Information processing system, information processing method, and information processing program

Abstract

By generating questions based on semantic criteria and prioritizing their necessity, this system enables the optimization and efficiency of document generation. [Solution] The information processing system is an information processing system that generates documents based on dialogue with a user using a generative AI model. The information processing system performs an acceptance process to receive user input information, including free-form text, from the user, and a deficiency inference process to evaluate whether there is any deficiency in the information necessary for creating the target document using the generative AI model, in light of the information perspective. If a deficiency is evaluated, the system performs a question generation and presentation process to generate a new question sentence to supplement the deficiency using the generative AI model and present the generated question sentence to the user, an answer acquisition process to record the answer as natural language text as dialogue history, and a document generation process to generate the target document as unstructured natural language text using the generative AI model based on the user input information and dialogue history.

Description

[[Technical Field]]

[0001] The present disclosure relates to an information processing system, an information processing method, and an information processing program. [[Background Art]]

[0002] Conventionally, there are technologies related to document generation.

[0003] As a technology related to document generation, for example, there is a technology related to creating job posting drafts based on interaction with a user. As an example of technology related to creating job posting drafts, there is a technology related to an information processing system or the like that can generate sentences tailored to job seekers (see Patent Document 1). The technology of Patent Document 1 discloses a process of inputting template candidates into a question information candidate generation model and causing the question information candidate generation model to output question information candidates. [[Prior Art Documents]] [[Patent Documents]]

[0004] [[Patent Document 1]] Japanese Patent No. 7618885 [[Summary of the Invention]] [[Problem to be Solved by the Invention]]

[0005] By the way, in conventional technologies related to document generation, pre-designed and accumulated question candidate templates are used for sentence generation methods. A document is generated by selecting and extracting template questions based on a user's attributes and input content, presenting them to the user, and fitting the obtained answers into predefined fixed phrases and slot structures (items).

[0006] However, using question templates presented the following challenges: For example, it required management costs to design branching flows in advance and to continuously maintain them. Also, because it relied on pre-determined and fixed question flows, it became difficult to flexibly engage in dialogue with unexpected cases, such as unique user backgrounds or diverse contexts. Furthermore, because the documents were generated step-by-step by fitting the obtained answers into standard templates or slots, the output documents tended to be uniform, making it difficult to create natural and expressive documents. In addition, unnecessary questions sometimes arose even when sufficient information had been entered.

[0007] The objective of this disclosure is to provide an information processing system, information processing method, and information processing program that enable the optimization and efficiency of document generation by generating questions for obtaining information necessary for document generation based on semantic criteria and performing the necessary actions as needed. [Means for solving the problem]

[0008] The information processing system disclosed herein is an information processing system that generates documents based on dialogue with a user using a generative AI model, and comprises at least one processor, the processor which, by reading a program, performs the following: reception processing to receive user input information, including free-form text, from a user; deficiency inference processing to evaluate whether there is a lack of information necessary for creating the target document by the generative AI model, using the user input information and the dialogue history based on the user input information as natural language text, in light of an information perspective that indicates a standard information perspective for creating the target document; question generation and presentation processing to generate a new question sentence to supplement the deficiency using the generative AI model if the deficiency inference processing evaluates that there is a deficiency, and present the generated question sentence to the user; answer acquisition processing to record the user's answer to the question sentence as natural language text as the dialogue history when an answer is obtained from the user; and document generation processing to generate the target document as an unstructured natural language sentence using the generative AI model based on the user input information and the dialogue history.

[0009] Furthermore, the information processing method disclosed herein is an information processing method in which at least one processor reads a program and generates a document based on a dialogue with a user using a generative AI model, wherein the computer receives user input information including free text from the user, uses the user input information and the dialogue history based on the user input information as natural language text through a deficiency inference process, evaluates whether there is a deficiency of information necessary for creating the target document using the generative AI model in light of an information perspective that indicates a standard information perspective for creating the target document, if a deficiency is evaluated by the deficiency inference process, generates a new question sentence to supplement the deficiency using the generative AI model, presents the generated question sentence to the user, records the user's answer to the question sentence as natural language text in the dialogue history when obtained, and generates the target document as an unstructured natural language sentence using the generative AI model based on the user input information and the dialogue history, and the computer executes the process.

[0010] Furthermore, the information processing program disclosed herein is an information processing program in which at least one processor reads the program and generates a document based on a dialogue with a user using a generative AI model, and the computer executes the following process: receiving user input information including free text from the user; using the user input information and the dialogue history based on the user input information as natural language text through a deficiency inference process; evaluating whether there is a deficiency in the information necessary for creating the target document using the generative AI model in light of an information perspective that indicates a standard information perspective for creating the target document; if a deficiency is evaluated by the deficiency inference process, the generative AI model generates a new question sentence to supplement the deficiency; presenting the generated question sentence to the user; and when the user's answer to the question sentence is obtained, recording it as natural language text in the dialogue history; and generating the target document as an unstructured natural language sentence using the generative AI model based on the user input information and the dialogue history. [Effects of the Invention]

[0011] The information processing system, information processing method, and information processing program disclosed herein provide the effect of optimizing and streamlining document generation by generating questions for obtaining information necessary for document generation based on semantic criteria and performing the necessary actions as needed. [Brief explanation of the drawing]

[0012] [Figure 1] Figure 1 is a block diagram showing the hardware configuration of the information processing system. [Figure 2] Figure 2 is a block diagram showing the functional configuration of the information processing system in this embodiment. [Figure 3] Figure 3 shows an example of a user interface for an input terminal used when interacting with a user using a generative AI model. [Figure 4] Figure 4 is a flowchart showing the processing flow by the information processing system. [Modes for carrying out the invention]

[0013] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. In each drawing, identical or equivalent components and parts are given the same reference numerals. Furthermore, the dimensional ratios in the drawings are exaggerated for illustrative purposes and may differ from actual ratios.

[0014] First, an overview of the embodiments of this disclosure will be provided. As described above, conventional document generation methods have various problems due to the use of templates. Therefore, in the embodiments of this disclosure, without designing and accumulating candidate questions or templates in advance, missing information is inferred from the user's input using a generation AI model, and question sentences are dynamically generated only when necessary, thereby generating documents as unstructured natural language sentences in a flexible and natural manner. In the following description of embodiments, an example will be given in which an information processing system generates documents created by a user, such as a resume, based on a dialogue with the user using a generation AI model. Note that the documents to be generated are not limited to resumes; the technology of this disclosure can also be applied to generating business reports, applications, proposals, or other explanatory documents in natural language.

[0015] Figure 1 is a block diagram showing the hardware configuration of the information processing system 100. As shown in Figure 1, the information processing system 100 includes a CPU (Central Processing Unit) 11, ROM (Read Only Memory) 12, RAM (Random Access Memory) 13, storage 14, input unit 15, display unit 16, and communication interface (I / F) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with one another. The information processing system 100 is not limited to a single server, but may be composed of multiple information processing devices connected to each other so as to be able to communicate with one another via a network N, and may be a configuration realized by cloud computing, for example. Furthermore, the information processing system 100 may also include an input terminal 150 operated by a user as part of its configuration.

[0016] The CPU 11 is a central processing unit that executes various programs and controls various components. Specifically, the CPU 11 reads a program from the ROM 12 or storage 14 and executes the program using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various calculations according to the program stored in the ROM 12 or storage 14. In this embodiment, the ROM 12 or storage 14 stores an information processing program.

[0017] ROM12 stores various programs and data. RAM13 temporarily stores programs or data as a working area. Storage14 consists of a storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various programs, including the operating system, and various data.

[0018] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used for various types of input. The display unit 16 is, for example, a liquid crystal display and displays various types of information. The display unit 16 may also function as the input unit 15 by employing a touch panel system.

[0019] The communication interface 17 is an interface for communicating with other devices such as terminals. For such communication, a wired communication standard such as Ethernet® or FDDI, or a wireless communication standard such as 4G, 5G, or Wi-Fi® may be used.

[0020] FIG. 2 is a block diagram showing the functional configuration of the information processing system 100 of the present embodiment. The information processing system 100 is connected to an input terminal 150 via a network N, and receives user interaction information from the input terminal 150. The information processing system 100 includes a storage unit 102, a reception processing unit 110, a deficiency inference processing unit 112, a question generation / presentation processing unit 114, an answer acquisition processing unit 116, a repetition unit 118, and a document generation processing unit 120. Note that the information processing system 100 may be configured to include the input terminal 150. The input terminal 150 is used by a user who is the creator of a document to be generated. The user who is the creator of the document to be generated herein is a user who is the user or a user who generates the document to be generated on behalf of the user, etc.

[0021] The storage unit 102 stores various types of information for creating a document to be generated. For example, setting data for configuring various input information (first input information to third input information) including instruction information such as prompts containing instruction content and constraints for a generative AI model, information perspectives (standard configuration) indicating perspectives of standard information for creating a document to be generated, user input information received as interaction data with a user who is the user, and an interaction history recording interactions with the user as natural language text are stored therein. The storage unit 102 may also store an information processing program for executing each process according to the present embodiment.

[0022] The information perspective will be described. The information perspective indicates perspectives of information that are generally easy to understand as being included in a document. The information perspective does not fix sentence structures, sentence examples, the number of paragraphs, description order, or expression content, and corresponds to guidelines that provide perspectives to be considered for a generative AI model. For example, when a resume is the document to be generated, the information perspective includes perspectives such as an overview of work, roles and scope of responsibility, points of ingenuity, and achievements and impacts. The information perspective (standard configuration) does not predefine sentence structure, sentence examples, expression content, paragraph configuration or description order unlike existing templates for question and answer, but means a conceptual configuration indicating perspectives of information to be included in the document to be generated.

[0023] In this description of the embodiment, information received from the user via the interactive user interface is referred to as user input information, while prompts and other elements used as input to the generated AI model are referred to as input information. Input information is not limited to prompts; it may also include extended interfaces such as API specifications, structured arguments, or function calls. This allows for flexible adaptation to future changes in the generated AI model specifications or extensions to the interface while maintaining the system's control structure.

[0024] The generative AI model used in this embodiment will now be described. For example, a general-purpose model such as a Large-Scale Language Model (LLM) can be used as the generative AI model. In this system, by providing the generative AI model with information perspectives and instruction information stored in the memory unit 102 as input information, specialized inference and question generation for specific purposes (e.g., resume creation) are achieved. This eliminates the need to build or train dedicated models for each document type or business domain, allowing for flexible and low-cost system configuration by utilizing existing general-purpose generative AI models. Furthermore, depending on the system's purpose and application, it is also possible to use a generative AI model that has been further trained (fine-tuned) to specialize in specific document formats or technical terms.

[0025] The reception processing unit 110 initiates a dialogue with the user via the input terminal 150 and receives user input information, including free-form text, from the user. Specifically, it receives user input information, including free-form text such as career history and skills, which the user has arbitrarily entered when creating the document to be generated. In this embodiment, the dialogue reception process does not rely on predetermined choices or a fixed question flow, and is characterized by its ability to initiate a dialogue based on the user's free input. The reception processing unit 110 receives the input information as natural language text and records the input information and the dialogue history based on that input information in the storage unit 102 for processing by the insufficiency inference processing unit 112, etc.

[0026] Figure 3 shows an example of the user interface for dialogue on the input terminal 150 when dialogue with a user takes place using a generative AI model. This figure shows an example of the user interface for dialogue displayed on the input terminal 150 according to this embodiment. The user interface for dialogue includes a dialogue history area 151 that displays the interaction with the user, and an input area 152 for the user to input free-form text. The input area 152 is not in a format where the user selects from predetermined options, but is configured as a text area where the user can arbitrarily input their own background, skills, etc. in natural language. When information is entered into the input area 152 by the user, the reception processing unit 110 receives it, displays its contents in the dialogue history area 151, stores it in the storage unit 102, and passes the information to the insufficient reasoning processing unit 112.

[0027] In this embodiment, the dialogue does not proceed mechanically according to a pre-prepared fixed list of questions or branching scenarios. Instead, the generating AI model sequentially interprets the meaning of the user's free-text input, evaluates the degree of information sufficiency, and then generates questions on the spot through semantic dialogue via the generating AI's inference. Specifically, the questions presented in the dialogue history area 151 are not extracted from a pre-held database, but are newly generated natural language text in real time by the question generation and presentation processing unit 114 using the generating AI model to supplement any missing items identified by the missing inference processing unit 112. This enables flexible in-depth information gathering that is difficult to achieve with a fixed flow, as it is tailored to each user's diverse background and unique context.

[0028] The deficiency inference processing unit 112 uses the user input information, including free-form text, received by the reception processing unit 110, and the dialogue history stored in the storage unit 102, as natural language text to evaluate whether there is any deficiency of information necessary for creating the document to be generated. Specifically, the deficiency inference processing unit 112 causes the generation AI model to infer the comprehensiveness or ease of understanding of the information contained in the current user input information and dialogue history, in light of the information perspectives stored in the storage unit 102. The deficiency inference processing unit 112 causes the generation AI model to determine the deficiency or excess of information from a semantic perspective by including evaluation criteria and inference instructions based on the information perspective as input information (first input information). If the inference results in an evaluation that there is an information deficiency, the process moves to the question generation / presentation processing unit 114 to perform processing to compensate for the deficiency. On the other hand, if the evaluation results in no information deficiency, the process moves to the document generation processing unit 120 without generating a question.

[0029] This section provides further details on comprehensiveness and comprehensibility, which are criteria used to determine insufficient information. Comprehensiveness refers to the degree to which predefined information perspectives for creating the target document (for example, each element such as job overview, role, innovations, and results in a resume) are included without omission in the user input information and dialogue history. The generating AI model is instructed to evaluate whether all necessary information items are present in light of the standard components of a document. Comprehensibility refers to the degree to which the descriptive content included in each information perspective is satisfied to the extent that it is concrete and objectively understandable to a third party. For example, it means that the generating AI model semantically evaluates whether it has sufficiently described not only the fact that a particular experience was "engaged," but also the specific challenges behind it, quantitative results using numerical data, or the motivations and causal relationships of a series of actions, which complement the context. By using these criteria, the insufficient information determination processing unit 112 can determine insufficient information not only based on the presence or absence of items, but also on the qualitative completeness of the generated document.

[0030] The question generation and presentation processing unit 114, when the deficiency inference processing unit 112 evaluates that there is a deficiency, generates a new question sentence using the generation AI model to supplement the deficiency, and presents the generated question sentence to the user via the dialogue user interface (dialogue history area 151) of the input terminal 150. Specifically, the question generation and presentation processing unit 114 includes instructions for textualization from an information perspective as input information (second input information) to the generation AI model, and also provides instructions to supplement the deficiency items identified by the deficiency inference processing unit 112, thereby dynamically creating a question sentence that is appropriate to the context of the dialogue without having to extract from a pre-stored set of question candidates. The presented question sentence functions as an auxiliary step to encourage the user to input new free-form text. Note that the second input information is constrained so that question generation is not performed if the deficiency inference processing unit 112 evaluates that there is no deficiency. The question generation and presentation process is positioned as an optional step to assist in the generation of the target document.

[0031] This section describes an example of how to configure the first and second input information as prompts to be input into the generating AI model. The following is an example of the first input information. "When creating a resume based on the information available so far, please infer any missing information necessary for the reader to fully understand the content. Analyze the user's input and dialogue history based on the following evaluation criteria. As evaluation criteria, assess the comprehensiveness of the information and its ease of understanding for third parties in light of predefined information perspectives (overview of work, role, innovations, and results). Then, perform an inference about missing information. In the inference, identify items that lack specificity or logical context in document generation, in light of the evaluation criteria." As described above, the first input information includes instructions for inferring the perspectives that are missing from the user input information and dialogue history in order to generate the target document, among the perspectives included in the information perspective.

[0032] The following is an example of the second input information. "If a deficiency is identified, please generate questions based on the information perspective. When generating questions, please create in-depth questions appropriate to the context without referring to the existing question candidate database. If it is evaluated that there is no information deficiency as a constraint, do not generate questions. If you do not generate questions, you may output 'No deficiencies' or an indication that no questions will be generated." As described above, the second input information includes instructions for textualization from an information perspective, and is constrained by the fact that it is not executed if the deficiency inference process determines that there are no deficiencies. In this way, the system determines whether it is possible to avoid creating a question.

[0033] The response acquisition processing unit 116 acquires the user's response to the question presented by the question generation / presentation processing unit 114, and records the response as dialogue history in the storage unit 102 as natural language text. Specifically, it stores the response entered via the input area 152 as dialogue history. Here, the dialogue history is recorded as raw natural language text without being structured by associating it with predefined items, slots, or keys. The recorded response is not immediately reflected in the generation of the target document, but is held in the storage unit 102 as provisional information and is used for subsequent re-evaluation by the insufficient reasoning processing unit 112 or for batch generation processing by the document generation processing unit 120.

[0034] The iteration unit 118 controls the system to repeatedly execute a series of processes performed by the reception processing unit 110, the deficiency inference processing unit 112, the question generation / presentation processing unit 114, and the answer acquisition processing unit 116 until sufficient information necessary for generating the target document is available. Specifically, it repeatedly asks questions to the user and acquires answers via the dialogue history area 151 of the dialogue interface until the deficiency inference processing unit 112 evaluates that there is no information deficiency, or until predetermined conditions (such as the number of dialogues or the degree of information sufficiency) are met. This makes it possible to dynamically accumulate natural language text information necessary for generating documents in bulk.

[0035] The document generation processing unit 120 generates a target document as unstructured natural language text using a generation AI model, based on user input information received by the reception processing unit 110 and the dialogue history accumulated by the answer acquisition processing unit 116. This process is executed all at once after integrating the dialogue history, including multiple questions and answers, when predetermined conditions are met through the repetition of the dialogue by the repetition unit 118. Specifically, the document generation processing unit 120 includes an instruction to document the dialogue history as unstructured natural language text as input information (third input information) to the generation AI model, and sets constraints such as not associating it with predefined items, slots, or keys, and not fitting it to predefined sentence structures or standard sentence templates. As a result, the generation AI model integrates all the information of the natural language text held as provisional information, constructs the sentence structure and expression based on the semantic content each time, and outputs the document. Furthermore, unstructured natural language sentences refer to natural language sentences in which a generative AI model generates sentence structure and expression each time based on semantic content, without being constrained by predefined templates, syntactic patterns, or slot structures. Such unstructured generated documents may differ each time in terms of the number of sentences, paragraph structure, expression, and order. Moreover, even with the same input, the syntax will not be the same, as the generative AI model constructs the sentence structure itself each time based on semantic content.

[0036] The following is an example of the third input information. Based on the user input and dialogue history to date, please organize the content into a natural-sounding document that is easy for the reader to understand, considering aspects such as the overview of the work, roles, innovations, and achievements, as if it were a general resume. Please judge whether the sentence structure, expression, number of paragraphs, and order of information are appropriate, referring to standard structures. Furthermore, as a constraint, please generate the text in a way that most naturally conveys the characteristics of the user's experience, without forcing it into specific headings, standard expressions, or order of content, or into any particular sentence structure or template. Also, please generate the text without mapping it to any predefined items, slots, or keys. As described above, the third input information includes instructions to write the question or answer as unstructured natural language text, and is constrained by the fact that it is not mapped to any predefined items, slots, or keys.

[0037] (Process flow) Next, the operation of the information processing system 100 will be explained. Figure 4 is a flowchart showing the processing flow by the information processing system 100. The CPU 11 reads a program from the ROM 12 or storage 14, loads it into the RAM 13, and executes it, thereby performing the document generation process. The CPU 11 functions as a part of the information processing system 100 to execute the following processes.

[0038] In step S100, the CPU 11 initiates a dialogue with the user using the generated AI model. Specifically, it performs a process to present an initial message to the user prompting them to enter free text via the dialogue user interface of the input terminal 150.

[0039] In step S102, the CPU 11 receives user input information from the input terminal 150, including free-form descriptions of work history and skills. The CPU 11 records the received user input information in the storage unit 102, and also generates and records a dialogue history based on the input information.

[0040] In step S104, the CPU 11 reads the first input information from the memory unit 102 and inputs the first input information to the generating AI model. This allows the CPU 11 to perform a deficiency inference process that evaluates whether there are any deficiencies in the current user input information and dialogue history from an information perspective, and generates an evaluation result. Here, the generating AI model is instructed to infer the comprehensiveness or ease of understanding of the information needed to create the document to be generated.

[0041] In step S106, the CPU 11 determines, based on the evaluation results, whether or not there is a lack of information necessary to create the document to be generated. If it is determined that there is a lack of information (step S106: YES), the process proceeds to step S108. On the other hand, if it is determined that there is no lack of information (step S106: NO), the process proceeds to step S112 to determine whether to continue the dialogue. Note that if there is no lack of information, the process may be terminated without going through step S112.

[0042] In step S108, the CPU 11 reads the second input information from the memory unit 102 and inputs it into the generated AI model to generate a new question sentence to fill in the identified missing items, and presents it to the user via the input terminal 150.

[0043] In step S110, the CPU 11 obtains the user's response (natural language text) to the presented question and records it in the storage unit 102 as dialogue history.

[0044] In step S112, the CPU 11 determines whether or not it is necessary to continue the dialogue. Specifically, in addition to the evaluation result of whether or not there are any deficiencies in step S106, it checks whether predetermined iteration conditions (number of dialogues, etc.) have been met. If it is determined that continuation is necessary, the process returns to step S102 and continues to acquire user input information. At this time, a notification may be issued prompting the user to input information other than questions. On the other hand, if it is determined that continuation is not necessary, the process proceeds to step S114.

[0045] In step S114, the CPU 11 reads third input information from the memory unit 102 based on the accumulated dialogue history and inputs the third input information to the generating AI model, thereby generating the target document as unstructured natural language sentences in a batch using the generating AI model.

[0046] As described above, the information processing system 100 according to this embodiment determines and generates questions for obtaining information necessary for document generation on a semantic basis, enabling optimization and efficiency of document generation. Furthermore, it realizes the generation of natural, flexible, and non-standard documents that are not constrained by templates.

[0047] Furthermore, because the generating AI model dynamically infers missing information from the user's input and generates questions without requiring prior design and storage of candidate questions or branching flows, flexible dialogue tailored to each user's diverse background and context becomes possible. Since the obtained answers are generated in batches as unstructured natural language sentences based on integrated information, rather than being fitted into fixed templates or slot structures, highly expressive and natural documents can be created instead of uniform ones. Additionally, management costs associated with designing and maintaining fixed question flows can be reduced. Moreover, by having the generating AI model evaluate whether or not there is missing information, unnecessary questions when sufficient information is available can be suppressed, enabling efficient document creation with minimal dialogue.

[0048] (modified version) A modified example will be described. In the embodiment described above, the case in which the user inputs user information in text format was used as an example, but the format of the input information is not limited to this. For example, the reception processing unit 110 may be configured to convert the user's spoken content into natural language text using speech recognition technology and acquire it.

[0049] Furthermore, multiple information perspectives (standard configurations) may be maintained depending on the type and purpose of the document to be generated, or additional perspectives may be added by the generation AI model. In the case of additions, for example, prior to the deficiency inference processing in step S104, the generation AI model considers and adds appropriate information perspectives for creating the document to be generated, based on the initial user input information and the content of the document to be generated. The generation AI model should be instructed to add appropriate information perspectives that take into account the semantic properties corresponding to the different generation context each time. For example, if the document to be generated is a resume, the AI ​​model should be instructed to add appropriate perspectives if any, taking into account the semantic properties at that time related to the role of a resume, such as highlighting one's career and skills, along with the user input information. If the document to be generated is a report, the AI ​​model should be instructed to add appropriate perspectives if any, taking into account the semantic properties at that time related to the role of a report, such as accurate communication of work progress and reporting frequency, along with the user input information. The deficiency inference processing unit 112 can be configured to evaluate whether or not there are deficiencies using the added information perspectives. This makes it possible to extract deficiencies flexibly and accurately in line with the content of the document to be generated.

[0050] Furthermore, a step may be included in which the results of the deficiency inference processing unit 112's evaluation of whether or not there is a deficiency are notified to the user, who is the creator of the document to be generated, and approval is obtained. Specifically, if the deficiency inference processing unit 112 identifies a deficiency in information, as a preliminary step before generating the question, a confirmation message such as "There is a deficiency in information regarding XX, may I ask additional questions?" is presented to the input terminal 150. The system is configured such that the question generation / presentation processing unit 114 generates the question only if the user gives an approval action such as "yes." This prevents the conversation from spreading in a direction unintended by the user.

[0051] Furthermore, a step may be added in which the user provides further revision instructions for the document generated by the document generation processing unit 120 using unstructured natural language sentences. For example, after the generated document is presented to the input terminal 150, the user inputs revision instructions in free text format, such as "Please reduce the use of technical terms and make it easier to understand" or "Please emphasize specific achievements." The document generation processing unit 120 receives these revision instructions as new input information and reconstructs the document using the generation AI model. Even in this case, by maintaining the characteristics of unstructured generation that do not depend on a specific template, it is possible to complete a document that is highly satisfactory to the user.

[0052] Furthermore, the technology disclosed herein is not limited to the embodiments described above, and various modifications and applications are possible without departing from the spirit of this invention.

[0053] Furthermore, although the present specification describes an embodiment in which the program is pre-installed, it is also possible to provide the program stored on a computer-readable recording medium. [Explanation of symbols]

[0054] 100 Information Processing Systems 102 Storage section 110 Reception Processing Unit 112 Insufficient Inference Processing Unit 114 Question generation and presentation processing unit 116 Response acquisition processing unit 118 Repeat section 120 Document Generation Processing Unit

Claims

1. An information processing system that generates documents based on interaction with a user using a generative AI model, It comprises at least one processor and memory, The aforementioned storage unit holds information perspectives and setting data for configuring the first input information, second input information, and third input information. The aforementioned information perspective represents information that indicates a standard information perspective for creating natural language text and generated documents. The aforementioned setting data includes, as a setting for the first input information, the setting of instructions for inferring deficiencies based on evaluation criteria for inferring which perspectives are missing from the information perspectives included in the information perspectives in order to generate the target document, in relation to the user input information and dialogue history. The aforementioned setting data includes, as settings for the second input information, instructions for textualization from the information perspective, and the setting of a first constraint that will not be executed if it is evaluated that there are no deficiencies by the deficiency inference process. The aforementioned setting data includes, as settings for the third input information, instructions for writing the user input information and the dialogue history, questions or answers into unstructured natural language sentences, and a setting for a second constraint that does not associate information with predefined items. The aforementioned processor, by reading the program, A reception process that receives user input information, including free-form text, from the user, The system uses the received user input information and the dialogue history based on said user input information as natural language text, reads the settings related to the information perspective and the first input information from the storage unit, sets first input information by combining the information perspective and the instruction for deficiency inference, uses the set first input information as input to the generation AI model, and performs a deficiency inference process to evaluate whether there is any deficiency of information necessary for creating the target document based on the output from the generation AI model. A question generation and presentation process that reads the settings related to the second input information from the memory unit, sets the second input information including the first constraint as an instruction, uses the set second input information as input to the generation AI model, and if the output from the generation AI model indicates that there is a deficiency through the deficiency inference process, generates a new question sentence to compensate for the deficiency, presents the generated question sentence to the user, and does not generate the question sentence if the deficiency inference process indicates that there is no deficiency. When the user's response to the aforementioned question is obtained, the response acquisition process records it as natural language text as the dialogue history, A document generation process that reads the settings related to the third input information from the memory unit, sets the third input information including the second constraint as an instruction, uses the set third input information as input to the generation AI model, and generates the target document as an unstructured natural language sentence based on the output from the generation AI model. An information processing system that performs [this action].

2. In the information processing system described in claim 1, An information processing system in which, in the setting of the first input information for the aforementioned deficiency inference processing, the first input information to be input to the generating AI model is set as an instruction to evaluate whether or not there is missing information, using at least one of the comprehensiveness of the information that should be included in the document to be generated and the ease of understanding as evaluation criteria.

3. In the information processing system described in claim 1, The aforementioned question generation and presentation process is executed only when a deficiency is determined to exist based on a conditional branching based on the result of the deficiency inference process's evaluation of whether or not there is a deficiency, and is not executed when the deficiency inference process determines that there is no deficiency, in this information processing system.

4. In the information processing system described in claim 1, In the document generation process, the third input information to be input to the generation AI model includes instructions to write the question or answer as a non-structured natural language sentence, and is set as a constraint that it will not be fitted into a predefined sentence structure or a fixed sentence template. There is an information processing system.

5. In the information processing system described in claim 1, The generation of the document to be generated includes an iterative unit that repeats the reception process, the deficiency inference process, the question generation / presentation process, and the answer acquisition process until the deficiency inference process evaluates that there is no deficiency of information necessary for creating the document to be generated, or until predetermined conditions including a predetermined number of dialogues or a certain degree of information sufficiency are met. The iterative unit repeatedly performs the question generation / presentation process and the answer acquisition process according to the result of the deficiency inference process, and controls the iterative process to converge until the information necessary for creating the target document is satisfied, and the iterative unit determines the end of the iterative process based on the result of the deficiency inference process and the predetermined conditions, The document generation process is an information processing system that, upon completion of the iterative process and fulfilling predetermined conditions, integrates the dialogue history, including multiple question sentences and answers, and then generates the target document in a batch.

6. In the information processing system described in claim 5, The document generation process does not immediately reflect the question and answer in the generation of the target document, but retains them as provisional information, stores the question and answer as natural language text in a memory unit, integrates the stored information as the dialogue history, and uses it to set the third input information when creating the target document, thereby reflecting it in the generation of the target document if the predetermined conditions are met.

7. In the information processing system described in claim 1, In the aforementioned deficiency inference process, the user who is the creator of the document to be generated is notified of the evaluation result regarding the presence or absence of deficiencies, and the user approves the evaluation that there are deficiencies. The aforementioned question generation and presentation process is an information processing system that generates the question text when the evaluation of "insufficient" is approved for the user.

8. In the information processing system described in claim 1, If the document to be generated is a resume, it shall include at least the outline of the work, roles, innovations, and results in the aforementioned information perspectives. If the documents to be generated are other than the resume, the information processing system includes work reports, application forms, proposals, and other explanatory documents.

9. An information processing method in which at least one processor reads a program and data stored in a memory unit to generate a document based on interaction with a user using a generation AI model, The aforementioned storage unit holds information perspectives and setting data for configuring the first input information, second input information, and third input information. The aforementioned information perspective represents information that indicates a standard information perspective for creating natural language text and generated documents. The aforementioned setting data includes, as a setting for the first input information, the setting of instructions for inferring deficiencies based on evaluation criteria for inferring which perspectives are missing from the information perspectives included in the information perspectives in order to generate the target document, in relation to the user input information and dialogue history. The aforementioned setting data includes, as settings for the second input information, instructions for textualization from the information perspective, and the setting of a first constraint that will not be executed if it is evaluated that there are no deficiencies by the deficiency inference process. The aforementioned setting data includes, as settings for the third input information, instructions for writing the user input information and the dialogue history, questions or answers into unstructured natural language sentences, and a setting for a second constraint that does not associate information with predefined items. We accept user input information, including free-form text, from users. The deficiency inference process uses the received user input information and the dialogue history based on said user input information as natural language text, reads the settings related to the information perspective and the first input information from the storage unit, sets first input information by combining the information perspective and the instruction for deficiency inference, uses the set first input information as input to the generation AI model, and evaluates whether there is any deficiency of information necessary for creating the target document based on the output from the generation AI model. The system reads the settings related to the second input information from the memory unit, sets the second input information including the first constraint as an instruction, uses the set second input information as input to the generating AI model, and if the output from the generating AI model indicates that there is a deficiency through the deficiency inference process, it generates a new question to compensate for the deficiency, presents the generated question to the user, and if the deficiency inference process indicates that there is no deficiency, it does not generate the question. When the user's response to the aforementioned question is obtained, it is recorded as the dialogue history in natural language text. The settings relating to the third input information are read from the memory unit, the third input information including the second constraint as an instruction is set, the set third input information is used as input to the generation AI model, and the target document is generated as an unstructured natural language sentence based on the output from the generation AI model. An information processing method in which a computer performs the processing.

10. An information processing program in which at least one processor reads a program and data stored in memory to generate a document based on interaction with a user using a generation AI model, The aforementioned storage unit holds information perspectives and setting data for configuring the first input information, second input information, and third input information. The aforementioned information perspective represents information that indicates a standard information perspective for creating natural language text and generated documents. The aforementioned setting data includes, as a setting for the first input information, the setting of instructions for inferring deficiencies based on evaluation criteria for inferring which perspectives are missing from the information perspectives included in the information perspectives in order to generate the target document, in relation to the user input information and dialogue history. The aforementioned setting data includes, as settings for the second input information, instructions for textualization from the information perspective, and the setting of a first constraint that will not be executed if it is evaluated that there are no deficiencies by the deficiency inference process. The aforementioned setting data includes, as settings for the third input information, instructions for writing the user input information and the dialogue history, questions or answers into unstructured natural language sentences, and a setting for a second constraint that does not associate information with predefined items. We accept user input information, including free-form text, from users. The deficiency inference process uses the received user input information and the dialogue history based on said user input information as natural language text, reads the settings related to the information perspective and the first input information from the storage unit, sets first input information by combining the information perspective and the instruction for deficiency inference, uses the set first input information as input to the generation AI model, and evaluates whether there is any deficiency of information necessary for creating the target document based on the output from the generation AI model. The system reads the settings related to the second input information from the memory unit, sets the second input information including the first constraint as an instruction, uses the set second input information as input to the generating AI model, and if the output from the generating AI model indicates that there is a deficiency through the deficiency inference process, it generates a new question to compensate for the deficiency, presents the generated question to the user, and if the deficiency inference process indicates that there is no deficiency, it does not generate the question. When the user's response to the aforementioned question is obtained, it is recorded as the dialogue history in natural language text. The settings relating to the third input information are read from the memory unit, the third input information including the second constraint as an instruction is set, the set third input information is used as input to the generation AI model, and the target document is generated as an unstructured natural language sentence based on the output from the generation AI model. An information processing program that instructs a computer to perform a task.

Images