by receiving, from the text generation device 20 via the network 40, the text generation instruction.

[0031] The answer generation device 30 inputs the obtained text generation instruction into a learned large language model (LLM) to generate an answer (text of a publicity-oriented article) for the obtained text generation instruction.

[0032] The answer generation device 30 transmits the generated text to the text generation device 20 via the network 40.

[0033] The answer generation device 30 obtains a text correction instruction by receiving the text correction instruction from the text generation device 20 via the network 40.

[0034] The answer generation device 30 inputs the obtained text correction instruction into the learned LLM to generate an answer (corrected text) for the obtained text correction instruction.

[0035] The answer generation device 30 transmits the generated corrected text to the text generation device 20 via the network 40.

[0036] In this embodiment, the answer generation device 30 is a device that provides an existing dialogue-type AI service (generative AI) using an LLM. The text generation device 20 transmits an instruction (text generation instruction, text correction instruction) to the answer generation device 30 using an API (Application Programming Interface) for using the dialogue-type AI service, and receives an answer (text, corrected text) for the instruction from the answer generation device 30. Alternatively, the answer generation device 30 may be a device that provides a dialogue-type AI service by training an LLM as described above using learning data such as text data published on the web.

[0037] <Configuration of User Device> FIG. 2 is a block diagram showing a hardware configuration example of a user device according to an embodiment of the present disclosure. The user device 10 may be realized by any computer (computer) or information processing device such as a tablet, a smartphone, or a notebook PC (Personal Computer).

[0038] For example, as shown in FIG. 2, the user device 10 may include a storage device 11, a processing device 12, a user interface (UI) device 13, a communication device 14, and a bus 15. The storage device 11, the processing device 12, the UI device 13, and the communication device 14 may be interconnected via the bus 15.

[0039] The programs or instructions for realizing the functions and processes described above and below of the user device 10 may be downloaded from some external device (such as a server) via a network or the like. Also, such programs or instructions may be provided from a removable storage medium such as a CD-ROM (Compact Disc Read Only Memory) or a flash memory.

[0040] The storage device 11 is realized by a RAM (Random Access Memory), a flash memory, a hard disk drive, etc., and stores files, data, etc. used for the execution of the installed programs or instructions together with the installed programs or instructions. The storage device 11 may include a non-transitory storage medium.

[0041] The processing device 12 may be implemented by, for example, a general-purpose processor or controller (circuit), or a dedicated processor or controller (circuit). When implemented by a general-purpose processor, the processing device 12 may be implemented by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc., which may be composed of one or more processor cores. In this case, the processing device 12 executes the functions and processes of the user device 10 described above and below according to the programs or instructions stored in the storage device 11, data such as the parameters used to execute the programs or instructions, etc. Such a program may be a program that causes the computer or the processing device 12 to function as the user device 10 according to the present disclosure when executed by the computer or the processing device 12. Further, such a program may be a program that causes the computer or the processing device 12 to execute the method related to text generation according to the present disclosure when executed by the computer or the processing device 12.

[0042] The UI device 13 may be composed of input devices such as a keyboard, a mouse, a camera, a microphone, output devices such as a display, a speaker, a headset, a printer, input / output devices such as a haptic device, a touch panel, etc., and realizes the interface between the user of the user device 10 and the user device 10. For example, the user of the user device 10 operates the user device 10 by operating the graphical user interface (GUI) displayed on the display or the touch panel using a keyboard, a mouse, etc.

[0043] The communication device 14 may be implemented by various communication circuits that execute communication processing with external devices, communication networks such as the Internet, LAN, VPN, etc.

[0044] The above-described hardware configuration of the user device 10 is merely an example, and the user device 10 according to the present disclosure may be implemented by other suitable hardware configurations.

[0045] FIG. 3 is a block diagram showing an example of the functional configuration of a user device according to an embodiment of the present disclosure.

[0046] For example, as shown in FIG. 3, the user device 10 includes an input reception unit 101, an input form acquisition unit 102, an input sentence output unit 103, a sentence acquisition unit 104, and a display control unit 105. The input reception unit 101, the input form acquisition unit 102, the input sentence output unit 103, the sentence acquisition unit 104, and the display control unit 105 may be implemented as hardware by the processing device 12, may be implemented as software, or may be implemented as a combination of hardware and software.

[0047] The input reception unit 101 receives an input by a user of the user device 10 (for example, pressing a button for transmitting a sentence generation request to the sentence generation device 20, a user input sentence input to an input form, pressing a button for transmitting the user input sentence to the sentence generation device 20, etc.). When the input is pressing a button (not shown) for transmitting a sentence generation request to the sentence generation device 20, the input reception unit 101 transmits the sentence generation request to the sentence generation device 20 via the communication device 14. When the input is pressing a button (for example, the "generate" button 1203 shown in FIG. 12) for transmitting the user input sentence to the sentence generation device 20, the input reception unit 101 outputs the received user input sentence to the input sentence output unit 103. The input reception unit may be referred to as an input acquisition unit.

[0048] When a text generation request is sent to the text generation device 20, the input form acquisition unit 102 acquires an input form by receiving, from the text generation device 20 via the communication device 14, screen data including the input form for generating the final text for publicity. The input form acquisition unit 102 outputs the screen data including the acquired input form to the display control unit 105. The screen data also includes data indicating a button for sending the user input text to the text generation device 20.

[0049] The input text output unit 103 sends the user input text output by the input reception unit 101 to the text generation device 20 via the communication device 14.

[0050] When the user input text is sent to the text generation device 20, the text acquisition unit 104 acquires the final text by receiving, from the text generation device 20 via the communication device 14, the final text based on the sent user input text. Also, when there are items for which the user of the user device 10 is entrusted with a correction judgment, the text acquisition unit 104 also receives (acquires) those items from the text generation device 20 together with the final text. Note that the final text (and items) are included in the screen data and acquired by the text acquisition unit 104. The text acquisition unit 104 outputs the screen data including the acquired final text and, if any, the items to the display control unit 105.

[0051] The display control unit 105 causes the display device (display unit) included in the UI device 13 to display the screen data including the input form output by the input form acquisition unit 102. The user input text from the user of the user device 10 is input into the input form thus displayed. The display control unit 105 causes the display device (display unit) to display the screen data including the final text and, if any, the items output by the text acquisition unit 104. An example of the display screen will be described later with reference to FIG. 12.

[0052] <Configuration of the text generation device> FIG. 4 is a block diagram showing an example of the hardware configuration of a text generation device according to an embodiment of the present disclosure. The text generation device 20 may be realized by a computing device (computer, calculator) such as a server or an information processing device.

[0053] For example, as shown in FIG. 4, the text generation device 20 may include a storage device 21, a processing device 22, a UI device 23, a communication device 24, and a bus 25. The storage device 21, the processing device 22, the UI device 23, and the communication device 24 may be interconnected via the bus 25.

[0054] The programs or instructions for realizing the functions and processes described above and below of the text generation device 20 may be downloaded from some external device (for example, a server) via a network or the like. Also, such programs or instructions may be provided from a removable storage medium such as a CD-ROM or a flash memory.

[0055] The storage device 21 is realized by a RAM, a flash memory, a hard disk drive, etc., and stores files, data, etc. used for the execution of the installed programs or instructions together with the installed programs or instructions. The storage device 21 may include a non-transitory storage medium.

[0056] The processing device 22 may be realized by, for example, a general-purpose processor or controller (circuit), or may be realized by a dedicated processor or controller (circuit). When realized by a general-purpose processor, the processing device 22 may be realized by one or more CPUs, GPUs, processing circuits, etc. that may be composed of one or more processor cores. In this case, the processing device 22 executes the functions and processes of the text generation device 20 described above and below according to the programs or instructions stored in the storage device 21, data such as parameters used to execute the programs or instructions, etc. Such a program may be a program that causes the computer or the processing device 22 to function as the text generation device 20 according to the present disclosure when executed by the computer or the processing device 22. Also, such a program may be a program that causes the computer or the processing device 22 to execute the method related to text generation according to the present disclosure when executed by the computer or the processing device 22.

[0057] The UI device 23 may be composed of input devices such as a keyboard, mouse, camera, microphone, etc., output devices such as a display, speaker, headset, printer, etc., input / output devices such as a haptic device, touch panel, etc., and realizes the interface between the user of the text generation device 20 and the text generation device 20. For example, the user operates the text generation device 20 by operating the GUI displayed on the display or touch panel using a keyboard, mouse, etc.

[0058] The communication device 24 may be realized by various communication circuits that execute communication processing with external devices, communication networks such as the Internet, LAN, VPN, etc.

[0059] The above-described hardware configuration of the text generation device 20 is merely an example, and the text generation device 20 according to the present disclosure may be realized by other appropriate hardware configurations.

[0060] FIG. 5 is a block diagram showing a functional configuration example of a text generation device according to an embodiment of the present disclosure.

[0061] For example, as shown in FIG. 5, the document generation device 20 includes an input form output unit 201, an input sentence acquisition unit 202, a draft generation unit 203, a draft correction unit 204, a document output unit 205, and a specialized information storage unit 211. The input form output unit 201, the input sentence acquisition unit 202, the draft generation unit 203, the draft correction unit 204, and the document output unit 205 may be realized as hardware, software, or a combination of hardware and software by the processing device 22. The specialized information storage unit 211 may be realized by the storage device 21.

[0062] The input form output unit 201 receives a document generation request from the user device 10 (the user thereof), and in response to the document generation request, transmits screen data including an input form to the user device 10 via the communication device 24.

[0063] The input sentence acquisition unit (input acquisition unit) 202 acquires a user input sentence by receiving the user input sentence from the user device 10 via the communication device 24. The input sentence acquisition unit 202 outputs the acquired user input sentence to the draft generation unit 203.

[0064] The draft generation unit 203 generates a draft generation instruction by converting information included in the input form transmitted to the user device 10 and the user input sentence into a draft generation instruction to be transmitted to the answer generation device 30 based on a predetermined conversion rule. The conversion rule is stored in the specialized information storage unit 211.

[0065] Hereinafter, with reference to FIG. 8, the user input sentence input to the input form, the conversion rule, and the generation of the draft generation instruction will be described.

[0066] (a) of FIG. 8 shows an example of a user input sentence entered in the input form 800. In the input form 800, headings such as "Product Overview and Features" 801A and "Background of Development" 801B starting with "■" are described. Also, in the input form 800, text boxes 802A, 802B, etc. for entering user input sentences are provided corresponding to the headings 801A, 801B, etc., respectively. The user of the user device 10 inputs, for example, according to the contents of the headings 801A, 801B, etc., into the text boxes 802A, 802B, etc., in a list form, the content that the user wants to include as a publicity-oriented sentence as the user input sentence.

[0067] (b) of FIG. 8 shows an example of a text generation instruction generated by converting information (heading) included in the input form and the input user input sentence into a text generation instruction 850 (prompt content for the generation AI) based on the conversion rule. For example, the text generation instruction 850 includes a header part 851, main body parts 852A and 852B, and a footer part 853. The header part 851 includes information provided by the text generation device 20 to improve the text generation accuracy. Specifically, the header part 851 includes information such as setting the role of the generation AI (for example, the answer processing unit 302 described later) of the answer generation device 30 (in this embodiment, as a public relations person, generating a title and text for a press release based on the user input sentence). Each of the main body parts 852A and 852B of the text generation instruction includes an item corresponding to the heading (indicated by "#") and a description regarding the content of the generated text, and is listed item by item so that the generation AI (answer processing unit 302) of the answer generation device 30 can easily recognize the description (user input sentence). In the example shown in FIG. 8, the user input sentence input to the heading 801A and the text box 802A corresponds to the main body part 852B, and the user input sentence input to the heading 801B and the text box 802B corresponds to the main body part 852A, and so on. The footer part 853 includes information provided by the text generation device 20 to improve the text generation accuracy. Specifically, the footer part 853 includes information such as conditions for the output content and format. The information described in the header part 851 and the footer part 853 may be fixedly provided, for example. Information described in such header part 851 and footer part 853, the headings, the correspondence between the headings (that is, the user input sentences input to the text boxes) and the main body part, the order of the main body part, etc. are stored in the specialized information storage unit 211 as conversion rules. Then, the text generation unit 203 refers to the conversion rules stored in the specialized information storage unit 211 and converts the information (heading) included in the input form and the input user input sentence into a text generation instruction.

[0068] The text generation unit 203 transmits the generated text generation instruction to the answer generation device 30 via the communication device 24.

[0069] The text generation unit 203 obtains (generates) a text by receiving, from the response generation device 30 via the communication device 24, a response to a text generation instruction (the text of a publicity-oriented sentence). The text generation unit 203 outputs the obtained (generated) text to the text correction unit 204.

[0070] In this way, by generating a text generation instruction so as to reflect the accuracy of the content and purport of the sentence, it is possible to improve the quality of the generated sentence (text), which contributes to the generation of a more appropriate sentence.

[0071] The text correction unit 204 generates a final text by performing correction on the text output by the text generation unit 203. The specific content of the correction will be described below.

[0072] First, the text correction unit 204 generates a text correction instruction for performing correction on the text, and transmits the generated text correction instruction to the response generation device 30 via the communication device 24. The text correction unit 204 obtains a corrected text by receiving, from the response generation device 30 via the communication device 24, a response (corrected text) to the text correction instruction. This process corresponds to the first-stage correction described above.

[0073] For example, the text correction unit 204 may generate a text correction instruction indicating that one or more of a plurality of types of corrections, such as correction of notation variations, correction of redundant expressions, correction of typos and omissions, etc., are to be performed on the text, and transmit it to the response generation device 30 to obtain a corrected text.

[0074] As a first example, as shown in FIG. 9(a), the text correction unit 204 may generate a text correction instruction indicating that a plurality of types of corrections are to be performed a plurality of times (for example, three times in the example shown in FIG. 9(a)) in order for each correction, and transmit it to the response generation device 30 to obtain a corrected text.

[0075] As a second example, as shown in FIG. 9(b), the text correction unit 204 may obtain a corrected text by generating a text correction instruction to perform a plurality of types of corrections in a batch a plurality of times (for example, three times in the example shown in FIG. 9(b)) for a plurality of types of corrections and transmitting the instruction to the response generation device 30.

[0076] As a third example, the text correction unit 204 may perform correction by repeatedly generating an instruction for the correction while changing the instruction for the correction to limited or specific conditions. The third example may be applied to each correction in the first example described above, or may be applied to each batch correction in the second example described above. Hereinafter, with reference to FIG. 10, an example in which the correction is a correction of character variation will be described. For example, it is assumed that the text generation unit 203 (and the response generation device 30) generates the text shown in FIG. 10(a). Next, the text correction unit 204 generates a simple text correction instruction shown in FIG. 10(b) and transmits it to the response generation device 30, and obtains a first intermediate corrected text shown in FIG. 10(c) from the response generation device 30. In the first intermediate corrected text, from the original text, "our company" (one place) is corrected to "Dai Nippon Printing Co., Ltd.", and "Dai Nippon Printing" (one place) is corrected to "Dai Nippon Printing Co., Ltd.", for a total of two corrections. Next, the text correction unit 204 generates a text correction instruction (an instruction to correct after indicating that it has been corrected once) shown in FIG. 10(d) and transmits it to the response generation device 30, and obtains a second intermediate corrected text shown in FIG. 10(e) from the response generation device 30. In the second intermediate corrected text, from the first intermediate corrected text, "LLM" (one place) is corrected to "large language model", for a total of one correction. Next, the text correction unit 204 generates a text correction instruction (an instruction to perform correction by focusing on portions that have not changed compared to the original text) shown in FIG. 10(f) and transmits it to the response generation device 30, and obtains a final corrected text shown in FIG. 10(g) from the response generation device 30. In the final corrected text, there are no correction portions from the original text. In this way, by repeatedly performing correction while changing the text correction instruction to limited or specific conditions, it is possible to reduce or prevent correction omission and improve the accuracy of correction.

[0077] As a fourth example, the text correction unit 204 may divide the text into predetermined chunks (for example, paragraphs), and perform the corrections described in the first to third examples above for each of the divided chunks.

[0078] Note that the combination of the number, order, type, etc. of corrections may be freely set according to the purpose (for example, manually by the system administrator or automatically by the text generation device 20 based on the field or the text format to be output). Thus, the generation of the text correction instruction to the acquisition of the corrected text may be repeatedly executed a plurality of times. In this case, the text generation accuracy is improved by the repeated text correction instructions. Alternatively, the generation of the text correction instruction to the acquisition of the corrected text may be executed only once.

[0079] Also, the same text correction instruction may be used in the corrections repeatedly performed a plurality of times. In this case, since different corrected texts may be generated in each correction, all the generated corrected texts may be displayed on the display unit of the user device 10, and the user of the user device 10 may be configured to select which corrected text to adopt.

[0080] Next, the text correction unit 204 performs rule-based correction on the (final) corrected text obtained as a result of performing the first-stage correction. This process corresponds to the second-stage correction described above.

[0081] For example, the text correction unit 204 may perform corrections of one or more contents set in the master and stored, for example, in the specialized information storage unit 211.

[0082] As a first example, with respect to the notation “~kudasai”, the text correction unit 204 may convert it to the notation “~ください”, and with respect to the notation “mezasu”, it may convert it to the notation “~めざし”. The text correction unit 204 may perform the conversion to the specific notation (recursively as necessary). Note that the specific notation may be set differently for each company. The above information may be stored in the specialized information storage unit 211 in the form of, for example, a conversion table.

[0083] As a second example, the text correction unit 204 may extract important terms from the user input text and the corrected text for comparison, and check whether they appear in both the user input text and the corrected text. The text correction unit 204 may extract, as important terms, terms that the user of the user device 10 has specified in advance (or, for example, based on TF-IDF (term frequency - inverse document frequency)) and that are desired to be included in the final text (for example, "large language model", "fine-tuning", "text correction", "operational efficiency improvement", and "DX promotion"), from the user input text and the corrected text for comparison. Suppose it is detected that the term "DX promotion" only does not appear in the corrected text. In this case, the text correction unit 204 may generate a text generation instruction with additional conditions (for example, to always include the term "DX promotion" in the text, and to change the DX promotion in the user input text to be enclosed in double quotes or square brackets), and may resend the generated text generation instruction to the answer generation device 30 to receive the text from the answer generation device 30, or may present it to the user of the user device 10 as an alert or the like. When receiving the text from the answer generation device 30, the above processing may be executed by the text generation unit 203.

[0084] As a third example, for proper nouns, technical terms, etc., the text correction unit 204 may perform correction as needed using the proper nouns, technical terms, etc. stored in the specialized information storage unit 211.

[0085] As a fourth example, the text correction unit 204 may present to the user of the user device 10, as an alert or the like, which of the similar terms such as "feature" and "characteristic", or "answer" and "respond" is appropriate.

[0086] The corrections described in the first example to the fourth example are merely illustrative, and other corrections may be performed. Also, the corrections described in the first example to the fourth example and other corrections not described may be arbitrarily combined and performed.

[0087] The text correction unit 204 generates, as the final text, the re-corrected text that is the result of performing rule-based correction.

[0088] The text correction unit 204 outputs the generated final text to the text output unit 205. For example, in the second example, the alert content when presenting the term "DX promotion" to the user of the user device 10, and the alert content in the fourth example, correspond to the items for which the user of the user device 10 described above is entrusted with the correction judgment. When the text correction unit 204 detects such items, it also outputs the detected items to the text output unit 205 together with the re-corrected text.

[0089] In this way, by performing correction so as to ensure the accuracy of terms and expressions in the text, it is possible to improve the quality of the generated text (final text), which contributes to the generation of more appropriate text.

[0090] The text output unit 205 transmits, via the communication device 24, the screen data including the final text output by the text correction unit 204 and the items if they exist to the user device 10.

[0091] The specialized information storage unit 211 stores the above conversion rules, specific notations, proper nouns, technical terms, similar terms, etc.

[0092] <Configuration of the answer generation device> FIG. 6 is a block diagram showing a hardware configuration example of an answer generation device according to an embodiment of the present disclosure. The answer generation device 30 may be realized by a computing device (computer, calculator) such as a server or an information processing device.

[0093] For example, as shown in FIG. 6, the answer generation device 30 may include a storage device 31, a processing device 32, a UI device 33, a communication device 34, and a bus 35. The storage device 31, the processing device 32, the UI device 33, and the communication device 34 may be interconnected via the bus 35.

[0094] The programs or instructions for implementing the functions and processes of the response generation device 30 described above and below may be downloaded from some external device (e.g., a server) via a network or the like. Further, such programs or instructions may be provided from a removable storage medium such as a CD-ROM or a flash memory.

[0095] The storage device 31 is realized by a RAM, a flash memory, a hard disk drive, etc., and stores the installed programs or instructions, as well as files, data, etc. used for the execution of the programs or instructions. The storage device 31 may include a non-transitory storage medium.

[0096] The processing device 32 may be realized by, for example, a general-purpose processor or a controller (circuit), or may be realized by a dedicated processor or a controller (circuit). When realized by a general-purpose processor, the processing device 32 may be realized by one or more CPUs, GPUs, processing circuits, etc. that may be composed of one or more processor cores. In this case, the processing device 32 executes the functions and processes of the response generation device 30 described above and below according to the programs or instructions stored in the storage device 31, data such as the parameters used to execute the programs or instructions, etc. Such a program may be a program for causing the computer or the processing device 32 to function as the response generation device 30 according to the present disclosure when executed by the computer or the processing device 32. Further, such a program may be a program for causing the computer or the processing device 32 to execute the method related to text generation according to the present disclosure when executed by the computer or the processing device 32.

[0097] The UI device 33 may be composed of input devices such as a keyboard, a mouse, a camera, and a microphone, output devices such as a display, a speaker, a headset, and a printer, input / output devices such as a haptic device and a touch panel, and realizes an interface between the user and the answer generation device 30. For example, the user operates the answer generation device 30 by operating the GUI displayed on the display or the touch panel using a keyboard, a mouse, etc.

[0098] The communication device 34 may be realized by various communication circuits that execute communication processing with external devices, communication networks such as the Internet, a LAN, and a VPN.

[0099] The above-described hardware configuration of the answer generation device 30 is merely an example, and the answer generation device 30 according to the present disclosure may be realized by other appropriate hardware configurations.

[0100] FIG. 7 is a block diagram showing a functional configuration example of an answer generation device according to an embodiment of the present disclosure.

[0101] For example, as shown in FIG. 7, the answer generation device 30 includes an instruction acquisition unit 301 and an answer processing unit 302. The instruction acquisition unit 301 and the answer processing unit 302 may be realized as hardware, software, or a combination of hardware and software by the processing device 32.

[0102] The instruction acquisition unit 301 acquires an instruction by receiving the instruction from the text generation device 20 via the communication device 34. The acquired instruction may include, for example, a text generation instruction and a text correction instruction. The instruction acquisition unit 301 outputs the acquired instruction to the answer processing unit 302.

[0103] The response processing unit 302 generates a response (a text for publicity or a corrected text) to an instruction (a text generation instruction or a text correction instruction) by inputting the instruction output by the instruction acquisition unit 301 into the LLM, which is a learned model. The response processing unit 302 transmits the generated response to the text generation device 20 via the communication device 34.

[0104] <Operation of the text generation system> Next, with reference to FIGS. 11A and 11B, an operation example of the text generation system 1 will be described.

[0105] FIGS. 11A and 11B are sequence diagrams showing operation examples of the text generation system according to an embodiment of the present disclosure.

[0106] In step S101, the user device 10 (input reception unit 101) receives a text generation request, which is an input from the user of the user device 10.

[0107] In step S102, the user device 10 (input reception unit 101) transmits the received text generation request to the text generation device 20.

[0108] When the text generation request is received by the text generation device 20, in step S103, the text generation device 20 (input form output unit 201) transmits screen data including an input form to the user device 10.

[0109] After the screen data is received by the user device 10 and displayed on the display unit, in step S104, the user device 10 (input reception unit 101) receives a user input text, which is an input from the user of the user device 10 to the displayed input form.

[0110] In step S105, the user device 10 (input text output unit 103) transmits the received user input text to the text generation device 20.

[0111] When the user input sentence is received by the text generation device 20, in step S106, the text generation device 20 (text generation unit 203) generates a text generation instruction by converting the information included in the input form transmitted to the user device 10 and the user input sentence into a text generation instruction based on a predetermined conversion rule.

[0112] In step S107, the text generation device 20 (text generation unit 203) transmits the generated text generation instruction to the response generation device 30.

[0113] When the text generation instruction is received by the response generation device 30, in step S108, the response generation device 30 (response processing unit 302) generates a text for a publicity-oriented article, which is a response to the text generation instruction, by inputting the text generation instruction into a learned LLM.

[0114] In step S109, the response generation device 30 (response processing unit 302) transmits the generated text to the text generation device 20.

[0115] When the text is received by the text generation device 20, in step S110, the text generation device 20 (text correction unit 204) generates a text correction instruction for performing correction on the text.

[0116] In step S111, the text generation device 20 (text correction unit 204) transmits the generated text correction instruction to the response generation device 30.

[0117] When the text correction instruction is received by the response generation device 30, in step S112, the response generation device 30 (response processing unit 302) generates a corrected text, which is a response to the text correction instruction, by inputting the text correction instruction into a learned LLM.

[0118] In step S113, the response generation device 30 (response processing unit 302) transmits the generated corrected text to the text generation device 20.

[0119] Steps S110 to S113 may be repeatedly executed.

[0120] When the corrected text is received by the text generation device 20, in step S114, the text generation device 20 (text correction unit 204) performs rule-based correction on the corrected text (re-corrects the corrected text) to generate the final text for publicity.

[0121] In step S115, the text generation device 20 (text output unit 205) transmits the screen data including the generated final text to the user device 10.

[0122] In addition, in step S114, when the text generation device 20 (text correction unit 204) detects an item for which the correction determination is entrusted to the user of the user device 10, in step S115, the text generation device 20 (text output unit 205) transmits the screen data including the generated final text and the detected item to the user device 10.

[0123] When the screen data is received by the user device 10, in step S116, the user device 10 (display control unit) causes the final text to be displayed on the display unit of the user device 10. In addition, when the detected item is also transmitted to the user device 10 in step S115, the user device 10 (display control unit) causes the detected item to be displayed on the display unit of the user device 10 together with the final text.

[0124] FIG. 12 is a diagram showing an example of a display screen including an input form and a final text displayed on a user device according to an embodiment of the present disclosure. FIG. 12 shows an example of a display screen displayed as a result of executing step S116 in FIG. 11B.

[0125] As shown in FIG. 12, a display screen example 1200 displayed on the display unit of the user device 10 includes headings 1201A and 1201B, text boxes 1202A and 1202B into which a user input sentence is input, a sentence generation button 1203, a final sentence display area 1204, and a detection item display area 1205. The headings 1201A and 1201B and the text boxes 1202A and 1202B form an input form. At the stage when a sentence generation request is sent to the sentence generation device 20, only the left side of the drawing (headings 1201A and 1201B, text boxes 1202A and 1202B, sentence generation button 1203, etc.) is displayed. When the sentence generation button 1203 is pressed, the right side of the screen (final sentence display area 1204, detection item display area 1205, etc.) is also displayed. The text generation instruction and the text correction instruction generated by the text generation device 20 are not displayed on the display screen example 1200.

[0126] The headings 1201A and 1201B and the text boxes 1202A and 1202B each have some differences, but correspond to the headings 801A and 802B and the text boxes 802A and 802B shown in FIG. 8(a).

[0127] The sentence generation button 1203 is a button for sending the user input sentence to the sentence generation device 20. When the sentence generation button 1203 is pressed, the user input sentences input into the text boxes 1202A and 1202B, etc. are sent to the sentence generation device 20 (step S105 in FIG. 11A), and then steps S106 to S116 in FIGS. 11A and 11B are executed.

[0128] In the final text display area 1204, the final text generated as a result of executing step S114 in FIG. 11B is displayed. Therefore, the final text displayed in the final text display area 1204 reflects the results of the first-stage proofreading and the second-stage proofreading. In the final text display area 1204, a final text in a form including at least a plurality of texts based on a plurality of user input texts, corresponding to a plurality of headings 1201A, 1201B, etc. corresponding to the plurality of input user input texts respectively corresponding to a plurality of text boxes 1202A, 1202B, etc., is displayed. This is because a final text in such a form is generated based on the text generation instruction generated by the text generation unit 203.

[0129] In the detection item display area 1205, when an item for which the user of the user device 10 is entrusted with a correction determination is detected in step S114 in FIG. 11B, the detected item is displayed.

[0130] As described above, according to the embodiment of the present disclosure, for a plurality of user input texts from a user, a text is generated in a form including at least a plurality of texts corresponding to a plurality of items associated with each of the plurality of user input texts, that is, a text divided into a plurality of texts according to a plurality of items is generated. Thereby, a more appropriate text based on the user input text can be generated.

[0131] <Modification Example> Hereinafter, a modification example of the above embodiment will be described.

[0132] [Modification Example 1] The text correction unit 204 may perform supplementary explanations or paraphrases for technical terms in the text or the corrected text using technical information. The text correction unit 204 may generate instructions for supplementary explanations or paraphrases, send the generated instructions to the answer generation device 30, and receive an answer to perform supplementary explanations or paraphrases, or may perform supplementary explanations or paraphrases based on rules using the technical information stored in the technical information storage unit 211. When an answer is received from the answer generation device 30, the above processing may be executed by the text generation unit 203. For example, in the case where fine-tuning corresponds to the above technical terms, as shown in Fig. 13(a), supplementary explanations may be inserted into the text, or as shown in Fig. 13(b), notations representing annotations may be added after the term, and supplementary explanations may be inserted at the end. By doing so, when a person who does not ultimately have specialized knowledge reads the generated text, it becomes easier for that person to understand the text.

[0133] [Modification Example 2] The text generation device 20 may store a text generation instruction, a text correction instruction, a text, a corrected text, a final text, etc. as log information, and may utilize the log information when generating a text generation instruction. For example, the user of the user device 10 may evaluate the log information of the text generation instruction (input) and the text (output) (for example, it may be recorded in the text generation device 20 as a high evaluation or a low evaluation). Then, the text generation unit 203 may add the input / output with a high evaluation from the user of the user device 10 as a good example (oneshot or fewshot) to the text generation instruction. For example, it is assumed that the text correction instruction shown in Fig. 14(a) and the text shown in Fig. 14(b) are highly evaluated by the user of the user device 10. In this case, as shown in Fig. 14(c), the text generation unit 203 may include the text correction instruction shown in Fig. 14(a) and the text shown in Fig. 14(b) as input / output examples, and may generate a text correction instruction including a description indicating that the text is generated with reference to the input / output examples in the header part. Conversely, the input / output evaluated as having a low evaluation may be added to the text correction instruction as a bad example. By doing so, the accuracy of text generation can be further improved, and the quality of the text can be further improved.

[0134] [Modification Example 3] Some or all of the functions of the user device 10, the text generation device 20, and / or the answer generation device 30 may be included in other devices. For example, all of the above-described functions of the text generation device 20 may be included in the user device 10. Also, for example, the functions related to answer generation of the answer generation device 30 may be included in the user device 10 or the text generation device 20.

[0135] [Modification Example 4] In the above-described embodiment, an example of generating a text for publicity has been described, but the present disclosure is not limited to this example. For example, the generated text may be a text described in various contracts, reports, FAQ documents, invention proposals, etc., and headings, conversion rules, specialized information, etc. corresponding to these documents may be set, stored, etc., and based on such headings, conversion rules, specialized information, etc., text generation, first-stage proofreading, second-stage proofreading, etc. may be performed.

[0136] [Other Modification Examples] The names of the above-described devices (user device 10, text generation device 20, etc.), functional units constituting the devices (input reception unit 101, input form acquisition unit 102), etc. are merely examples and may be changed as appropriate. Also, two or more of the functional units constituting the device may be combined and integrated, or one functional unit may be divided into a plurality of functional units. For example, the text generation unit 203 and the text proofreading unit 204 may be integrated into a functional unit such as a text generation unit. Also, the transmission function and / or reception function possessed by each functional unit may be summarized as a notification unit, a notified unit, a communication unit, a transmission / reception unit, a transmission unit, a reception unit, etc.

[0137] The order of the steps in the flowchart shown and described above is merely an example and may be changed as appropriate, or some steps may be executed in parallel.

[0138] As described above, according to one aspect of the present disclosure, a more appropriate text can be generated based on the user input text.

[0139] The embodiments have been described above with reference to the drawings, but the present disclosure is not limited to such examples. It is obvious that those skilled in the art can conceive of various modification examples or correction examples within the scope described in the claims. Such modification examples or correction examples are also understood to belong to the technical scope of the present disclosure. Also, within the scope not departing from the gist of the present disclosure, each component or each matter in the embodiments (including modification examples or correction examples) may be arbitrarily combined.

Explanation of Signs

[0140] 1 Article generation system 10 User device 101 Input reception unit 102 Input form acquisition unit 103 Input sentence output unit 104 Article acquisition unit 105 Display control unit 20 Article generation device 201 Input form output unit 202 Input sentence acquisition unit 203 Draft generation unit 204 Draft correction unit 205 Article output unit 211 Specialized information storage unit 30 Answer generation device 301 Instruction acquisition unit 302 Answer processing unit

Claims

1. An input acquisition unit that acquires a plurality of user input sentences respectively associated with a plurality of items from a user; Based on the plurality of user input sentences, generate a first instruction sentence for generating a first sentence in a format including at least a plurality of sentences corresponding to the plurality of items, and input the first instruction sentence into a learned model to generate the first sentence. A sentence generation unit; An information processing apparatus comprising:

2. The sentence generation unit generates a second instruction sentence for performing correction on the first sentence, inputs the second instruction sentence into the learned model, and generates a second sentence obtained by correcting the first sentence. The information processing apparatus according to claim 1.

3. The sentence generation unit generates a third instruction sentence for performing the correction on the second sentence, inputs the third instruction sentence into the learned model, and generates a third sentence obtained by correcting the second sentence. The information processing apparatus according to claim 2.

4. The sentence generation unit generates a fourth instruction sentence for performing the correction on an uncorrected portion of the second sentence, inputs the fourth instruction sentence into the learned model, and generates a fourth sentence obtained by correcting the second sentence. The information processing apparatus according to claim 2.

5. The correction includes at least one of correction of writing variations, correction of redundant expressions, and correction of spelling mistakes and omissions. The information processing apparatus according to claim 2.

6. The sentence generation unit performs rule-based correction on the second sentence and generates a fifth sentence obtained by correcting the second sentence. The information processing apparatus according to claim 2.

7. The rule-based correction includes converting a first predetermined term in the second sentence into a second predetermined term. The information processing apparatus according to claim 6.

8. The rule-based correction includes detecting a term that exists in the user input sentence and does not exist in the second sentence. The information processing apparatus according to claim 6.

9. Acquiring a plurality of user input sentences respectively associated with a plurality of items from a user; Generating a first instruction sentence for generating a first sentence in a format including at least a plurality of sentences corresponding to the plurality of items based on the plurality of user input sentences; Inputting the first instruction sentence into a learned model to generate the first sentence; A program for causing an information processing apparatus to execute a process including:

Images