Processing device, processing program, processing method, and processing system

Abstract

[Problem] To more efficiently generate input information to be input to a trained machine learning model. [Solution] The present invention acquires category selection information capable of identifying a category selected from one or more categories on the basis of an operation input from a user, outputs template candidate information including candidates of a template utilized for generating the input information on the basis of the acquired category selection information, acquires template selection information capable of identifying a template selected from the template candidate information on the basis of an operation input from the user, outputs template information including parameter information utilized for generating the input information on the basis of the acquired template selection information, and acquires the input information generated by receiving a change to the parameter information on the basis of an operation input from the user.

Description

Processing apparatus, processing program, processing method, and processing system

[0001] This disclosure relates to a processing device, processing program, processing method, and processing system used for the use of a trained model.

[0002] Conventionally, systems have been known that generate prompts for input into large-scale language models, etc., for tasks such as document generation, and that can obtain predetermined response information by inputting said prompts into the large-scale language model. For example, Patent Document 1 describes a system comprising: an operation log acquisition unit that acquires operation logs of monitored terminals used for business operations; an operation log classification unit that classifies the operation logs acquired by the operation log acquisition unit by comparing them with business elements that serve as classification indicators for the business operations for each predetermined time frame; and a document creation unit that creates business support documents, including daily reports, based on the classification results classified by the operation log classification unit, wherein the document creation unit has a large-scale language model that probabilistically predicts how likely words and sentences given in prompts are to occur in natural language by combining one or more types of processing in natural language processing such as morphological analysis, syntactic analysis, semantic analysis, contextual analysis, and intent analysis, analyzes the prompts, and predicts and generates documents based on the content of the analyzed prompts. The description includes a "business support document creation device" comprising: a language model unit; a document creation prompt generation unit that outputs a prompt to the large-scale language model unit that instructs the creation of the business support document as a document based on the classification results; a business pattern formation unit that forms a business pattern for a certain period of time of business performed on the monitored terminal based on the classification results in the operation log classification unit; a business change detection unit that detects a qualitative change in the business performed on the monitored terminal based on the amount of change between the business patterns that are repeatedly formed at each of the certain periods; and a warning information output unit that, when a qualitative change in the business is detected, outputs the monitored terminal that caused the qualitative change, along with the content of the qualitative change, as warning information to the administrator terminal.

[0003] Patent No. 7572760

[0004] Therefore, based on the above technologies, an object of the present disclosure is to provide a processing device, a processing program, a processing method, and a processing system that can generate input information to be input to a learned machine learning model more efficiently according to various embodiments.

[0005] According to one aspect of the present disclosure, there is provided "a processing device including at least one processor, wherein the at least one processor is configured to generate input information to be input to at least any one of one or more learned models configured to output response information when arbitrary input information is input, obtain category selection information capable of specifying at least one category selected from one or more categories based on an operation input from a user, output template candidate information including candidates for one or more templates to be used for generating the input information based on the obtained category selection information, obtain template selection information capable of specifying at least one template selected from the template candidate information based on an operation input from the user, output template information including at least one parameter information to be used for generating the input information based on the obtained template selection information, and execute a process for obtaining the input information generated by receiving a change to the at least one parameter information based on an operation input from the user."

[0006] According to one aspect of the present disclosure, a processing program is provided which causes the at least one processor to function in order to generate input information to be input into at least one of one or more trained models configured such that response information is output when arbitrary input information is input, by obtaining category selection information that can identify at least one category selected from one or more categories based on user input, outputting template candidate information that includes one or more template candidates used to generate the input information based on the obtained category selection information, obtaining template selection information that can identify at least one template selected from the template candidate information based on user input, outputting template information that includes at least one parameter information used to generate the input information based on the obtained template selection information, and receiving changes to the at least one parameter information based on user input, thereby executing a process for obtaining the input information.

[0007] According to one aspect of the present disclosure, a processing method is provided for a computer having at least one processor, the method being executed by the at least one processor, and includes the steps of: acquiring category selection information that can identify at least one category selected from one or more categories based on user input, in order to generate input information to be input into at least one of one or more trained models configured such that response information is output when arbitrary input information is input; outputting template candidate information that includes one or more template candidate information to be used to generate the input information based on the acquired category selection information, and acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information to be used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting a change to the at least one parameter information based on user input.

[0008] According to one aspect of this disclosure, a processing system is provided which includes "the processing device described above and a terminal device connected to the processing device via a communication network and configured to transmit category selection information and template selection information to the processing device via the communication network."

[0009] According to this disclosure, it is possible to provide a processing device, a processing program, a processing method, and a processing system that can generate input information to be input to a trained machine learning model more efficiently.

[0010] The effects described above are merely illustrative for the sake of explanation and are not limiting. In addition to, or in lieu of, any other effects described herein or that would be obvious to those skilled in the art may be achieved.

[0011] Figure 1 is a block diagram showing the configuration of a processing system 1 according to one embodiment of the present disclosure. Figure 2A is a block diagram showing the configuration of a server device 100 according to one embodiment of the present disclosure. Figure 2B is a block diagram showing the configuration of a terminal device 200 according to one embodiment of the present disclosure. Figure 3A is a diagram conceptually showing a template management table stored in the server device 100 according to one embodiment of the present disclosure. Figure 3B is a diagram conceptually showing a user management table stored in the server device 100 according to one embodiment of the present disclosure. Figure 3C is a diagram conceptually showing a trained model management table stored in the server device 100 according to one embodiment of the present disclosure. Figure 4 is a diagram conceptually showing an example of the trained model selection process in the server device 100 according to one embodiment of the present disclosure. Figure 5 is a diagram showing a processing sequence executed in the processing system 1 according to one embodiment of the present disclosure. Figure 6 is a diagram showing the processing flow of the process executed in the server device 100 according to one embodiment of the present disclosure. Figure 7A is a diagram showing an example of a category selection screen 10 output in the terminal device 200 according to an embodiment of the present disclosure. Figure 7B is a diagram showing an example of a template selection screen 20 output in the terminal device 200 according to an embodiment of the present disclosure. Figure 7C shows an example of an input information generation screen 30 output by the terminal device 200 in an embodiment of the present disclosure. Figure 8 shows an example of a response information screen 40 output by the terminal device 200 in an embodiment of the present disclosure.

[0012] Various embodiments of the present invention will be described below with reference to the attached drawings. Note that common components in the drawings are denoted by the same reference numerals. Also, please note that components shown in one drawing may be omitted in another drawing for the sake of clarity. Furthermore, please note that the attached drawings are not necessarily drawn to an exact scale.

[0013] The various systems, methods, and apparatus described in this disclosure should not be construed as limiting in any way. In practice, this disclosure is directed to any novel features and aspects of each of the various embodiments disclosed, combinations of these various embodiments, and combinations of some of these various embodiments. The various systems, methods, and apparatus described in this disclosure are not limited to any particular aspect, particular feature, or combination of such particular aspects and particular features, and the things and methods described in this disclosure do not require that one or more particular effects exist or problems are solved. Furthermore, various features or aspects of the various embodiments described in this disclosure, or some of such features or aspects, may be used in combination with each other.

[0014] While the operation of some of the various methods disclosed in this disclosure is described in a specific order for convenience, this method of description should be understood to include the possibility of rearranging the order of the operations unless a specific order is required by the following specific sentences. For example, multiple operations described in order may, in some cases, be rearranged or performed simultaneously. Furthermore, for the sake of simplification, the accompanying drawings do not show various ways in which the various matters and methods described in this disclosure may be used in conjunction with other matters and methods.

[0015] Any operating theories, scientific principles, or other theoretical descriptions presented in connection with the apparatus or method of this disclosure are provided for the purpose of better understanding and are not intended to limit the technical scope. The apparatus and method in the appended claims are not limited to apparatus and method that operate in the manner described by such operating theories.

[0016] Any of the various methods disclosed herein can be implemented using a plurality of computer-executable instructions stored on one or more computer-readable media, and can be executed on a computer. The one or more media may be non-transient computer-readable storage media such as, for example, at least one optical media disk, a plurality of volatile memory components, or a plurality of non-volatile memory components. The plurality of volatile memory components include, for example, DRAM or SRAM. The plurality of non-volatile memory components include, for example, hard drives and solid-state drives (SSDs). Furthermore, the computer includes any computer available on the market, including, for example, smartphones and other mobile devices having computing hardware.

[0017] Any of the multiple computer-executable instructions for implementing the technology disclosed herein may be stored in one or more computer-readable media (e.g., non-temporary computer-readable storage media) along with any data generated and used during implementations of the various embodiments disclosed herein. Such multiple computer-executable instructions may, for example, be part of a separate software application, or part of a software application accessed or downloaded via a web browser or other software application (such as a remote computing application). Such software may be executed, for example, on a single local computer (as a process run on any suitable computer available on the market), or in a network environment (e.g., the Internet, a wide area network, a local area network, a client-server network (such as a cloud computing network), or other such network) using one or more network computers.

[0018] For clarity, only specific selected aspects of various software-based implementations are described. Other details that are well known in the art are omitted. For example, the technology disclosed in this disclosure is not limited to any particular computer language or program. For example, the technology disclosed in this disclosure may be executed by software written in C, C++, Java®, or any other suitable programming language. Similarly, the technology disclosed in this disclosure is not limited to any particular computer or type of hardware. Specific details of suitable computers and hardware are well known and do not need to be described in detail in this disclosure.

[0019] Furthermore, any of the various embodiments of such software (including, for example, a set of computer-executable instructions for causing a computer to perform any of the various methods disclosed herein) may be uploaded, downloaded, or accessed remotely by preferred means of communication. Such preferred means of communication include, for example, the Internet, the World Wide Web, intranets, software applications, cables (including fiber optic cables), magnetic communications, electromagnetic communications (including RF communications, microwave communications, and infrared communications), electronic communications, or other such means of communication.

[0020] 1. Schematic diagram 1 of the processing system 1 is a block diagram showing the configuration of the processing system 1 according to one embodiment of the present disclosure. According to Figure 1, the processing system 1 includes at least a server device 100 and a terminal device 200, and each device is connected to communicate via a wired or wireless network. The server device 100, for example, acquires category selection information and template selection information from the terminal device 200, while outputting template candidate information and template information to the terminal device 200. The terminal device 200, for example, is configured to be held by a user, and receives template candidate information and template information from the server device 100, while transmitting category selection information and template selection information to the server device 100.

[0021] Such a processing system 1 is used to generate input information to be input to a trained machine learning model in order to obtain output of arbitrary response information from the trained model. Specifically, in order to generate input information to be input to at least one of one or more trained models configured to output response information when arbitrary input information is input, the processing system 1 acquires category selection information that allows it to identify at least one category selected from one or more categories based on user input. The processing system 1 also outputs template candidate information that includes one or more template candidates used to generate input information based on the acquired category selection information, and acquires template selection information that allows it to identify at least one template selected from the template candidate information based on user input. The processing system 1 also outputs template information that includes at least one parameter information used to generate input information based on the acquired template selection information, and acquires the generated input information by accepting changes to at least one parameter information based on user input.

[0022] Therefore, processing system 1 can generate input information to be input to a trained model by accepting category selection, template selection, and parameter information changes. Generally, input information (e.g., prompts) input to a large-scale language model, which is one type of trained model, is difficult to make into sentences appropriate for individual users, and its quality often depends on the user's individual experience and ability. However, processing system 1 makes it possible to generate input information more simply and efficiently by generating the input information as described above.

[0023] In this disclosure, "processing device" means a server device 100, a terminal device 200, or a combination thereof. That is, the following description will focus on the case where the server device 100 functions as a processing device, but the terminal device 200 can also function as a processing device in the same way. Furthermore, in this disclosure, the storage and processing performed by the processing device may be distributed to other terminal devices or other server devices. In other words, the processing device is not limited to those consisting of a single enclosure, but includes the server device 100, the terminal device 200, other server devices, other terminal devices, or a combination thereof.

[0024] Furthermore, in this disclosure, the "trained model" can be any model that outputs response information when arbitrary input information is input. Such a trained model is a model that has undergone arbitrary machine learning. As an example, a trained model can be generated by preparing multiple combinations of training data and the correct labels assigned to said training data, and inputting these combinations into a learner to machine learn the correct patterns. Examples of such learning models include neural networks, convolutional neural networks, multilayer Herceptons (MLP), LSTM (Long Short Term Memory), GRU (Gated Recurrent Unit), GNN (Graph Neural Network), Transformers, and other neural network-based methods; gradient boosting decision trees (GBDT) such as LightGBM (Light Gradient Boosting Machine), XGBoost, and CatBoost; and various learning models using ridge regression, logistic regression, support vector regression (SVR), nearest neighbor, decision trees, regression trees, random forests, or combinations thereof.

[0025] Furthermore, pre-trained models include generative pre-trained models such as Large Language Models (LLMs), which are generated by deep learning using large amounts of text data. Examples of such pre-trained models include BERT (Bidirectional Encoder Representations from Transformers), GPT (Generative Pre-trained Transformer), Gemini, DALL-E, Midjourney, or combinations thereof.

[0026] As illustrated above, the trained model will output response information when arbitrary input information is provided. The input information can be in various formats, including text, images (including both video and still images), audio, document files, print layouts, links, or combinations thereof. Similarly, the output response information can be in various formats, including text, images (including both video and still images), audio, document files, print layouts, links, or combinations thereof. For the sake of explanation, the following will describe the case where the input information is in text format and the response information is in text format, but it is certainly not limited to this case.

[0027] 2. Configuration diagram 2A of the server device 100 is a block diagram showing the configuration of a server device 100 according to one embodiment of the present disclosure. According to Figure 2A, the server device 100 includes a processor 111, a memory 112, and a communication interface 113. Each of these components is electrically connected to the others via control lines and data lines. The server device 100 does not need to have all of the components shown in Figure 2A; it is possible to omit some components or add other components. For example, it is possible to use an external memory connected via communication as memory, a database (DB) device, or other server devices. It is also possible to distribute and execute some processing with processing devices including other server devices. In other words, the server device 100 is not limited to a single device, but also includes cases where it is distributed across multiple devices depending on the handling of information and the processing load.

[0028] The processor 111 functions as a control unit that controls other components of the processing system 1 based on the processing program stored in the memory 112. Based on the processing program stored in the memory 112, the processor 111 executes processing related to the generation of input information to be input to the trained model. Specifically, the following processes are executed based on a processing program stored in memory 112: "a process to acquire category selection information via the communication interface 113 that allows identification of at least one category selected from one or more categories based on user input, in order to generate input information to be input into at least one of one or more trained models configured to output response information when arbitrary input information is input"; "a process to output template candidate information via the communication interface 113 that includes one or more template candidates used to generate input information based on the acquired category selection information"; "a process to acquire template selection information via the communication interface 113 that allows identification of at least one template selected from the template candidate information based on user input"; "a process to output template information via the communication interface 113 that includes at least one parameter information used to generate input information based on the acquired template selection information"; "a process to acquire input information generated by accepting a change to at least one parameter information based on user input via the communication interface 113"; and "a process to acquire response information from at least one of the trained models to which the input information was input by inputting the acquired input information into at least one of one or more trained models." The processor 111 is mainly composed of one or more CPUs, but may be combined with a GPU, FPGA, etc. as appropriate.

[0029] Memory 112 is composed of RAM, ROM, non-volatile memory, HDD, SSD, etc., and functions as a storage unit that stores various information. Memory 112 stores instruction commands for various controls of the processing system 1 according to this embodiment as processing programs. Specifically, memory 112 stores processing programs for processor 111 to execute, such as: "a process to acquire category selection information via communication interface 113 that allows identification of at least one category selected from one or more categories based on user input, in order to generate input information to be input into at least one of one or more trained models configured to output response information when arbitrary input information is input"; "a process to output template candidate information via communication interface 113 that includes one or more template candidates used to generate input information based on the acquired category selection information"; "a process to acquire template selection information via communication interface 113 that allows identification of at least one template selected from the template candidate information based on user input"; "a process to output template information via communication interface 113 that includes at least one parameter information used to generate input information based on the acquired template selection information"; "a process to acquire input information generated by accepting a change to at least one parameter information based on user input via communication interface 113"; and "a process to acquire response information from at least one trained model to which the input information has been input by inputting the acquired input information into at least one of one or more trained models." In addition to the program, memory 112 also stores various information stored in the template management table, user management table, or trained model management table. Note that this information does not necessarily need to be constantly stored in memory 112 within the server device 100; it may be stored in a remotely located database (DB) device. In that case, the database device is also included in memory 112.

[0030] The communication interface 113 functions as a notification unit for sending and receiving various information with a terminal device 200 connected via a wired or wireless network. Examples of the communication interface 113 include wired communication connectors such as USB and SCSI, wireless communication transceivers such as wireless LAN, Bluetooth®, LTE, and infrared, and various connection terminals for printed circuit boards and flexible circuit boards. For example, the communication interface 113 receives category selection information or template selection information from the terminal device 200, or sends template candidate information, template information, or response information to the terminal device 200.

[0031] 3. Diagram 2B of the configuration of the terminal device 200 is a block diagram showing the configuration of a terminal device 200 according to one embodiment of the present disclosure. According to Figure 2B, the terminal device 200 includes a processor 211, a memory 212, an input interface 213, an output interface 214, and a communication interface 215. Each of these components is electrically connected to the others via control lines and data lines. Note that the terminal device 200 does not need to have all of the components shown in Figure 2B; it is possible to omit some components or add other components. The terminal device 200 can be any device that can communicate with the server device 100 via a wired or wireless network, and smartphones, tablet devices, laptop PCs, desktop PCs, etc., can be used as terminal devices 200. Note that if there are multiple users as described above, multiple terminal devices 200 will be used for each user, but each terminal device 200 may be a different type of terminal device. Furthermore, it is not necessarily required that there be one terminal device 200 for each user; multiple users may use one terminal device 200, or one user may use multiple terminal devices 200.

[0032] The processor 211 functions as a control unit that controls other components of the processing system 1 based on a processing program stored in the memory 212. The processor 211 executes processing related to the generation of input information based on the processing program stored in the memory 212. Specifically, this includes: "a process of receiving user operation input via the input interface 213 and generating category selection information that can identify at least one category selected from one or more categories in order to generate input information to be input into one or more trained models"; "a process of transmitting the generated category selection information to the server device 100 via the communication interface 215"; "a process of receiving template candidate information including one or more template candidates from the server device 100 via the communication interface 215, receiving user operation input via the input interface 213 and generating template selection information that can identify at least one template selected from the template candidate information"; and "via the communication interface 215, The processor 111 is mainly composed of one or more CPUs, but a GPU or FPGA may be combined as appropriate.

[0033] Memory 212 is composed of RAM, ROM, non-volatile memory, HDD, SSD, etc., and functions as a storage unit. Memory 212 stores instruction commands for various control of the processing system 1 according to this embodiment as programs. Specifically, memory 212 performs the following processes: "receiving user operation input via input interface 213 and generating category selection information that can identify at least one category selected from one or more categories in order to generate input information to be input to one or more trained models"; "transmitting the generated category selection information to the server device 100 via communication interface 215"; "receiving template candidate information including one or more template candidates from the server device 100 via communication interface 215, receiving user operation input via input interface 213, and generating template selection information that can identify at least one template selected from the template candidate information"; and "communication interface 215 The system stores processing programs for the processor 211 to execute, such as "a process of sending generated template selection information to the server device 100 via the communication interface 215", "a process of receiving template information including at least one parameter from the server device 100 via the communication interface 215, receiving user operation input via the input interface 213, and generating input information by accepting changes to at least one parameter", "a process of sending the generated input information to the server device 100 via the communication interface 215", and "a process of receiving response information obtained by inputting the input information into at least one of one or more trained models via the communication interface 215, and outputting the response information via the output interface 214".

[0034] The input interface 213 functions as an input unit that receives user input to the terminal device 200. Examples of the input interface 213 include physical key buttons and a touch panel having an input coordinate system corresponding to the display coordinate system of the display. In the case of a touch panel, icons are displayed on the display, and the user makes a selection for each icon by making an input via the touch panel. The method for detecting the user's input via the touch panel can be any method, such as capacitive or resistive. The input interface 213 does not always need to be physically provided on the terminal device 200 and may be connected as needed via a wired or wireless network. Therefore, in addition to the above, a mouse or keyboard can also be used as the input interface 213.

[0035] The output interface 214 functions as an output unit for outputting various types of information, such as category selection screens, template selection screens, input information generation screens, or answer information screens. An example of the output interface 214 is a display composed of an LCD panel, an organic EL display, or a plasma display. However, the terminal device 200 itself does not necessarily need to be equipped with a display. For example, an interface for connecting to a display that can be connected to the terminal device 200 via a wired or wireless network can also function as the output interface 214 for outputting display data to the display.

[0036] The communication interface 215 functions as a communication unit for sending and receiving various types of information with the server device 100, which is connected via a wired or wireless network. Examples of the communication interface 215 include wired communication connectors such as USB and SCSI, wireless communication transceivers such as wireless LAN, Bluetooth®, LTE, and infrared, and various connection terminals for printed circuit boards and flexible circuit boards. For example, the communication interface 215 receives template candidate information, template information, or answer information from the server device 100, or transmits category selection information or template selection information to the server device 100.

[0037] 4. Various types of information used in processing in processing system 1 (1) Template management table Figure 3A is a conceptual diagram showing a template management table stored in a server device 100 according to one embodiment of the present disclosure. According to Figure 3A, category information, template ID information, parameter information, prompt information, and model information are stored in the template management table in association with each other.

[0038] "Category information" is information that indicates the category of the input information that is input into the trained model. Examples of such category information include various categories such as business, marketing, parenting, career development, job change, job hunting, finance, investment, health, beauty, fitness, lifestyle, hobbies, entertainment, technology, IT, law, social norms, or creative. Category information may be set in advance by the administrator of processing system 1, or it may be set based on the user's request or selection. Note that in Figure 3A, two categories of category information, "business" and "parenting," are shown for the sake of explanation, but of course, there may be only one of these categories, or there may be three or more.

[0039] "Template ID information" is information used to identify template information that includes at least parameter information and prompt information, and is unique to each template information. As shown in Figure 3A, one or more template ID pieces are stored in association with category information. That is, for each category, one or more template pieces that can be used for various purposes are stored. Each template piece is registered, for example, for each task in each category identified by each category information. In other words, template information is prepared to correspond to various tasks such as summarization, drafting, research, analysis, translation, revision, dialogue, programming, presentation material creation, image generation, reasoning, deduction, or idea generation. In Figure 3A, the category information shown in A1 (Business) is associated with the template ID pieces B1-1, B1-2, B1-3, and B1-4, and each template ID piece is associated with the template information corresponding to the drafting tasks shown in "Product Description," "Press Release," "Quotation," and "Approval Document." Furthermore, Figure 3A shows that the category information indicated by A2 (childcare) is associated with the template ID information B2-1, B2-2, and B2-3, and that each template ID information is associated with template information corresponding to the idea generation task indicated by "menu," the research task indicated by "hospital," and the document creation task indicated by "notification form creation." Naturally, each category information may be associated with only one template ID information, or with multiple template ID information. In addition, although not specifically illustrated in Figure 3A, other categories may store template ID information to identify template information corresponding to other tasks. Such template ID information is generated each time new template information is registered, and each template information may be registered in advance by the administrator, or it may be registered based on user requests or selections.

[0040] "Parameter information" is information stored in association with each template ID information and is used to generate input information that is input to the trained model. That is, each template information identified by the template ID information contains one or more parameter information. Such parameter information can be represented in any format, such as text, numbers, symbols, images, sounds, or combinations thereof. Parameter information is user-modifiable information and can be changed, for example, based on user input to obtain the desired input information. In Figure 3A, two to five parameter information items are associated with the template ID information, but of course, only one parameter information item may be associated, or multiple parameter information items may be associated. Parameter information may be pre-registered by the administrator for each template information identified by the template ID information, or it may be registered based on user requests or selections.

[0041] "Prompt information" is information, often referred to as a question statement, that is stored in association with each template ID and is used to generate input information that is fed into the trained model. In other words, one or more prompt information entries are included for each template information identified by the template ID. Such prompt information can be presented in any format, such as text, numbers, symbols, images, audio, or combinations thereof. It is desirable that user modification of prompt information be restricted, and if it is modified, it is desirable that it be stored separately from the original template information as new template information. In Figure 3A, one prompt information entry is associated with the template ID, but of course, any number of prompt information entries may be associated. Prompt information may be pre-registered by the administrator for each template information identified by the template ID, or it may be registered based on user requests or selections.

[0042] "Model information" is information for identifying a trained model into which input information generated based on template information is input. Such model information can include, for example, various types of information such as model ID information for identifying each trained model, a name indicating each trained model, information indicating the storage location of each trained model, or a combination thereof.

[0043] Here, FIG. 4 is a diagram conceptually showing an example of a selection process of a trained model in the server device 100 according to an embodiment of the present disclosure. Specifically, FIG. 4 is a diagram showing the connection relationship between each processing device storing one or more trained models into which input information generated based on template information is input and the server device 100. According to FIG. 4, an arbitrary number of processing devices 300n such as a processing device 300A capable of processing based on the trained model A, a processing device 300B capable of processing based on the trained model B, and a processing device 300C capable of processing based on the trained model C are connected to the server device 100. When input information is generated, the server device 100 transmits the input information and a processing request to the processing device storing the trained model specified by the model information in FIG. 3A, and acquires response information from the processing device.

[0044] Although FIG. 4 describes the case where each process by the trained models including the trained models A to C is executed by each processing device, it is of course possible that the server device 100 reads each trained model from the memory and the server device 100 executes it. Also, the number of trained models to be executed does not have to be one for one input information, and multiple trained models may be executed for one input information, and multiple response information or response information obtained by ensemble processing of multiple output information may be acquired.

[0045] Also, for each trained model, it is possible to use any trained model such as a trained model obtained by machine learning of the combination of the learning data and the correct label exemplified above, or a generative trained model.

[0046] Returning to FIG. 3A again, as shown in FIG. 4, the model information stores information specifying which of one or more learned models including the learned model A, the learned model B, and the learned model C the input information is input to. The model information may be registered in advance by an administrator. As described in FIG. 4, the model information in FIG. 3A may store information specifying only one learned model, or may store information specifying a plurality of learned models. Further, when information specifying a plurality of learned models is stored, the model information may store information specifying a method of generating response information, such as whether to output a plurality of output information output from each learned model as response information as it is, or to perform ensemble processing on the plurality of output information output from each learned model and output it as response information. Also, in the example of FIG. 3A, the model information is stored for each template ID information, but different model information may be stored for each category information, different model information may be stored for each user, or the same model information may be stored regardless of the template or category.

[0047] Each piece of information stored in the template management table is typically stored in the memory 112 of the server device 100, but of course, it may be stored in a database device or other server devices. Also, each piece of information shown in FIG. 3A is an example of the information stored in the template management table, and of course, other information may be stored.

[0048] Note that FIG. 3A shows the case where a one - level category is set, but a multi - level classification may be made, such as setting sub - categories such as publicity, accounting, or personnel under the category, or further setting a finer classification under the sub - category.

[0049] Furthermore, although not specifically shown in Figure 3A, display priority information may be set for each category, each template, and each parameter. For example, if there are category information A1 to A10, a display priority may be set in advance for each category information. Similarly, if there are template ID information B1-1 to B1-4, a display priority may be set in advance for each template ID. Similarly, if there are parameter information C1-1-1 to C1-1-3, a display priority may be set in advance for each parameter. By making such settings, for example, when displaying category icons corresponding to each category information on the category selection screen 10 (Figure 7A), the processor 211 of the terminal device 200 may arrange the category icons in an order based on the display priority. Similarly, when displaying template selection areas corresponding to each template candidate information on the template selection screen 20 (Figure 7B), the processor 211 of the terminal device 200 may arrange the template selection areas in an order based on the display priority. Furthermore, when displaying parameter information display areas corresponding to each parameter information on the input information generation screen 30 (Figure 7C), the processor 211 of the terminal device 200 may arrange the parameter information display areas in an order based on the display priority.

[0050] Furthermore, by making such settings, for example, when displaying category icons corresponding to each category information on the category selection screen 10 (Figure 7A), the processor 211 of the terminal device 200 may arrange, for example, the top three category icons according to the display priority. Also, when displaying template selection areas corresponding to each template candidate information on the template selection screen 20 (Figure 7B), the processor 211 of the terminal device 200 may arrange, for example, the top three template selection areas according to the display priority. Also, when displaying parameter information display areas corresponding to each parameter information on the input information generation screen 30 (Figure 7C), the processor 211 of the terminal device 200 may arrange, for example, the top three parameter information display areas according to the display priority.

[0051] (2) User Management Table Figure 3B is a conceptual diagram showing a user management table stored in a server device 100 according to one embodiment of the present disclosure. According to Figure 3B, the user management table stores user name information, attribute information, model specification information, etc., associated with user ID information.

[0052] "User ID information" is unique information for each user, used to identify each user. Examples of such user ID information include various pieces of information such as arbitrary identification information assigned by the server device 100, email addresses, SNS accounts, telephone numbers, or combinations thereof. User ID information is generated by the server device 100, for example, each time a request for new registration as a user of a service provided by the processing system 1 via the terminal device 200 is received.

[0053] "Username information" is information that indicates the name of each user. Such username information can include various types of information, such as the user's name, nickname, SNS account name, or a combination of these. Username information is, for example, information specified by the user when registering as a new user on the aforementioned service.

[0054] "Attribute information" refers to information that indicates the attributes of each user. Examples of such attribute information include various types of information such as authentication information, gender, age, occupation, address, telephone number, email address, SNS account, workplace, membership rank of the above service, past usage history of the above service, search and browsing history of websites, etc., or combinations thereof. In this embodiment, as shown in Figure 4, one or more trained models are selected and input information is entered, but attribute information can be used to select the trained model. Attribute information is, for example, information specified by the user when registering as a new user for the above service, or information collected by the server device 100.

[0055] "Model specification information" is information used to identify which of one or more trained models will receive the input information generated based on template information. Examples of such model specification information include model ID information to identify each trained model, a name indicating each trained model, information indicating the storage location of each trained model, or a combination of these. As shown in Figure 4, the model specification information stores information that identifies which of one or more trained models, including trained model A, trained model B, and trained model C, will receive the input information. This model specification information is generated when the processor 211 of the terminal device 200 receives user input via the input interface 213 and selects the trained model desired by the user from among one or more trained models. In other words, in this embodiment, it is also possible for the user to select the trained model to receive the input. Note that, similar to the model information in Figure 3A, information that identifies only one trained model may be stored, or information that identifies multiple trained models may be stored. Furthermore, if information identifying multiple pre-trained models is stored, the model specification information may also include information specifying how the response information is generated, such as whether to output the multiple output pieces from each pre-trained model as response information as is, or to perform ensemble processing on the multiple output pieces from each pre-trained model and output them as response information.

[0056] Each piece of information stored in the user management table is typically stored in the memory 112 of the server device 100, but it may, of course, be stored in a database device or other server devices. Also, the information shown in Figure 3B is just an example of the information stored in the user management table, and of course, other information may be stored as well.

[0057] (3) Model Management Table Figure 3C is a conceptual diagram showing a model management table stored in a server device 100 according to one embodiment of the present disclosure. According to Figure 3C, condition information is stored in association with model ID information.

[0058] "Model ID information" is unique information for each trained model into which input information is received, and is used to identify each trained model. In this embodiment, as shown in Figure 4, it is possible to use multiple trained models, and such model ID information is assigned to each trained model by the server device 100 each time that trained model is newly registered as an available model.

[0059] "Conditional information" refers to the conditions under which input information is entered and response information is obtained for each trained model identified by the model ID information; in other words, it indicates the conditions for using each trained model. Here, each trained model differs in the information used for machine learning, the machine learning method, performance, specifications, or the learner used for machine learning. For example, each trained model may differ in whether the input information is used as training data, whether confidential information such as personal information can be entered, or whether it can be used by minors. Therefore, it is possible to set conditions that can be determined based on the user's attribute information as conditional information. Examples of such conditional information include whether the user's age is above a certain age, whether the user's address is in a certain country, whether the user's workplace allows the use of trained models, whether the service's lowest rank is above a certain rank, or a combination of these, and so on.

[0060] Each piece of information stored in the model management table is typically stored in the memory 112 of the server device 100, but it may, of course, be stored in a database device or other server devices. Also, the information shown in Figure 3C is just an example of the information stored in the model management table, and of course, other information may be stored there as well.

[0061] 5. Processing Sequence Executed in Processing System 1 Figure 5 is a diagram showing the processing sequence executed in Processing System 1 according to one embodiment of the present disclosure. Specifically, Figure 5 is a diagram showing a series of processing flows related to the generation of input information to be input to a trained model.

[0062] According to Figure 5, the processor 211 of the terminal device 200 receives user input via the input interface 213 and starts an application program for using the service provided via the processing system 1 (S11). The processor 211 then receives user input via the input interface 213 and inputs user ID information, which has been registered in advance as a user who can use the service, and authentication information for authenticating the user (S12). When the user ID information and authentication information are input, the processor 211 transmits the user ID information and so on (T11) to the server device 100 via the communication interface 215.

[0063] When the processor 111 of the server device 100 receives user ID information and authentication information from the terminal device 200 via the communication interface 113, it authenticates whether the user is a legitimate user of the service by referring to the user ID information and attribute information in the user management table (S13). If the processor 111 is authenticated as a legitimate user, it reads user information from the user management table based on the user ID information, including attribute information, model specification information, or username information of the user. The processor 111 then transmits the read user information (T12) to the terminal device 200 that sent the user ID information via the communication interface 113.

[0064] When the processor 211 of the terminal device 200 receives user information via the communication interface 215, it generates a category selection screen 10 for selecting category information that identifies the category of template information to be used as a template, based on the received user information, in order to generate input information to be input into at least one of one or more trained models configured to output response information when arbitrary input information is input.Then the processor 211 outputs the generated category selection screen 10 to the display via the output interface 214 (S14).

[0065] Here, Figure 7A shows an example of a category selection screen 10 output by the terminal device 200 in an embodiment of the present disclosure. Specifically, Figure 7A shows an example of a category selection screen 10 output in S14 of Figure 5. According to Figure 7A, the category selection screen 10 includes a category selection area 11, a free word input area 12, a template selection area 13, and a My Page button 17.

[0066] The category selection area 11 includes pre-prepared category icons 14 that correspond to category information stored in the template management table. Specifically, a total of eight icons are output as category icons 14, including an icon corresponding to the category information for A1 (Business) and an icon corresponding to the category information for A2 (Childcare). Each category icon 14 is selected by receiving user input for any of the category icons via the input interface 213.

[0067] In Figure 7A, eight icons are shown as category icons 14, but naturally, a different number of icons may be output depending on the number of categories stored in the category information. Furthermore, instead of outputting icons corresponding to all category information stored in the template management table, category icons 14 may output icons corresponding to some of the category information selected according to the user's attribute information. This allows the user to select the desired category information more efficiently.

[0068] The free word input area 12 also includes an input box 15 and a send button (not shown). The input box 15 outputs any string selected by the user through the input interface 213. The string entered in the input box 15 is then sent by receiving user input to the send button (not shown) through the input interface 213. An example of such a string is a string that includes keywords related to category information or template information, such as "I want to create a quotation for business use." Once such a string is entered and the send button is pressed, the user transitions to a chat screen, where input information can be generated in a chat format.

[0069] Furthermore, the template selection area 13 includes one or more template selection icons 16 that are pre-configured to correspond to templates identified by the template information. Such template selection icons are output in accordance with template information that has been pre-selected according to the frequency with which templates are used by the user in question or by multiple users including that user. In other words, template information that is frequently used can be selected more efficiently without going through processes such as category selection.

[0070] Furthermore, the My Page button 17 is a button for transitioning to the My Page screen. That is, by accepting user input for the My Page button 17 via the input interface 213, it is possible to check and change user information such as model specification information, membership rank, age, address, workplace, or past usage history of the service (template information used in the past or input information generated in the past).

[0071] As shown in Figure 7A, when generating input information, the user can efficiently select their desired category via the category selection screen 10.

[0072] Returning to Figure 5, when the category selection screen 10 illustrated in Figure 7A is displayed, the processor 211 of the terminal device 200 selects one or more category icons 14 contained in the category selection area 11 of the category selection screen 10 based on the user's operation input received via the input interface 213 (S15). In the example in Figure 7A, the category icon A1 is selected. The processor 211 then generates category selection information (for example, category information for A1) to identify the category corresponding to the selected category icon 14, and transmits the generated category selection information (T12) along with the user ID information to the server device 100 via the communication interface 215.

[0073] When the processor 111 of the server device 100 receives category selection information from the terminal device 200 via the communication interface 113, it generates template candidate information (S16) that will be used to generate input information, based on the received category selection information (for example, category information for A1). Specifically, the processor 111 refers to the template management table based on the received category selection information (for example, category information for A1) and reads the template ID information associated with the category information specified by the category selection information. The processor 111 then generates template candidate information with the templates specified by the read template ID information as candidates. For example, if category information for A1 is received as category selection information, the processor 111 generates template candidate information with the templates specified by the template ID information for B1-1, B1-2, B1-3, and B1-4 as candidates.

[0074] Furthermore, the processor 111 does not need to consider all templates identified by the template ID information associated with the category information as candidates. The processor 111 may only read out templates identified by the template ID information specified in advance based on the user's attribute information (for example, templates B1-1 and B1-2 depending on the user's workplace) as candidates. By pre-selecting templates in this way, the user can generate input information more efficiently.

[0075] Once template candidate information is generated, the processor 111 of the server device 100 transmits the template candidate information (T13) to the terminal device 200 that sent the category selection information via the communication interface 113.

[0076] In this explanation, we described the case where a category is selected on the category selection screen 10 in Figure 7A, the category selection information is sent from the terminal device 200 to the server device 100, template candidate information is generated based on the category selection information and sent from the server device 100 to the terminal device 200. However, for example, if subcategories such as public relations, accounting, or human resources are set under the category in the template management table in Figure 3A, then the server device 100 may send subcategory candidate information indicating subcategory candidates based on the category selection information to the terminal device 200, and then the terminal device 200 may send subcategory selection information indicating the selected subcategory to the server device 100.

[0077] When the processor 211 of the terminal device 200 receives template candidate information via the communication interface 215, it generates a template selection screen 20 for selecting template candidate information to be used as a template based on the received template candidate information. The processor 211 then outputs the generated template selection screen 20 to the display via the output interface 214 (S17).

[0078] Here, Figure 7B shows an example of a template selection screen 20 output in the terminal device 200 in an embodiment of the present disclosure. Specifically, Figure 7B shows an example of a template selection screen 20 output in S17 of Figure 5. According to Figure 7B, the template selection screen 20 includes template selection areas 21 to 24, along with information indicating that a business category has been selected.

[0079] The template selection areas 21 to 24 are output in accordance with the template ID information that identifies each template included in the template candidate information. In other words, the example in Figure 7B shows a case where the template candidate information includes templates corresponding to the template ID information of B1-1, B1-2, B1-3, and B1-4, and each area is output in accordance with each of these templates. Therefore, each template selection area 21 to 24 includes, in addition to the icon output in accordance with each template ID information, the title (product description, release statement, quotation, or approval document) and summary of each template associated with each template ID information. Each template selection area 21 to 24 is selected by receiving user input for any of the areas via the input interface 213.

[0080] As shown in Figure 7B, by accepting operation input for the template selection areas 21 to 24 on the template selection screen 20, the user can select the desired template more efficiently. Furthermore, by outputting the title and summary information of each template in each template selection area, the user can select the desired template even more efficiently.

[0081] Returning to Figure 5, when the template selection screen 20 illustrated in Figure 7B is displayed, the processor 211 of the terminal device 200 selects one of the template selection areas 21 to 24 corresponding to one or more template candidate information included in the template selection screen 20, based on the user's operation input received via the input interface 213 (S18). As a result, the processor 211 generates template selection information including at least one template ID information corresponding to the selected template selection area. The processor 211 then transmits the generated template selection information (T14) along with the user ID information to the server device 100 via the communication interface 215.

[0082] When the processor 111 of the server device 100 receives template selection information from the terminal device 200 via the communication interface 113, it generates template information based on the template ID information contained in the received template selection information (S19). Specifically, the processor 111 refers to the template management table based on the template ID information contained in the received template selection information and reads the parameter information and prompt information associated with the said template ID information. Then, the processor 111 generates template information including the read parameter information and prompt information. For example, if template ID information B1-1 is received as template selection information, the processor 111 generates template information including parameter information C1-1-1, C1-1-2 and C1-3, and prompt information D1-1.

[0083] Once template candidate information is generated, the processor 111 of the server device 100 transmits the generated template information (T15) to the terminal device 200 that sent the template selection information via the communication interface 113.

[0084] When the processor 211 of the terminal device 200 receives template information via the communication interface 215, it generates an input information generation screen 30 for generating input information based on the received template information. The processor 211 then outputs the generated input information generation screen 30 to the display via the output interface 214 (S20).

[0085] Here, Figure 7C shows an example of an input information generation screen 30 output in the terminal device 200 in an embodiment of the present disclosure. Specifically, Figure 7C shows an example of an input information generation screen 30 output in S20 of Figure 5. According to Figure 7C, the input information generation screen 30 includes a template information area 31, a prompt information display area 32, and a parameter information display area 33.

[0086] The template information area 31 includes the title and summary of the template selected on the template selection screen 20. For example, in the example in Figure 7C, the template identified by the template ID information B1-1 is selected, and it is shown that the title of the template is "Product Description" and its summary is "xxx xxx xxx".

[0087] Furthermore, in the prompt information display area 32, below the display of "Prompt: Main Text," the following text is displayed: "You are a first-class advertising writer. Based on the following features and benefits of the following product, please create an attractive product description for the following target audience." This indicates that the text information was stored as prompt information for D1-1. Such prompt information is stored by the administrator by registering in advance prompts that are suitable for generating product descriptions. In other words, users do not need to input or create suitable prompts themselves, enabling more efficient generation of prompt information.

[0088] Furthermore, the prompt information display area 32 includes an edit icon 34. That is, when user input for the edit icon 34 is received via the input interface 213, the user can edit the prompt information output in the prompt information display area 32. Details of the processing when such editing is performed will be explained in Figure 6.

[0089] The parameter information display area 33 includes each prompt information (parameter title and its content) associated with the template ID information selected as a template. Specifically, as shown in Figure 7C, the parameter information display area 33 includes the title "Parameter 1: Product type and name" and the content "Bath pillow 'Bath Pillow'" as parameter information identified by the prompt information C1-1-1, the title "Parameter 2: Product features and benefits" and the content "High relaxation effect" as parameter information identified by the prompt information C1-1-2, and the title "Parameter 3: Target group" and the content "People who spend a long time in the bath" as parameter information identified by the prompt information C1-1-3.

[0090] Furthermore, each parameter information display area 33 includes change icons 35a to 35c at the edge of the area where the content of each prompt information is displayed. When user input is received for any of the change icons 35a to 35c via the input interface 213, the user can make changes to the content of the parameter output in the parameter information display area 33. For example, when user input is received for change icon 35a, the content of the prompt titled "Parameter 1: Product type and name," "Bath pillow 'Bath Pillow'," is displayed in an editable format. Then, when user input is received via the input interface 213, the content of the prompt is changed by the user entering an arbitrary string. The same applies when user input is received for change icons 35b and 35c.

[0091] As shown in Figure 7C, the user can efficiently check the contents of the prompt set as a template. Furthermore, the user can easily change parameters via the change icons 35a to 35c. This makes it possible to easily generate input information simply by changing parameters.

[0092] Returning to Figure 5, when the input information generation screen 30 illustrated in Figure 7C is output, the processor 211 of the terminal device 200 modifies at least one parameter information (parameter content) included in the template information based on the user's operation input received via the input interface 213 (S21). Specifically, the processor 211 selects the change icon in the parameter information display area of ​​the parameter information to be changed based on the user's operation input received via the input interface 213, and also accepts the input of an arbitrary string. The processor 211 then stores the accepted arbitrary string as the modified parameter information. For example, although not specifically shown, the modified parameter information may include the following.

[0093] - Change the parameter content of parameter information C1-1-1 to "Sports Soap 'Sposoap'" - Change the parameter content of parameter information C1-1-2 to "Contains moisturizing ingredients for sun protection" - Change the parameter content of parameter information C1-1-3 to "People who enjoy outdoor sports"

[0094] Although not specifically shown in Figure 5, as explained in Figure 7C, the processor 211 of the terminal device 200 can also edit prompt information by receiving user input via the input interface 213.

[0095] Next, the processor 211 of the terminal device 200 sends a response generation request to the server device 100 via the communication interface 215, which includes the user ID information and the modified parameter information (T16). Note that changing the parameter information in Figure 5 is not necessarily required; if it is not changed, the parameter information included in the template information sent in T15 may be included in the response generation request as is.

[0096] When the processor 111 of the server device 100 receives a response generation request including parameter information from the terminal device 200 via the communication interface 113, it acquires the input information by generating input information to be input to one of the trained models based on the received parameter information (S22). The acquired input information is stored in association with user ID information. Specifically, the processor 111 generates input information to be input to one of the trained models based on prompt information identified based on template ID information included as template selection information (edited prompt information if the prompt information has been edited) and the received parameter information. An example of such input information is as follows.

[0097] <Example of input information> "You are a top-notch advertising writer. Based on the following features and benefits of the product below, please create an attractive product description for the following target audience. • Product type and name: Sports soap "Sposoap" • Product features and benefits: Contains moisturizing ingredients for sun protection • Target audience: People who enjoy outdoor sports"

[0098] Then, the processor 111 of the server device 100 executes the answer generation process (S23) by inputting the generated input information into a trained model selected based on at least one of the model information in the template management table, the model specification information in the user management table, and the condition information in the model management table. Through this answer generation process, the processor 111 obtains answer information for the input information from one of the trained models. Once the processor 111 obtains the answer information, it transmits the obtained answer information (T17) along with the input information to the terminal device 200 that sent the answer generation request via the communication interface 113. Details of the answer generation process will be explained in Figure 6.

[0099] When the processor 211 of the terminal device 200 receives response information via the communication interface 215, it generates a response information screen 40 based on the received response information. The processor 211 then outputs the generated response information screen 40 to the display via the output interface 214 (S24).

[0100] Here, Figure 8 shows an example of the response information screen 40 output by the terminal device 200 in an embodiment of the present disclosure. Specifically, Figure 8 shows an example of the response information screen 40 output in S24 of Figure 5. According to Figure 8, the response information screen 40 includes a prompt information area 41, a response area 42, an additional question area 43, a message input area 44, and a model switching area 45.

[0101] The prompt information area 41 includes the prompt information contained in the input information input to the trained model in S23 of Figure 5. In other words, by checking the prompt information area 41, it is possible to confirm the prompt information that was actually input to the trained model. Although not specifically shown in Figure 8, parameter information included in the input information may also be included in the same way as the prompt information.

[0102] The response area 42 contains the received response information. Specifically, although not shown in Figure 8, the response area 42 includes a product description obtained as response information from a trained model according to the input information. Therefore, the user can view the obtained response information while comparing it with the input information, making it possible to check the response information more efficiently.

[0103] The additional question area 43 and the message input area 44 are areas used to ask further questions about the generated answer information and obtain updated answer information. Specifically, the additional question area 43 includes selection icons for selecting each candidate information based on candidate additional question information (not shown in Figure 3A) that has been set in advance and associated with template ID information. For example, candidate additional question information may include: - Candidate 1: Please describe the target group more clearly. - Candidate 2: Please describe the product features first. - Candidate 3: Please use the product name at least twice. - Candidate 4: Please keep the text to 100 characters or less. Each candidate information is associated with a selection icon. Therefore, when a user input is received via the input interface 213 and one of the selection icons is selected, the candidate information corresponding to that selection icon is sent to the server device 100 as prompt information. The server device 100 then inputs the prompt information as input information to the trained model again, and the updated answer information is obtained.

[0104] Furthermore, the message input area 44 includes an input box into which any string can be entered. That is, when the input box is selected after receiving user input via the input interface 213, and the user enters any string, the entered information (for example, the string "Change the product name to sports soap and further limit the target audience to women.") is sent to the server device 100 as prompt information. The server device 100 then inputs the prompt information back into the trained model as input information, thereby obtaining updated response information.

[0105] The model switching area 45 includes model selection icons for the user to select which trained model receives the generated input information. In this embodiment, as shown in Figure 4 for example, multiple trained models with different characteristics are available, and the trained model to be input is selected based on at least one of the model specification information in the user management table and the condition information in the model management table. The model selection icons are provided corresponding to each of the multiple trained models, and by selecting any of the model selection icons, it is possible to switch the trained model to which the input information is received. Specifically, when an operation input from the user is received via the input interface 213 and any of the model selection icons are selected, the model selection information corresponding to that model selection icon is transmitted to the server device 100. Then, the input information generated in S22 of Figure 5 is input again to the trained model selected by the model selection information by the server device 100, and updated answer information is obtained. As a result, it is possible to obtain answer information different from the answer information output in S24 of Figure 5 from a trained model different from the trained model from which the answer information was obtained, and the user can select the desired answer information from a wider variety of answer information.

[0106] Returning to Figure 5, the processing sequence is now complete.

[0107] Thus, since the processing system 1 generates input information for input to the trained model using pre-prepared prompt information and parameter information, the user can generate input information more efficiently. Furthermore, since the parameter information can be changed according to the user's wishes, the user can easily generate a variety of input information.

[0108] 6. The processing flow diagram 6, which is executed in the server device 100, is a diagram showing the processing flow executed in the server device 100 according to one embodiment of the present disclosure. Specifically, it is a diagram showing the processing flow of the response generation process executed in S23 of Figure 5A. This processing flow is mainly performed by the processor 111 of the server device 100 reading and executing a program stored in the memory 112.

[0109] As shown in Figure 6, the processor 111 generates input information based on the received parameter information, as explained in S22 of Figure 5, and the processing flow is started (S111). Here, as explained in Figure 7C, the prompt information can be edited when user input is received for the edit icon 34 in the prompt information display area 32 of the input information generation screen 30. For example, as shown in Figure 7C, the prompt information "You are a top-notch advertising writer. For the following product, please create an attractive product description for the following target audience, taking into account the following features and advantages." can be edited to "You are a top-notch advertising writer. For the following product, please create an attractive product description of the following target audience within 100 characters, taking into account the following features and advantages, for use in SNS advertising." Therefore, the processor 111 determines whether the prompt information has been edited based on whether or not it has received the edited prompt information along with the parameter information (S112). If the prompt information has not been edited, the processor 111 proceeds to the process in S114 without executing the process in S113.

[0110] On the other hand, if the prompt information has been edited, the processor 111 generates new template ID information and stores the edited prompt information and the parameter information received together with the prompt information as a new template in the template management table, associating them with the template ID information (S113). At this time, the category information may be set to be the same as the category information associated with the prompt information before editing. In other words, when the prompt information is edited, the processor 111 restricts the original prompt information from being rewritten while storing it as a new template associated with the template ID information. Therefore, it becomes possible to appropriately protect the template information before editing. On the other hand, in subsequent processing, the processor 111 can add the template identified by the newly generated template ID information as one or more template candidates and select it.

[0111] Next, the processor 111 refers to the template management table based on the template ID information of the template identified by the template selection information in Figure 5, and reads the model information associated with the template ID information (S114). Then, the processor 111 determines whether or not model information identifying one or more trained models is set as the model information (S115). If model information is set, the processor 111 proceeds to the process in S119 without executing the processes from S116 onwards.

[0112] On the other hand, if, for example, the model information reads out that any pre-trained model is acceptable, or if no model information is set due to a configuration error, the processor 111 refers to the user management table based on the user ID information received along with prompt information in S16 of Figure 5. The processor 111 then reads the model specification information associated with the user ID information (S116). The processor 111 determines whether or not model information specifying one or more pre-trained models is set as the model specification information (S117). If model specification information is set, the processor 111 proceeds to the process in S119 without executing the process in S118.

[0113] On the other hand, if, for example, the model specification information reads out that any pre-trained model is acceptable, or if no model specification information is set due to a setting error, the processor 111 refers to the condition information in the model management table. Then, the processor 111 refers to the user management table based on the user ID information received along with prompt information, etc., in S16 of Figure 5, and reads the attribute information associated with the user ID information. The processor 111 compares the condition information and attribute information set as conditions for the pre-trained models identified by each model ID information to identify the available pre-trained models. For example, if the condition information for each pre-trained model is set as "the user is 18 years of age or older", the processor 111 refers to the user's attribute information to determine whether the condition is met. Then, the processor 111 selects at least one pre-trained model that satisfies all the conditions (S118). If there are multiple pre-trained models that satisfy the conditions, it is also possible to further narrow down the pre-trained models to be selected by using other attribute information, for example, by selecting a pre-trained model that has been used frequently based on the user's past usage history.

[0114] The processor 111 sets at least one trained model selected by any of the model information in S114, the model specification information in S116, and the condition information in S118 as the trained model to input the input information generated in S111 (S119). Here, as described above, the trained model selected in advance as model information in the template management table is preferentially selected as the trained model to input the input information. In other words, input information is preferentially input to the trained model selected as model information rather than the trained model specified by the model specification information set by the user or the like. For example, the selection of a trained model is expected to affect the quality of the response information, but by setting the optimal trained model for the input information in advance and preferentially inputting to that trained model, it becomes possible to obtain higher quality response information.

[0115] The processor 111 inputs the input information generated in S111 to the configured trained model (S120). Specifically, as shown in Figure 4, if the configured trained model is stored in a processing unit connected via a communication network, the processor 111 transmits the input information to the processing unit via the communication interface 113. Also, if the configured trained model is stored in memory 112 or the like, the processor 111 inputs the input information to the trained model.

[0116] Then, the processor 111 executes the trained model into which the input information has been input, and obtains the answer information from the trained model (S121). If multiple trained models are set in S119, the processor 111 may obtain the answer information from each of the multiple trained models and transmit it directly to the terminal device 200. Alternatively, the processor 111 may obtain the output information from each of the multiple trained models, perform ensemble processing on this output information to obtain the answer information, and transmit it directly to the terminal device 200.

[0117] The processing flow is now terminated.

[0118] As shown in Figure 6, when prompt information is edited, the template is restricted from being rewritten, and by registering it as a new template, the user can select a suitable template from the next time onward. Furthermore, by selecting a trained model for inputting input information based on at least one of the model information, model specification information, and condition information, the user can obtain response information more efficiently.

[0119] In this embodiment, we can provide a processing device, a processing program, a processing method, and a processing system that can generate input information to be input to a trained machine learning model more efficiently.

[0120] 7. Modifications (A) Input Information and Response Information Figures 1 to 8 describe the case in which text-formatted input information is generated using text-formatted prompt information and text-formatted parameter information, and text-formatted response information is obtained based on said input information. However, as described above, input information can be in various formats other than text, such as image format (including both video and still images), audio format, document file format, print layout format, link format, or a combination thereof. Similarly, response information can be obtained in various formats other than text, such as image format (including both video and still images), audio format, document file format, print layout format, link format, or a combination thereof.

[0121] For example, processor 111 generates input information that includes a text-based prompt message, "You are a top-notch designer. Based on the following product images, please create an attractive advertising image for the following target audience, taking into account the following features and benefits," and parameter information in image format, including images of the products. Then, processor 111 inputs this input information into one of the trained models to obtain an advertising image in image format.

[0122] Another example is when processor 111 generates input information that includes a text-based prompt message, "You are a first-class composer. Based on the following product images, please create music for an advertisement that will appeal to the following target audience, taking into account the following features and benefits," and parameter information in image format, including images of the products. Then, processor 111 inputs this input information into one of the trained models to obtain music data in audio format.

[0123] (B) Category selection, template selection, and parameter changes Figures 1 to 8 describe the case where input information is generated by receiving user input via the input interface 213 on each of the screens: the category selection screen 10 in Figure 7A, the template selection screen 20 in Figure 7B, and the input information generation screen 30 in Figure 7C. However, it is also possible to generate input information in a chat format by entering an arbitrary string into the input box of the free word input area 12 in Figure 7A, for example.

[0124] For example, the processor 211 of the terminal device 200 inputs the string "I want to create a quotation for business use" into an input box based on user input received via the input interface 213. The processor 211 then sends this string to the server device 100 via the communication interface 215, and the processor 111 of the server device 100 inputs the input string into an arbitrary pre-trained model (for example, a large-scale language model) and obtains response information. The server device 100 then sends the response information to the terminal device 200 via the communication interface 113, and the processor 211 of the terminal device 200 outputs this response information on the chat screen via the output interface 214. This process of inputting strings and generating input information is repeated an arbitrary number of times to generate input information to be input into the pre-trained model.

[0125] (C) Regarding the generation of input information, Figures 1 to 8 describe the case in which the server device 100 receives parameter information from the terminal device 200 (T16 in Figure 5) and generates input information (S22 in Figure 5). However, it is also possible to generate the input information in the terminal device 200 instead. Specifically, when the parameter information is changed via the input information generation screen 30, the processor 211 of the terminal device 200 generates input information based on the prompt information output on the input information generation screen 30 and the changed parameter information. Then, the processor 211 transmits the generated input information to the server device 100 via the communication interface 215, and the server device 100 executes the processing from S23 onwards in Figure 5 based on the received input information. This makes it possible to process in the same way as the processing shown in Figure 5.

[0126] (D) Regarding the processing unit, Figures 1 to 8 describe the case where the server device 100 functions as the processing unit. However, it is also possible for the terminal device 200 to function as the processing unit instead of the server device 100, or in combination with the server device 100. For example, at least one of the processes S16, S19, S22, and S23 in Figure 5 is performed by the processor 211 of the terminal device 200 executing a program stored in memory 212. In this case, the template management table, user management table, and model management table may be stored in memory 212 as appropriate, or stored in a database device connected via a communication network and read out as appropriate according to the progress of the processing. Furthermore, the processing related to the execution of each trained model may be performed by another processing unit connected via a communication network.

[0127] 8. Addendum The configurations of each embodiment described above are illustrated below. (1) A processing device comprising at least one processor, wherein the at least one processor is configured to perform the following processes in order to generate input information to be input to at least one of one or more trained models configured such that response information is output when arbitrary input information is input: acquiring category selection information that can identify at least one category selected from one or more categories based on user input; outputting template candidate information that includes one or more template candidates used to generate the input information based on the acquired category selection information; acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting changes to the at least one parameter information based on user input.

[0128] (2) The processing apparatus according to (1), wherein the at least one processor is configured to input the acquired input information to at least one of the one or more trained models, thereby performing a process to acquire the response information from at least one of the trained models to which the input information has been input.

[0129] (3) The processing device described in (2) above, wherein the input information is input to a pre-set trained model from among the one or more trained models.

[0130] (4) The processing device described in (2) or (3) above, wherein the input information is input to a trained model specified by the user from among the one or more trained models.

[0131] (5) The processing device according to any one of (2) to (4) above, wherein the input information is preferentially input to a pre-set

[0132] (6) The processing device according to any one of the above items (2) to (5), wherein the input information is input to a trained model selected from the one or more trained models based on the user's attribute information.

[0133] (7) The processing device described in (6) above, wherein the attribute information is the age of the user.

[0134] (8) The processing apparatus according to any one of (1) to (7) above, wherein the change to at least one parameter information is made based on an operation input by the user for inputting an arbitrary string.

[0135] (9) The processing apparatus according to any one of items (1) to (8) above, wherein the template information further includes prompt information.

[0136] (10) The processing apparatus according to (9) above, wherein the input information includes the prompt information and the modified at least one parameter information.

[0137] (11) The processing apparatus according to (9) or (10) above, wherein if the prompt information is edited based on the user's operation input, a template identified by the template information including the edited prompt information is added to the candidate of one or more templates.

[0138] (12) A computer comprising at least one processor, wherein the at least one processor is configured to perform the following processes in order to generate input information to be input to at least one of one or more trained models configured such that response information is output when arbitrary input information is input: acquiring category selection information that can identify at least one category selected from one or more categories based on user input; outputting template candidate information that includes one or more template candidates used to generate the input information based on the acquired category selection information; acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting changes to the at least one parameter information based on user input.

[0139] (13) A processing method performed by at least one processor in a computer comprising at least one processor, the processing method comprising: acquiring category selection information that can identify at least one category selected from one or more categories based on user input, in order to generate input information to be input into at least one of one or more trained models configured such that response information is output when arbitrary input information is input; outputting template candidate information that includes one or more template candidates to be used to generate the input information based on the acquired category selection information, and acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information to be used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting a change to the at least one parameter information based on user input.

[0140] (14) A processing system comprising: a processing device described in any of (1) to (11) above; and a terminal device connected to the processing device via a communication network and configured to transmit category selection information and template selection information to the processing device via the communication network.

[0141] The embodiments and variations of this disclosure are presented as examples only and are not intended to limit the scope of this disclosure. These embodiments and variations can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of this disclosure. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents.

[0142] 1 Processing system 100 Server device 200 Terminal device

Claims

1. A processing device comprising at least one processor, wherein the at least one processor is configured to perform the following processes in order to generate input information to be input to at least one of one or more trained models configured such that response information is output when arbitrary input information is input: acquiring category selection information that can identify at least one category selected from one or more categories based on user input; outputting template candidate information that includes one or more template candidates used to generate the input information based on the acquired category selection information; acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting changes to the at least one parameter information based on user input.

2. The apparatus according to claim 1, wherein the at least one processor is configured to input the acquired input information to at least one of the one or more trained models, thereby performing a process to acquire the response information from at least one of the trained models to which the input information has been input.

3. The processing apparatus according to claim 2, wherein the input information is input to a pre-set trained model from among the one or more trained models.

4. The processing apparatus according to claim 2, wherein the input information is input to a trained model specified by the user from among the one or more trained models.

5. The processing apparatus according to claim 2, wherein the input information is preferentially input to a pre-set trained model from among one or more trained models, for a trained model specified by the user from among the one or more trained models.

6. The processing apparatus according to claim 2, wherein the input information is input to a trained model selected from the one or more trained models based on the user's attribute information.

7. The processing apparatus according to claim 6, wherein the attribute information is the age of the user.

8. The processing apparatus according to claim 1, wherein the change to at least one parameter information is performed based on an operation input by the user for inputting an arbitrary string.

9. The apparatus according to claim 1, wherein the template information further includes prompt information.

10. The processing apparatus according to claim 9, wherein the input information includes the prompt information and the modified at least one parameter information.

11. The processing apparatus according to claim 9, wherein if the prompt information is edited based on the user's operation input, a template identified by the template information containing the edited prompt information is added to the candidate of one or more templates.

12. A computer comprising at least one processor, wherein the at least one processor is configured to perform the following processes for generating input information to be input to at least one of one or more trained models configured such that response information is output when arbitrary input information is input: acquiring category selection information that can identify at least one category selected from one or more categories based on user input; outputting template candidate information that includes one or more template candidates used to generate the input information based on the acquired category selection information; acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting changes to the at least one parameter information based on user input.

13. A processing method performed by at least one processor in a computer having at least one processor, the processing method comprising: acquiring category selection information that can identify at least one category selected from one or more categories based on user input, in order to generate input information to be input to at least one of one or more trained models configured such that response information is output when arbitrary input information is input; outputting template candidate information that includes one or more template candidates used to generate the input information based on the acquired category selection information; acquiring template selection information that can identify at least one template selected from the template candidate information based on user input; outputting template information that includes at least one parameter information used to generate the input information based on the acquired template selection information; and acquiring the input information generated by accepting a change to the at least one parameter information based on user input.

14. A processing system comprising: a processing device according to claim 1; and a terminal device connected to the processing device via a communication network and configured to transmit category selection information and template selection information to the processing device via the communication network.

Images