Information processing methods, programs, and information processing systems

Abstract

This invention provides an information processing method, program, and information processing system that facilitate obtaining desired processing results. [Solution] In an information processing system comprising a server as an information processing device connected to each other via a network for mutual communication, and a user terminal, the information processing method executed by the information processing device includes: a processing instruction acquisition step S100 for acquiring processing instruction information that instructs processing; a step processing step S110 for executing processing corresponding to the processing instruction information in a plurality of related subprocesses; a step processing result acquisition step S112 for acquiring the processing results in the step processing step; and processing result acquisition steps S114, S116 for inputting the processing results of the step processing result acquisition step into a trained model and obtaining output from the trained model.

Description

Technical Field

[0006] , ,

[0007] , ,

[0001] The present invention relates to an information processing method, a program, and an information processing system.

Background Art

[0002] Citation Document 1 discloses an information processing method for selecting an appropriate database based on a user's question and generating a prompt for inputting into a language model based on the selected database and the content of the question.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the case of the above prior art, the language model generates an answer according to the generated prompt, but the answer from this language model may not be what the user intends, and there is room for improvement in this regard.

[0005] In consideration of the above facts, an object of the present invention is to make it easier to obtain a desired processing result.

Means for Solving the Problems

[0006] According to an information processing method according to an embodiment, it is an information processing method executed by an information processing apparatus, including a processing instruction acquisition step of acquiring processing instruction information for instructing processing, a step processing step of executing the processing corresponding to the processing instruction information in a plurality of sub-processes related to each other, a step processing result acquisition step of acquiring a processing result in the step processing step, and a processing result acquisition step of inputting the processing result of the step processing result acquisition step into a learned model and obtaining an output from the learned model.

[0007] According to a program of one embodiment, the information processing device is made to execute an information processing method that includes: a processing instruction acquisition step of acquiring processing instruction information that instructs processing; a step processing step of executing processing corresponding to the processing instruction information in a plurality of related subprocesses; a step processing result acquisition step of acquiring the processing results in the step processing step; and a processing result acquisition step of inputting the processing results of the step processing result acquisition step to a trained model and obtaining output from the trained model.

[0008] According to an information processing system according to one embodiment, the information processing system executed by the information processing device includes: a processing instruction acquisition unit that acquires processing instruction information that instructs processing; a step processing unit that executes processing corresponding to the processing instruction information in a plurality of related subprocesses; a step processing result acquisition unit that acquires the processing results in the step processing steps; and a processing result acquisition unit that inputs the processing results of the step processing result acquisition unit to a trained model and obtains output from the trained model. [Effects of the Invention]

[0009] According to one embodiment, it is possible to easily obtain the desired processing result. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows an example of the configuration of an information processing system according to the first embodiment. [Figure 2] This figure shows an example of the server hardware configuration according to the first embodiment. [Figure 3] This figure shows an example of the functional configuration of a server according to the first embodiment. [Figure 4] This flowchart shows an example of a process performed by the information processing system according to the first embodiment. [Figure 5] This is an explanatory diagram showing a part of the outline of the first embodiment. [Figure 6] This is an explanatory diagram showing a part of the outline of the first embodiment. [Figure 7] This is an explanatory diagram showing a part of the outline of the first embodiment. [Figure 8] This figure shows an example of the functional configuration of a server according to the second embodiment. [Figure 9] A flowchart illustrating an example of the processing performed by the information processing system according to the second embodiment. [Figure 10] This is an explanatory diagram showing a part of the outline of the second embodiment. [Modes for carrying out the invention]

[0011] (First Embodiment) The first embodiment of the information processing system according to the present invention will be described below with reference to Figures 1 to 7. In each figure, identical or equivalent components and parts are given the same reference numerals. Also, the dimensional ratios in the drawings are exaggerated for illustrative purposes and may differ from the actual ratios.

[0012] (System Overview) First, an overview of the information processing system 10 according to this embodiment will be described. The information processing system 10 according to this embodiment is a system in which, upon input of a prompt as processing instruction information that instructs processing, a trained model generates a processing result in response to the prompt and outputs the processing result. That is, when a prompt is input by user U, the information processing system 10 obtains the usage information necessary for processing from a predetermined information source based on the prompt, performs processing using a trained model including a language model, and generates a processing result. Although the prompt is defined as an instruction in natural language, it is not limited to this and can also include structured data (e.g., JSON, XML, CSV), programming code, mathematical formulas and symbolic notation, specific template formats, image and audio data, tags and metadata, and even assembly language and binary data, and these formats can be used.

[0013] When the information processing system 10 receives a prompt, it performs processing in multiple subprocesses based on that prompt before processing using a trained model including a language model. As an example, consider the case where the user provides the prompt "Create a summary of product reviews." First, based on the prompt, the system collects text data of product reviews and performs a cleaning process to remove unnecessary HTML tags and special characters. This is the first subprocess. Next, it extracts the main topics from the clean text data using a topic modeling algorithm. This is the second subprocess. Subsequently, it generates multiple summary candidates based on the extracted topics. Here, it creates multiple summaries from different perspectives using an existing summarization algorithm. This is the third subprocess. Finally, it evaluates the generated summary candidates and filters out redundant and irrelevant summaries. During this process, it evaluates the quality using evaluation metrics (e.g., ROUGE score). This is the fourth subprocess. After going through these subprocesses, the results obtained from the subprocesses are input into a trained model including a language model to generate the final summary. The trained model including the language model generates a highly accurate summary based on the data that has undergone these preprocessing steps.

[0014] Furthermore, each subprocess in the information processing system 10 is configured as an interchangeable module. This makes it easy to replace specific processing steps with different algorithms or methods, improving the overall flexibility and adaptability of the system.

[0015] Furthermore, these modules can be represented as nodes N (also expressed as Nx, where x is a number), and it is possible to output a graph (see graph G in Figures 5 and 6) in which the connections between nodes are represented as edges EG. This makes it possible to visualize the processing flow and intuitively understand the data flow and dependencies between each step. Each subprocess is related to the others, and the subprocess itself may utilize various trained models, including language models, or it may utilize rule-based algorithms. Furthermore, it may include processes that are executed by the user operating the terminal. For example, in the data preprocessing stage, text cleaning can be performed using a trained natural language processing model. Also, in the topic extraction stage, the LDA (Latent Dirichlet Allocation) algorithm for topic modeling can be used. In the summary candidate generation stage, summary rules can be manually set using a rule-based algorithm. Furthermore, in the evaluation and filtering stages of summary candidates, the user can operate the terminal and manually set evaluation criteria. By including such user-involved processes in the subprocesses, the transparency of the process is increased, and the user can feel satisfied with the output results. By executing the step-by-step processing steps in this example, the following effects can be obtained. First, data preprocessing removes noise, and topic extraction highlights relevant information, ultimately making it easier for the trained model to generate highly accurate summaries. Next, dividing the process into subprocesses and performing parallel processing shortens the overall processing time, allowing for rapid handling of large amounts of data. Furthermore, using multiple subprocesses allows each process to be optimized independently, enabling flexible adaptation to different use cases and datasets. In other words, it becomes possible to generate a variety of processing results. In this way, by utilizing step-by-step processing, the accuracy and efficiency of information processing by the trained model are improved, making it possible to provide users with high-quality and ultimately desired processing results. The details of the information processing system 10 are described below. In addition, by performing step-by-step processing steps, it is possible to execute the processing by imitating the step-by-step processing like the human brain. Therefore, not only from the perspective of transparency, but also the way of thinking can be traced. Further, by making it possible to add a process executed by the user operating the terminal to the sub-process of step-by-step processing and deliberately leaving room for human intervention, the user who receives the processing result including the process with human intervention is more likely to feel convinced. Also, the graph G may be generated by the system based on a prompt, or may be configured to use a graph created by a person such as a user. Furthermore, it may be configured to use a template in which the graph is standardized (that is, the standardization of sub-processes). By using the graph based on this standardized template, it becomes easier to tune the processing result. This is because each sub-process is clearly defined and separated as an independent module, and the impact of changes and optimizations to individual sub-processes on other parts can be reduced. As a result, efficient tuning focusing on specific processing steps becomes possible, and overall performance improvement can be easily achieved. <8000092>

[0016] (System Configuration) FIG. 1 is a diagram showing an example of the configuration of an information processing system 10 according to the present embodiment. As shown in FIG. 1, the information processing system 10 according to the present embodiment includes a server 12 as an information processing device and a user terminal 14 that are communicably connected to each other via a network N. The network N is, for example, a wired LAN (Local Area Network), a wireless LAN, the Internet, a public line network, a mobile data communication network, or a combination thereof.

[0017] The user terminal 14 is an example of an information processing device that performs operations for inputting and displaying various information by the user U. The user terminal 14 may be a PC (Personal Computer), a smartphone, a tablet terminal, a server device, a microcomputer, a wearable device, or a combination thereof.

[0018] Server 12 is an example of an information processing device that acquires information input from the user terminal 14, processes that information, and outputs the results. Server 12 may be a PC (Personal Computer), a smartphone, a tablet terminal, a server device, a microcomputer, or a combination thereof. The specific configuration and operation of Server 12 will be described later.

[0019] (Hardware configuration) Figure 2 is a block diagram showing the hardware configuration of server 12. Server 12 comprises a processor 120, memory 122, storage 124, communication interface 126, input / output interface 128, and drive device 134, all of which are connected to each other via bus B for communication.

[0020] The processor 120 controls the configuration of the server 12 and realizes the functions of the server 12 by loading various programs stored in the storage 124 into the memory 122 and executing them. The programs executed by the processor 120 include, but are not limited to, the OS (Operating System) and the program 220 described later. By the processor 120 executing these programs, a part of the state visualization method according to this embodiment is realized. The processor 120 is, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), a DSP (Digital Signal Processor), or a combination thereof.

[0021] Memory 122 is, for example, ROM (Read Only Memory), RAM (Random Access Memory), or a combination thereof. ROM is, for example, PROM (Programmable ROM), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), or a combination thereof. RAM is, for example, DRAM (Dynamic RAM), SRAM (Static RAM), MRAM (Magnetoresistive RAM), or a combination thereof.

[0022] Storage 124 stores the OS, various programs described later, and various data. Storage 124 is, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), SCM (Storage Class Memories), or a combination of these.

[0023] The communication interface 126 is an interface for connecting the server 12 to external devices, including the user terminal 14 and the imaging device 16, via the network N, and for controlling communication. The communication interface 126 is, for example, an adapter compliant with Bluetooth®, Wi-Fi®, ZigBee®, Ethernet®, or optical communication (e.g., Fibre Channel).

[0024] The input / output interface 128 is an interface for connecting input devices 132 and output devices 130 to the server 12. Input devices 132 include, for example, a mouse, keyboard, touch panel, microphone, scanner, camera, various sensors, operation buttons, or a combination thereof. Output devices 130 include, for example, a display, projector, printer, speaker, vibrator, or a combination thereof.

[0025] The drive device 134 reads and writes data to the disk media 136. The drive device 134 is, for example, a magnetic disk drive, an optical disk drive, a magneto-optical disk drive, or a combination thereof. The disk media 109 is, for example, a CD (Compact Disc), a DVD (Digital Versatile Disc), an FD (Floppy Disk), an MO (Magneto-Optical disk), a BD (Blu-ray® Disc), or a combination thereof.

[0026] In this embodiment, the program may be written to memory 122 or storage 124 during the manufacturing stage of server 12, or it may be provided to server 12 via network N, or it may be provided to server 12 via a non-temporary, computer-readable recording medium such as disk media 136.

[0027] Furthermore, since the hardware configuration of the user terminal 14 is substantially the same as that of the server 12 described above, a detailed explanation will be omitted.

[0028] (Functional Configuration) Next, the functional configuration of server 12 will be described. Figure 3 shows an example of the functional configuration of server 12. When executing various programs, server 12 uses the above hardware resources to realize various functions. Server 12 has a storage unit 22, a control unit 24, and a communication unit 20 as the functional configuration that server 12 realizes. Each functional configuration is realized when the processor 120 reads and executes a program 220 stored in memory 122 or storage 124.

[0029] Figure 4 is a diagram illustrating the processes performed by the information processing system 10. The processing performed by the information processing system 10 is as shown in Figure 4. At least the processing in step S100 corresponds to the "processing instruction acquisition step" described in claim 1. Also, at least the processing in step S110 corresponds to the "stage processing step" described in claim 1. Furthermore, at least the processing in step S112 corresponds to the "stage processing result acquisition step" described in claim 1. Furthermore, at least steps S114 and S116 correspond to the "processing result acquisition step" described in claim 1. In addition, the processing in step S104 determines whether or not to perform the stage processing step based on a graph existing in the graph DB 228, as an example, by specification from the user or the system. If stage processing is to be performed that is not based on an existing graph (step S104: NO), the process proceeds to step S106 to determine a new subprocess and generate a graph (graphing). Furthermore, the processing in step S118 determines whether or not there has been an operation by the user or the system to modify the output graph.

[0030] Figure 5 is a schematic diagram illustrating an example of a part of the processing flow performed in the information processing system 10. The processes executed by the information processing system 10 are as shown in Figure 5. The subprocesses indicated by nodes N1 and N2 are, as an example, subprocesses that use a Large Language Model (LLM). The subprocess indicated by node N3 is, as an example, a subprocess that uses various pre-trained models, not limited to language models. The subprocess indicated by node N4 is, as an example, a subprocess that uses a rule-based algorithm. The subprocess indicated by node N5 is, as an example, a subprocess that uses a process operated by a user operating a terminal. The subprocess indicated by node N6 is, as an example, a subprocess that uses a Small Language Model (SLM). Furthermore, this diagram shows a configuration in which the processing results of subprocesses are input to the appropriate pre-trained model via an AI gateway. This allows for centralized management of multiple AI models and services, enabling users and applications to access these resources uniformly and efficiently. In addition, data security is ensured by applying consistent security protocols and authentication mechanisms. Moreover, the scalability of the entire system is improved by efficiently processing large volumes of requests and data. Monitoring and logging functions allow for real-time monitoring of the usage status of AI models and services, enabling early detection and response to problems. Orchestration functions allow for the coordination of multiple AI services and models, enabling efficient management of complex workflows. Note that while this diagram shows a configuration in which the processing results of subprocesses are input to the appropriate pre-trained model via an AI gateway, the system is not limited to this configuration; the results may also be input to pre-trained models such as language models without using an AI gateway. Note that the large-scale language models mentioned above may include multimodal large-scale language models.

[0031] The training data, input data, and output data for the Large Language Model (LLM) and Multi-modal Large Language Model (Multi-modal LLM) related to the present invention are described below. First, let's discuss the training data for LLM. LLM is pre-trained using a vast amount of text data. This training data includes natural language text collected from diverse sources such as books, articles, web pages, forum posts, technical documents, and dialogue data. Such diverse datasets provide LLM with a broad vocabulary, grammatical structure, and contextual understanding capabilities. The selection and preprocessing of training data directly impacts the model's performance, making data quality and diversity crucial factors. On the other hand, multimodal LLMs are pre-trained using diverse data formats, including images, audio, and video, in addition to text data. This training data includes books and articles, as well as image datasets (e.g., ImageNet), audio datasets (e.g., LibriSpeech), and video datasets (e.g., UCF101). Such diverse datasets give multimodal LLMs the ability to understand and process not only text but also visual and auditory information in an integrated manner. Next, we will discuss the input data for LLM and multimodal LLM. In the case of LLM, the input data consists of text prompts and queries provided by the user. This input data is provided in natural language and can take various forms, such as specific questions, commands, requests for summarizing information, and translation requests. For example, one example of input data is a prompt such as, "Summarize the following sentence: 'AI technology is rapidly evolving and is expected to have applications in various industries.'" On the other hand, in the case of multimodal LLM, the input data includes not only text but also prompts and queries in various forms such as images, audio, and video. This input data can take various forms, such as questions and commands in natural language, requests for image caption generation, requests for audio-to-text conversion, and requests for video content summarization. For example, prompts such as "Describe the content of this image" and requests such as "Convert the following audio to text" are examples of input data. Finally, let's discuss the output data for LLM and multimodal LLM. In the case of LLM, the output data is the response to the input data and the resulting text. The output data can be diverse, including answers to user prompts, summaries, translations, and generated sentences. For example, as output data for the aforementioned summarization request, a summary sentence such as "AI technology is developing rapidly and is expected to have industrial applications" might be generated. On the other hand, in the case of multimodal LLM, the output data includes not only text but also data in various formats such as images, audio, and video. The output data is wide-ranging, including responses to user prompts, summaries, translations, generated text, image captions, audio-to-text conversion results, and video content summaries. For example, in response to the aforementioned image description request, the output data might include text such as, "This image depicts a green meadow under a blue sky." As described above, the LLM and multimodal LLM related to the present invention are pre-trained on a vast amount of diverse training data and can generate high-quality output data in response to diverse input data such as text, images, audio, and video provided by the user. This enables efficient and effective natural language processing and multimodal natural language processing in a wide range of application fields.

[0032] Furthermore, the training data, input data, and output data of the Small Language Model (SLM) related to the present invention will be described below. First, let's discuss the training data for SLM. SLM is pre-trained using relatively small amounts of text data. This training data includes books, articles, technical documents, and dialogue data specific to a particular domain or application. Although the amount of training data for SLM is smaller compared to LLM, it can focus on specific purposes and application ranges, allowing it to achieve high efficiency and accuracy for specific tasks. The selection and preprocessing of training data directly impacts the model's performance, making data quality and suitability crucial factors. Next, let's discuss the input data for SLM. Input data consists of text prompts and queries provided by the user. This input data is provided in natural language and can take various forms, such as specific questions, commands, requests for information summaries, and translation requests. In the case of SLM, the input data is usually related to a specific domain or application. For example, prompts such as "Please provide a summary of the following technical document" or "Please provide the definition of this technical term" are examples of input data. Finally, let's discuss the output data of SLM. Output data is the response to the input data and the resulting text. Output data can range from answers to user prompts, summaries, translations, and generated sentences. Because SLM output data is specialized for specific domains and applications, it is expected to have high accuracy and relevance. For example, as output data for the aforementioned technical document summary request, a summary sentence such as "This document describes an overview of the latest AI technologies and their applications" might be generated. As described above, the SLM related to the present invention is pre-trained on relatively small amounts of training data specific to a particular domain or application, and can generate high-quality output data in response to natural language input data provided by the user. This enables efficient and effective natural language processing in specific tasks and application fields.

[0033] Figure 6 is a schematic diagram illustrating an example of a part of the processing flow in the step-by-step processing steps performed in the information processing system 10. The processing performed by the information processing system 10 is as shown in Figure 6. Specifically, the left figure in Figure 6 is an example of a user interface in which the user enters prompts, and the right figure is an example of a graph G generated based on the entered prompts. Details of the subprocesses performed by each node N in Figure 6 are shown in Figure 6. Although not shown in the diagram, it is possible to select a specific node N or edge EG from this graph G and replace (exchange) it with other nodes N or edge EGs.

[0034] Figure 7 is a schematic diagram illustrating a part of the user interface for inputting supplementary information to each node (subprocess) in the step-by-step processing performed in the information processing system 10. The processing performed by the information processing system 10 is as shown in Figure 7. Specifically, an input field is provided for selecting and specifying supplementary information from text data or predetermined options, according to the processing content, purpose, and method of each subprocess. The system receives the supplementary information entered in the input field and executes each subprocess processing based on that supplementary information.

[0035] (Second embodiment) Next, an information processing system 50 according to the second embodiment of the present invention will be described with reference to Figure 8-10. The information processing system 50 according to the second embodiment has the same basic configuration as the first embodiment, and is characterized in that when a prompt is input, it acquires usage information necessary for processing from a predetermined information source based on the prompt, performs step-by-step processing including the usage information and processing by a trained model, and generates a processing result. Components identical to those in the first embodiment are denoted by the same reference numerals and their descriptions are omitted.

[0036] (Functional Configuration) The functional configuration of the server 52 as an information processing device in the information processing system 50 will be described. Figure 8 is a diagram showing an example of the functional configuration of the server 52. When executing various programs, the server 52 uses the same hardware resources as the server 12 in the first embodiment to realize various functions. The server 52 has a communication unit 20, a storage unit 60, and a control unit 70 as the functional configuration that the server 52 realizes. Each functional configuration is realized by the processor 120 reading and executing a program 600 stored in the memory 122 or storage 124.

[0037] The communication unit 20 is implemented by the communication interface 126. The communication unit 20 sends and receives information with the user terminal 14 via the network N. The communication unit 20 receives information input from the user terminal 14. The communication unit 20 also sends information to the user terminal 14 and receives requests from user U from the user terminal 14.

[0038] The memory unit 22 is implemented by memory 122 and storage 124. The memory unit 22 stores the program 220, the trained model 222, the prompt, the usage information DB 602, and the processing result 228.

[0039] The trained model 222 is composed of at least one trained machine learning model. This trained model 222 is, for example, a large-scale language model, trained on a large amount of text data from articles, books, and websites on the internet. When text data is input as a prompt, it generates and outputs text data corresponding to that prompt. In this embodiment, the large-scale language model in the trained model 222 determines the direction of the output result based on the content of the prompt. This output result refers to the expected output result.

[0040] Here, direction refers to the nature or characteristics of the output result (hereinafter simply referred to as "nature"), and the trained model 222 makes the following two determinations as an example. The first property is the "rigorous, source-based" property (hereinafter simply referred to as "rigorous"), and the output result in line with this property is based on clear information extracted or derived from specific sources. On the other hand, the second property is the "broad and creative" property (hereinafter simply referred to as "creativity"), and the output result in line with this property is not limited to a specific source, but draws insights from a wide range of information and provides new perspectives and interpretations. By determining the direction, the trained model 222 determines the degree to which the output result exhibits the contradictory properties of "rigor" and "creativity". That is, if the prompt instructs "Please answer based on the content of the previous management meeting", the trained model 222 determines that an output result with the rigorous nature based on sources such as the minutes of the previous management meeting is required, and determines the direction of the output result to be "rigorous". On the other hand, if the prompt instructs the user to "answer considering recent economic and technological trends," the trained model 222 determines that a creative output result, not limited to specific information sources, is required, and determines the direction of the output result to be "creative." It should be noted that the above-mentioned determination of "rigor" and "creativity" is merely an example, and it goes without saying that the direction will be appropriately determined based on the content of other prompts as well. Furthermore, the direction may be determined based on other properties. Moreover, instead of simply determining the presence or absence of each property, the degree to which the property is required may be determined numerically, such as "rigor 80, creativity 20." Furthermore, the output result may be divided into multiple parts, and the properties of each part may be determined. A concrete example is to divide the output result into a first part that provides a factual answer and a second part that provides consideration of the first part, and determine the first part as "rigor" and the second part as "creativity."

[0041] Furthermore, the trained model 222 determines candidate information to be used for processing based on the prompt. Specifically, the trained model 222 evaluates the characteristics of the information required for processing based on the prompt (such as whether the information comes from general common sense or from specialized knowledge), and if it is determined that the characteristics require acquisition from an external information source, it searches for the information source and determines the information required for processing as a candidate for information to be used. The information source may be the memory unit 22 where various types of information are stored, or it may be an external information source such as a book or information on the internet. The information to be used is various types of information, including text information stored in these information sources, and the details will be described later.

[0042] Furthermore, the trained model 222 generates at least a portion of the indicator information of the candidate usage information. At least a portion of this indicator information is, for example, information freshness information, and this freshness information changes depending on the content and processing of the usage information. In other words, freshness information is, to put it another way, the "value of the information on the time axis," and can be broadly divided into three patterns: a pattern in which the value decreases over time (hereinafter simply referred to as the "decrease pattern"), a pattern in which the value remains constant over time (hereinafter simply referred to as the "constant pattern"), and a pattern in which the value increases over time (hereinafter simply referred to as the "increase pattern"). Each pattern will be explained below. Note that "information value" here represents how much the information contributes to the processing result desired by user U, and a high information value indicates a high contribution to the output of the desired processing result.

[0043] There are two patterns of decline in value: one where the value decreases at a roughly constant rate (see Figure 5(A)) and another where the value decreases rapidly (see Figure 5(B)). Information that declines at a roughly constant rate includes general news information, product information with regularly released new models, and event results. This type of information is time-sensitive, as its value gradually decreases over time. Information that declines rapidly includes stock market information, weather forecasts, and news related to rapidly changing events such as accidents and disasters.

[0044] Information that follows a certain pattern (see Figure 5(C)) specifically includes information on historical facts and scientific principles. This type of information tends to change in value relatively little over time.

[0045] There are two patterns of value increase: one where the value increases at a roughly constant rate (see Figure 5(D)) and another where the value increases rapidly (see Figure 5(E)). Information that increases at a roughly constant rate includes historical data, classical literature, and information about limited-production goods and antique books—information that may not initially be considered valuable, but whose importance and value are re-recognized over time, leading to a rise in value. Information that increases rapidly includes information about past works of art and collectibles, predictive analysis and forecasts, specific technologies and theories, and vintage goods—information that initially received little attention, but whose value increases rapidly due to events such as the death of the author or the later success of past predictions. Note that the specific information for each pattern is not limited to those mentioned above.

[0046] Furthermore, the three patterns described above are just examples, and other patterns may be included, such as patterns whose value fluctuates periodically, like information about regular events. Patterns that combine the above-mentioned patterns may also be included. That is, information whose value increases and then becomes constant or decreases after a predetermined time, information whose value was constant and then increases or decreases after a predetermined time, and information whose value decreases and then becomes constant or increases after a predetermined time. In other words, the trained model 222 can also be described as "evaluating usage information whose value changes in response to acquired prompts using a metric called freshness information."

[0047] The prompt is information indicating the prompt entered by user U. The prompt is in natural language and, for example, is stored linked to user U's account information.

[0048] The usage information in the usage information DB602 is the usage information necessary for processing. As mentioned above, this usage information in the usage information DB602 is information obtained from information sources such as books, the internet, and the storage unit 22, and includes text information and image information. The usage information includes both unstructured data and structured data. Here, "unstructured data" refers to data that is stored without being structured in a predefined way. On the other hand, "structured data" refers to data that exists in specific fields within files or records. "Semi-structured data" is also included in "structured data". The usage information can be stored in a data lake (not shown) in the storage unit 22, and can be retrieved from this data lake using the usage information index information described later.

[0049] The processing result 228 is information that shows the result of executing a process using the usage information based on the prompt. The processing result 228 is various types of information, including text information and image information, and is stored in association with the prompt and usage information.

[0050] The control unit 24 is realized when the processor 120 reads and executes the program 220 from the memory 122 (see Figure 2) and works in cooperation with other hardware components. The control unit 24 includes an information acquisition unit 242, an information processing unit 244, and an output unit 246.

[0051] The information acquisition unit 242 acquires various types of information necessary for processing in the information processing unit 244. Specifically, it acquires usage information, including prompts to instruct processing and unstructured data necessary for processing. The information acquisition unit 242 also acquires indicator information for the usage information (details will be described later). This acquired information may be information stored in the storage unit 22 or may be acquired from an external data source, etc.

[0052] The information processing unit 244 performs the following: determining the direction of the output result based on the content of the prompt, acquiring and generating indicator information of the information used, determining the degree of agreement between the direction of the output result and the information used, and executing processing based on the prompt. The direction of the output result is determined by using the trained model 222 to determine the properties of the expected output result corresponding to the prompt entered by the user U, as described above.

[0053] The information processing unit 244 generates freshness information, one of the indicator information for usage information, using the trained model 222 as described above. The information processing unit 244 also has access to metadata and log data, which are indicator information for usage information acquired by the information acquisition unit 242. Metadata is information that describes the characteristics and attributes of the data, such as the content, source, creator, and timestamp of the usage information. Log data is information that is recorded along with the time generated each time a certain operation is performed, such as the date and time when the usage information was recorded or accessed, the person who performed the operation, request and response information, and error information. This metadata and log data are pre-associated with the usage information and made available for use. The information processing unit 244 uses at least one of the indicator information, freshness information, metadata, and log data to determine the consistency between the usage information and the direction of the output result (details will be described later).

[0054] The information processing unit 244 determines the degree of agreement between the direction of the output result and the information used. That is, it determines the degree of agreement between the direction and the candidate information used, based on the indicator information of the information used, to determine whether the candidate information used matches the processing based on the prompt. For example, if the direction is determined to be "strict," the degree of agreement of the information used is determined using the indicator information from perspectives such as whether the candidate information used is accurate, complete, up-to-date, reliable, clear, traceable, and consistent.

[0055] In other words, accuracy is determined based on criteria such as whether the update date and time of the usage information is recent, whether the record of that date and time is accurate, whether the registrant is a highly reliable person or organization, whether the metadata conforms to a predefined format or standard, whether there are any signs of system malfunction in the log data by searching for error messages or abnormal patterns, and whether there are any inconsistencies with the content when there are rising or falling patterns in the freshness information by tracking the factors that caused those patterns.

[0056] Whether or not the data is complete is determined based on criteria such as whether all necessary metadata elements are present, whether all log data to be collected is present, and whether it includes important elements or bits to capture the overall picture of fluctuations in information value, such as the specific time period and circumstances under which an increase or decrease in value occurred from the freshness information.

[0057] Whether something is up-to-date is determined based on criteria such as whether the metadata reflects the latest information, whether the log data is real-time or up-to-date, whether the freshness information reflects the latest changes in information value, and whether the information value is above a predetermined value.

[0058] Whether or not data is highly reliable is determined based on criteria such as the reliability of the data source and the evaluation of the metadata generation process, whether log data is collected using reliable means and is free from the risk of tampering, and whether the freshness information maintains its value in little variation across all patterns.

[0059] Whether or not information is clear is determined based on criteria such as whether the metadata is defined according to clear descriptions and conventions, whether the log data is recorded in a consistent format and can be clearly interpreted, and whether the factors causing a decrease, constancy, or increase in informational value can be clearly interpreted from the freshness information.

[0060] Whether or not data is traceable is determined based on criteria such as whether the data source is clearly indicated in the metadata, whether it is possible to query that data source, whether each log entry can be traced over time from the log data (whether time information is included), and whether the changes in the value of each piece of information can be traced over time from the freshness information, that is, whether the time when the change occurred is clearly indicated.

[0061] Whether or not consistency exists is determined based on criteria such as whether there is consistency between metadata and whether they conform to defined rules and standards, whether log data is generated in a consistent format and pattern, and whether the fluctuation in information value is consistent from freshness information. Note that the above criteria are just examples, and other criteria may also be used for determination.

[0062] If the degree of agreement between the direction of the output result and the usage information is above a predetermined level, the information processing unit 244 uses the trained model 222 to perform processing corresponding to the prompt based on the usage information. On the other hand, if the degree of agreement between the direction of the output result and the usage information is not above a predetermined level, the information processing unit 244 searches for usage information again and obtains candidate usage information as a search result via the information acquisition unit 242.

[0063] The output unit 246 controls the information processing unit 244 to output the processing result to the user terminal 14.

[0064] Figure 9 is a diagram illustrating the processes performed by the information processing system 50. The processing performed by the information processing system 50 is as shown in Figure 9. Note that the processing from step S300 to step S308 corresponds to the "usage information acquisition step" described in claim 8.

[0065] Figure 10 is a schematic diagram illustrating an example of a part of the processing flow performed in the information processing system 50. The processes performed by the information processing system 50 are as shown in Figure 10.

[0066] According to the information processing system 50 of this embodiment, the system performs a processing instruction acquisition step to acquire a prompt (processing instruction information) that instructs processing, a usage information acquisition step to acquire usage information necessary for processing from an information source based on the processing instruction information, and a processing step to execute processing corresponding to the prompt using a trained model 222 based on the usage information. As a result, processing is performed using the usage information based on the prompt, and the desired processing result can be obtained effectively.

[0067] Furthermore, since the information processing system 10 acquires usage information in the usage information acquisition step based on the direction of the output determined from the prompt, it can execute processing corresponding to the prompt based on appropriate usage information. This makes it possible to obtain the desired processing result more effectively.

[0068] Although the first and second embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above, and any modifications, improvements, etc., that can achieve the objectives of the present invention are included in the present invention.

[0069] Furthermore, the series of processes described above can be executed by hardware or by software. In other words, the functional configuration is merely illustrative and not particularly limited. That is, it is sufficient for the information processing system to have the functionality to execute the series of processes described above as a whole, and there is no particular limitation on what functional blocks are used to realize this functionality. Also, the location of the functional blocks is not particularly limited and can be arbitrary. For example, the functional blocks of a server may be delegated to a user terminal, etc. Conversely, the functional blocks of a user terminal may be delegated to a server, etc. Moreover, a single functional block may be composed of hardware alone, software alone, or a combination of both.

[0070] For example, when a series of processes are executed by software, the programs that make up that software are installed on a computer or other device from a network or storage medium. The computer may be a computer built into dedicated hardware. Alternatively, the computer may be a computer capable of performing various functions by installing various programs, such as a server, a general-purpose smartphone, or a personal computer.

[0071] Furthermore, for example, a recording medium containing such a program may consist not only of removable media (not shown) distributed separately from the main unit of the device to provide the program to the user, but also of a recording medium provided to the user in a state where it is pre-installed in the main unit of the device.

[0072] In this specification, the step of describing a program to be recorded on a recording medium includes not only processes that are performed chronologically in that order, but also processes that are not necessarily performed chronologically, but are executed in parallel or individually. Furthermore, in this specification, the term "system" refers to an overall system composed of multiple devices, means, etc.

[0073] In other words, a program to which the present invention applies only needs to have the following configuration, and can take various forms. In other words, the program to which the present invention applies is used in an information processing device. A process instruction acquisition step to obtain process instruction information that instructs the process, A step-by-step processing step in which processing corresponding to the processing instruction information is executed in a plurality of related subprocesses, The process results in the aforementioned step-by-step processing step are obtained in the step-by-step processing result acquisition step, The process result acquisition step involves inputting the processing result of the aforementioned step-by-step processing result acquisition step into the trained model and obtaining output from the trained model, Execute the process that includes this.

[0074] This allows for improved accuracy and efficiency of information processing by trained models through the use of multi-stage processing steps, enabling the provision of high-quality and ultimately desired processing results to users.

[0075] Furthermore, while the above-described step-by-step processing is configured to be executed by multiple related subprocesses, it is not limited to this configuration, and may also be configured to be executed by a single subprocess.

[0076] <Note> This embodiment includes the following disclosures.

[0077] (Note 1) An information processing method performed by an information processing device, A process instruction acquisition step to obtain process instruction information that instructs the process, A step-by-step processing step in which processing corresponding to the processing instruction information is executed in a plurality of related subprocesses, The process results in the aforementioned step-by-step processing step are obtained in the step-by-step processing result acquisition step, The process result acquisition step involves inputting the processing result of the aforementioned step-by-step processing result acquisition step into the trained model and obtaining output from the trained model, Information processing methods including

[0078] (Note 2) A graph is generated and output in which the subprocesses in the aforementioned step-by-step processing are represented as nodes and the relationships between the subprocesses are represented as edges. The information processing method described in Appendix 1.

[0079] (Note 3) The aforementioned processing step executes processing based on a template in which the relationships between the multiple subprocesses are pre-defined. The information processing methods described in Appendix 1 and Appendix 2.

[0080] (Note 4) The graph is output by a trained model that has been trained to output the graph when the processing instruction information is input. The information processing method described in Appendix 2.

[0081] (Note 5) At least a portion of the aforementioned subprocess includes processing that is performed by the user operating the terminal, The information processing method described in any one of the items in Appendix 1 to Appendix 4.

[0082] (Note 6) Each of the aforementioned subprocesses is configured as an interchangeable module. The information processing method described in any one of the items in Appendix 1 to Appendix 5.

[0083] (Note 7) At least a portion of the aforementioned subprocess includes a supplemental information input receiving process that receives supplemental information related to the aforementioned process. The information processing method described in any one of the items in Appendix 1 to Appendix 6.

[0084] (Note 8) Based on the processing instruction information obtained in the processing instruction acquisition step, a usage information acquisition step is performed to acquire usage information necessary for processing from the information source, The aforementioned processing step further includes processing corresponding to the processing instruction information based on the usage information, The information processing method described in any one of the items in Appendix 1 to Appendix 7.

[0085] (Note 9) In an information processing device, A process instruction acquisition step to obtain process instruction information that instructs the process, A step-by-step processing step in which processing corresponding to the processing instruction information is executed in a plurality of related subprocesses, The process results in the aforementioned step-by-step processing step are obtained in the step-by-step processing result acquisition step, The process result acquisition step involves inputting the processing result of the aforementioned step-by-step processing result acquisition step into the trained model and obtaining output from the trained model, A program for executing an information processing method that includes this method.

[0086] (Note 10) An information processing system executed by an information processing device, A processing instruction acquisition unit that acquires processing instruction information to instruct the processing, A step-by-step processing unit that executes the processing corresponding to the processing instruction information in a plurality of related subprocesses, A step-processing result acquisition unit that acquires the processing results in the aforementioned step-processing step, A processing result acquisition unit inputs the processing result of the aforementioned step-by-step processing result acquisition unit to a trained model and obtains output from the trained model, An information processing system having [Explanation of Symbols]

[0087] 10. Information Processing Systems 12. User terminals (information processing devices) 14 Servers (Information Processing Devices) 50 Information Processing Systems

Claims

1. An information processing method performed by an information processing device, A process instruction acquisition step to obtain process instruction information that instructs the process, A step-by-step processing step in which processing corresponding to the processing instruction information is executed in a plurality of related subprocesses, The process results in the aforementioned step-by-step processing step are obtained in the step-by-step processing result acquisition step, The process result acquisition step involves inputting the processing result of the aforementioned step-by-step processing result acquisition step into the trained model and obtaining output from the trained model, Information processing methods including

2. A graph is generated and output in which the subprocesses in the aforementioned step-by-step processing are represented as nodes and the relationships between the subprocesses are represented as edges. The information processing method according to claim 1.

3. The aforementioned processing step executes processing based on a template in which the relationships between the multiple subprocesses are pre-defined. The information processing method according to claim 1.

4. The graph is output by a trained model that has been trained to output the graph when the processing instruction information is input. The information processing method according to claim 2.

5. At least a portion of the aforementioned subprocess includes processing that is performed when a user operates a terminal, The information processing method according to claim 1.

6. Each of the aforementioned subprocesses is configured as an interchangeable module. The information processing method according to claim 1.

7. At least a portion of the aforementioned subprocess includes a supplemental information input receiving process that receives supplemental information related to the aforementioned process. The information processing method according to claim 1.

8. Based on the processing instruction information obtained in the processing instruction acquisition step, a usage information acquisition step is performed to acquire usage information necessary for processing from the information source, The aforementioned processing step further includes processing corresponding to the processing instruction information based on the usage information, The information processing method described in claim 1.

9. In an information processing device, A process instruction acquisition step to obtain process instruction information that instructs the process, A step-by-step processing step in which processing corresponding to the processing instruction information is executed in a plurality of related subprocesses, The process results in the aforementioned step-by-step processing step are obtained in the step-by-step processing result acquisition step, The process result acquisition step involves inputting the processing result of the aforementioned step-by-step processing result acquisition step into the trained model and obtaining output from the trained model, A program for executing an information processing method that includes this method.

10. An information processing system executed by an information processing device, A processing instruction acquisition unit that acquires processing instruction information to instruct the processing, A step-by-step processing unit that executes the processing corresponding to the processing instruction information in a plurality of related subprocesses, A step-processing result acquisition unit that acquires the processing results in the aforementioned step-processing step, A processing result acquisition unit inputs the processing result of the aforementioned step-by-step processing result acquisition unit to a trained model and obtains output from the trained model, An information processing system having

Images