system

Abstract

Provide a system. A means for receiving a request from a user; means for sending identical input to a plurality of text generation engines and collecting the text generated therefrom; a means for having each text generation engine evaluate the collected text and aggregating the evaluation results; a means of recommending the highest-rated text generation engines to users; and A system including:

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, a large number of text generation engines have been developed, each with different algorithms and characteristics. It is difficult for users to select the optimal generation engine from among them. This difficulty arises because the degree to which a particular generation engine produces appropriate output for a particular input text varies. Therefore, there is a need for a method that allows users to easily find the generation engine that best suits their needs. [Means for solving the problem]

[0005] In order to solve the above-mentioned problems, the present invention provides the following means. We propose a system including: means for accepting a request from a user; means for sending the same input to multiple text generation engines and collecting the text generated from them; means for having each text generation engine evaluate the collected text and aggregating the evaluation results; and means for recommending the most highly rated text generation engine to the user. The system also includes means for calculating an overall evaluation score for the collected text based on the evaluation results from each text generation engine and identifying the text generation engine that has obtained the highest evaluation score; and means for aggregating the evaluation results by adding the evaluation scores of each text generation engine with the output of other text generation engines to perform an overall evaluation. This allows users to easily find the text generation engine that best suits their needs.

[0006] "User" refers to the entity that operates the system and evaluates and selects the text generation engine.

[0007] "Request" refers to an instruction sent by a user to the system asking for evaluation and text generation by a generation engine.

[0008] "Text generation engine" refers to software or a device that uses specific algorithms to automatically generate new text based on input text.

[0009] "Input" refers to the text or portion thereof to be generated that is provided by the user at the time of the request.

[0010] "Collecting" refers to compiling the texts generated by multiple text generation engines into a single dataset.

[0011] "Evaluation" refers to the act of a text generation engine judging the output text of another text generation engine based on criteria such as quality, appropriateness, and relevance.

[0012] "Evaluation result" refers to the score given by a text generation engine to the output text of another generation engine.

[0013] "Aggregation" refers to the process of combining multiple evaluation results to calculate an overall evaluation score.

[0014] "Recommending" refers to the act of the system presenting the most suitable text generation engine to the user based on the aggregated results.

[0015] "System" refers to a set of devices and software that receives a user's request and automatically performs a series of operations, including the evaluation and selection process of text generation engines.

[0016] The "overall evaluation score" refers to the overall score calculated based on the evaluation results given by each text generation engine to the output text of other generation engines. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8]FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0018] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0019] First, the terms used in the following description will be explained.

[0020] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), and an APU (Accelerated Processing Unit).

[0021] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0022] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0023] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0024] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0025] [First embodiment]

[0026] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0027] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0028] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0029] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0030] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0031] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0032] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0033] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0034] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0035] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0036] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0037] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0038] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[0039] Specific examples of program processing

[0040] Step 1: Accepting a request from a user

[0041] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[0042] Step 2: Distributing input to the text generation engine

[0043] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[0044] Step 3: Text generation

[0045] Each text generation engine generates text based on the input it receives. The generated text is returned to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy with occasional rain today," and generation engine C generates "There's a high chance of rain today."

[0046] Step 4: Submit a text rating request

[0047] The server sends the collected output of each generation engine back to each generation engine, and sends a request to evaluate the output text of other generation engines. For example, generation engine A receives a request to evaluate the output text of generation engines B and C.

[0048] Step 5: Peer assessment of the text

[0049] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, generation engine A evaluates the generation result of B and gives it a score of 80, and evaluates the generation result of C and gives it a score of 70. Generation engine B also evaluates the generation results of A and C in the same way and returns their respective scores to the server.

[0050] Step 6: Aggregating the evaluation results

[0051] The server aggregates the reputation scores received from all the generation engines. For example, let's say the collected reputation scores are aggregated as follows:

[0052] Generation Engine A scores: 80 points for B, 70 points for C

[0053] Generation Engine B scores: 75 points for A, 85 points for C

[0054] Generation Engine C scores: 65 points for A, 90 points for B

[0055] Step 7: Determine and recommend the best generation engine

[0056] The server calculates the overall evaluation score and identifies the generation engine with the highest evaluation. For example, generation engine B has the highest overall evaluation score of 170 points. This result is recommended to the user.

[0057] Specific example explanation

[0058] 1. Input text: The user enters the text "What's the weather like today?"

[0059] 2. Text generation: Each generation engine (A, B, C) produces a different output.

[0060] Generation Engine A: "It's sunny today."

[0061] Generation Engine B: "It's cloudy with occasional rain today."

[0062] Generation Engine C: "There is a high chance of rain today."

[0063] 3. Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[0064] 4. Rating Aggregation: The server aggregates the rating scores and determines that engine B has the highest rating.

[0065] 5. Recommendation to the user: The server notifies the user of the best generation engine B.

[0066] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[0067] The processing flow will be explained below.

[0068] Step 1:

[0069] A user uses a terminal to send a request including an input text to be evaluated to a server. The request includes information indicating that the user wants to evaluate a text generation engine that the user wants to use.

[0070] Step 2:

[0071] The server receives a request from a user, obtains the input text included in the request, and sends it to multiple text generation engines (e.g., text generation engine A, text generation engine B, and text generation engine C).

[0072] Step 3:

[0073] Each text generation engine generates new text based on the input text received from the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[0074] Step 4:

[0075] Each text generation engine sends the generated text back to the server, which collects these results.

[0076] Step 5:

[0077] The server resends the collected generation results to each of text generation engines A, B, and C, and sends a request to evaluate the generation results of the other generation engines.

[0078] Step 6:

[0079] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points, while text generation engine C's generation, "There is a high chance of rain today," gives it a score of 70 points.

[0080] Step 7:

[0081] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[0082] Step 8:

[0083] The server aggregates the evaluation scores received from all text generation engines, for example, as follows:

[0084] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[0085] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[0086] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[0087] Step 9:

[0088] The server calculates the overall evaluation score based on the aggregated evaluation scores. For example, it calculates it as follows:

[0089] Overall score for Text Generation Engine A = 75 points + 65 points = 140 points

[0090] Overall score for Text Generation Engine B = 80 points + 90 points = 170 points

[0091] Overall score for text generation engine C = 70 points + 85 points = 155 points

[0092] Step 10:

[0093] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[0094] Step 11:

[0095] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[0096] Example 1

[0097] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0098] Conventional text generation systems rely on the performance of a single text generation engine, resulting in inconsistent quality of generated text and not necessarily optimal generation results. Furthermore, there is no established method for comparing and evaluating the output of multiple text generation engines, making it difficult to determine which engine best suits a user's needs. This makes it difficult for users to find the optimal generation engine, affecting their satisfaction with the generated results.

[0099] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0100] In this invention, the server includes means for accepting requests from a user, means for sending the same input to a plurality of text generation engines and collecting the texts generated therefrom, means for sending the collected texts to each text generation engine to have them evaluate each other and collecting the evaluation results, means for aggregating the evaluation results, and means for recommending the most highly rated text generation engine to the user. This enables the user to easily find the text generation engine that is most suitable for the text generation result they desire.

[0101] "Means for accepting requests from users" refers to the function of sending text or requests entered by a user using a terminal to a server and having the server receive them.

[0102] "Means for sending the same input to multiple text generation engines and collecting the text generated from them" refers to a function in which a server sends a specific input text to multiple different text generation engines and collects the generated results.

[0103] "Means for sending the collected text to each text generation engine to have them evaluate each other, and collecting the evaluation results" refers to the function of the server sending the collected text to each text generation engine again and having them evaluate the generation results of the other generation engines.

[0104] "Means for aggregating evaluation results" refers to the function of aggregating the evaluation results returned from each text generation engine and calculating the total score and average score.

[0105] The "means for recommending the most highly rated text generation engine to the user" refers to a function for identifying the text generation engine that has received the highest score based on the evaluation results and notifying the user of that information.

[0106] "Calculating an overall evaluation score" refers to calculating an overall evaluation point based on the evaluation scores from each text generation engine.

[0107] "Mutual evaluation" refers to an evaluation method in which each text generation engine evaluates the results of the other engines' generation and returns the evaluation results in numerical form.

[0108] "Text generation engine" refers to an algorithm or software that generates natural language text based on specific input.

[0109] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[0110] Specific implementation methods

[0111] 1. Hardware and software configuration

[0112] This system consists of a server running on the cloud, user devices (PCs, smartphones, tablets, etc.), and multiple text generation engines, including OpenAI (registered trademark) GPT-3 (registered trademark) and Google (registered trademark) BERT.

[0113] 2. Accepting a request from a user

[0114] A user uses a device to input a specific prompt sentence (e.g., "What's the weather like today?") via the system's web interface or mobile app. The device sends this request to the server, which accepts the request.

[0115] 3. Input distribution to multiple text generation engines

[0116] The server sends the received input text to multiple pre-configured text generation engines, each of which analyzes the received input text and returns the generated text to the server.

[0117] 4. Text Generation

[0118] Each text generation engine (for example, engine A, engine B, and engine C) sends its generated text to the server based on the input "What's the weather like today?". As a specific example, engine A's generated text is "It's sunny today," engine B's is "It's cloudy with occasional rain today," and engine C's is "There's a high chance of rain today."

[0119] 5. Submitting a Text Rating Request

[0120] The server sends the collected output of each generation engine to each generation engine again, and has them evaluate the output text of other generation engines. For example, it sends a request to engine A to evaluate the output text of engines B and C.

[0121] 6. Peer evaluation of texts

[0122] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, engine A gives a score of 80 to the generation result of engine B and a score of 70 to engine C. Engine B similarly evaluates the output of engines A and C and returns their respective scores to the server.

[0123] 7. Calculation of evaluation results

[0124] The server aggregates the ratings received from all the generation engines and identifies the generation engine with the highest overall rating score. For example, if the following rating results are obtained:

[0125] Generation Engine A scores: 80 points for B, 70 points for C

[0126] Generation Engine B scores: 75 points for A, 85 points for C

[0127] Generation Engine C scores: 65 points for A, 90 points for B

[0128] 8. Determining and recommending the best generation engine

[0129] The server sums up the evaluation scores and recommends the highest-rated generation engine (e.g., engine B in this case) to the user.

[0130] Specific example explanation

[0131] Input text: The user enters the text "What's the weather like today?"

[0132] Text generation: Each generation engine (Engine A, Engine B, Engine C) produces a different output.

[0133] Generation Engine A: "It's sunny today."

[0134] Generation Engine B: "It's cloudy with occasional rain today."

[0135] Generation Engine C: "There is a high chance of rain today."

[0136] Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[0137] Rating aggregation: The server aggregates the rating scores and determines that engine B received the highest rating.

[0138] Recommendation to user: The server notifies the user of the best generation engine B.

[0139] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[0140] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0141] Step 1:

[0142] A user uses a terminal to input the input text "What's the weather like today?" into the system. This input text is sent from the user's terminal to the server. In this case, the prompt sentence ("What's the weather like today?") becomes the input data.

[0143] Step 2:

[0144] The server distributes the user's input text to multiple text generation engines (generative AI models). The server sends the same input text to each text generation engine (e.g., Engine A, Engine B, and Engine C). At this stage, the user's prompt sentence is sent as input to each generation engine.

[0145] Step 3:

[0146] Each text generation engine receives input text and generates text using a generation algorithm. For example, Engine A generates "It's sunny today," Engine B generates "It's cloudy with occasional rain today," and Engine C generates "There's a high chance of rain today." Each generation result is sent back to the server. Here, the generative AI model performs data calculations based on the input text and outputs the text generation result.

[0147] Step 4:

[0148] The server again sends the collected output text of the generation engines to each generation engine, and the generation results of the other generation engines are evaluated against each other. Specifically, the server sends the generation results of engines B and C to engine A, the generation results of engines A and C to engine B, and the generation results of engines A and B to engine C. At this stage, the different generation results are sent to each engine as input data for evaluation.

[0149] Step 5:

[0150] Each generation engine evaluates the output text of other generation engines and returns the evaluation score to the server. For example, Engine A gives Engine B's generation result 80 points and Engine C 70 points. Similarly, Engine B gives Engine A 75 points and Engine C 85 points. Engine C also evaluates the other generation results and returns the score to the server. Here, the score is output based on the evaluation data for the input generation result.

[0151] Step 6:

[0152] The server aggregates the ratings received from all the generation engines, and calculates the overall rating score for each engine, resulting in something like this:

[0153] Generation Engine A scores: 80 points for B, 70 points for C

[0154] Generation Engine B scores: 75 points for A, 85 points for C

[0155] Generation Engine C scores: 65 points for A, 90 points for B

[0156] At this stage, the aggregated evaluation scores are output.

[0157] Step 7:

[0158] The server will aggregate the evaluation scores and identify the generation engine that received the highest score. For example, let's say generation engine B received the highest overall evaluation score of 170 points. The server will notify the user of this result. Here, the generation engine with the highest evaluation score will be output.

[0159] This series of processes allows the user to find the optimal text generation engine.

[0160] (Application example 1)

[0161] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0162] With conventional text generation technology, it is difficult to select the optimal engine from the many natural language generation engines available, making it impossible to obtain text of the quality desired by users. Furthermore, especially on online shopping sites, the quality of text generated when generating product descriptions and customer reviews significantly affects the user experience, so there is a demand for higher quality text generation.

[0163] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0164] In this invention, the server includes means for accepting requests from users, means for sending the same input to multiple natural language generation engines and collecting the text generated therefrom, means for having each natural language generation engine evaluate the collected text and aggregating the evaluation results, means for recommending the most highly rated natural language generation engine to the user, and means for generating product descriptions using an optimal natural language generation engine based on the collected text, thereby enabling users to obtain high-quality product descriptions and reviews using an optimal quality text generation engine.

[0165] "Means for accepting requests from users" refers to a device or interface that allows the server to accept information or requests entered by users.

[0166] "Multiple natural language generation engines" are multiple computer programs that generate sentences based on different algorithms or models.

[0167] "Means for sending identical input and collecting the text generated from them" refers to a device or software that allows a server to send the same input data to multiple natural language generation engines and collect the output results.

[0168] The "means for causing each natural language generation engine to evaluate the collected text" refers to a device or software that instructs each generation engine to evaluate the text generated by the other generation engines.

[0169] The "means for aggregating the evaluation results" refers to a device or software that compiles and analyzes the evaluation results from each generation engine and calculates an overall evaluation.

[0170] The "means for recommending the most highly rated natural language generation engine to the user" is a device or software that notifies the user of the best performing generation engine based on the evaluation results.

[0171] A "means for generating a product description" is a device or software that uses a specific generation engine to generate text describing the details and characteristics of a product based on collected text.

[0172] The present invention relates to a system that utilizes multiple natural language generation engines based on user input and recommends the most suitable engine from among them.

[0173] This system consists of a terminal that accepts requests entered by users, a server that sends the same input to multiple generation engines and collects and evaluates the text generated by each generation engine, and a system that recommends and provides the optimal generation engine and generated text to the user.

[0174] The main components of the system are:

[0175] 1. How to accept requests from users:

[0176] Sends information entered by the user at the terminal (e.g., product description prompts) to the server.

[0177] 2. A means of sending identical input to multiple natural language generation engines and collecting the text generated by them:

[0178] The server receives input from the user and sends it to multiple generation engines such as generation engine A, generation engine B, and generation engine C.

[0179] 3. A method for evaluating the collected text using each natural language generation engine and aggregating the evaluation results:

[0180] The server collects the generated text returned by each generation engine and has the other generation engines evaluate the text.

[0181] 4. How to recommend the highest-rated natural language generation engines to users:

[0182] The evaluation results are compiled, and the server recommends the generation engine with the highest score to the user.

[0183] 5. Using the collected text to generate product descriptions using the best natural language generation engine:

[0184] The server uses an optimal generation engine to generate a high-quality product description and presents it to the user.

[0185] Hardware and Software

[0186] Hardware: Smartphones, tablets, PCs, servers

[0187] Software: Natural language generation API, web frameworks such as Python and Flask, HTTP request libraries (such as requests)

[0188] Data processing and calculation

[0189] The server's actions include:

[0190] Data collection: Collect the generated text from each generation engine.

[0191] Data evaluation: Have each generation engine evaluate the text generated by other generation engines.

[0192] Score calculation: Calculate the overall evaluation score based on the evaluation results.

[0193] Engine recommendation: Recommend the highest scoring generation engine to the user.

[0194] Specific examples

[0195] As an example of a specific prompt sentence, the user may enter, "Please tell me the features of my new smartphone." In this case, the server will perform the following process.

[0196] Generator A: "This smartphone features fast charging."

[0197] Generator B: "It has the latest OS and is waterproof."

[0198] Generator C: "The large battery capacity and high-quality camera are attractive features."

[0199] The server collects the output of each generation engine and evaluates them to select the optimal generation engine and text, which it then provides to the user. In this way, the system generates high-quality text and improves user satisfaction.

[0200] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0201] Step 1:

[0202] A user uses a terminal to enter and send a request.

[0203] Input: A prompt asking the user for a product description (e.g., "What are the features of my new smartphone?")

[0204] Action: The terminal sends this input to the server. Output: The server accepts the prompt.

[0205] Step 2:

[0206] The server sends the received prompts to multiple natural language generation engines (generation engine A, generation engine B, generation engine C, etc.).

[0207] Input: The prompt the server received from the user

[0208] Operation: The server sends the same prompt to generation engine A, generation engine B, and generation engine C. Output: Text generation begins in each generation engine.

[0209] Step 3:

[0210] Each generation engine generates text based on the prompts it receives and sends that text back to the server.

[0211] Input: The prompt received by each generation engine

[0212] Operation: Generation Engine A, Generation Engine B, and Generation Engine C each generate text using their own algorithm. Output: Generated text (Example: Generation Engine A: "This smartphone features fast charging.")

[0213] Step 4:

[0214] The server again sends the collected text to each generation engine and sends a request to evaluate the output text of the other generation engines.

[0215] Input: Text generated by each generation engine

[0216] Operation: The server asks generation engine A to evaluate the texts generated by generation engines B and C. Similarly, it requests evaluations for all other combinations. Output: Each generation engine returns an evaluation score for the texts generated by the other generation engines.

[0217] Step 5:

[0218] Each generation engine evaluates the output text of the other generation engines and returns its evaluation score to the server.

[0219] Input: Text to be evaluated (generated by other generation engines)

[0220] Operation: The evaluation engine evaluates the received text and calculates a score. Output: Evaluation score (e.g., generation engine A gives generation engine B's text a score of 80).

[0221] Step 6:

[0222] The server aggregates the rating scores received from all the generation engines.

[0223] Input: Evaluation scores sent by all generation engines

[0224] Operation: The server aggregates the evaluation scores for each generation engine and calculates the overall evaluation score. Output: Overall evaluation score (e.g., generation engine B received the highest overall score).

[0225] Step 7:

[0226] The server calculates an overall rating score and identifies the generation engine with the highest rating.

[0227] Input: Overall evaluation score

[0228] Behavior: The server selects the best engine and its generated text. Output: The best engine (e.g. engine B).

[0229] Step 8:

[0230] The server generates product descriptions and reviews using an optimal generation engine and provides the final text to the user.

[0231] Input: The optimized generation engine and the generated text

[0232] Behavior: The server uses the best generation engine to finalize the text containing the product description and send it to the user. Output: High-quality product description text provided to the user (e.g., "This smartphone has fast charging capabilities and the latest OS.")

[0233] This processing step enables the system to recommend the most appropriate natural language generation engine for the user's request and generate and provide the most appropriate text.

[0234] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0235] The present invention relates to a system that allows multiple text generation engines to evaluate each other's outputs, recognizes the user's emotions, and takes this information into account to recommend the most suitable generation engine. This system is composed of the following elements: a means for accepting user requests, a means for sending input to multiple text generation engines, a means for evaluating the collected text, a means for aggregating the evaluation results of the generation engines, an emotion engine that recognizes the user's emotions, and a means for recommending the most suitable generation engine to the user.

[0236] Specific examples of program processing

[0237] Step 1: Accepting a request from a user

[0238] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[0239] Step 2: Recognizing user emotions with the emotion engine

[0240] The server uses the input text from the user and the past dialogue history to have the emotion engine recognize the user's emotion. For example, if the input text includes "I'm very tired today," the emotion engine will recognize the user's emotional state as "fatigue."

[0241] Step 3: Distributing input to the text generation engine

[0242] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[0243] Step 4: Text Generation

[0244] Each text generation engine generates new text based on the input text received from the server. The generated text is returned to the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[0245] Step 5: Submit a text rating request

[0246] The server sends the collected output of each generation engine to each generation engine again, and sends a request to each generation engine to evaluate the generation results of other generation engines. It also evaluates how well the generated text matches the user's emotion, based on the user's emotion recognized by the emotion engine.

[0247] Step 6: Peer assessment of the text

[0248] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "It's cloudy with occasional rain today," and assigns a score taking into account the evaluation of the emotion engine. It also evaluates the result of text generation engine C's generation, "There is a high chance of rain today."

[0249] Step 7: Return your evaluation score

[0250] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[0251] Step 8: Aggregating the evaluation results

[0252] The server aggregates the evaluation scores received from all text generation engines and also integrates the emotional relevance scores from the emotion engine. For example, the evaluation scores are collected as follows:

[0253] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[0254] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[0255] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[0256] Step 9: Calculating the overall score

[0257] The server calculates the overall evaluation score based on the aggregated evaluation scores and emotional compatibility scores. For example, it calculates it as follows:

[0258] Overall evaluation score for text generation engine A: 75 points (evaluation score) + 65 points (emotional compatibility) = 140 points

[0259] Overall evaluation score for Text Generation Engine B: 80 points (evaluation score) + 90 points (emotional relevance) = 170 points

[0260] Overall evaluation score for text generation engine C: 70 points (evaluation score) + 85 points (emotional relevance) = 155 points

[0261] Step 10: Determine the best generation engine

[0262] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[0263] Step 11: User Recommendations

[0264] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[0265] In this way, the system of the present invention makes it possible to easily find the most suitable text generation engine while taking into account the emotional state of the user.

[0266] The processing flow will be explained below.

[0267] Step 1:

[0268] The user uses a terminal to send a request to the server, including the input text to be evaluated. The request also includes the content and purpose of the text generation desired by the user.

[0269] Step 2:

[0270] The server receives a request from a user, obtains the input text included in the request, and simultaneously sends a request to the emotion engine to recognize the user's emotion from the input text.

[0271] Step 3:

[0272] The emotion engine analyzes the user's input text and recognizes the user's current emotional state. For example, from the text "I'm very tired today," the emotion engine recognizes the emotional state as "fatigue." The emotion engine then sends the recognition result back to the server.

[0273] Step 4:

[0274] The server sends the received input text to multiple text generation engines (for example, text generation engine A, text generation engine B, and text generation engine C). Each of the text generation engines generates new text based on the input text.

[0275] Step 5:

[0276] Each text generation engine sends the newly generated text back to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy today with occasional rain," and generation engine C generates "There's a high chance of rain today."

[0277] Step 6:

[0278] The server again sends the collected generated text to each text generation engine and sends a request to evaluate the generated results of the other generation engines. It also sends the generated text to the emotion engine and sends a request to evaluate how well it matches the recognized user emotion.

[0279] Step 7:

[0280] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points. It also gives the result of text generation engine C, "There is a high chance of rain today," a score of 70 points.

[0281] Step 8:

[0282] The emotion engine evaluates the suitability of each generated text to the user's emotion and sends the score back to the server. For example, a high score is given to the generated result that best suits the emotional state of "fatigue."

[0283] Step 9:

[0284] The server aggregates the evaluation scores received from each generation engine and the emotional relevance scores from the emotion engine. For example, it calculates a total evaluation score of the evaluation scores and emotional relevance scores of each text generation engine.

[0285] Step 10:

[0286] The server identifies the best-rated text generation engine based on the aggregated overall rating score. For example, if text generation engine B has the highest rating with an overall rating score of 170, this engine is considered the best.

[0287] Step 11:

[0288] The server recommends the optimal text generation engine B to the user. The recommendation result is notified to the user's terminal, and the output of the selected generation engine is provided to the user.

[0289] Example 2

[0290] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0291] Conventional text generation systems often provide generated text and data without considering the user's emotional state, resulting in low user satisfaction. Furthermore, using only a single generative model can lead to variations in the quality of generated results. Furthermore, when using multiple generative models, there is no way to evaluate and compare the generated results, making it difficult to select an efficient generative model.

[0292] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0293] In this invention, the server includes means for accepting requests from a user, means for sending the same input to multiple generative models and collecting data generated from them, means for having each generative model evaluate the collected data and aggregating the evaluation results, means for recommending the most highly rated generative model to the user, means for recognizing the user's emotions, and means for correcting the evaluation results based on the emotions. This makes it possible to select the most appropriate generative model and recommend it to the user, taking into account the user's emotional state.

[0294] "User" means a person or institution that receives data generated using the system.

[0295] A "request" is an instruction or input data that a user sends to a system.

[0296] A "generative model" is an algorithm or software used to generate text or other data.

[0297] "Data" refers to the text or other forms of information output by a generative model.

[0298] "Collection" refers to the process by which the server compiles the data output from the generative model.

[0299] "Evaluation" refers to the process in which a generative model judges the quality and appropriateness of the results generated by other models and assigns them a score.

[0300] "Evaluation results" refer to the scores and feedback obtained after the evaluation process.

[0301] "Aggregation" refers to the process of organizing and calculating the results of multiple evaluations.

[0302] "Recommendation" refers to the act of the server proposing the optimal generative model to the user.

[0303] "Emotion" refers to the emotional state that the user expresses in the input text.

[0304] An "emotion engine" is software or algorithms for recognizing emotions from user input data.

[0305] "Emotional compatibility" is an index that indicates how well the generated text matches the user's emotions.

[0306] "Correction" refers to the process of adjusting the evaluation results by taking into account the emotional compatibility.

[0307] This invention relates to a system for selecting and recommending an optimal model from multiple generative AI models while taking into account the emotional state of a user. This system comprises a means for accepting requests from a user, a means for sending input to multiple generative models, a means for evaluating collected data, a means for aggregating the evaluation results of the generative models, a means for recognizing the user's emotions, and a means for notifying the user of the recommendation results.

[0308] Specifically, this system is implemented using the following hardware and software: The hardware includes the user's device (e.g., PC, smartphone) and a server, while the software includes a generative model (e.g., GPT-3, BERT, Transformer), an emotion engine (e.g., IBM Watson® Tone Analyzer), and scripts for the evaluation algorithm.

[0309] The server receives requests from users and sends them to multiple generative models. Each generative model generates new text based on the input text and returns the results to the server. The server aggregates these generation results and uses an evaluation algorithm to have the generation results of each model evaluated against each other.

[0310] Furthermore, the emotion engine recognizes the user's emotion and corrects the evaluation of the generated results based on that emotion. Based on the corrected evaluation results, the server selects the optimal generation model and recommends the result to the user.

[0311] As a concrete example, consider the case where a user sends a request to the system saying, "I'm very tired today. I'd like to know the weather forecast." In this case, the device sends this request to the server, which then forwards it to multiple generative models. Each generative model generates the corresponding text and returns the result to the server. For example, generative model A might generate a result such as "It's sunny today," generative model B might generate a result such as "It's cloudy today with occasional rain," and generative model C might generate a result such as "There's a high chance of rain today."

[0312] The server collects these generated results and configures each model to evaluate the other generated results. The evaluation of each generated result is adjusted based on the user's "fatigue" emotion recognized by the emotion engine. Finally, the server aggregates these evaluation results and recommends the highest-rated generation model to the user.

[0313] In this way, the system is able to provide optimal generated results that take into account the user's emotional state.

[0314] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0315] Step 1:

[0316] The user uses the terminal to send the input text to be evaluated as a request. For example, the user might enter, "I'm very tired today. I'd like to know the weather forecast." The terminal then sends this request to the server. The input is the user's request text, and the output is the request sent to the server.

[0317] Step 2:

[0318] The server recognizes the user's emotion based on the received request. The input is the request text from the user, and the output is the recognized emotion information (e.g., fatigue). The server uses an emotion engine (e.g., IBM Watson Tone Analyzer) to analyze the input text, and the emotion engine recognizes emotions such as "fatigue."

[0319] Step 3:

[0320] The server sends the received input text to multiple generative models. For example, the text "I'm very tired today. I'd like to know the weather forecast." is sent to Model A (GPT-3), Model B (BERT), and Model C (Transformer). The input is the user's request text, and the output is the text sent to each model.

[0321] Step 4:

[0322] Each generative model generates new text based on the input text received from the server. The input is the request text to the generative model, and the output is the generated text. For example, model A generates "It's sunny today," model B generates "It's cloudy with occasional rain today," and model C generates "There's a high chance of rain today," and sends them back to the server.

[0323] Step 5:

[0324] The server sends the collected output of each generative model back to each model and sends a request to evaluate the generated results of other models. The input is the generated text from each generative model, and the output is the sending of an evaluation request. In addition, based on the emotion "fatigue" recognized by the emotion engine, the server also evaluates how well the generated results match that emotion.

[0325] Step 6:

[0326] Each generative model receives the output of other generative models and evaluates its quality and appropriateness. The input is the generated text and recognized sentiment of other generative models, and the output is an evaluation score. For example, model A evaluates the generated result of model B and assigns a score (e.g., 80 points). It also makes corrections based on the sentiment fit.

[0327] Step 7:

[0328] Once each generative model has completed the evaluation, it sends its evaluation score back to the server. The input is the evaluation score, and the output is the score sent to the server. Specifically, model A returns score information to the server, such as "Model B = 80 points, Model C = 70 points."

[0329] Step 8:

[0330] The server aggregates the evaluation scores received from all generative models and also integrates the emotional relevance score from the emotion engine. The input is the evaluation score and emotional relevance score of each generative model, and the output is an overall evaluation score. For example, the overall evaluation score is calculated by adding the evaluation score and the emotional relevance.

[0331] Step 9:

[0332] The server calculates the final overall evaluation score based on the aggregated evaluation scores and emotional relevance scores. The input is the aggregated evaluation scores and emotional relevance scores, and the output is the final overall evaluation score. For example, we calculate the overall evaluation score of Model A.

[0333] Step 10:

[0334] The server identifies the generative model with the highest score based on the overall evaluation score. The input is the overall evaluation score, and the output is the identified generative model. In this case, model B with the highest score is selected as the optimal generative model.

[0335] Step 11:

[0336] The server recommends the optimal generative model to the user. The input is the optimal generative model, and the output is a recommendation notification to the user. The server returns the recommendation result to the user's device and notifies the user.

[0337] (Application example 2)

[0338] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0339] In recent years, it has become increasingly important to provide users with information appropriate to their emotional state when consuming news or content. However, existing text generation engines simply generate text without considering the user's emotions, which prevents them from increasing user satisfaction. Furthermore, there are limited means for relatively evaluating the quality of generated text, making it difficult to select the optimal text generation engine. Therefore, a system is needed that considers the user's emotions and selects and recommends the optimal engine from among multiple text generation engines.

[0340] The specification process by the specification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for accepting a request from a user, means for sending the same input to a plurality of text generation engines and collecting texts generated therefrom, means for having each text generation engine evaluate the collected text and aggregating the evaluation results, an emotion engine for recognizing the user's emotion, means for evaluating how well the generated text matches the user's emotion, and means for recommending the most highly rated text generation engine to the user based on the emotion match. This makes it possible to select a text generation engine that is most suitable for the user's emotional state and provide the user with text generated by that engine.

[0341] "Means for accepting requests from a user" refers to an interface or device for accepting user input or requests.

[0342] "Multiple text generation engines" refers to multiple systems or models that generate text using different algorithms or techniques.

[0343] "Means for sending identical inputs and collecting the text generated from them" refers to a mechanism or process for sending user requests to multiple text generation engines and collecting the text generated by those engines.

[0344] "Means for having each text generation engine evaluate the collected text and aggregating the evaluation results" refers to a mechanism for having each text generation engine evaluate the quality and relevance of the collected text and integrating the evaluation results.

[0345] "Emotion engine that recognizes user emotions" refers to an engine or algorithm that analyzes and recognizes emotions from a user's text and behavioral history.

[0346] "Means for assessing how well generated text matches the user's emotions" refers to a process or device for assessing how well generated text matches the user's current emotional state.

[0347] "Means for recommending the most highly rated text generation engine to a user based on emotional compatibility" refers to a mechanism for selecting the most suitable text generation engine and recommending it to a user, taking into account the user's emotional compatibility.

[0348] The present invention relates to a system that provides content by selecting an optimal engine from among a plurality of text generation engines in consideration of a user's feelings. Specific embodiments for carrying out the present invention are described below.

[0349] System Overview

[0350] This system consists of a user terminal, a server, multiple text generation engines, and an emotion engine. The user sends a request, and the system generates text according to the request and recommends the generation engine that best suits the user's emotion.

[0351] Major hardware and software

[0352] 1. User Device

[0353] A device (smartphone, tablet, etc.) for accepting user input.

[0354] 2. Server

[0355] It hosts the text generation and sentiment engines, and processes and evaluates requests.

[0356] 3. Emotion Engine

[0357] An engine for information analysis and emotion recognition. It can use "AWS (registered trademark) Rekognition" and "Google Cloud Natural Language."

[0358] 4. Text Generation Engine

[0359] Generate text using multiple generation engines (e.g., GPT-3, BERT).

[0360] Processing Details

[0361] The server first receives a request from the user. This request contains information about the content the user desires. The emotion engine then recognizes the user's emotion and generates metrics based on that emotion. This emotion data is obtained by analyzing past requests, interaction history, and other factors.

[0362] The server then sends the user's input text to multiple text generation engines and collects the generated text from each engine. The collected text is then sent to each engine again to have it evaluate the output of other generation engines. Furthermore, the emotion engine evaluates the emotional relevance of the generated text and integrates the evaluation score into the evaluation score of each text generation engine.

[0363] The server finally aggregates all the evaluation scores and calculates an overall evaluation score, taking into account the emotional compatibility. The text generation engine with the highest overall evaluation score is recommended to the user.

[0364] Specific examples

[0365] For example, if a user sends a request saying, "I want to see more happy news today," the emotion engine extracts the emotion data of "fun." Generation engine A generates "happy stories about cats," generation engine B generates "latest superhero movie reviews," and generation engine C generates "new restaurant information." Each generation engine evaluates the other's generation results, and the engine with the highest score, taking into account emotional compatibility, is determined and recommended to the user.

[0366] Prompt Sentence Examples

[0367] Example of a user request: "I want to see more exciting news today."

[0368] Example of an emotion rating: "fun"

[0369] Generated text by Generation Engine A: "A happy story about a cat"

[0370] Generated text by Generation Engine B: "Latest superhero movie reviews"

[0371] Generated text by generation engine C: "New restaurant information"

[0372] Through this system, it is possible to select a text generation engine that best suits the user's emotional state and provide the user with the text generated by that engine.

[0373] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0374] Step 1:

[0375] The server receives a request from the user terminal. This request contains information about the content the user desires. As input, the user sends the text "I want to see more entertaining news today." This text reaches the server and triggers the next processing step.

[0376] Step 2:

[0377] The server sends the received request text to the emotion engine. The emotion engine recognizes the user's emotion based on the user's text and past dialogue history. For example, the emotion engine recognizes emotions such as "happy," "sad," and "tired." Specifically, for the input text "I want to see more happy news today," the emotion engine extracts "happy." Emotion recognition data is generated and used in the next step.

[0378] Step 3:

[0379] The server sends the user's input text to multiple text generation engines. A generative AI model is used for this process. For example, generation engine A, generation engine B, and generation engine C receive the same input text, "I want to see more fun news today," and each generates new text using its own algorithm. As output, generation engine A sends "Happy stories about cats," generation engine B sends "Latest superhero movie reviews," and generation engine C sends "New restaurant information" back to the server.

[0380] Step 4:

[0381] The server then sends the collected generated texts to each text generation engine, which then evaluates the other's generated results. In this process, each generation engine evaluates the quality and relevance of the text received from the other engines. Specifically, generation engine A evaluates the texts from generation engines B and C, and generation engine B performs the same evaluation. At this stage, an evaluation score is generated.

[0382] Step 5:

[0383] The server evaluates the emotional relevance of the generated text using the emotion engines. For example, it evaluates whether "a happy story about a cat" is highly relevant to the emotion of "fun." This emotional relevance score is also integrated into the evaluation score of each text generation engine. This evaluation score becomes part of the final overall evaluation score.

[0384] Step 6:

[0385] The server aggregates all the evaluation scores and calculates an overall evaluation score. This overall evaluation score includes the evaluation scores from each engine and the emotional compatibility score. Specifically, let's say that generation engine A receives an overall evaluation score of 80 points, generation engine B receives an overall evaluation score of 90 points, and generation engine C receives an overall evaluation score of 85 points.

[0386] Step 7:

[0387] The server identifies the most highly rated text generation engine based on the overall evaluation score. Here, engine B has the highest score of 90, and is therefore recognized as the most suitable text generation engine.

[0388] Step 8:

[0389] The server recommends the most suitable text generation engine B to the user. The recommendation result is returned to the user's terminal and notified to the user. Specifically, the user receives an article titled "Latest Superhero Movie Reviews" and can view it.

[0390] In this way, it is possible to select the most suitable text generation engine while taking into consideration the emotional state of the user, and to provide appropriate content to the user.

[0391] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0392] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (registered trademark) (Internet search engine).<URL: https: / / openai.com / blog / chatgpt> ), Gemini (registered trademark) (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0393] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0394] [Second embodiment]

[0395] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0396] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0397] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0398] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0399] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0400] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0401] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0402] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0403] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0404] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0405] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0406] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0407] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[0408] Specific examples of program processing

[0409] Step 1: Accepting a request from a user

[0410] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[0411] Step 2: Distributing input to the text generation engine

[0412] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[0413] Step 3: Text generation

[0414] Each text generation engine generates text based on the input it receives. The generated text is returned to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy with occasional rain today," and generation engine C generates "There's a high chance of rain today."

[0415] Step 4: Submit a text rating request

[0416] The server sends the collected output of each generation engine back to each generation engine, and sends a request to evaluate the output text of other generation engines. For example, generation engine A receives a request to evaluate the output text of generation engines B and C.

[0417] Step 5: Peer assessment of the text

[0418] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, generation engine A evaluates the generation result of B and gives it a score of 80, and evaluates the generation result of C and gives it a score of 70. Generation engine B also evaluates the generation results of A and C in the same way and returns their respective scores to the server.

[0419] Step 6: Aggregating the evaluation results

[0420] The server aggregates the reputation scores received from all the generation engines. For example, let's say the collected reputation scores are aggregated as follows:

[0421] Generation Engine A scores: 80 points for B, 70 points for C

[0422] Generation Engine B scores: 75 points for A, 85 points for C

[0423] Generation Engine C scores: 65 points for A, 90 points for B

[0424] Step 7: Determine and recommend the best generation engine

[0425] The server calculates the overall evaluation score and identifies the generation engine with the highest evaluation. For example, generation engine B has the highest overall evaluation score of 170 points. This result is recommended to the user.

[0426] Specific example explanation

[0427] 1. Input text: The user enters the text "What's the weather like today?"

[0428] 2. Text generation: Each generation engine (A, B, C) produces a different output.

[0429] Generation Engine A: "It's sunny today."

[0430] Generation Engine B: "It's cloudy with occasional rain today."

[0431] Generation Engine C: "There is a high chance of rain today."

[0432] 3. Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[0433] 4. Rating Aggregation: The server aggregates the rating scores and determines that engine B has the highest rating.

[0434] 5. Recommendation to the user: The server notifies the user of the best generation engine B.

[0435] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[0436] The processing flow will be explained below.

[0437] Step 1:

[0438] A user uses a terminal to send a request including an input text to be evaluated to a server. The request includes information indicating that the user wants to evaluate a text generation engine that the user wants to use.

[0439] Step 2:

[0440] The server receives a request from a user, obtains the input text included in the request, and sends it to multiple text generation engines (e.g., text generation engine A, text generation engine B, and text generation engine C).

[0441] Step 3:

[0442] Each text generation engine generates new text based on the input text received from the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[0443] Step 4:

[0444] Each text generation engine sends the generated text back to the server, which collects these results.

[0445] Step 5:

[0446] The server resends the collected generation results to each of text generation engines A, B, and C, and sends a request to evaluate the generation results of the other generation engines.

[0447] Step 6:

[0448] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points, while text generation engine C's generation, "There is a high chance of rain today," gives it a score of 70 points.

[0449] Step 7:

[0450] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[0451] Step 8:

[0452] The server aggregates the evaluation scores received from all text generation engines, for example, as follows:

[0453] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[0454] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[0455] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[0456] Step 9:

[0457] The server calculates the overall evaluation score based on the aggregated evaluation scores. For example, it calculates it as follows:

[0458] Overall score for Text Generation Engine A = 75 points + 65 points = 140 points

[0459] Overall score for Text Generation Engine B = 80 points + 90 points = 170 points

[0460] Overall score for text generation engine C = 70 points + 85 points = 155 points

[0461] Step 10:

[0462] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[0463] Step 11:

[0464] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[0465] Example 1

[0466] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0467] Conventional text generation systems rely on the performance of a single text generation engine, resulting in inconsistent quality of generated text and not necessarily optimal generation results. Furthermore, there is no established method for comparing and evaluating the output of multiple text generation engines, making it difficult to determine which engine best suits a user's needs. This makes it difficult for users to find the optimal generation engine, affecting their satisfaction with the generated results.

[0468] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0469] In this invention, the server includes means for accepting requests from a user, means for sending the same input to a plurality of text generation engines and collecting the texts generated therefrom, means for sending the collected texts to each text generation engine to have them evaluate each other and collecting the evaluation results, means for aggregating the evaluation results, and means for recommending the most highly rated text generation engine to the user. This enables the user to easily find the text generation engine that is most suitable for the text generation result they desire.

[0470] "Means for accepting requests from users" refers to the function of sending text or requests entered by a user using a terminal to a server and having the server receive them.

[0471] "Means for sending the same input to multiple text generation engines and collecting the text generated from them" refers to a function in which a server sends a specific input text to multiple different text generation engines and collects the generated results.

[0472] "Means for sending the collected text to each text generation engine to have them evaluate each other, and collecting the evaluation results" refers to the function of the server sending the collected text to each text generation engine again and having them evaluate the generation results of the other generation engines.

[0473] "Means for aggregating evaluation results" refers to the function of aggregating the evaluation results returned from each text generation engine and calculating the total score and average score.

[0474] The "means for recommending the most highly rated text generation engine to the user" refers to a function for identifying the text generation engine that has received the highest score based on the evaluation results and notifying the user of that information.

[0475] "Calculating an overall evaluation score" refers to calculating an overall evaluation point based on the evaluation scores from each text generation engine.

[0476] "Mutual evaluation" refers to an evaluation method in which each text generation engine evaluates the results of the other engines' generation and returns the evaluation results in numerical form.

[0477] "Text generation engine" refers to an algorithm or software that generates natural language text based on specific input.

[0478] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[0479] Specific implementation methods

[0480] 1. Hardware and software configuration

[0481] This system consists of a server running on the cloud, user devices (PCs, smartphones, tablets, etc.), and multiple text generation engines, including OpenAI GPT-3 and Google BERT.

[0482] 2. Accepting a request from a user

[0483] A user uses a device to input a specific prompt sentence (e.g., "What's the weather like today?") via the system's web interface or mobile app. The device sends this request to the server, which accepts the request.

[0484] 3. Input distribution to multiple text generation engines

[0485] The server sends the received input text to multiple pre-configured text generation engines, each of which analyzes the received input text and returns the generated text to the server.

[0486] 4. Text Generation

[0487] Each text generation engine (for example, engine A, engine B, and engine C) sends its generated text to the server based on the input "What's the weather like today?". As a specific example, engine A's generated text is "It's sunny today," engine B's is "It's cloudy with occasional rain today," and engine C's is "There's a high chance of rain today."

[0488] 5. Submitting a Text Rating Request

[0489] The server sends the collected output of each generation engine to each generation engine again, and has them evaluate the output text of other generation engines. For example, it sends a request to engine A to evaluate the output text of engines B and C.

[0490] 6. Peer evaluation of texts

[0491] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, engine A gives a score of 80 to the generation result of engine B and a score of 70 to engine C. Engine B similarly evaluates the output of engines A and C and returns their respective scores to the server.

[0492] 7. Calculation of evaluation results

[0493] The server aggregates the ratings received from all the generation engines and identifies the generation engine with the highest overall rating score. For example, if the following rating results are obtained:

[0494] Generation Engine A scores: 80 points for B, 70 points for C

[0495] Generation Engine B scores: 75 points for A, 85 points for C

[0496] Generation Engine C scores: 65 points for A, 90 points for B

[0497] 8. Determining and recommending the best generation engine

[0498] The server sums up the evaluation scores and recommends the highest-rated generation engine (e.g., engine B in this case) to the user.

[0499] Specific example explanation

[0500] Input text: The user enters the text "What's the weather like today?"

[0501] Text generation: Each generation engine (Engine A, Engine B, Engine C) produces a different output.

[0502] Generation Engine A: "It's sunny today."

[0503] Generation Engine B: "It's cloudy with occasional rain today."

[0504] Generation Engine C: "There is a high chance of rain today."

[0505] Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[0506] Rating aggregation: The server aggregates the rating scores and determines that engine B received the highest rating.

[0507] Recommendation to user: The server notifies the user of the best generation engine B.

[0508] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[0509] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0510] Step 1:

[0511] A user uses a terminal to input the input text "What's the weather like today?" into the system. This input text is sent from the user's terminal to the server. In this case, the prompt sentence ("What's the weather like today?") becomes the input data.

[0512] Step 2:

[0513] The server distributes the user's input text to multiple text generation engines (generative AI models). The server sends the same input text to each text generation engine (e.g., Engine A, Engine B, and Engine C). At this stage, the user's prompt sentence is sent as input to each generation engine.

[0514] Step 3:

[0515] Each text generation engine receives input text and generates text using a generation algorithm. For example, Engine A generates "It's sunny today," Engine B generates "It's cloudy with occasional rain today," and Engine C generates "There's a high chance of rain today." Each generation result is sent back to the server. Here, the generative AI model performs data calculations based on the input text and outputs the text generation result.

[0516] Step 4:

[0517] The server again sends the collected output text of the generation engines to each generation engine, and the generation results of the other generation engines are evaluated against each other. Specifically, the server sends the generation results of engines B and C to engine A, the generation results of engines A and C to engine B, and the generation results of engines A and B to engine C. At this stage, the different generation results are sent to each engine as input data for evaluation.

[0518] Step 5:

[0519] Each generation engine evaluates the output text of other generation engines and returns the evaluation score to the server. For example, Engine A gives Engine B's generation result 80 points and Engine C 70 points. Similarly, Engine B gives Engine A 75 points and Engine C 85 points. Engine C also evaluates the other generation results and returns the score to the server. Here, the score is output based on the evaluation data for the input generation result.

[0520] Step 6:

[0521] The server aggregates the ratings received from all the generation engines, and calculates the overall rating score for each engine, resulting in something like this:

[0522] Generation Engine A scores: 80 points for B, 70 points for C

[0523] Generation Engine B scores: 75 points for A, 85 points for C

[0524] Generation Engine C scores: 65 points for A, 90 points for B

[0525] At this stage, the aggregated evaluation scores are output.

[0526] Step 7:

[0527] The server will aggregate the evaluation scores and identify the generation engine that received the highest score. For example, let's say generation engine B received the highest overall evaluation score of 170 points. The server will notify the user of this result. Here, the generation engine with the highest evaluation score will be output.

[0528] This series of processes allows the user to find the optimal text generation engine.

[0529] (Application example 1)

[0530] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0531] With conventional text generation technology, it is difficult to select the optimal engine from the many natural language generation engines available, making it impossible to obtain text of the quality desired by users. Furthermore, especially on online shopping sites, the quality of text generated when generating product descriptions and customer reviews significantly affects the user experience, so there is a demand for higher quality text generation.

[0532] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0533] In this invention, the server includes means for accepting requests from users, means for sending the same input to multiple natural language generation engines and collecting the text generated therefrom, means for having each natural language generation engine evaluate the collected text and aggregating the evaluation results, means for recommending the most highly rated natural language generation engine to the user, and means for generating product descriptions using an optimal natural language generation engine based on the collected text, thereby enabling users to obtain high-quality product descriptions and reviews using an optimal quality text generation engine.

[0534] "Means for accepting requests from users" refers to a device or interface that allows the server to accept information or requests entered by users.

[0535] "Multiple natural language generation engines" are multiple computer programs that generate sentences based on different algorithms or models.

[0536] "Means for sending identical input and collecting the text generated from them" refers to a device or software that allows a server to send the same input data to multiple natural language generation engines and collect the output results.

[0537] The "means for causing each natural language generation engine to evaluate the collected text" refers to a device or software that instructs each generation engine to evaluate the text generated by the other generation engines.

[0538] The "means for aggregating the evaluation results" refers to a device or software that compiles and analyzes the evaluation results from each generation engine and calculates an overall evaluation.

[0539] The "means for recommending the most highly rated natural language generation engine to the user" is a device or software that notifies the user of the best performing generation engine based on the evaluation results.

[0540] A "means for generating a product description" is a device or software that uses a specific generation engine to generate text describing the details and characteristics of a product based on collected text.

[0541] The present invention relates to a system that utilizes multiple natural language generation engines based on user input and recommends the most suitable engine from among them.

[0542] This system consists of a terminal that accepts requests entered by users, a server that sends the same input to multiple generation engines and collects and evaluates the text generated by each generation engine, and a system that recommends and provides the optimal generation engine and generated text to the user.

[0543] The main components of the system are:

[0544] 1. How to accept requests from users:

[0545] Sends information entered by the user at the terminal (e.g., product description prompts) to the server.

[0546] 2. A means of sending identical input to multiple natural language generation engines and collecting the text generated by them:

[0547] The server receives input from the user and sends it to multiple generation engines such as generation engine A, generation engine B, and generation engine C.

[0548] 3. A method for evaluating the collected text using each natural language generation engine and aggregating the evaluation results:

[0549] The server collects the generated text returned by each generation engine and has the other generation engines evaluate the text.

[0550] 4. How to recommend the highest-rated natural language generation engines to users:

[0551] The evaluation results are compiled, and the server recommends the generation engine with the highest score to the user.

[0552] 5. Using the collected text to generate product descriptions using the best natural language generation engine:

[0553] The server uses an optimal generation engine to generate a high-quality product description and presents it to the user.

[0554] Hardware and Software

[0555] Hardware: Smartphones, tablets, PCs, servers

[0556] Software: Natural language generation API, web frameworks such as Python and Flask, HTTP request libraries (such as requests)

[0557] Data processing and calculation

[0558] The server's actions include:

[0559] Data collection: Collect the generated text from each generation engine.

[0560] Data evaluation: Have each generation engine evaluate the text generated by other generation engines.

[0561] Score calculation: Calculate the overall evaluation score based on the evaluation results.

[0562] Engine recommendation: Recommend the highest scoring generation engine to the user.

[0563] Specific examples

[0564] As an example of a specific prompt sentence, the user may enter, "Please tell me the features of my new smartphone." In this case, the server will perform the following process.

[0565] Generator A: "This smartphone features fast charging."

[0566] Generator B: "It has the latest OS and is waterproof."

[0567] Generator C: "The large battery capacity and high-quality camera are attractive features."

[0568] The server collects the output of each generation engine and evaluates them to select the optimal generation engine and text, which it then provides to the user. In this way, the system generates high-quality text and improves user satisfaction.

[0569] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0570] Step 1:

[0571] A user uses a terminal to enter and send a request.

[0572] Input: A prompt asking the user for a product description (e.g., "What are the features of my new smartphone?")

[0573] Action: The terminal sends this input to the server. Output: The server accepts the prompt.

[0574] Step 2:

[0575] The server sends the received prompts to multiple natural language generation engines (generation engine A, generation engine B, generation engine C, etc.).

[0576] Input: The prompt the server received from the user

[0577] Operation: The server sends the same prompt to generation engine A, generation engine B, and generation engine C. Output: Text generation begins in each generation engine.

[0578] Step 3:

[0579] Each generation engine generates text based on the prompts it receives and sends that text back to the server.

[0580] Input: The prompt received by each generation engine

[0581] Operation: Generation Engine A, Generation Engine B, and Generation Engine C each generate text using their own algorithm. Output: Generated text (Example: Generation Engine A: "This smartphone features fast charging.")

[0582] Step 4:

[0583] The server again sends the collected text to each generation engine and sends a request to evaluate the output text of the other generation engines.

[0584] Input: Text generated by each generation engine

[0585] Operation: The server asks generation engine A to evaluate the texts generated by generation engines B and C. Similarly, it requests evaluations for all other combinations. Output: Each generation engine returns an evaluation score for the texts generated by the other generation engines.

[0586] Step 5:

[0587] Each generation engine evaluates the output text of the other generation engines and returns its evaluation score to the server.

[0588] Input: Text to be evaluated (generated by other generation engines)

[0589] Operation: The evaluation engine evaluates the received text and calculates a score. Output: Evaluation score (e.g., generation engine A gives generation engine B's text a score of 80).

[0590] Step 6:

[0591] The server aggregates the rating scores received from all the generation engines.

[0592] Input: Evaluation scores sent by all generation engines

[0593] Operation: The server aggregates the evaluation scores for each generation engine and calculates the overall evaluation score. Output: Overall evaluation score (e.g., generation engine B received the highest overall score).

[0594] Step 7:

[0595] The server calculates an overall rating score and identifies the generation engine with the highest rating.

[0596] Input: Overall evaluation score

[0597] Behavior: The server selects the best engine and its generated text. Output: The best engine (e.g. engine B).

[0598] Step 8:

[0599] The server generates product descriptions and reviews using an optimal generation engine and provides the final text to the user.

[0600] Input: The optimized generation engine and the generated text

[0601] Behavior: The server uses the best generation engine to finalize the text containing the product description and send it to the user. Output: High-quality product description text provided to the user (e.g., "This smartphone has fast charging capabilities and the latest OS.")

[0602] This processing step enables the system to recommend the most appropriate natural language generation engine for the user's request and generate and provide the most appropriate text.

[0603] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0604] The present invention relates to a system that allows multiple text generation engines to evaluate each other's outputs, recognizes the user's emotions, and takes this information into account to recommend the most suitable generation engine. This system is composed of the following elements: a means for accepting user requests, a means for sending input to multiple text generation engines, a means for evaluating the collected text, a means for aggregating the evaluation results of the generation engines, an emotion engine that recognizes the user's emotions, and a means for recommending the most suitable generation engine to the user.

[0605] Specific examples of program processing

[0606] Step 1: Accepting a request from a user

[0607] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[0608] Step 2: Recognizing user emotions with the emotion engine

[0609] The server uses the input text from the user and the past dialogue history to have the emotion engine recognize the user's emotion. For example, if the input text includes "I'm very tired today," the emotion engine will recognize the user's emotional state as "fatigue."

[0610] Step 3: Distributing input to the text generation engine

[0611] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[0612] Step 4: Text Generation

[0613] Each text generation engine generates new text based on the input text received from the server. The generated text is returned to the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[0614] Step 5: Submit a text rating request

[0615] The server sends the collected output of each generation engine to each generation engine again, and sends a request to each generation engine to evaluate the generation results of other generation engines. It also evaluates how well the generated text matches the user's emotion, based on the user's emotion recognized by the emotion engine.

[0616] Step 6: Peer assessment of the text

[0617] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "It's cloudy with occasional rain today," and assigns a score taking into account the evaluation of the emotion engine. It also evaluates the result of text generation engine C's generation, "There is a high chance of rain today."

[0618] Step 7: Return your evaluation score

[0619] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[0620] Step 8: Aggregating the evaluation results

[0621] The server aggregates the evaluation scores received from all text generation engines and also integrates the emotional relevance scores from the emotion engine. For example, the evaluation scores are collected as follows:

[0622] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[0623] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[0624] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[0625] Step 9: Calculating the overall score

[0626] The server calculates the overall evaluation score based on the aggregated evaluation scores and emotional compatibility scores. For example, it calculates it as follows:

[0627] Overall evaluation score for text generation engine A: 75 points (evaluation score) + 65 points (emotional compatibility) = 140 points

[0628] Overall evaluation score for Text Generation Engine B: 80 points (evaluation score) + 90 points (emotional relevance) = 170 points

[0629] Overall evaluation score for text generation engine C: 70 points (evaluation score) + 85 points (emotional relevance) = 155 points

[0630] Step 10: Determine the best generation engine

[0631] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[0632] Step 11: User Recommendations

[0633] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[0634] In this way, the system of the present invention makes it possible to easily find the most suitable text generation engine while taking into account the emotional state of the user.

[0635] The processing flow will be explained below.

[0636] Step 1:

[0637] The user uses a terminal to send a request to the server, including the input text to be evaluated. The request also includes the content and purpose of the text generation desired by the user.

[0638] Step 2:

[0639] The server receives a request from a user, obtains the input text included in the request, and simultaneously sends a request to the emotion engine to recognize the user's emotion from the input text.

[0640] Step 3:

[0641] The emotion engine analyzes the user's input text and recognizes the user's current emotional state. For example, from the text "I'm very tired today," the emotion engine recognizes the emotional state as "fatigue." The emotion engine then sends the recognition result back to the server.

[0642] Step 4:

[0643] The server sends the received input text to multiple text generation engines (for example, text generation engine A, text generation engine B, and text generation engine C). Each of the text generation engines generates new text based on the input text.

[0644] Step 5:

[0645] Each text generation engine sends the newly generated text back to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy today with occasional rain," and generation engine C generates "There's a high chance of rain today."

[0646] Step 6:

[0647] The server again sends the collected generated text to each text generation engine and sends a request to evaluate the generated results of the other generation engines. It also sends the generated text to the emotion engine and sends a request to evaluate how well it matches the recognized user emotion.

[0648] Step 7:

[0649] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points. It also gives the result of text generation engine C, "There is a high chance of rain today," a score of 70 points.

[0650] Step 8:

[0651] The emotion engine evaluates the suitability of each generated text to the user's emotion and sends the score back to the server. For example, a high score is given to the generated result that best suits the emotional state of "fatigue."

[0652] Step 9:

[0653] The server aggregates the evaluation scores received from each generation engine and the emotional relevance scores from the emotion engine. For example, it calculates a total evaluation score of the evaluation scores and emotional relevance scores of each text generation engine.

[0654] Step 10:

[0655] The server identifies the best-rated text generation engine based on the aggregated overall rating score. For example, if text generation engine B has the highest rating with an overall rating score of 170, this engine is considered the best.

[0656] Step 11:

[0657] The server recommends the optimal text generation engine B to the user. The recommendation result is notified to the user's terminal, and the output of the selected generation engine is provided to the user.

[0658] Example 2

[0659] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0660] Conventional text generation systems often provide generated text and data without considering the user's emotional state, resulting in low user satisfaction. Furthermore, using only a single generative model can lead to variations in the quality of generated results. Furthermore, when using multiple generative models, there is no way to evaluate and compare the generated results, making it difficult to select an efficient generative model.

[0661] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0662] In this invention, the server includes means for accepting requests from a user, means for sending the same input to multiple generative models and collecting data generated from them, means for having each generative model evaluate the collected data and aggregating the evaluation results, means for recommending the most highly rated generative model to the user, means for recognizing the user's emotions, and means for correcting the evaluation results based on the emotions. This makes it possible to select the most appropriate generative model and recommend it to the user, taking into account the user's emotional state.

[0663] "User" means a person or institution that receives data generated using the system.

[0664] A "request" is an instruction or input data that a user sends to a system.

[0665] A "generative model" is an algorithm or software used to generate text or other data.

[0666] "Data" refers to the text or other forms of information output by a generative model.

[0667] "Collection" refers to the process by which the server compiles the data output from the generative model.

[0668] "Evaluation" refers to the process in which a generative model judges the quality and appropriateness of the results generated by other models and assigns them a score.

[0669] "Evaluation results" refer to the scores and feedback obtained after the evaluation process.

[0670] "Aggregation" refers to the process of organizing and calculating the results of multiple evaluations.

[0671] "Recommendation" refers to the act of the server proposing the optimal generative model to the user.

[0672] "Emotion" refers to the emotional state that the user expresses in the input text.

[0673] An "emotion engine" is software or algorithms for recognizing emotions from user input data.

[0674] "Emotional compatibility" is an index that indicates how well the generated text matches the user's emotions.

[0675] "Correction" refers to the process of adjusting the evaluation results by taking into account the emotional compatibility.

[0676] This invention relates to a system for selecting and recommending an optimal model from multiple generative AI models while taking into account the emotional state of a user. This system comprises a means for accepting requests from a user, a means for sending input to multiple generative models, a means for evaluating collected data, a means for aggregating the evaluation results of the generative models, a means for recognizing the user's emotions, and a means for notifying the user of the recommendation results.

[0677] Specifically, this system is implemented using the following hardware and software: The hardware includes the user's device (e.g., PC, smartphone) and a server, while the software includes a generative model (e.g., GPT-3, BERT, Transformer), an emotion engine (e.g., IBM Watson Tone Analyzer), and scripts for the evaluation algorithm.

[0678] The server receives requests from users and sends them to multiple generative models. Each generative model generates new text based on the input text and returns the results to the server. The server aggregates these generation results and uses an evaluation algorithm to have the generation results of each model evaluated against each other.

[0679] Furthermore, the emotion engine recognizes the user's emotion and corrects the evaluation of the generated results based on that emotion. Based on the corrected evaluation results, the server selects the optimal generation model and recommends the result to the user.

[0680] As a concrete example, consider the case where a user sends a request to the system saying, "I'm very tired today. I'd like to know the weather forecast." In this case, the device sends this request to the server, which then forwards it to multiple generative models. Each generative model generates the corresponding text and returns the result to the server. For example, generative model A might generate a result such as "It's sunny today," generative model B might generate a result such as "It's cloudy today with occasional rain," and generative model C might generate a result such as "There's a high chance of rain today."

[0681] The server collects these generated results and configures each model to evaluate the other generated results. The evaluation of each generated result is adjusted based on the user's "fatigue" emotion recognized by the emotion engine. Finally, the server aggregates these evaluation results and recommends the highest-rated generation model to the user.

[0682] In this way, the system is able to provide optimal generated results that take into account the user's emotional state.

[0683] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0684] Step 1:

[0685] The user uses the terminal to send the input text to be evaluated as a request. For example, the user might enter, "I'm very tired today. I'd like to know the weather forecast." The terminal then sends this request to the server. The input is the user's request text, and the output is the request sent to the server.

[0686] Step 2:

[0687] The server recognizes the user's emotion based on the received request. The input is the request text from the user, and the output is the recognized emotion information (e.g., fatigue). The server uses an emotion engine (e.g., IBM Watson Tone Analyzer) to analyze the input text, and the emotion engine recognizes emotions such as "fatigue."

[0688] Step 3:

[0689] The server sends the received input text to multiple generative models. For example, the text "I'm very tired today. I'd like to know the weather forecast." is sent to Model A (GPT-3), Model B (BERT), and Model C (Transformer). The input is the user's request text, and the output is the text sent to each model.

[0690] Step 4:

[0691] Each generative model generates new text based on the input text received from the server. The input is the request text to the generative model, and the output is the generated text. For example, model A generates "It's sunny today," model B generates "It's cloudy with occasional rain today," and model C generates "There's a high chance of rain today," and sends them back to the server.

[0692] Step 5:

[0693] The server sends the collected output of each generative model back to each model and sends a request to evaluate the generated results of other models. The input is the generated text from each generative model, and the output is the sending of an evaluation request. In addition, based on the emotion "fatigue" recognized by the emotion engine, the server also evaluates how well the generated results match that emotion.

[0694] Step 6:

[0695] Each generative model receives the output of other generative models and evaluates its quality and appropriateness. The input is the generated text and recognized sentiment of other generative models, and the output is an evaluation score. For example, model A evaluates the generated result of model B and assigns a score (e.g., 80 points). It also makes corrections based on the sentiment fit.

[0696] Step 7:

[0697] Once each generative model has completed the evaluation, it sends its evaluation score back to the server. The input is the evaluation score, and the output is the score sent to the server. Specifically, model A returns score information to the server, such as "Model B = 80 points, Model C = 70 points."

[0698] Step 8:

[0699] The server aggregates the evaluation scores received from all generative models and also integrates the emotional relevance score from the emotion engine. The input is the evaluation score and emotional relevance score of each generative model, and the output is an overall evaluation score. For example, the overall evaluation score is calculated by adding the evaluation score and the emotional relevance.

[0700] Step 9:

[0701] The server calculates the final overall evaluation score based on the aggregated evaluation scores and emotional relevance scores. The input is the aggregated evaluation scores and emotional relevance scores, and the output is the final overall evaluation score. For example, we calculate the overall evaluation score of Model A.

[0702] Step 10:

[0703] The server identifies the generative model with the highest score based on the overall evaluation score. The input is the overall evaluation score, and the output is the identified generative model. In this case, model B with the highest score is selected as the optimal generative model.

[0704] Step 11:

[0705] The server recommends the optimal generative model to the user. The input is the optimal generative model, and the output is a recommendation notification to the user. The server returns the recommendation result to the user's device and notifies the user.

[0706] (Application example 2)

[0707] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0708] In recent years, it has become increasingly important to provide users with information appropriate to their emotional state when consuming news or content. However, existing text generation engines simply generate text without considering the user's emotions, which prevents them from increasing user satisfaction. Furthermore, there are limited means for relatively evaluating the quality of generated text, making it difficult to select the optimal text generation engine. Therefore, a system is needed that considers the user's emotions and selects and recommends the optimal engine from among multiple text generation engines.

[0709] The specification process by the specification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for accepting a request from a user, means for sending the same input to a plurality of text generation engines and collecting texts generated therefrom, means for having each text generation engine evaluate the collected text and aggregating the evaluation results, an emotion engine for recognizing the user's emotion, means for evaluating how well the generated text matches the user's emotion, and means for recommending the most highly rated text generation engine to the user based on the emotion match. This makes it possible to select a text generation engine that is most suitable for the user's emotional state and provide the user with text generated by that engine.

[0710] "Means for accepting requests from a user" refers to an interface or device for accepting user input or requests.

[0711] "Multiple text generation engines" refers to multiple systems or models that generate text using different algorithms or techniques.

[0712] "Means for sending identical inputs and collecting the text generated from them" refers to a mechanism or process for sending user requests to multiple text generation engines and collecting the text generated by those engines.

[0713] "Means for having each text generation engine evaluate the collected text and aggregating the evaluation results" refers to a mechanism for having each text generation engine evaluate the quality and relevance of the collected text and integrating the evaluation results.

[0714] "Emotion engine that recognizes user emotions" refers to an engine or algorithm that analyzes and recognizes emotions from a user's text and behavioral history.

[0715] "Means for assessing how well generated text matches the user's emotions" refers to a process or device for assessing how well generated text matches the user's current emotional state.

[0716] "Means for recommending the most highly rated text generation engine to a user based on emotional compatibility" refers to a mechanism for selecting the most suitable text generation engine and recommending it to a user, taking into account the user's emotional compatibility.

[0717] The present invention relates to a system that provides content by selecting an optimal engine from among a plurality of text generation engines in consideration of a user's feelings. Specific embodiments for carrying out the present invention are described below.

[0718] System Overview

[0719] This system consists of a user terminal, a server, multiple text generation engines, and an emotion engine. The user sends a request, and the system generates text according to the request and recommends the generation engine that best suits the user's emotion.

[0720] Major hardware and software

[0721] 1. User Device

[0722] A device (smartphone, tablet, etc.) for accepting user input.

[0723] 2. Server

[0724] It hosts the text generation and sentiment engines, and processes and evaluates requests.

[0725] 3. Emotion Engine

[0726] An engine for information analysis and emotion recognition. It can use "AWS Rekognition" and "Google Cloud Natural Language."

[0727] 4. Text Generation Engine

[0728] Generate text using multiple generation engines (e.g., GPT-3, BERT).

[0729] Processing Details

[0730] The server first receives a request from the user. This request contains information about the content the user desires. The emotion engine then recognizes the user's emotion and generates metrics based on that emotion. This emotion data is obtained by analyzing past requests, interaction history, and other factors.

[0731] The server then sends the user's input text to multiple text generation engines and collects the generated text from each engine. The collected text is then sent to each engine again to have it evaluate the output of other generation engines. Furthermore, the emotion engine evaluates the emotional relevance of the generated text and integrates the evaluation score into the evaluation score of each text generation engine.

[0732] The server finally aggregates all the evaluation scores and calculates an overall evaluation score, taking into account the emotional compatibility. The text generation engine with the highest overall evaluation score is recommended to the user.

[0733] Specific examples

[0734] For example, if a user sends a request saying, "I want to see more happy news today," the emotion engine extracts the emotion data of "fun." Generation engine A generates "happy stories about cats," generation engine B generates "latest superhero movie reviews," and generation engine C generates "new restaurant information." Each generation engine evaluates the other's generation results, and the engine with the highest score, taking into account emotional compatibility, is determined and recommended to the user.

[0735] Prompt Sentence Examples

[0736] Example of a user request: "I want to see more exciting news today."

[0737] Example of an emotion rating: "fun"

[0738] Generated text by Generation Engine A: "A happy story about a cat"

[0739] Generated text by Generation Engine B: "Latest superhero movie reviews"

[0740] Generated text by generation engine C: "New restaurant information"

[0741] Through this system, it is possible to select a text generation engine that best suits the user's emotional state and provide the user with the text generated by that engine.

[0742] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0743] Step 1:

[0744] The server receives a request from the user terminal. This request contains information about the content the user desires. As input, the user sends the text "I want to see more entertaining news today." This text reaches the server and triggers the next processing step.

[0745] Step 2:

[0746] The server sends the received request text to the emotion engine. The emotion engine recognizes the user's emotion based on the user's text and past dialogue history. For example, the emotion engine recognizes emotions such as "happy," "sad," and "tired." Specifically, for the input text "I want to see more happy news today," the emotion engine extracts "happy." Emotion recognition data is generated and used in the next step.

[0747] Step 3:

[0748] The server sends the user's input text to multiple text generation engines. A generative AI model is used for this process. For example, generation engine A, generation engine B, and generation engine C receive the same input text, "I want to see more fun news today," and each generates new text using its own algorithm. As output, generation engine A sends "Happy stories about cats," generation engine B sends "Latest superhero movie reviews," and generation engine C sends "New restaurant information" back to the server.

[0749] Step 4:

[0750] The server then sends the collected generated texts to each text generation engine, which then evaluates the other's generated results. In this process, each generation engine evaluates the quality and relevance of the text received from the other engines. Specifically, generation engine A evaluates the texts from generation engines B and C, and generation engine B performs the same evaluation. At this stage, an evaluation score is generated.

[0751] Step 5:

[0752] The server evaluates the emotional relevance of the generated text using the emotion engines. For example, it evaluates whether "a happy story about a cat" is highly relevant to the emotion of "fun." This emotional relevance score is also integrated into the evaluation score of each text generation engine. This evaluation score becomes part of the final overall evaluation score.

[0753] Step 6:

[0754] The server aggregates all the evaluation scores and calculates an overall evaluation score. This overall evaluation score includes the evaluation scores from each engine and the emotional compatibility score. Specifically, let's say that generation engine A receives an overall evaluation score of 80 points, generation engine B receives an overall evaluation score of 90 points, and generation engine C receives an overall evaluation score of 85 points.

[0755] Step 7:

[0756] The server identifies the most highly rated text generation engine based on the overall evaluation score. Here, engine B has the highest score of 90, and is therefore recognized as the most suitable text generation engine.

[0757] Step 8:

[0758] The server recommends the most suitable text generation engine B to the user. The recommendation result is returned to the user's terminal and notified to the user. Specifically, the user receives an article titled "Latest Superhero Movie Reviews" and can view it.

[0759] In this way, it is possible to select the most suitable text generation engine while taking into consideration the emotional state of the user, and to provide appropriate content to the user.

[0760] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0761] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0762] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[0763] [Third embodiment]

[0764] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0765] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[0766] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0767] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[0768] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0769] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0770] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0771] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0772] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0773] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0774] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0775] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[0776] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[0777] Specific examples of program processing

[0778] Step 1: Accepting a request from a user

[0779] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[0780] Step 2: Distributing input to the text generation engine

[0781] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[0782] Step 3: Text generation

[0783] Each text generation engine generates text based on the input it receives. The generated text is returned to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy with occasional rain today," and generation engine C generates "There's a high chance of rain today."

[0784] Step 4: Submit a text rating request

[0785] The server sends the collected output of each generation engine back to each generation engine, and sends a request to evaluate the output text of other generation engines. For example, generation engine A receives a request to evaluate the output text of generation engines B and C.

[0786] Step 5: Peer assessment of the text

[0787] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, generation engine A evaluates the generation result of B and gives it a score of 80, and evaluates the generation result of C and gives it a score of 70. Generation engine B also evaluates the generation results of A and C in the same way and returns their respective scores to the server.

[0788] Step 6: Aggregating the evaluation results

[0789] The server aggregates the reputation scores received from all the generation engines. For example, let's say the collected reputation scores are aggregated as follows:

[0790] Generation Engine A scores: 80 points for B, 70 points for C

[0791] Generation Engine B scores: 75 points for A, 85 points for C

[0792] Generation Engine C scores: 65 points for A, 90 points for B

[0793] Step 7: Determine and recommend the best generation engine

[0794] The server calculates the overall evaluation score and identifies the generation engine with the highest evaluation. For example, generation engine B has the highest overall evaluation score of 170 points. This result is recommended to the user.

[0795] Specific example explanation

[0796] 1. Input text: The user enters the text "What's the weather like today?"

[0797] 2. Text generation: Each generation engine (A, B, C) produces a different output.

[0798] Generation Engine A: "It's sunny today."

[0799] Generation Engine B: "It's cloudy with occasional rain today."

[0800] Generation Engine C: "There is a high chance of rain today."

[0801] 3. Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[0802] 4. Rating Aggregation: The server aggregates the rating scores and determines that engine B has the highest rating.

[0803] 5. Recommendation to the user: The server notifies the user of the best generation engine B.

[0804] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[0805] The processing flow will be explained below.

[0806] Step 1:

[0807] A user uses a terminal to send a request including an input text to be evaluated to a server. The request includes information indicating that the user wants to evaluate a text generation engine that the user wants to use.

[0808] Step 2:

[0809] The server receives a request from a user, obtains the input text included in the request, and sends it to multiple text generation engines (e.g., text generation engine A, text generation engine B, and text generation engine C).

[0810] Step 3:

[0811] Each text generation engine generates new text based on the input text received from the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[0812] Step 4:

[0813] Each text generation engine sends the generated text back to the server, which collects these results.

[0814] Step 5:

[0815] The server resends the collected generation results to each of text generation engines A, B, and C, and sends a request to evaluate the generation results of the other generation engines.

[0816] Step 6:

[0817] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points, while text generation engine C's generation, "There is a high chance of rain today," gives it a score of 70 points.

[0818] Step 7:

[0819] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[0820] Step 8:

[0821] The server aggregates the evaluation scores received from all text generation engines, for example, as follows:

[0822] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[0823] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[0824] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[0825] Step 9:

[0826] The server calculates the overall evaluation score based on the aggregated evaluation scores. For example, it calculates it as follows:

[0827] Overall score for Text Generation Engine A = 75 points + 65 points = 140 points

[0828] Overall score for Text Generation Engine B = 80 points + 90 points = 170 points

[0829] Overall score for text generation engine C = 70 points + 85 points = 155 points

[0830] Step 10:

[0831] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[0832] Step 11:

[0833] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[0834] Example 1

[0835] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0836] Conventional text generation systems rely on the performance of a single text generation engine, resulting in inconsistent quality of generated text and not necessarily optimal generation results. Furthermore, there is no established method for comparing and evaluating the output of multiple text generation engines, making it difficult to determine which engine best suits a user's needs. This makes it difficult for users to find the optimal generation engine, affecting their satisfaction with the generated results.

[0837] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0838] In this invention, the server includes means for accepting requests from a user, means for sending the same input to a plurality of text generation engines and collecting the texts generated therefrom, means for sending the collected texts to each text generation engine to have them evaluate each other and collecting the evaluation results, means for aggregating the evaluation results, and means for recommending the most highly rated text generation engine to the user. This enables the user to easily find the text generation engine that is most suitable for the text generation result they desire.

[0839] "Means for accepting requests from users" refers to the function of sending text or requests entered by a user using a terminal to a server and having the server receive them.

[0840] "Means for sending the same input to multiple text generation engines and collecting the text generated from them" refers to a function in which a server sends a specific input text to multiple different text generation engines and collects the generated results.

[0841] "Means for sending the collected text to each text generation engine to have them evaluate each other, and collecting the evaluation results" refers to the function of the server sending the collected text to each text generation engine again and having them evaluate the generation results of the other generation engines.

[0842] "Means for aggregating evaluation results" refers to the function of aggregating the evaluation results returned from each text generation engine and calculating the total score and average score.

[0843] The "means for recommending the most highly rated text generation engine to the user" refers to a function for identifying the text generation engine that has received the highest score based on the evaluation results and notifying the user of that information.

[0844] "Calculating an overall evaluation score" refers to calculating an overall evaluation point based on the evaluation scores from each text generation engine.

[0845] "Mutual evaluation" refers to an evaluation method in which each text generation engine evaluates the results of the other engines' generation and returns the evaluation results in numerical form.

[0846] "Text generation engine" refers to an algorithm or software that generates natural language text based on specific input.

[0847] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[0848] Specific implementation methods

[0849] 1. Hardware and software configuration

[0850] This system consists of a server running on the cloud, user devices (PCs, smartphones, tablets, etc.), and multiple text generation engines, including OpenAI GPT-3 and Google BERT.

[0851] 2. Accepting a request from a user

[0852] A user uses a device to input a specific prompt sentence (e.g., "What's the weather like today?") via the system's web interface or mobile app. The device sends this request to the server, which accepts the request.

[0853] 3. Input distribution to multiple text generation engines

[0854] The server sends the received input text to multiple pre-configured text generation engines, each of which analyzes the received input text and returns the generated text to the server.

[0855] 4. Text Generation

[0856] Each text generation engine (for example, engine A, engine B, and engine C) sends its generated text to the server based on the input "What's the weather like today?". As a specific example, engine A's generated text is "It's sunny today," engine B's is "It's cloudy with occasional rain today," and engine C's is "There's a high chance of rain today."

[0857] 5. Submitting a Text Rating Request

[0858] The server sends the collected output of each generation engine to each generation engine again, and has them evaluate the output text of other generation engines. For example, it sends a request to engine A to evaluate the output text of engines B and C.

[0859] 6. Peer evaluation of texts

[0860] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, engine A gives a score of 80 to the generation result of engine B and a score of 70 to engine C. Engine B similarly evaluates the output of engines A and C and returns their respective scores to the server.

[0861] 7. Calculation of evaluation results

[0862] The server aggregates the ratings received from all the generation engines and identifies the generation engine with the highest overall rating score. For example, if the following rating results are obtained:

[0863] Generation Engine A scores: 80 points for B, 70 points for C

[0864] Generation Engine B scores: 75 points for A, 85 points for C

[0865] Generation Engine C scores: 65 points for A, 90 points for B

[0866] 8. Determining and recommending the best generation engine

[0867] The server sums up the evaluation scores and recommends the highest-rated generation engine (e.g., engine B in this case) to the user.

[0868] Specific example explanation

[0869] Input text: The user enters the text "What's the weather like today?"

[0870] Text generation: Each generation engine (Engine A, Engine B, Engine C) produces a different output.

[0871] Generation Engine A: "It's sunny today."

[0872] Generation Engine B: "It's cloudy with occasional rain today."

[0873] Generation Engine C: "There is a high chance of rain today."

[0874] Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[0875] Rating aggregation: The server aggregates the rating scores and determines that engine B received the highest rating.

[0876] Recommendation to user: The server notifies the user of the best generation engine B.

[0877] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[0878] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0879] Step 1:

[0880] A user uses a terminal to input the input text "What's the weather like today?" into the system. This input text is sent from the user's terminal to the server. In this case, the prompt sentence ("What's the weather like today?") becomes the input data.

[0881] Step 2:

[0882] The server distributes the user's input text to multiple text generation engines (generative AI models). The server sends the same input text to each text generation engine (e.g., Engine A, Engine B, and Engine C). At this stage, the user's prompt sentence is sent as input to each generation engine.

[0883] Step 3:

[0884] Each text generation engine receives input text and generates text using a generation algorithm. For example, Engine A generates "It's sunny today," Engine B generates "It's cloudy with occasional rain today," and Engine C generates "There's a high chance of rain today." Each generation result is sent back to the server. Here, the generative AI model performs data calculations based on the input text and outputs the text generation result.

[0885] Step 4:

[0886] The server again sends the collected output text of the generation engines to each generation engine, and the generation results of the other generation engines are evaluated against each other. Specifically, the server sends the generation results of engines B and C to engine A, the generation results of engines A and C to engine B, and the generation results of engines A and B to engine C. At this stage, the different generation results are sent to each engine as input data for evaluation.

[0887] Step 5:

[0888] Each generation engine evaluates the output text of other generation engines and returns the evaluation score to the server. For example, Engine A gives Engine B's generation result 80 points and Engine C 70 points. Similarly, Engine B gives Engine A 75 points and Engine C 85 points. Engine C also evaluates the other generation results and returns the score to the server. Here, the score is output based on the evaluation data for the input generation result.

[0889] Step 6:

[0890] The server aggregates the ratings received from all the generation engines, and calculates the overall rating score for each engine, resulting in something like this:

[0891] Generation Engine A scores: 80 points for B, 70 points for C

[0892] Generation Engine B scores: 75 points for A, 85 points for C

[0893] Generation Engine C scores: 65 points for A, 90 points for B

[0894] At this stage, the aggregated evaluation scores are output.

[0895] Step 7:

[0896] The server will aggregate the evaluation scores and identify the generation engine that received the highest score. For example, let's say generation engine B received the highest overall evaluation score of 170 points. The server will notify the user of this result. Here, the generation engine with the highest evaluation score will be output.

[0897] This series of processes allows the user to find the optimal text generation engine.

[0898] (Application example 1)

[0899] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0900] With conventional text generation technology, it is difficult to select the optimal engine from the many natural language generation engines available, making it impossible to obtain text of the quality desired by users. Furthermore, especially on online shopping sites, the quality of text generated when generating product descriptions and customer reviews significantly affects the user experience, so there is a demand for higher quality text generation.

[0901] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0902] In this invention, the server includes means for accepting requests from users, means for sending the same input to multiple natural language generation engines and collecting the text generated therefrom, means for having each natural language generation engine evaluate the collected text and aggregating the evaluation results, means for recommending the most highly rated natural language generation engine to the user, and means for generating product descriptions using an optimal natural language generation engine based on the collected text, thereby enabling users to obtain high-quality product descriptions and reviews using an optimal quality text generation engine.

[0903] "Means for accepting requests from users" refers to a device or interface that allows the server to accept information or requests entered by users.

[0904] "Multiple natural language generation engines" are multiple computer programs that generate sentences based on different algorithms or models.

[0905] "Means for sending identical input and collecting the text generated from them" refers to a device or software that allows a server to send the same input data to multiple natural language generation engines and collect the output results.

[0906] The "means for causing each natural language generation engine to evaluate the collected text" refers to a device or software that instructs each generation engine to evaluate the text generated by the other generation engines.

[0907] The "means for aggregating the evaluation results" refers to a device or software that compiles and analyzes the evaluation results from each generation engine and calculates an overall evaluation.

[0908] The "means for recommending the most highly rated natural language generation engine to the user" is a device or software that notifies the user of the best performing generation engine based on the evaluation results.

[0909] A "means for generating a product description" is a device or software that uses a specific generation engine to generate text describing the details and characteristics of a product based on collected text.

[0910] The present invention relates to a system that utilizes multiple natural language generation engines based on user input and recommends the most suitable engine from among them.

[0911] This system consists of a terminal that accepts requests entered by users, a server that sends the same input to multiple generation engines and collects and evaluates the text generated by each generation engine, and a system that recommends and provides the optimal generation engine and generated text to the user.

[0912] The main components of the system are:

[0913] 1. How to accept requests from users:

[0914] Sends information entered by the user at the terminal (e.g., product description prompts) to the server.

[0915] 2. A means of sending identical input to multiple natural language generation engines and collecting the text generated by them:

[0916] The server receives input from the user and sends it to multiple generation engines such as generation engine A, generation engine B, and generation engine C.

[0917] 3. A method for evaluating the collected text using each natural language generation engine and aggregating the evaluation results:

[0918] The server collects the generated text returned by each generation engine and has the other generation engines evaluate the text.

[0919] 4. How to recommend the highest-rated natural language generation engines to users:

[0920] The evaluation results are compiled, and the server recommends the generation engine with the highest score to the user.

[0921] 5. Using the collected text to generate product descriptions using the best natural language generation engine:

[0922] The server uses an optimal generation engine to generate a high-quality product description and presents it to the user.

[0923] Hardware and Software

[0924] Hardware: Smartphones, tablets, PCs, servers

[0925] Software: Natural language generation API, web frameworks such as Python and Flask, HTTP request libraries (such as requests)

[0926] Data processing and calculation

[0927] The server's actions include:

[0928] Data collection: Collect the generated text from each generation engine.

[0929] Data evaluation: Have each generation engine evaluate the text generated by other generation engines.

[0930] Score calculation: Calculate the overall evaluation score based on the evaluation results.

[0931] Engine recommendation: Recommend the highest scoring generation engine to the user.

[0932] Specific examples

[0933] As an example of a specific prompt sentence, the user may enter, "Please tell me the features of my new smartphone." In this case, the server will perform the following process.

[0934] Generator A: "This smartphone features fast charging."

[0935] Generator B: "It has the latest OS and is waterproof."

[0936] Generator C: "The large battery capacity and high-quality camera are attractive features."

[0937] The server collects the output of each generation engine and evaluates them to select the optimal generation engine and text, which it then provides to the user. In this way, the system generates high-quality text and improves user satisfaction.

[0938] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0939] Step 1:

[0940] A user uses a terminal to enter and send a request.

[0941] Input: A prompt asking the user for a product description (e.g., "What are the features of my new smartphone?")

[0942] Action: The terminal sends this input to the server. Output: The server accepts the prompt.

[0943] Step 2:

[0944] The server sends the received prompts to multiple natural language generation engines (generation engine A, generation engine B, generation engine C, etc.).

[0945] Input: The prompt the server received from the user

[0946] Operation: The server sends the same prompt to generation engine A, generation engine B, and generation engine C. Output: Text generation begins in each generation engine.

[0947] Step 3:

[0948] Each generation engine generates text based on the prompts it receives and sends that text back to the server.

[0949] Input: The prompt received by each generation engine

[0950] Operation: Generation Engine A, Generation Engine B, and Generation Engine C each generate text using their own algorithm. Output: Generated text (Example: Generation Engine A: "This smartphone features fast charging.")

[0951] Step 4:

[0952] The server again sends the collected text to each generation engine and sends a request to evaluate the output text of the other generation engines.

[0953] Input: Text generated by each generation engine

[0954] Operation: The server asks generation engine A to evaluate the texts generated by generation engines B and C. Similarly, it requests evaluations for all other combinations. Output: Each generation engine returns an evaluation score for the texts generated by the other generation engines.

[0955] Step 5:

[0956] Each generation engine evaluates the output text of the other generation engines and returns its evaluation score to the server.

[0957] Input: Text to be evaluated (generated by other generation engines)

[0958] Operation: The evaluation engine evaluates the received text and calculates a score. Output: Evaluation score (e.g., generation engine A gives generation engine B's text a score of 80).

[0959] Step 6:

[0960] The server aggregates the rating scores received from all the generation engines.

[0961] Input: Evaluation scores sent by all generation engines

[0962] Operation: The server aggregates the evaluation scores for each generation engine and calculates the overall evaluation score. Output: Overall evaluation score (e.g., generation engine B received the highest overall score).

[0963] Step 7:

[0964] The server calculates an overall rating score and identifies the generation engine with the highest rating.

[0965] Input: Overall evaluation score

[0966] Behavior: The server selects the best engine and its generated text. Output: The best engine (e.g. engine B).

[0967] Step 8:

[0968] The server generates product descriptions and reviews using an optimal generation engine and provides the final text to the user.

[0969] Input: The optimized generation engine and the generated text

[0970] Behavior: The server uses the best generation engine to finalize the text containing the product description and send it to the user. Output: High-quality product description text provided to the user (e.g., "This smartphone has fast charging capabilities and the latest OS.")

[0971] This processing step enables the system to recommend the most appropriate natural language generation engine for the user's request and generate and provide the most appropriate text.

[0972] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0973] The present invention relates to a system that allows multiple text generation engines to evaluate each other's outputs, recognizes the user's emotions, and takes this information into account to recommend the most suitable generation engine. This system is composed of the following elements: a means for accepting user requests, a means for sending input to multiple text generation engines, a means for evaluating the collected text, a means for aggregating the evaluation results of the generation engines, an emotion engine that recognizes the user's emotions, and a means for recommending the most suitable generation engine to the user.

[0974] Specific examples of program processing

[0975] Step 1: Accepting a request from a user

[0976] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[0977] Step 2: Recognizing user emotions with the emotion engine

[0978] The server uses the input text from the user and the past dialogue history to have the emotion engine recognize the user's emotion. For example, if the input text includes "I'm very tired today," the emotion engine will recognize the user's emotional state as "fatigue."

[0979] Step 3: Distributing input to the text generation engine

[0980] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[0981] Step 4: Text Generation

[0982] Each text generation engine generates new text based on the input text received from the server. The generated text is returned to the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[0983] Step 5: Submit a text rating request

[0984] The server sends the collected output of each generation engine to each generation engine again, and sends a request to each generation engine to evaluate the generation results of other generation engines. It also evaluates how well the generated text matches the user's emotion, based on the user's emotion recognized by the emotion engine.

[0985] Step 6: Peer assessment of the text

[0986] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "It's cloudy with occasional rain today," and assigns a score taking into account the evaluation of the emotion engine. It also evaluates the result of text generation engine C's generation, "There is a high chance of rain today."

[0987] Step 7: Return your evaluation score

[0988] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[0989] Step 8: Aggregating the evaluation results

[0990] The server aggregates the evaluation scores received from all text generation engines and also integrates the emotional relevance scores from the emotion engine. For example, the evaluation scores are collected as follows:

[0991] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[0992] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[0993] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[0994] Step 9: Calculating the overall score

[0995] The server calculates the overall evaluation score based on the aggregated evaluation scores and emotional compatibility scores. For example, it calculates it as follows:

[0996] Overall evaluation score for text generation engine A: 75 points (evaluation score) + 65 points (emotional compatibility) = 140 points

[0997] Overall evaluation score for Text Generation Engine B: 80 points (evaluation score) + 90 points (emotional relevance) = 170 points

[0998] Overall evaluation score for text generation engine C: 70 points (evaluation score) + 85 points (emotional relevance) = 155 points

[0999] Step 10: Determine the best generation engine

[1000] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[1001] Step 11: User Recommendations

[1002] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[1003] In this way, the system of the present invention makes it possible to easily find the most suitable text generation engine while taking into account the emotional state of the user.

[1004] The processing flow will be explained below.

[1005] Step 1:

[1006] The user uses a terminal to send a request to the server, including the input text to be evaluated. The request also includes the content and purpose of the text generation desired by the user.

[1007] Step 2:

[1008] The server receives a request from a user, obtains the input text included in the request, and simultaneously sends a request to the emotion engine to recognize the user's emotion from the input text.

[1009] Step 3:

[1010] The emotion engine analyzes the user's input text and recognizes the user's current emotional state. For example, from the text "I'm very tired today," the emotion engine recognizes the emotional state as "fatigue." The emotion engine then sends the recognition result back to the server.

[1011] Step 4:

[1012] The server sends the received input text to multiple text generation engines (for example, text generation engine A, text generation engine B, and text generation engine C). Each of the text generation engines generates new text based on the input text.

[1013] Step 5:

[1014] Each text generation engine sends the newly generated text back to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy today with occasional rain," and generation engine C generates "There's a high chance of rain today."

[1015] Step 6:

[1016] The server again sends the collected generated text to each text generation engine and sends a request to evaluate the generated results of the other generation engines. It also sends the generated text to the emotion engine and sends a request to evaluate how well it matches the recognized user emotion.

[1017] Step 7:

[1018] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points. It also gives the result of text generation engine C, "There is a high chance of rain today," a score of 70 points.

[1019] Step 8:

[1020] The emotion engine evaluates the suitability of each generated text to the user's emotion and sends the score back to the server. For example, a high score is given to the generated result that best suits the emotional state of "fatigue."

[1021] Step 9:

[1022] The server aggregates the evaluation scores received from each generation engine and the emotional relevance scores from the emotion engine. For example, it calculates a total evaluation score of the evaluation scores and emotional relevance scores of each text generation engine.

[1023] Step 10:

[1024] The server identifies the best-rated text generation engine based on the aggregated overall rating score. For example, if text generation engine B has the highest rating with an overall rating score of 170, this engine is considered the best.

[1025] Step 11:

[1026] The server recommends the optimal text generation engine B to the user. The recommendation result is notified to the user's terminal, and the output of the selected generation engine is provided to the user.

[1027] Example 2

[1028] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1029] Conventional text generation systems often provide generated text and data without considering the user's emotional state, resulting in low user satisfaction. Furthermore, using only a single generative model can lead to variations in the quality of generated results. Furthermore, when using multiple generative models, there is no way to evaluate and compare the generated results, making it difficult to select an efficient generative model.

[1030] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1031] In this invention, the server includes means for accepting requests from a user, means for sending the same input to multiple generative models and collecting data generated from them, means for having each generative model evaluate the collected data and aggregating the evaluation results, means for recommending the most highly rated generative model to the user, means for recognizing the user's emotions, and means for correcting the evaluation results based on the emotions. This makes it possible to select the most appropriate generative model and recommend it to the user, taking into account the user's emotional state.

[1032] "User" means a person or institution that receives data generated using the system.

[1033] A "request" is an instruction or input data that a user sends to a system.

[1034] A "generative model" is an algorithm or software used to generate text or other data.

[1035] "Data" refers to the text or other forms of information output by a generative model.

[1036] "Collection" refers to the process by which the server compiles the data output from the generative model.

[1037] "Evaluation" refers to the process in which a generative model judges the quality and appropriateness of the results generated by other models and assigns them a score.

[1038] "Evaluation results" refer to the scores and feedback obtained after the evaluation process.

[1039] "Aggregation" refers to the process of organizing and calculating the results of multiple evaluations.

[1040] "Recommendation" refers to the act of the server proposing the optimal generative model to the user.

[1041] "Emotion" refers to the emotional state that the user expresses in the input text.

[1042] An "emotion engine" is software or algorithms for recognizing emotions from user input data.

[1043] "Emotional compatibility" is an index that indicates how well the generated text matches the user's emotions.

[1044] "Correction" refers to the process of adjusting the evaluation results by taking into account the emotional compatibility.

[1045] This invention relates to a system for selecting and recommending an optimal model from multiple generative AI models while taking into account the emotional state of a user. This system comprises a means for accepting requests from a user, a means for sending input to multiple generative models, a means for evaluating collected data, a means for aggregating the evaluation results of the generative models, a means for recognizing the user's emotions, and a means for notifying the user of the recommendation results.

[1046] Specifically, this system is implemented using the following hardware and software: The hardware includes the user's device (e.g., PC, smartphone) and a server, while the software includes a generative model (e.g., GPT-3, BERT, Transformer), an emotion engine (e.g., IBM Watson Tone Analyzer), and scripts for the evaluation algorithm.

[1047] The server receives requests from users and sends them to multiple generative models. Each generative model generates new text based on the input text and returns the results to the server. The server aggregates these generation results and uses an evaluation algorithm to have the generation results of each model evaluated against each other.

[1048] Furthermore, the emotion engine recognizes the user's emotion and corrects the evaluation of the generated results based on that emotion. Based on the corrected evaluation results, the server selects the optimal generation model and recommends the result to the user.

[1049] As a concrete example, consider the case where a user sends a request to the system saying, "I'm very tired today. I'd like to know the weather forecast." In this case, the device sends this request to the server, which then forwards it to multiple generative models. Each generative model generates the corresponding text and returns the result to the server. For example, generative model A might generate a result such as "It's sunny today," generative model B might generate a result such as "It's cloudy today with occasional rain," and generative model C might generate a result such as "There's a high chance of rain today."

[1050] The server collects these generated results and configures each model to evaluate the other generated results. The evaluation of each generated result is adjusted based on the user's "fatigue" emotion recognized by the emotion engine. Finally, the server aggregates these evaluation results and recommends the highest-rated generation model to the user.

[1051] In this way, the system is able to provide optimal generated results that take into account the user's emotional state.

[1052] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1053] Step 1:

[1054] The user uses the terminal to send the input text to be evaluated as a request. For example, the user might enter, "I'm very tired today. I'd like to know the weather forecast." The terminal then sends this request to the server. The input is the user's request text, and the output is the request sent to the server.

[1055] Step 2:

[1056] The server recognizes the user's emotion based on the received request. The input is the request text from the user, and the output is the recognized emotion information (e.g., fatigue). The server uses an emotion engine (e.g., IBM Watson Tone Analyzer) to analyze the input text, and the emotion engine recognizes emotions such as "fatigue."

[1057] Step 3:

[1058] The server sends the received input text to multiple generative models. For example, the text "I'm very tired today. I'd like to know the weather forecast." is sent to Model A (GPT-3), Model B (BERT), and Model C (Transformer). The input is the user's request text, and the output is the text sent to each model.

[1059] Step 4:

[1060] Each generative model generates new text based on the input text received from the server. The input is the request text to the generative model, and the output is the generated text. For example, model A generates "It's sunny today," model B generates "It's cloudy with occasional rain today," and model C generates "There's a high chance of rain today," and sends them back to the server.

[1061] Step 5:

[1062] The server sends the collected output of each generative model back to each model and sends a request to evaluate the generated results of other models. The input is the generated text from each generative model, and the output is the sending of an evaluation request. In addition, based on the emotion "fatigue" recognized by the emotion engine, the server also evaluates how well the generated results match that emotion.

[1063] Step 6:

[1064] Each generative model receives the output of other generative models and evaluates its quality and appropriateness. The input is the generated text and recognized sentiment of other generative models, and the output is an evaluation score. For example, model A evaluates the generated result of model B and assigns a score (e.g., 80 points). It also makes corrections based on the sentiment fit.

[1065] Step 7:

[1066] Once each generative model has completed the evaluation, it sends its evaluation score back to the server. The input is the evaluation score, and the output is the score sent to the server. Specifically, model A returns score information to the server, such as "Model B = 80 points, Model C = 70 points."

[1067] Step 8:

[1068] The server aggregates the evaluation scores received from all generative models and also integrates the emotional relevance score from the emotion engine. The input is the evaluation score and emotional relevance score of each generative model, and the output is an overall evaluation score. For example, the overall evaluation score is calculated by adding the evaluation score and the emotional relevance.

[1069] Step 9:

[1070] The server calculates the final overall evaluation score based on the aggregated evaluation scores and emotional relevance scores. The input is the aggregated evaluation scores and emotional relevance scores, and the output is the final overall evaluation score. For example, we calculate the overall evaluation score of Model A.

[1071] Step 10:

[1072] The server identifies the generative model with the highest score based on the overall evaluation score. The input is the overall evaluation score, and the output is the identified generative model. In this case, model B with the highest score is selected as the optimal generative model.

[1073] Step 11:

[1074] The server recommends the optimal generative model to the user. The input is the optimal generative model, and the output is a recommendation notification to the user. The server returns the recommendation result to the user's device and notifies the user.

[1075] (Application example 2)

[1076] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1077] In recent years, it has become increasingly important to provide users with information appropriate to their emotional state when consuming news or content. However, existing text generation engines simply generate text without considering the user's emotions, which prevents them from increasing user satisfaction. Furthermore, there are limited means for relatively evaluating the quality of generated text, making it difficult to select the optimal text generation engine. Therefore, a system is needed that considers the user's emotions and selects and recommends the optimal engine from among multiple text generation engines.

[1078] The specification process by the specification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for accepting a request from a user, means for sending the same input to a plurality of text generation engines and collecting texts generated therefrom, means for having each text generation engine evaluate the collected text and aggregating the evaluation results, an emotion engine for recognizing the user's emotion, means for evaluating how well the generated text matches the user's emotion, and means for recommending the most highly rated text generation engine to the user based on the emotion match. This makes it possible to select a text generation engine that is most suitable for the user's emotional state and provide the user with text generated by that engine.

[1079] "Means for accepting requests from a user" refers to an interface or device for accepting user input or requests.

[1080] "Multiple text generation engines" refers to multiple systems or models that generate text using different algorithms or techniques.

[1081] "Means for sending identical inputs and collecting the text generated from them" refers to a mechanism or process for sending user requests to multiple text generation engines and collecting the text generated by those engines.

[1082] "Means for having each text generation engine evaluate the collected text and aggregating the evaluation results" refers to a mechanism for having each text generation engine evaluate the quality and relevance of the collected text and integrating the evaluation results.

[1083] "Emotion engine that recognizes user emotions" refers to an engine or algorithm that analyzes and recognizes emotions from a user's text and behavioral history.

[1084] "Means for assessing how well generated text matches the user's emotions" refers to a process or device for assessing how well generated text matches the user's current emotional state.

[1085] "Means for recommending the most highly rated text generation engine to a user based on emotional compatibility" refers to a mechanism for selecting the most suitable text generation engine and recommending it to a user, taking into account the user's emotional compatibility.

[1086] The present invention relates to a system that provides content by selecting an optimal engine from among a plurality of text generation engines in consideration of a user's feelings. Specific embodiments for carrying out the present invention are described below.

[1087] System Overview

[1088] This system consists of a user terminal, a server, multiple text generation engines, and an emotion engine. The user sends a request, and the system generates text according to the request and recommends the generation engine that best suits the user's emotion.

[1089] Major hardware and software

[1090] 1. User Device

[1091] A device (smartphone, tablet, etc.) for accepting user input.

[1092] 2. Server

[1093] It hosts the text generation and sentiment engines, and processes and evaluates requests.

[1094] 3. Emotion Engine

[1095] An engine for information analysis and emotion recognition. It can use "AWS Rekognition" and "Google Cloud Natural Language."

[1096] 4. Text Generation Engine

[1097] Generate text using multiple generation engines (e.g., GPT-3, BERT).

[1098] Processing Details

[1099] The server first receives a request from the user. This request contains information about the content the user desires. The emotion engine then recognizes the user's emotion and generates metrics based on that emotion. This emotion data is obtained by analyzing past requests, interaction history, and other factors.

[1100] The server then sends the user's input text to multiple text generation engines and collects the generated text from each engine. The collected text is then sent to each engine again to have it evaluate the output of other generation engines. Furthermore, the emotion engine evaluates the emotional relevance of the generated text and integrates the evaluation score into the evaluation score of each text generation engine.

[1101] The server finally aggregates all the evaluation scores and calculates an overall evaluation score, taking into account the emotional compatibility. The text generation engine with the highest overall evaluation score is recommended to the user.

[1102] Specific examples

[1103] For example, if a user sends a request saying, "I want to see more happy news today," the emotion engine extracts the emotion data of "fun." Generation engine A generates "happy stories about cats," generation engine B generates "latest superhero movie reviews," and generation engine C generates "new restaurant information." Each generation engine evaluates the other's generation results, and the engine with the highest score, taking into account emotional compatibility, is determined and recommended to the user.

[1104] Prompt Sentence Examples

[1105] Example of a user request: "I want to see more exciting news today."

[1106] Example of an emotion rating: "fun"

[1107] Generated text by Generation Engine A: "A happy story about a cat"

[1108] Generated text by Generation Engine B: "Latest superhero movie reviews"

[1109] Generated text by generation engine C: "New restaurant information"

[1110] Through this system, it is possible to select a text generation engine that best suits the user's emotional state and provide the user with the text generated by that engine.

[1111] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1112] Step 1:

[1113] The server receives a request from the user terminal. This request contains information about the content the user desires. As input, the user sends the text "I want to see more entertaining news today." This text reaches the server and triggers the next processing step.

[1114] Step 2:

[1115] The server sends the received request text to the emotion engine. The emotion engine recognizes the user's emotion based on the user's text and past dialogue history. For example, the emotion engine recognizes emotions such as "happy," "sad," and "tired." Specifically, for the input text "I want to see more happy news today," the emotion engine extracts "happy." Emotion recognition data is generated and used in the next step.

[1116] Step 3:

[1117] The server sends the user's input text to multiple text generation engines. A generative AI model is used for this process. For example, generation engine A, generation engine B, and generation engine C receive the same input text, "I want to see more fun news today," and each generates new text using its own algorithm. As output, generation engine A sends "Happy stories about cats," generation engine B sends "Latest superhero movie reviews," and generation engine C sends "New restaurant information" back to the server.

[1118] Step 4:

[1119] The server then sends the collected generated texts to each text generation engine, which then evaluates the other's generated results. In this process, each generation engine evaluates the quality and relevance of the text received from the other engines. Specifically, generation engine A evaluates the texts from generation engines B and C, and generation engine B performs the same evaluation. At this stage, an evaluation score is generated.

[1120] Step 5:

[1121] The server evaluates the emotional relevance of the generated text using the emotion engines. For example, it evaluates whether "a happy story about a cat" is highly relevant to the emotion of "fun." This emotional relevance score is also integrated into the evaluation score of each text generation engine. This evaluation score becomes part of the final overall evaluation score.

[1122] Step 6:

[1123] The server aggregates all the evaluation scores and calculates an overall evaluation score. This overall evaluation score includes the evaluation scores from each engine and the emotional compatibility score. Specifically, let's say that generation engine A receives an overall evaluation score of 80 points, generation engine B receives an overall evaluation score of 90 points, and generation engine C receives an overall evaluation score of 85 points.

[1124] Step 7:

[1125] The server identifies the most highly rated text generation engine based on the overall evaluation score. Here, engine B has the highest score of 90, and is therefore recognized as the most suitable text generation engine.

[1126] Step 8:

[1127] The server recommends the most suitable text generation engine B to the user. The recommendation result is returned to the user's terminal and notified to the user. Specifically, the user receives an article titled "Latest Superhero Movie Reviews" and can view it.

[1128] In this way, it is possible to select the most suitable text generation engine while taking into consideration the emotional state of the user, and to provide appropriate content to the user.

[1129] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1130] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1131] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1132] [Fourth embodiment]

[1133] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1134] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1135] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1136] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1137] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1138] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1139] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1140] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1141] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1142] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1143] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1144] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1145] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1146] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[1147] Specific examples of program processing

[1148] Step 1: Accepting a request from a user

[1149] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[1150] Step 2: Distributing input to the text generation engine

[1151] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[1152] Step 3: Text generation

[1153] Each text generation engine generates text based on the input it receives. The generated text is returned to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy with occasional rain today," and generation engine C generates "There's a high chance of rain today."

[1154] Step 4: Submit a text rating request

[1155] The server sends the collected output of each generation engine back to each generation engine, and sends a request to evaluate the output text of other generation engines. For example, generation engine A receives a request to evaluate the output text of generation engines B and C.

[1156] Step 5: Peer assessment of the text

[1157] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, generation engine A evaluates the generation result of B and gives it a score of 80, and evaluates the generation result of C and gives it a score of 70. Generation engine B also evaluates the generation results of A and C in the same way and returns their respective scores to the server.

[1158] Step 6: Aggregating the evaluation results

[1159] The server aggregates the reputation scores received from all the generation engines. For example, let's say the collected reputation scores are aggregated as follows:

[1160] Generation Engine A scores: 80 points for B, 70 points for C

[1161] Generation Engine B scores: 75 points for A, 85 points for C

[1162] Generation Engine C scores: 65 points for A, 90 points for B

[1163] Step 7: Determine and recommend the best generation engine

[1164] The server calculates the overall evaluation score and identifies the generation engine with the highest evaluation. For example, generation engine B has the highest overall evaluation score of 170 points. This result is recommended to the user.

[1165] Specific example explanation

[1166] 1. Input text: The user enters the text "What's the weather like today?"

[1167] 2. Text generation: Each generation engine (A, B, C) produces a different output.

[1168] Generation Engine A: "It's sunny today."

[1169] Generation Engine B: "It's cloudy with occasional rain today."

[1170] Generation Engine C: "There is a high chance of rain today."

[1171] 3. Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[1172] 4. Rating Aggregation: The server aggregates the rating scores and determines that engine B has the highest rating.

[1173] 5. Recommendation to the user: The server notifies the user of the best generation engine B.

[1174] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[1175] The processing flow will be explained below.

[1176] Step 1:

[1177] A user uses a terminal to send a request including an input text to be evaluated to a server. The request includes information indicating that the user wants to evaluate a text generation engine that the user wants to use.

[1178] Step 2:

[1179] The server receives a request from a user, obtains the input text included in the request, and sends it to multiple text generation engines (e.g., text generation engine A, text generation engine B, and text generation engine C).

[1180] Step 3:

[1181] Each text generation engine generates new text based on the input text received from the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[1182] Step 4:

[1183] Each text generation engine sends the generated text back to the server, which collects these results.

[1184] Step 5:

[1185] The server resends the collected generation results to each of text generation engines A, B, and C, and sends a request to evaluate the generation results of the other generation engines.

[1186] Step 6:

[1187] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points, while text generation engine C's generation, "There is a high chance of rain today," gives it a score of 70 points.

[1188] Step 7:

[1189] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[1190] Step 8:

[1191] The server aggregates the evaluation scores received from all text generation engines, for example, as follows:

[1192] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[1193] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[1194] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[1195] Step 9:

[1196] The server calculates the overall evaluation score based on the aggregated evaluation scores. For example, it calculates it as follows:

[1197] Overall score for Text Generation Engine A = 75 points + 65 points = 140 points

[1198] Overall score for Text Generation Engine B = 80 points + 90 points = 170 points

[1199] Overall score for text generation engine C = 70 points + 85 points = 155 points

[1200] Step 10:

[1201] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[1202] Step 11:

[1203] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[1204] Example 1

[1205] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1206] Conventional text generation systems rely on the performance of a single text generation engine, resulting in inconsistent quality of generated text and not necessarily optimal generation results. Furthermore, there is no established method for comparing and evaluating the output of multiple text generation engines, making it difficult to determine which engine best suits a user's needs. This makes it difficult for users to find the optimal generation engine, affecting their satisfaction with the generated results.

[1207] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1208] In this invention, the server includes means for accepting requests from a user, means for sending the same input to a plurality of text generation engines and collecting the texts generated therefrom, means for sending the collected texts to each text generation engine to have them evaluate each other and collecting the evaluation results, means for aggregating the evaluation results, and means for recommending the most highly rated text generation engine to the user. This enables the user to easily find the text generation engine that is most suitable for the text generation result they desire.

[1209] "Means for accepting requests from users" refers to the function of sending text or requests entered by a user using a terminal to a server and having the server receive them.

[1210] "Means for sending the same input to multiple text generation engines and collecting the text generated from them" refers to a function in which a server sends a specific input text to multiple different text generation engines and collects the generated results.

[1211] "Means for sending the collected text to each text generation engine to have them evaluate each other, and collecting the evaluation results" refers to the function of the server sending the collected text to each text generation engine again and having them evaluate the generation results of the other generation engines.

[1212] "Means for aggregating evaluation results" refers to the function of aggregating the evaluation results returned from each text generation engine and calculating the total score and average score.

[1213] The "means for recommending the most highly rated text generation engine to the user" refers to a function for identifying the text generation engine that has received the highest score based on the evaluation results and notifying the user of that information.

[1214] "Calculating an overall evaluation score" refers to calculating an overall evaluation point based on the evaluation scores from each text generation engine.

[1215] "Mutual evaluation" refers to an evaluation method in which each text generation engine evaluates the results of the other engines' generation and returns the evaluation results in numerical form.

[1216] "Text generation engine" refers to an algorithm or software that generates natural language text based on specific input.

[1217] This invention relates to a system that allows multiple text generation engines to evaluate each other's outputs and recommends the optimal generation engine to a user. This system mainly consists of the following elements: a means for accepting requests from a user, a means for sending input to multiple text generation engines, a means for having collected texts evaluated, a means for aggregating the evaluation results, and a means for recommending the optimal generation engine.

[1218] Specific implementation methods

[1219] 1. Hardware and software configuration

[1220] This system consists of a server running on the cloud, user devices (PCs, smartphones, tablets, etc.), and multiple text generation engines, including OpenAI GPT-3 and Google BERT.

[1221] 2. Accepting a request from a user

[1222] A user uses a device to input a specific prompt sentence (e.g., "What's the weather like today?") via the system's web interface or mobile app. The device sends this request to the server, which accepts the request.

[1223] 3. Input distribution to multiple text generation engines

[1224] The server sends the received input text to multiple pre-configured text generation engines, each of which analyzes the received input text and returns the generated text to the server.

[1225] 4. Text Generation

[1226] Each text generation engine (for example, engine A, engine B, and engine C) sends its generated text to the server based on the input "What's the weather like today?". As a specific example, engine A's generated text is "It's sunny today," engine B's is "It's cloudy with occasional rain today," and engine C's is "There's a high chance of rain today."

[1227] 5. Submitting a Text Rating Request

[1228] The server sends the collected output of each generation engine to each generation engine again, and has them evaluate the output text of other generation engines. For example, it sends a request to engine A to evaluate the output text of engines B and C.

[1229] 6. Peer evaluation of texts

[1230] Each generation engine evaluates the output text of the other generation engines and returns the evaluation score to the server. For example, engine A gives a score of 80 to the generation result of engine B and a score of 70 to engine C. Engine B similarly evaluates the output of engines A and C and returns their respective scores to the server.

[1231] 7. Calculation of evaluation results

[1232] The server aggregates the ratings received from all the generation engines and identifies the generation engine with the highest overall rating score. For example, if the following rating results are obtained:

[1233] Generation Engine A scores: 80 points for B, 70 points for C

[1234] Generation Engine B scores: 75 points for A, 85 points for C

[1235] Generation Engine C scores: 65 points for A, 90 points for B

[1236] 8. Determining and recommending the best generation engine

[1237] The server sums up the evaluation scores and recommends the highest-rated generation engine (e.g., engine B in this case) to the user.

[1238] Specific example explanation

[1239] Input text: The user enters the text "What's the weather like today?"

[1240] Text generation: Each generation engine (Engine A, Engine B, Engine C) produces a different output.

[1241] Generation Engine A: "It's sunny today."

[1242] Generation Engine B: "It's cloudy with occasional rain today."

[1243] Generation Engine C: "There is a high chance of rain today."

[1244] Peer evaluation: Each generation engine evaluates the output of the other generation engines and sends a score back to the server.

[1245] Rating aggregation: The server aggregates the rating scores and determines that engine B received the highest rating.

[1246] Recommendation to user: The server notifies the user of the best generation engine B.

[1247] In this way, the system of the present invention allows a user to easily find the text generation engine that is most suitable for the text generation results that he or she desires.

[1248] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1249] Step 1:

[1250] A user uses a terminal to input the input text "What's the weather like today?" into the system. This input text is sent from the user's terminal to the server. In this case, the prompt sentence ("What's the weather like today?") becomes the input data.

[1251] Step 2:

[1252] The server distributes the user's input text to multiple text generation engines (generative AI models). The server sends the same input text to each text generation engine (e.g., Engine A, Engine B, and Engine C). At this stage, the user's prompt sentence is sent as input to each generation engine.

[1253] Step 3:

[1254] Each text generation engine receives input text and generates text using a generation algorithm. For example, Engine A generates "It's sunny today," Engine B generates "It's cloudy with occasional rain today," and Engine C generates "There's a high chance of rain today." Each generation result is sent back to the server. Here, the generative AI model performs data calculations based on the input text and outputs the text generation result.

[1255] Step 4:

[1256] The server again sends the collected output text of the generation engines to each generation engine, and the generation results of the other generation engines are evaluated against each other. Specifically, the server sends the generation results of engines B and C to engine A, the generation results of engines A and C to engine B, and the generation results of engines A and B to engine C. At this stage, the different generation results are sent to each engine as input data for evaluation.

[1257] Step 5:

[1258] Each generation engine evaluates the output text of other generation engines and returns the evaluation score to the server. For example, Engine A gives Engine B's generation result 80 points and Engine C 70 points. Similarly, Engine B gives Engine A 75 points and Engine C 85 points. Engine C also evaluates the other generation results and returns the score to the server. Here, the score is output based on the evaluation data for the input generation result.

[1259] Step 6:

[1260] The server aggregates the ratings received from all the generation engines, and calculates the overall rating score for each engine, resulting in something like this:

[1261] Generation Engine A scores: 80 points for B, 70 points for C

[1262] Generation Engine B scores: 75 points for A, 85 points for C

[1263] Generation Engine C scores: 65 points for A, 90 points for B

[1264] At this stage, the aggregated evaluation scores are output.

[1265] Step 7:

[1266] The server will aggregate the evaluation scores and identify the generation engine that received the highest score. For example, let's say generation engine B received the highest overall evaluation score of 170 points. The server will notify the user of this result. Here, the generation engine with the highest evaluation score will be output.

[1267] This series of processes allows the user to find the optimal text generation engine.

[1268] (Application example 1)

[1269] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1270] With conventional text generation technology, it is difficult to select the optimal engine from the many natural language generation engines available, making it impossible to obtain text of the quality desired by users. Furthermore, especially on online shopping sites, the quality of text generated when generating product descriptions and customer reviews significantly affects the user experience, so there is a demand for higher quality text generation.

[1271] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1272] In this invention, the server includes means for accepting requests from users, means for sending the same input to multiple natural language generation engines and collecting the text generated therefrom, means for having each natural language generation engine evaluate the collected text and aggregating the evaluation results, means for recommending the most highly rated natural language generation engine to the user, and means for generating product descriptions using an optimal natural language generation engine based on the collected text, thereby enabling users to obtain high-quality product descriptions and reviews using an optimal quality text generation engine.

[1273] "Means for accepting requests from users" refers to a device or interface that allows the server to accept information or requests entered by users.

[1274] "Multiple natural language generation engines" are multiple computer programs that generate sentences based on different algorithms or models.

[1275] "Means for sending identical input and collecting the text generated from them" refers to a device or software that allows a server to send the same input data to multiple natural language generation engines and collect the output results.

[1276] The "means for causing each natural language generation engine to evaluate the collected text" refers to a device or software that instructs each generation engine to evaluate the text generated by the other generation engines.

[1277] The "means for aggregating the evaluation results" refers to a device or software that compiles and analyzes the evaluation results from each generation engine and calculates an overall evaluation.

[1278] The "means for recommending the most highly rated natural language generation engine to the user" is a device or software that notifies the user of the best performing generation engine based on the evaluation results.

[1279] A "means for generating a product description" is a device or software that uses a specific generation engine to generate text describing the details and characteristics of a product based on collected text.

[1280] The present invention relates to a system that utilizes multiple natural language generation engines based on user input and recommends the most suitable engine from among them.

[1281] This system consists of a terminal that accepts requests entered by users, a server that sends the same input to multiple generation engines and collects and evaluates the text generated by each generation engine, and a system that recommends and provides the optimal generation engine and generated text to the user.

[1282] The main components of the system are:

[1283] 1. How to accept requests from users:

[1284] Sends information entered by the user at the terminal (e.g., product description prompts) to the server.

[1285] 2. A means of sending identical input to multiple natural language generation engines and collecting the text generated by them:

[1286] The server receives input from the user and sends it to multiple generation engines such as generation engine A, generation engine B, and generation engine C.

[1287] 3. A method for evaluating the collected text using each natural language generation engine and aggregating the evaluation results:

[1288] The server collects the generated text returned by each generation engine and has the other generation engines evaluate the text.

[1289] 4. How to recommend the highest-rated natural language generation engines to users:

[1290] The evaluation results are compiled, and the server recommends the generation engine with the highest score to the user.

[1291] 5. Using the collected text to generate product descriptions using the best natural language generation engine:

[1292] The server uses an optimal generation engine to generate a high-quality product description and presents it to the user.

[1293] Hardware and Software

[1294] Hardware: Smartphones, tablets, PCs, servers

[1295] Software: Natural language generation API, web frameworks such as Python and Flask, HTTP request libraries (such as requests)

[1296] Data processing and calculation

[1297] The server's actions include:

[1298] Data collection: Collect the generated text from each generation engine.

[1299] Data evaluation: Have each generation engine evaluate the text generated by other generation engines.

[1300] Score calculation: Calculate the overall evaluation score based on the evaluation results.

[1301] Engine recommendation: Recommend the highest scoring generation engine to the user.

[1302] Specific examples

[1303] As an example of a specific prompt sentence, the user may enter, "Please tell me the features of my new smartphone." In this case, the server will perform the following process.

[1304] Generator A: "This smartphone features fast charging."

[1305] Generator B: "It has the latest OS and is waterproof."

[1306] Generator C: "The large battery capacity and high-quality camera are attractive features."

[1307] The server collects the output of each generation engine and evaluates them to select the optimal generation engine and text, which it then provides to the user. In this way, the system generates high-quality text and improves user satisfaction.

[1308] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1309] Step 1:

[1310] A user uses a terminal to enter and send a request.

[1311] Input: A prompt asking the user for a product description (e.g., "What are the features of my new smartphone?")

[1312] Action: The terminal sends this input to the server. Output: The server accepts the prompt.

[1313] Step 2:

[1314] The server sends the received prompts to multiple natural language generation engines (generation engine A, generation engine B, generation engine C, etc.).

[1315] Input: The prompt the server received from the user

[1316] Operation: The server sends the same prompt to generation engine A, generation engine B, and generation engine C. Output: Text generation begins in each generation engine.

[1317] Step 3:

[1318] Each generation engine generates text based on the prompts it receives and sends that text back to the server.

[1319] Input: The prompt received by each generation engine

[1320] Operation: Generation Engine A, Generation Engine B, and Generation Engine C each generate text using their own algorithm. Output: Generated text (Example: Generation Engine A: "This smartphone features fast charging.")

[1321] Step 4:

[1322] The server again sends the collected text to each generation engine and sends a request to evaluate the output text of the other generation engines.

[1323] Input: Text generated by each generation engine

[1324] Operation: The server asks generation engine A to evaluate the texts generated by generation engines B and C. Similarly, it requests evaluations for all other combinations. Output: Each generation engine returns an evaluation score for the texts generated by the other generation engines.

[1325] Step 5:

[1326] Each generation engine evaluates the output text of the other generation engines and returns its evaluation score to the server.

[1327] Input: Text to be evaluated (generated by other generation engines)

[1328] Operation: The evaluation engine evaluates the received text and calculates a score. Output: Evaluation score (e.g., generation engine A gives generation engine B's text a score of 80).

[1329] Step 6:

[1330] The server aggregates the rating scores received from all the generation engines.

[1331] Input: Evaluation scores sent by all generation engines

[1332] Operation: The server aggregates the evaluation scores for each generation engine and calculates the overall evaluation score. Output: Overall evaluation score (e.g., generation engine B received the highest overall score).

[1333] Step 7:

[1334] The server calculates an overall rating score and identifies the generation engine with the highest rating.

[1335] Input: Overall evaluation score

[1336] Behavior: The server selects the best engine and its generated text. Output: The best engine (e.g. engine B).

[1337] Step 8:

[1338] The server generates product descriptions and reviews using an optimal generation engine and provides the final text to the user.

[1339] Input: The optimized generation engine and the generated text

[1340] Behavior: The server uses the best generation engine to finalize the text containing the product description and send it to the user. Output: High-quality product description text provided to the user (e.g., "This smartphone has fast charging capabilities and the latest OS.")

[1341] This processing step enables the system to recommend the most appropriate natural language generation engine for the user's request and generate and provide the most appropriate text.

[1342] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1343] The present invention relates to a system that allows multiple text generation engines to evaluate each other's outputs, recognizes the user's emotions, and takes this information into account to recommend the most suitable generation engine. This system is composed of the following elements: a means for accepting user requests, a means for sending input to multiple text generation engines, a means for evaluating the collected text, a means for aggregating the evaluation results of the generation engines, an emotion engine that recognizes the user's emotions, and a means for recommending the most suitable generation engine to the user.

[1344] Specific examples of program processing

[1345] Step 1: Accepting a request from a user

[1346] The user uses the terminal to send an input text to be evaluated as a request, which the terminal sends to the server, which accepts the request.

[1347] Step 2: Recognizing user emotions with the emotion engine

[1348] The server uses the input text from the user and the past dialogue history to have the emotion engine recognize the user's emotion. For example, if the input text includes "I'm very tired today," the emotion engine will recognize the user's emotional state as "fatigue."

[1349] Step 3: Distributing input to the text generation engine

[1350] The server sends the received input text to multiple text generation engines. For example, the same input text is sent to text generation engine A, text generation engine B, and text generation engine C.

[1351] Step 4: Text Generation

[1352] Each text generation engine generates new text based on the input text received from the server. The generated text is returned to the server. For example, text generation engine A generates "It's sunny today," text generation engine B generates "It's cloudy with occasional rain today," and text generation engine C generates "There's a high chance of rain today."

[1353] Step 5: Submit a text rating request

[1354] The server sends the collected output of each generation engine to each generation engine again, and sends a request to each generation engine to evaluate the generation results of other generation engines. It also evaluates how well the generated text matches the user's emotion, based on the user's emotion recognized by the emotion engine.

[1355] Step 6: Peer assessment of the text

[1356] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "It's cloudy with occasional rain today," and assigns a score taking into account the evaluation of the emotion engine. It also evaluates the result of text generation engine C's generation, "There is a high chance of rain today."

[1357] Step 7: Return your evaluation score

[1358] Once each text generation engine has completed its evaluation, it sends its evaluation score back to the server.

[1359] Step 8: Aggregating the evaluation results

[1360] The server aggregates the evaluation scores received from all text generation engines and also integrates the emotional relevance scores from the emotion engine. For example, the evaluation scores are collected as follows:

[1361] Evaluation score of Text Generation Engine A: Text Generation Engine B = 80 points, Text Generation Engine C = 70 points

[1362] Evaluation score for Text Generation Engine B: Text Generation Engine A = 75 points, Text Generation Engine C = 85 points

[1363] Evaluation score for Text Generation Engine C: Text Generation Engine A = 65 points, Text Generation Engine B = 90 points

[1364] Step 9: Calculating the overall score

[1365] The server calculates the overall evaluation score based on the aggregated evaluation scores and emotional compatibility scores. For example, it calculates it as follows:

[1366] Overall evaluation score for text generation engine A: 75 points (evaluation score) + 65 points (emotional compatibility) = 140 points

[1367] Overall evaluation score for Text Generation Engine B: 80 points (evaluation score) + 90 points (emotional relevance) = 170 points

[1368] Overall evaluation score for text generation engine C: 70 points (evaluation score) + 85 points (emotional relevance) = 155 points

[1369] Step 10: Determine the best generation engine

[1370] The server identifies the most highly rated text generation engine based on the overall evaluation score. In this case, engine B has the highest score of 170 points, so it is selected as the optimal engine.

[1371] Step 11: User Recommendations

[1372] The server recommends the optimal text generation engine B to the user, and returns the recommendation result to the user's terminal to notify the user.

[1373] In this way, the system of the present invention makes it possible to easily find the most suitable text generation engine while taking into account the emotional state of the user.

[1374] The processing flow will be explained below.

[1375] Step 1:

[1376] The user uses a terminal to send a request to the server, including the input text to be evaluated. The request also includes the content and purpose of the text generation desired by the user.

[1377] Step 2:

[1378] The server receives a request from a user, obtains the input text included in the request, and simultaneously sends a request to the emotion engine to recognize the user's emotion from the input text.

[1379] Step 3:

[1380] The emotion engine analyzes the user's input text and recognizes the user's current emotional state. For example, from the text "I'm very tired today," the emotion engine recognizes the emotional state as "fatigue." The emotion engine then sends the recognition result back to the server.

[1381] Step 4:

[1382] The server sends the received input text to multiple text generation engines (for example, text generation engine A, text generation engine B, and text generation engine C). Each of the text generation engines generates new text based on the input text.

[1383] Step 5:

[1384] Each text generation engine sends the newly generated text back to the server. For example, suppose generation engine A generates "It's sunny today," generation engine B generates "It's cloudy today with occasional rain," and generation engine C generates "There's a high chance of rain today."

[1385] Step 6:

[1386] The server again sends the collected generated text to each text generation engine and sends a request to evaluate the generated results of the other generation engines. It also sends the generated text to the emotion engine and sends a request to evaluate how well it matches the recognized user emotion.

[1387] Step 7:

[1388] Each text generation engine receives the text generated by the other generation engines and evaluates its quality and appropriateness. For example, text generation engine A evaluates the result of text generation engine B's generation, "Today it's cloudy with occasional rain," and gives it a score of 80 points. It also gives the result of text generation engine C, "There is a high chance of rain today," a score of 70 points.

[1389] Step 8:

[1390] The emotion engine evaluates the suitability of each generated text to the user's emotion and sends the score back to the server. For example, a high score is given to the generated result that best suits the emotional state of "fatigue."

[1391] Step 9:

[1392] The server aggregates the evaluation scores received from each generation engine and the emotional relevance scores from the emotion engine. For example, it calculates a total evaluation score of the evaluation scores and emotional relevance scores of each text generation engine.

[1393] Step 10:

[1394] The server identifies the best-rated text generation engine based on the aggregated overall rating score. For example, if text generation engine B has the highest rating with an overall rating score of 170, this engine is considered the best.

[1395] Step 11:

[1396] The server recommends the optimal text generation engine B to the user. The recommendation result is notified to the user's terminal, and the output of the selected generation engine is provided to the user.

[1397] Example 2

[1398] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1399] Conventional text generation systems often provide generated text and data without considering the user's emotional state, resulting in low user satisfaction. Furthermore, using only a single generative model can lead to variations in the quality of generated results. Furthermore, when using multiple generative models, there is no way to evaluate and compare the generated results, making it difficult to select an efficient generative model.

[1400] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1401] In this invention, the server includes means for accepting requests from a user, means for sending the same input to multiple generative models and collecting data generated from them, means for having each generative model evaluate the collected data and aggregating the evaluation results, means for recommending the most highly rated generative model to the user, means for recognizing the user's emotions, and means for correcting the evaluation results based on the emotions. This makes it possible to select the most appropriate generative model and recommend it to the user, taking into account the user's emotional state.

[1402] "User" means a person or institution that receives data generated using the system.

[1403] A "request" is an instruction or input data that a user sends to a system.

[1404] A "generative model" is an algorithm or software used to generate text or other data.

[1405] "Data" refers to the text or other forms of information output by a generative model.

[1406] "Collection" refers to the process by which the server compiles the data output from the generative model.

[1407] "Evaluation" refers to the process in which a generative model judges the quality and appropriateness of the results generated by other models and assigns them a score.

[1408] "Evaluation results" refer to the scores and feedback obtained after the evaluation process.

[1409] "Aggregation" refers to the process of organizing and calculating the results of multiple evaluations.

[1410] "Recommendation" refers to the act of the server proposing the optimal generative model to the user.

[1411] "Emotion" refers to the emotional state that the user expresses in the input text.

[1412] An "emotion engine" is software or algorithms for recognizing emotions from user input data.

[1413] "Emotional compatibility" is an index that indicates how well the generated text matches the user's emotions.

[1414] "Correction" refers to the process of adjusting the evaluation results by taking into account the emotional compatibility.

[1415] This invention relates to a system for selecting and recommending an optimal model from multiple generative AI models while taking into account the emotional state of a user. This system comprises a means for accepting requests from a user, a means for sending input to multiple generative models, a means for evaluating collected data, a means for aggregating the evaluation results of the generative models, a means for recognizing the user's emotions, and a means for notifying the user of the recommendation results.

[1416] Specifically, this system is implemented using the following hardware and software: The hardware includes the user's device (e.g., PC, smartphone) and a server, while the software includes a generative model (e.g., GPT-3, BERT, Transformer), an emotion engine (e.g., IBM Watson Tone Analyzer), and scripts for the evaluation algorithm.

[1417] The server receives requests from users and sends them to multiple generative models. Each generative model generates new text based on the input text and returns the results to the server. The server aggregates these generation results and uses an evaluation algorithm to have the generation results of each model evaluated against each other.

[1418] Furthermore, the emotion engine recognizes the user's emotion and corrects the evaluation of the generated results based on that emotion. Based on the corrected evaluation results, the server selects the optimal generation model and recommends the result to the user.

[1419] As a concrete example, consider the case where a user sends a request to the system saying, "I'm very tired today. I'd like to know the weather forecast." In this case, the device sends this request to the server, which then forwards it to multiple generative models. Each generative model generates the corresponding text and returns the result to the server. For example, generative model A might generate a result such as "It's sunny today," generative model B might generate a result such as "It's cloudy today with occasional rain," and generative model C might generate a result such as "There's a high chance of rain today."

[1420] The server collects these generated results and configures each model to evaluate the other generated results. The evaluation of each generated result is adjusted based on the user's "fatigue" emotion recognized by the emotion engine. Finally, the server aggregates these evaluation results and recommends the highest-rated generation model to the user.

[1421] In this way, the system is able to provide optimal generated results that take into account the user's emotional state.

[1422] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1423] Step 1:

[1424] The user uses the terminal to send the input text to be evaluated as a request. For example, the user might enter, "I'm very tired today. I'd like to know the weather forecast." The terminal then sends this request to the server. The input is the user's request text, and the output is the request sent to the server.

[1425] Step 2:

[1426] The server recognizes the user's emotion based on the received request. The input is the request text from the user, and the output is the recognized emotion information (e.g., fatigue). The server uses an emotion engine (e.g., IBM Watson Tone Analyzer) to analyze the input text, and the emotion engine recognizes emotions such as "fatigue."

[1427] Step 3:

[1428] The server sends the received input text to multiple generative models. For example, the text "I'm very tired today. I'd like to know the weather forecast." is sent to Model A (GPT-3), Model B (BERT), and Model C (Transformer). The input is the user's request text, and the output is the text sent to each model.

[1429] Step 4:

[1430] Each generative model generates new text based on the input text received from the server. The input is the request text to the generative model, and the output is the generated text. For example, model A generates "It's sunny today," model B generates "It's cloudy with occasional rain today," and model C generates "There's a high chance of rain today," and sends them back to the server.

[1431] Step 5:

[1432] The server sends the collected output of each generative model back to each model and sends a request to evaluate the generated results of other models. The input is the generated text from each generative model, and the output is the sending of an evaluation request. In addition, based on the emotion "fatigue" recognized by the emotion engine, the server also evaluates how well the generated results match that emotion.

[1433] Step 6:

[1434] Each generative model receives the output of other generative models and evaluates its quality and appropriateness. The input is the generated text and recognized sentiment of other generative models, and the output is an evaluation score. For example, model A evaluates the generated result of model B and assigns a score (e.g., 80 points). It also makes corrections based on the sentiment fit.

[1435] Step 7:

[1436] Once each generative model has completed the evaluation, it sends its evaluation score back to the server. The input is the evaluation score, and the output is the score sent to the server. Specifically, model A returns score information to the server, such as "Model B = 80 points, Model C = 70 points."

[1437] Step 8:

[1438] The server aggregates the evaluation scores received from all generative models and also integrates the emotional relevance score from the emotion engine. The input is the evaluation score and emotional relevance score of each generative model, and the output is an overall evaluation score. For example, the overall evaluation score is calculated by adding the evaluation score and the emotional relevance.

[1439] Step 9:

[1440] The server calculates the final overall evaluation score based on the aggregated evaluation scores and emotional relevance scores. The input is the aggregated evaluation scores and emotional relevance scores, and the output is the final overall evaluation score. For example, we calculate the overall evaluation score of Model A.

[1441] Step 10:

[1442] The server identifies the generative model with the highest score based on the overall evaluation score. The input is the overall evaluation score, and the output is the identified generative model. In this case, model B with the highest score is selected as the optimal generative model.

[1443] Step 11:

[1444] The server recommends the optimal generative model to the user. The input is the optimal generative model, and the output is a recommendation notification to the user. The server returns the recommendation result to the user's device and notifies the user.

[1445] (Application example 2)

[1446] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1447] In recent years, it has become increasingly important to provide users with information appropriate to their emotional state when consuming news or content. However, existing text generation engines simply generate text without considering the user's emotions, which prevents them from increasing user satisfaction. Furthermore, there are limited means for relatively evaluating the quality of generated text, making it difficult to select the optimal text generation engine. Therefore, a system is needed that considers the user's emotions and selects and recommends the optimal engine from among multiple text generation engines.

[1448] The specification process by the specification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for accepting a request from a user, means for sending the same input to a plurality of text generation engines and collecting texts generated therefrom, means for having each text generation engine evaluate the collected text and aggregating the evaluation results, an emotion engine for recognizing the user's emotion, means for evaluating how well the generated text matches the user's emotion, and means for recommending the most highly rated text generation engine to the user based on the emotion match. This makes it possible to select a text generation engine that is most suitable for the user's emotional state and provide the user with text generated by that engine.

[1449] "Means for accepting requests from a user" refers to an interface or device for accepting user input or requests.

[1450] "Multiple text generation engines" refers to multiple systems or models that generate text using different algorithms or techniques.

[1451] "Means for sending identical inputs and collecting the text generated from them" refers to a mechanism or process for sending user requests to multiple text generation engines and collecting the text generated by those engines.

[1452] "Means for having each text generation engine evaluate the collected text and aggregating the evaluation results" refers to a mechanism for having each text generation engine evaluate the quality and relevance of the collected text and integrating the evaluation results.

[1453] "Emotion engine that recognizes user emotions" refers to an engine or algorithm that analyzes and recognizes emotions from a user's text and behavioral history.

[1454] "Means for assessing how well generated text matches the user's emotions" refers to a process or device for assessing how well generated text matches the user's current emotional state.

[1455] "Means for recommending the most highly rated text generation engine to a user based on emotional compatibility" refers to a mechanism for selecting the most suitable text generation engine and recommending it to a user, taking into account the user's emotional compatibility.

[1456] The present invention relates to a system that provides content by selecting an optimal engine from among a plurality of text generation engines in consideration of a user's feelings. Specific embodiments for carrying out the present invention are described below.

[1457] System Overview

[1458] This system consists of a user terminal, a server, multiple text generation engines, and an emotion engine. The user sends a request, and the system generates text according to the request and recommends the generation engine that best suits the user's emotion.

[1459] Major hardware and software

[1460] 1. User Device

[1461] A device (smartphone, tablet, etc.) for accepting user input.

[1462] 2. Server

[1463] It hosts the text generation and sentiment engines, and processes and evaluates requests.

[1464] 3. Emotion Engine

[1465] An engine for information analysis and emotion recognition. It can use "AWS Rekognition" and "Google Cloud Natural Language."

[1466] 4. Text Generation Engine

[1467] Generate text using multiple generation engines (e.g., GPT-3, BERT).

[1468] Processing Details

[1469] The server first receives a request from the user. This request contains information about the content the user desires. The emotion engine then recognizes the user's emotion and generates metrics based on that emotion. This emotion data is obtained by analyzing past requests, interaction history, and other factors.

[1470] The server then sends the user's input text to multiple text generation engines and collects the generated text from each engine. The collected text is then sent to each engine again to have it evaluate the output of other generation engines. Furthermore, the emotion engine evaluates the emotional relevance of the generated text and integrates the evaluation score into the evaluation score of each text generation engine.

[1471] The server finally aggregates all the evaluation scores and calculates an overall evaluation score, taking into account the emotional compatibility. The text generation engine with the highest overall evaluation score is recommended to the user.

[1472] Specific examples

[1473] For example, if a user sends a request saying, "I want to see more happy news today," the emotion engine extracts the emotion data of "fun." Generation engine A generates "happy stories about cats," generation engine B generates "latest superhero movie reviews," and generation engine C generates "new restaurant information." Each generation engine evaluates the other's generation results, and the engine with the highest score, taking into account emotional compatibility, is determined and recommended to the user.

[1474] Prompt Sentence Examples

[1475] Example of a user request: "I want to see more exciting news today."

[1476] Example of an emotion rating: "fun"

[1477] Generated text by Generation Engine A: "A happy story about a cat"

[1478] Generated text by Generation Engine B: "Latest superhero movie reviews"

[1479] Generated text by generation engine C: "New restaurant information"

[1480] Through this system, it is possible to select a text generation engine that best suits the user's emotional state and provide the user with the text generated by that engine.

[1481] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1482] Step 1:

[1483] The server receives a request from the user terminal. This request contains information about the content the user desires. As input, the user sends the text "I want to see more entertaining news today." This text reaches the server and triggers the next processing step.

[1484] Step 2:

[1485] The server sends the received request text to the emotion engine. The emotion engine recognizes the user's emotion based on the user's text and past dialogue history. For example, the emotion engine recognizes emotions such as "happy," "sad," and "tired." Specifically, for the input text "I want to see more happy news today," the emotion engine extracts "happy." Emotion recognition data is generated and used in the next step.

[1486] Step 3:

[1487] The server sends the user's input text to multiple text generation engines. A generative AI model is used for this process. For example, generation engine A, generation engine B, and generation engine C receive the same input text, "I want to see more fun news today," and each generates new text using its own algorithm. As output, generation engine A sends "Happy stories about cats," generation engine B sends "Latest superhero movie reviews," and generation engine C sends "New restaurant information" back to the server.

[1488] Step 4:

[1489] The server then sends the collected generated texts to each text generation engine, which then evaluates the other's generated results. In this process, each generation engine evaluates the quality and relevance of the text received from the other engines. Specifically, generation engine A evaluates the texts from generation engines B and C, and generation engine B performs the same evaluation. At this stage, an evaluation score is generated.

[1490] Step 5:

[1491] The server evaluates the emotional relevance of the generated text using the emotion engines. For example, it evaluates whether "a happy story about a cat" is highly relevant to the emotion of "fun." This emotional relevance score is also integrated into the evaluation score of each text generation engine. This evaluation score becomes part of the final overall evaluation score.

[1492] Step 6:

[1493] The server aggregates all the evaluation scores and calculates an overall evaluation score. This overall evaluation score includes the evaluation scores from each engine and the emotional compatibility score. Specifically, let's say that generation engine A receives an overall evaluation score of 80 points, generation engine B receives an overall evaluation score of 90 points, and generation engine C receives an overall evaluation score of 85 points.

[1494] Step 7:

[1495] The server identifies the most highly rated text generation engine based on the overall evaluation score. Here, engine B has the highest score of 90, and is therefore recognized as the most suitable text generation engine.

[1496] Step 8:

[1497] The server recommends the most suitable text generation engine B to the user. The recommendation result is returned to the user's terminal and notified to the user. Specifically, the user receives an article titled "Latest Superhero Movie Reviews" and can view it.

[1498] In this way, it is possible to select the most suitable text generation engine while taking into consideration the emotional state of the user, and to provide appropriate content to the user.

[1499] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[1500] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1501] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[1502] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[1503] FIG. 9 illustrates an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and behaviors arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[1504] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[1505] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[1506] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[1507] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[1508] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[1509] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[1510] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[1511] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[1512] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[1513] It is not necessary to store all of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[1514] The hardware resource for executing a specific process can be any of the following types of processors: A processor, for example, is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[1515] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Also, the hardware resource that executes the specific processing may be a single processor.

[1516] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[1517] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[1518] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[1519] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[1520] The following is further disclosed regarding the above embodiment.

[1521] (Claim 1)

[1522] means for accepting requests from users;

[1523] means for sending identical input to a plurality of text generation engines and collecting the text generated therefrom;

[1524] a means for having each text generation engine evaluate the collected text and aggregating the evaluation results;

[1525] a means of recommending the highest-rated text generation engines to users; and

[1526] A system including:

[1527] (Claim 2)

[1528] 2. The system according to claim 1, further comprising means for calculating an overall evaluation score for the collected text based on the evaluation results from each text generation engine, and identifying the text generation engine that has obtained the highest evaluation score.

[1529] (Claim 3)

[1530] 2. The system according to claim 1, further comprising means for adding up evaluation scores of the outputs of other text generation engines by each text generation engine in the evaluation result aggregation means to perform a comprehensive evaluation.

[1531] "Example 1"

[1532] (Claim 1)

[1533] means for accepting requests from users;

[1534] means for sending identical input to a plurality of text generation engines and collecting the text generated therefrom;

[1535] a means for transmitting the collected text to each text generation engine to cause them to evaluate each other, and collecting the evaluation results;

[1536] a means for aggregating the evaluation results;

[1537] a means of recommending the highest-rated text generation engines to users; and

[1538] A system including:

[1539] (Claim 2)

[1540] 2. The system according to claim 1, further comprising means for calculating an overall evaluation score for the collected text based on mutual evaluation results from each text generation engine, and identifying the text generation engine that has obtained the highest overall evaluation score.

[1541] (Claim 3)

[1542] 2. The system according to claim 1, further comprising means for adding up the evaluation scores of each text generation engine with respect to other text generation engines in the evaluation result aggregation means to perform a comprehensive evaluation.

[1543] "Application Example 1"

[1544] (Claim 1)

[1545] means for accepting requests from users;

[1546] means for sending identical input to a plurality of natural language generation engines and collecting the text generated therefrom;

[1547] A means for having each natural language generation engine evaluate the collected text and aggregating the evaluation results;

[1548] A means of recommending the highest-rated natural language generation engines to users; and

[1549] A means for generating product descriptions based on the collected text using an optimal natural language generation engine;

[1550] A system including:

[1551] (Claim 2)

[1552] The system according to claim 1, further comprising means for calculating an overall evaluation score for the collected text based on the evaluation results from each natural language generation engine, and identifying the natural language generation engine that has obtained the highest evaluation score.

[1553] (Claim 3)

[1554] 2. The system according to claim 1, further comprising means for adding up evaluation scores of the outputs of other natural language generation engines by each natural language generation engine in the evaluation result aggregation means to perform an overall evaluation.

[1555] "Example 2: Combining Emotion Engines"

[1556] (Claim 1)

[1557] means for accepting requests from users;

[1558] means for sending identical inputs to multiple generative models and collecting the data generated therefrom;

[1559] A means for evaluating the collected data using each generative model and aggregating the evaluation results;

[1560] A means of recommending the most highly rated generative model to the user;

[1561] means for recognizing a user's emotion;

[1562] A means for correcting the evaluation result based on the emotion;

[1563] A system including:

[1564] (Claim 2)

[1565] The system of claim 1, further comprising means for calculating an overall evaluation score for the collected data based on the evaluation results from each generative model, and identifying the generative model that has obtained the highest evaluation score.

[1566] (Claim 3)

[1567] 2. The system according to claim 1, further comprising means for adding up the evaluation scores of the outputs of other generative models by each generative model in the evaluation result aggregation means to perform an overall evaluation.

[1568] "Application example 2 when combining emotion engines"

[1569] (Claim 1)

[1570] means for accepting requests from users;

[1571] means for sending identical input to a plurality of text generation engines and collecting the text generated therefrom;

[1572] a means for having each text generation engine evaluate the collected text and aggregating the evaluation results;

[1573] an emotion engine that recognizes the user's emotions;

[1574] a means for evaluating how well the generated text matches the user's sentiment;

[1575] A means for recommending the most highly rated text generation engine to a user based on emotional compatibility; and

[1576] A system including:

[1577] (Claim 2)

[1578] 2. The system of claim 1, further comprising: means for calculating an overall evaluation score for the collected text based on the evaluation results from each text generation engine, integrating the user's emotional compatibility score, and identifying the text generation engine that has obtained the highest evaluation score.

[1579] (Claim 3)

[1580] 2. The system according to claim 1, further comprising means for adding up evaluation scores of the outputs of other text generation engines by each text generation engine, and performing a comprehensive evaluation by taking into account the evaluation of the emotion engine, in the evaluation result aggregation means. [Explanation of symbols]

[1581] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. means for accepting requests from users; means for sending identical input to a plurality of text generation engines and collecting the text generated therefrom; a means for having each text generation engine evaluate the collected text and aggregating the evaluation results; a means of recommending the highest-rated text generation engines to users; and A system including:

2. The system according to claim 1 , further comprising means for calculating a comprehensive evaluation score for the collected text based on the evaluation results from each text generation engine, and for identifying the text generation engine that has obtained the highest evaluation score.

3. 2. The system according to claim 1, further comprising means for adding up the evaluation scores of the outputs of the other text generation engines by each text generation engine in the evaluation result aggregation means to perform a comprehensive evaluation.

Images