System

Abstract

A system is provided.SOLUTION: A system, comprising: a generative artificial intelligence model; means for combining time series data and corresponding natural language data and using the generative artificial intelligence model to make predictions; and means for using rigorous gradient descent to improve prediction accuracy.SELECTED DRAWING: Figure 1

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] Conventional stock price and exchange rate prediction systems mainly make predictions based solely on numerical sequences and do not consider external factors or natural language data, resulting in low prediction accuracy. Because stock price and exchange rate fluctuations are often heavily dependent on news and related document information, predictive models that do not consider this information have difficulty achieving practical accuracy. The present invention aims to improve prediction accuracy through a rigorous gradient descent method using a generative artificial intelligence model that incorporates natural language data. [Means for solving the problem]

[0005] The present invention provides a system that uses a generative artificial intelligence model to create a dataset that combines time series data with corresponding news and related document information, and then uses this dataset to make predictions. Specifically, the system combines a Transformer model and a time series prediction model as the generative artificial intelligence model, and includes means for inputting time series data and natural language data to the model and predicting the next time series data. Furthermore, the system improves prediction accuracy by retraining using a gradient descent method to improve the model based on the difference between the prediction results and the actual data. This system is capable of achieving more accurate predictions that take external factors into account.

[0006] A "generative artificial intelligence model" is a type of artificial intelligence technology that learns patterns from large amounts of data and is used to generate new data and make predictions.

[0007] "Time series data" refers to data recorded at regular time intervals that shows fluctuations in stock prices, exchange rates, etc.

[0008] "Natural language data" is data based on the language used by humans on a daily basis, and includes news articles and analytical reports.

[0009] A "prompt" is a piece of data input to a generative artificial intelligence model that instructs the model to begin a specific generation or prediction task.

[0010] "Gradient descent" is an optimization algorithm used to train machine learning models, and is a procedure for adjusting parameters to minimize the model's error.

[0011] A "Transformer model" is a type of deep learning model that exhibits high performance in natural language processing and other generative tasks, and can efficiently capture dependencies between different parts of the input data using a self-attention mechanism.

[0012] "LSTM" stands for Long Short-Term Memory and is a recurrent neural network (RNN) suitable for processing time series data with long-term dependencies.

[0013] An "integrated dataset" is a dataset that combines time series data and associated natural language data and is used to train predictive models.

[0014] "Retraining" is the process of retraining a generative artificial intelligence model to improve its performance based on newly acquired data and error information.

[0015] "Prediction accuracy" is an index that indicates how accurately a prediction model can make predictions on actual data. [Brief explanation of the drawings]

[0016] [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

[0017] 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.

[0018] First, the terms used in the following description will be explained.

[0019] 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).

[0020] 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.

[0021] 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.

[0022] 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.

[0023] 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."

[0024] [First embodiment]

[0025] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0026] 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.

[0027] 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).

[0028] 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.

[0029] 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.

[0030] 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.

[0031] 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.

[0032] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0033] 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.

[0034] 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.

[0035] 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.

[0036] 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."

[0037] The present invention relates to a system for predicting stock prices and exchange rates using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[0038] System program and processing description

[0039] server

[0040] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The server then formats the data and creates an integrated dataset that combines the time-series data with corresponding news and related document information. This integrated dataset is used as input data for making predictions using a transformer model (e.g., a model for automatic learning) and a time-series prediction model (e.g., LSTM or RNN).

[0041] Next, the server builds a generative artificial intelligence model. Specifically, it loads a "Transformer model" and adds an "LSTM" layer to it to define it as an integrated neural network. It then trains this model using the integrated dataset. For training, it prepares batches of data using a data loader, inputs this data into the model, and generates predictions. It calculates an error based on the difference between the prediction results and the actual data, backpropagates this error to calculate the gradient, and updates the model parameters. This procedure is repeated over multiple epochs (learning cycles).

[0042] The server evaluates the model during the training process. For evaluation, it uses validation data and test data and calculates the difference between the prediction and the actual data as an indicator (e.g., MSE). If the model meets certain performance criteria, the server saves it and provides it to external parties as an API.

[0043] Terminal

[0044] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to forecast. This data is then sent to the server, which makes a forecast request.

[0045] The server uses the received data to input data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period. The prediction results are sent back to the terminal, which displays them to the user, who can use them as a reference when making trading and investment decisions.

[0046] Specific examples

[0047] User

[0048] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the terminal. Next, the user requests a "USD / JPY forecast for the next week." The terminal then sends this data to the server and makes a prediction request.

[0049] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. The forecast results are then sent back to the terminal. The terminal then displays the received forecast results to the user, who can then begin trading based on that information.

[0050] The above is a detailed description of an embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[0051] The processing flow will be explained below.

[0052] Step 1:

[0053] The server collects time-series data on stock prices and exchange rates, and retrieves past data from APIs and databases. For example, it retrieves and stores the past year's USD / JPY exchange rate data.

[0054] Step 2:

[0055] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0056] Step 3:

[0057] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[0058] Step 4:

[0059] The server loads the Transformer model and adds and integrates an LSTM layer to build a generative artificial intelligence model.

[0060] Step 5:

[0061] The server splits the combined dataset into three parts: training, validation, and testing. For example, 80% is for training, 10% is for validation, and 10% is for testing.

[0062] Step 6:

[0063] The server prepares the batch data using a data loader and is configured to load the training data in fixed batch sizes.

[0064] Step 7:

[0065] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated for multiple epochs.

[0066] Step 8:

[0067] The server evaluates the performance of the trained model using the validation data, calculating the difference between the predictions and the actual data as a metric (e.g., MSE) to confirm the model's performance.

[0068] Step 9:

[0069] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[0070] Step 10:

[0071] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[0072] Step 11:

[0073] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[0074] Step 12:

[0075] The terminal transmits the data entered by the user to the server and makes a prediction request.

[0076] Step 13:

[0077] The server inputs the received data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period.

[0078] Step 14:

[0079] The server returns the prediction result to the terminal.

[0080] Step 15:

[0081] The terminal displays the prediction results received from the server to the user, who can then make investment and trading decisions based on the prediction results.

[0082] Example 1

[0083] 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."

[0084] Conventional financial market forecasting systems have relied on a single data source, resulting in low forecast accuracy. Furthermore, the lack of a method for effectively linking natural language data and time series data limits the accuracy of forecasts. Furthermore, the training and evaluation methods for generative AI models have not been fully optimized, preventing the full potential of forecasting performance.

[0085] 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.

[0086] In this invention, the server includes a means for collecting time-series data of stock prices and exchange rates and natural language data, a means for creating an integrated data set by associating the collected time-series data with the natural language data, and a means for constructing a generative artificial intelligence model that combines a Transformer model and a time-series prediction model, thereby enabling highly accurate predictions.

[0087] "Time series data of stock prices and exchange rates" refers to numerical information on stock prices and exchange rates recorded at specific time intervals (e.g., daily, hourly, etc.).

[0088] "Natural language data" is text data written in the language that people use on a daily basis, such as news articles, analytical reports, and social media posts.

[0089] An "integrated dataset" is a single dataset that links time series data and natural language data based on date and time and relevance, and is used to train artificial intelligence models.

[0090] A "generative artificial intelligence model" is an artificial intelligence model designed to make predictions or generate data from input data, and specifically includes transformer models and time series prediction models.

[0091] A "Transformer model" is a type of machine learning for data, and is a neural network model used in particular for natural language processing.

[0092] A "time series prediction model" is an algorithm or model that uses time series data as input to predict future values, and specifically includes LSTM and RNN.

[0093] "Calculating gradients by backpropagating errors" is the process of calculating the gradients of the parameters of an artificial intelligence model using the difference (error) between the predicted value and the actual value, and updating the model parameters.

[0094] "Evaluating a model" is the process of measuring the performance of a trained generative artificial intelligence model using validation data and evaluating its predictive accuracy.

[0095] "Providing the prediction results to the user" means displaying the prediction values ​​generated by the generative artificial intelligence model to the user so that the user can make decisions based on that information.

[0096] The present invention relates to a system for predicting stock prices and exchange rates, and is a system that achieves highly accurate predictions using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[0097] server

[0098] The server operates as follows: First, it uses the APIs and databases of financial data providers to collect time-series data on stock prices and exchange rates. It also uses news APIs to obtain natural language data such as related news articles and analytical reports. Specific data collection uses financial data APIs (e.g., Yahoo! Finance API, Alpha Vantage API) and news APIs.

[0099] The server then processes the collected time series data and natural language data, aligning them by date and time, to create an integrated dataset that links the time series data with the corresponding news articles. This integrated dataset is then used as input data for making predictions using transformer models (e.g., BERT models) and time series prediction models (e.g., LSTM, RNN).

[0100] The server then builds a hybrid neural network that combines a Transformer model with an LSTM layer and trains the model using the integrated dataset. In this invention, we use the Hugging Face Transformers library as the Transformer model. The training process involves using PyTorch to prepare the data loader and generate predictions. The error is calculated from the difference between the prediction result and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. This procedure is repeated multiple times.

[0101] During the training process, the server also evaluates the model. For evaluation, it uses a validation dataset and calculates the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error). If the model meets certain performance criteria, it saves it and provides it to external parties as an API.

[0102] Terminal

[0103] The terminal provides an interface that accepts input from the user. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. This data is sent to the server, which issues a prediction request. The server then uses this data to make a prediction using a generative artificial intelligence model and sends the prediction results back to the terminal. The terminal then displays the prediction results, which the user can use as a reference when making trading and investment decisions.

[0104] User

[0105] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into a terminal. Then, the user requests a "prediction of the USD / JPY for the next week." The terminal sends this data to the server and makes a prediction request. The server inputs the received data into a generative AI model, which predicts the USD / JPY exchange rate for the next week. The prediction results are sent back to the terminal, which displays them to the user. The user then makes trading decisions based on this information. For example, the following prompt sentences can be input into the generative AI model:

[0106] What is your forecast for the dollar-yen exchange rate for the next week?

[0107] Data and news articles from the past week are below.

[0108] Exchange data: [List of data]

[0109] News articles: [list of articles]

[0110] The above is a detailed description of an embodiment of the present invention. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[0111] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0112] Step 1: Data collection

[0113] server

[0114] The server collects stock and exchange time series data from financial data providers' APIs and databases, and also uses news APIs to retrieve natural language data such as related news articles and analytical reports.

[0115] input

[0116] Financial data API, news API.

[0117] output

[0118] Time series data and news article data.

[0119] Specific actions

[0120] The server calls a financial data API to retrieve stock prices and exchange rates for the past few months, and similarly calls a news API to retrieve relevant news articles.

[0121] Step 2: Data Refinement and Integration

[0122] server

[0123] The server then formats the collected time series data and natural language data by aligning the data by date and time, and creating an integrated dataset that links a series of time series data with the corresponding news articles.

[0124] input

[0125] Time series data, news article data.

[0126] output

[0127] Integrated dataset.

[0128] Specific actions

[0129] The server uses the Pandas library to convert the time series data and news article data into data frames, then merges them based on date and time to create a unified dataset.

[0130] Step 3: Model Building

[0131] server

[0132] The server builds a hybrid neural network that combines a Transformer model and an LSTM layer, using a common natural language processing library and a time series forecasting model.

[0133] input

[0134] Natural language processing library, time series forecasting library.

[0135] output

[0136] Hybrid neural network model.

[0137] Specific actions

[0138] The server loads a Transformer model and adds an LSTM layer to it to define a hybrid neural network.

[0139] Step 4: Model training

[0140] server

[0141] The server uses the combined dataset to train the hybrid neural network model, using a data loader to prepare batches of data, which are then fed into the model to generate predictions.

[0142] input

[0143] Integrated datasets, hybrid neural network models.

[0144] output

[0145] The trained neural network model.

[0146] Specific actions

[0147] The server uses PyTorch to prepare a data loader and input batches of data into the model. The error is calculated from the difference between the predicted results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters.

[0148] Step 5: Model evaluation

[0149] server

[0150] The server evaluates the model during the training process, using a validation dataset to calculate the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error).

[0151] input

[0152] Validation dataset, trained neural network model.

[0153] output

[0154] Evaluation results (MSE).

[0155] Specific actions

[0156] The server uses the validation dataset to calculate the difference between the prediction and the actual value, and calculates the MSE to evaluate the model's predictive accuracy.

[0157] Step 6: Processing a prediction request

[0158] Terminal

[0159] The terminal receives input from the user about past exchange data and news articles, and sends this to the server. The server then uses this information to make predictions for the next period and sends the results back to the terminal.

[0160] input

[0161] User input data: historical exchange data, news articles.

[0162] output

[0163] Prediction results.

[0164] Specific actions

[0165] The terminal receives data from the user and sends it to the server via an HTTP request. The server inputs the received data into the model, performs exchange rate predictions for the next period, and sends the results back to the terminal.

[0166] Step 7: View the prediction results

[0167] Terminal

[0168] The terminal displays the prediction results returned from the server to the user, who then makes trading and investment decisions based on these prediction results.

[0169] input

[0170] Prediction results from the server.

[0171] output

[0172] The prediction results displayed on the screen.

[0173] Specific actions

[0174] The terminal displays the prediction results received from the server on the screen so that the user can easily check them. The user can then start trading based on this information.

[0175] (Application example 1)

[0176] 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."

[0177] Conventional systems do not provide sufficient information for investment decisions, particularly in terms of real-time forecasts and related information in financial markets. Furthermore, they are unable to provide personalized information tailored to individual user needs, creating a need for a means to efficiently provide information useful for users' investment decisions.

[0178] 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.

[0179] In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data and making predictions using the generative artificial intelligence model, a means for improving prediction accuracy using rigorous gradient descent, a means for providing information related to financial investments, and a means for providing a personalized notification function based on user interests, thereby enabling users to receive highly accurate forecast information in real time and efficiently obtain related information useful for their investment decisions.

[0180] A "generative artificial intelligence model" is an artificial intelligence model that combines time series data and natural language data to make predictions using a neural network.

[0181] "Time series data" refers to a continuous record of data collected at regular intervals, such as stock prices or currency exchange data.

[0182] "Natural language data" refers to text data written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[0183] "Exact gradient descent" is a computational technique that iteratively updates the model parameters to minimize error during the neural network training process.

[0184] "Information related to financial investment" refers to information to support investment decisions, including stock price forecasts, related news articles, and analytical reports.

[0185] "Personalized notification function" is a function that automatically delivers specific information based on the user's individual interests and concerns.

[0186] A "Transformer model" is a neural network model widely used in natural language processing, and is a model for understanding and generating language data.

[0187] A "time series prediction model" is a neural network model that uses time series data as input to predict future data points.

[0188] A "content distribution service" is a service that provides related information to users in real time.

[0189] This invention relates to a system that uses generative artificial intelligence models to predict stock prices and exchange rates, and provides information related to financial investments in real time. The system includes two main components, a server and a terminal, and provides an interface that is easy for users to use.

[0190] server

[0191] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The collected data is then formatted through preprocessing to create an integrated dataset that combines the time-series data with corresponding news and related document information.

[0192] Next, the server builds a generative artificial intelligence model. Specifically, it loads a Transformer model and adds an LSTM layer to it to define it as an integrated neural network. This model is trained using the integrated dataset. A data loader is used to prepare batch data, which is then input into the model to generate predictions. An error is calculated based on the difference between the generated prediction results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. The model is evaluated using validation and test data, and if it meets certain performance criteria, the server saves the model and provides it externally as an API.

[0193] Terminal

[0194] The device provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict into the device. This data is sent to the server, and a prediction request is made. The server uses the received data to input data into a generative artificial intelligence model, which predicts exchange rates and stock prices for the next period. The prediction results are sent back to the device, which displays them to the user. A personalized notification function based on the user's interests is also provided.

[0195] User

[0196] For example, a user inputs the exchange rate data for a specific currency from the past week and related news articles into their device. Next, the user requests a "currency forecast for the next week." The device sends this data to the server and makes a prediction request. The server inputs the received data into a generative artificial intelligence model and calculates a forecast for the next week's exchange rate. The prediction results are then sent back to the device, which displays them to the user. Personalized notifications are also provided based on the user's interests.

[0197] Usage examples and prompt statements

[0198] As a concrete example, if a user wants to get a forecast for the next week's exchange rate for a specific currency (e.g., the yen), the prompt text is as follows:

[0199] curl -X GET "https: / / api.example.com / stocks?ticker=USDJPY"

[0200] curl -X GET "https: / / api.example.com / news?ticker=USDJPY"

[0201] This prompt shows an API request to gather exchange and news data for a specific currency. This data can be used by the application to make predictions and provide the user with a currency forecast for the coming week.

[0202] The above is a specific embodiment for carrying out the present invention of this system.

[0203] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0204] Step 1:

[0205] The terminal receives input from the user, including past exchange data and corresponding news articles. This allows the terminal to obtain data for the period for which the user wishes to make a prediction. The input data is then sent to the server.

[0206] Step 2:

[0207] The server performs data processing based on the received exchange data and news articles. This process combines time series data and natural language data to create an integrated dataset. The processed data is then created.

[0208] Step 3:

[0209] The server uses the formatted data to input data into a generative artificial intelligence model. Specifically, predictions are made using a Transformer model and an LSTM layer. Data is input into the model, and prediction results are output.

[0210] Step 4:

[0211] The server calculates the difference between the predicted results and the actual data, and then retrains the generative AI model using rigorous gradient descent. This process improves the accuracy of predictions, and a retrained model is created.

[0212] Step 5:

[0213] The server then uses the trained model to make another prediction and returns the prediction results to the device, where they are displayed.

[0214] Step 6:

[0215] The device displays the received prediction results to the user, who can then make investment decisions based on this information.The device also provides personalized notifications based on the user's interests.

[0216] Step 7:

[0217] The user decides on the next investment action based on the displayed prediction results and notified information, and the user's investment action is executed.

[0218] 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.

[0219] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. Detailed embodiments for implementing this system are described below.

[0220] System program and processing description

[0221] server

[0222] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. For example, it retrieves and stores data on the dollar-yen exchange rate for the past year. It also collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0223] The server then transforms the time series data and natural language data, matches them with corresponding news articles and analytical reports to create an integrated dataset, and loads a Transformer model, adding an LSTM layer to integrate them to build a generative artificial intelligence model.

[0224] The server uses this combined dataset to train the model, preparing batches of data using a data loader. The model generates predictions for each batch of data, calculates the error based on the difference between the predictions and the actual data, and updates the model parameters using gradient descent. This procedure is repeated over multiple epochs (learning cycles).

[0225] During the training process, the server evaluates the model's performance using validation data and finally using test data. Models that achieve satisfactory performance are saved and made public as an API.

[0226] The server also incorporates an emotion engine that recognizes the user's emotions by analyzing the user's facial expressions, voice tone, and text messages, and evaluates the user's emotions in real time.

[0227] Terminal

[0228] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict. In addition, the emotion engine obtains the user's current emotion data. This data is sent to the server, and a prediction request is made.

[0229] The server uses the received data to input it into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the prediction based on the user's emotional data recognized by the emotion engine, further improving the accuracy of the prediction.

[0230] The server sends the prediction results back to the device, which then displays them to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with peace of mind.

[0231] Specific examples

[0232] User

[0233] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine then recognizes the user's current emotion from their tone of voice and facial expression. Next, the user requests a "USD / JPY forecast for the next week." The device then sends this data to the server and makes a prediction request.

[0234] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The server then sends the forecast results back to the terminal. The terminal then displays the received forecast results to the user, who then begins trading based on that information.

[0235] Furthermore, the device provides appropriate feedback based on the user's emotional data, making it easier for the user to understand the prediction results and use the device with confidence.

[0236] The above is a detailed embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide more reliable information to users. Users can use these prediction results to make more appropriate trading and investment decisions.

[0237] The processing flow will be explained below.

[0238] Step 1:

[0239] The server collects time series data on stock prices and exchange rates, and retrieves and stores the past year's USD / JPY exchange rate data via API and database.

[0240] Step 2:

[0241] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0242] Step 3:

[0243] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[0244] Step 4:

[0245] The server loads the Transformer model and adds an LSTM layer to integrate it to build a generative artificial intelligence model. Specifically, the output of the Transformer model is input to the LSTM, and the output is used as a prediction value to define a generative artificial intelligence model.

[0246] Step 5:

[0247] The server splits the combined dataset into three parts: training, validation, and testing: 80% of the dataset is stored for training, 10% for validation, and 10% for testing.

[0248] Step 6:

[0249] The server prepares the batch data using a data loader. It uses PyTorch's DataLoader class and configures it to load training data in fixed batch sizes.

[0250] Step 7:

[0251] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated over multiple epochs (learning cycles).

[0252] Step 8:

[0253] The server evaluates the performance of the trained model using the validation data. It calculates the difference between the predictions and the actual data as an evaluation metric (e.g., MSE) to confirm the model's performance.

[0254] Step 9:

[0255] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[0256] Step 10:

[0257] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[0258] Step 11:

[0259] The server incorporates an emotion engine, which analyzes the user's tone of voice, facial expressions, and input text messages to evaluate the user's emotions in real time.

[0260] Step 12:

[0261] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[0262] Step 13:

[0263] The terminal receives data input by the user and transmits it to the server. In addition, the terminal acquires the user's emotion data in real time and transmits it to the server.

[0264] Step 14:

[0265] The server inputs the received time series data and natural language data into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the predictions based on the user's emotional data recognized by the emotion engine.

[0266] Step 15:

[0267] The server returns the prediction result and adjusted reliability information to the terminal.

[0268] Step 16:

[0269] The device displays the prediction results and reliability information received from the server to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with confidence.

[0270] These are the specific processing steps of the system that combines the emotion engine. Users can use the prediction results to make appropriate trading and investment decisions that take emotion into account.

[0271] Example 2

[0272] 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."

[0273] Current stock price and exchange rate prediction systems use time series data and limited natural language data, which limits their prediction accuracy. Furthermore, because they do not take into account the user's emotional state, users may lack confidence in or understand the predictions. This makes it difficult for users to make appropriate investment and trading decisions based on the prediction results.

[0274] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data, a means for making a prediction using the generative artificial intelligence model, a means for improving prediction accuracy using strict gradient descent, a means for collecting user emotion data and analyzing it using an emotion engine, a means for dynamically adjusting the reliability of the prediction based on the analyzed emotion data, and a means for displaying the prediction result to the user. This enables highly accurate prediction using the generative artificial intelligence model, and by taking the user emotion data into consideration, the reliability of the prediction is further increased, allowing the user to use the prediction result with peace of mind.

[0275] A "generative artificial intelligence model" refers to an advanced machine learning algorithm that takes time series data or natural language data as input and performs predictive or generative tasks.

[0276] "Time series data" is a continuous set of data that changes over time, and includes numerical information collected at specific time intervals, such as stock prices or foreign exchange prices.

[0277] "Natural language data" refers to text information written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[0278] "Gradient descent" refers to an algorithm that calculates the gradient of the error function and gradually updates the parameters based on that in order to optimize the parameters of a machine learning model.

[0279] An "emotion engine" is a technology for analyzing a user's emotional state, and takes input data such as facial expressions, voice tone, and text messages.

[0280] "Dynamic adjustment" means that the system automatically changes its settings and behavior in response to changing data and conditions in real time.

[0281] "Prediction reliability" is an indicator that shows the accuracy and credibility of the prediction results output by a generative artificial intelligence model.

[0282] A "Transformer model" is a type of neural network that can perform natural language processing tasks using large datasets, particularly those that utilize attention mechanisms.

[0283] An "LSTM layer" is a recurrent neural network layer with long short-term memory units, which can memorize long-term dependencies.

[0284] A "data loader" is a program or library for efficiently processing batches of data during model training.

[0285] An "epoch" is a unit of machine learning training that refers to passing the entire dataset through a model once.

[0286] This invention combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. An embodiment of this system will be described below.

[0287] server

[0288] The server runs on high-performance hardware and programming languages ​​such as Python. The server first collects time-series data on stock prices and exchange rates through an API. For example, it uses the Yahoo Finance API to obtain data on the dollar-yen exchange rate for the past year and stores it in a database. It also collects natural language data such as news articles and analytical reports using news feeds and RSS readers. This data is stored as text files.

[0289] Next, the server preprocesses the time series data and natural language data to create an integrated dataset. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data is also cleaned by removing unnecessary characters and tags, and then tokenized and vectorized. This creates an integrated dataset that combines time series data and natural language data.

[0290] Using the combined dataset, the server builds and trains a generative artificial intelligence model. It uses a model that combines a transformer model and an LSTM layer, and uses the TensorFlow library for training. It prepares batches of data using a data loader, and separates the data into training, validation, and test data. It calculates the gradient of the error function and updates the model parameters using gradient descent.

[0291] Furthermore, the server collects user emotional data and analyzes it using an emotion engine. It uses a camera and microphone to capture the user's facial expressions and voice tone, and analyzes emotions in real time. Based on this emotional data, the reliability of predictions is dynamically adjusted.

[0292] Terminal

[0293] The terminal provides a user interface and collects prediction requests and emotion data. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. The emotion engine analyzes the user's facial expressions and tone of voice to collect emotion data in real time. This data is then sent to the server, which then makes a prediction request to the server.

[0294] User

[0295] The user inputs, for example, the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine recognizes the user's current emotion from their tone of voice and facial expression, and transmits it to the server. The user inputs the following:

[0296] Based on the past week's USD / JPY exchange rate data and related news articles, please predict next week's USD / JPY exchange rate.

[0297] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The forecast results are sent back to the device, which displays them to the user. The user can then make investment and trading decisions based on the forecast results.

[0298] The above is an embodiment of the present invention. This system combines a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide users with reliable information. Users can use these prediction results to make more appropriate trading and investment decisions.

[0299] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0300] Step 1: Data collection

[0301] The server first uses an API (general term) to collect time-series data on stock prices and exchange rates. Specifically, to obtain, for example, the dollar-yen exchange rate data for the past year, the appropriate API endpoint is called and the data is downloaded in CSV format. This data is then stored in the server's database. Furthermore, news articles for a specified period are collected using a news feed or RSS reader and saved as text files.

[0302] Input: API endpoint, RSS feed URL

[0303] Output: Time series data (CSV), natural language data (text file)

[0304] Step 2: Data preprocessing and integration

[0305] The server formats the collected time series data and natural language data. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data undergoes cleaning (removal of unnecessary tags and characters), tokenization, and vectorization. The time series data and natural language data are then matched based on date to create an integrated dataset.

[0306] Input: time series data, natural language data

[0307] Output: Unified dataset

[0308] Step 3: Building the model

[0309] The server uses the combined dataset to build a generative artificial intelligence model, which combines a transformer model with an LSTM layer, and uses the TensorFlow library in Python to define, compile, and initialize the model.

[0310] Input: Unified dataset

[0311] Output: An initialized generative artificial intelligence model

[0312] Step 4: Train the model

[0313] The server uses a data loader to split the training data into batches and train the model. The model generates predictions on the training data and calculates the error based on the difference between the predictions and the actual data. To minimize the error, the model's parameters are updated using gradient descent. This process is repeated for multiple epochs.

[0314] Input: Initialized generative artificial intelligence model, training data

[0315] Output: A trained generative artificial intelligence model

[0316] Step 5: Evaluate and save the model

[0317] The server evaluates the model's performance using the validation and test data. It calculates evaluation metrics (e.g., RMSE and MAE) and checks whether the model meets the specified performance criteria. If it does, it saves the model to storage and exposes it as an API.

[0318] Input: A trained generative AI model, validation data, and test data

[0319] Output: Saved generative AI model, exposed API

[0320] Step 6: Collect and analyze emotion data

[0321] The device uses a camera and microphone to collect the user's emotional data. This allows the device to capture the user's facial expressions and vocal tone in real time and transmit them to the server. The server then uses an emotion engine to analyze this data and evaluate the user's emotional state.

[0322] Input: Facial expression data, voice tone data

[0323] Output: Parsed emotion data

[0324] Step 7: Request and make predictions

[0325] The device collects past data and news articles for the period the user wants to predict, and sends them along with emotion data to the server. The server then inputs this data into a generative AI model to predict stock prices and exchange rates for the next period. The reliability of the prediction is dynamically adjusted based on the user's emotion data.

[0326] Input: Historical data, news articles, sentiment data

[0327] Output: Prediction results

[0328] Step 8: Displaying prediction results and providing feedback

[0329] The server sends the prediction results to the device, which then displays them to the user. Furthermore, feedback is generated based on the user's emotional data, providing additional information to deepen the user's understanding of the prediction results. This allows the user to use the prediction results with confidence.

[0330] Input: Prediction results, analyzed emotion data

[0331] Output: Prediction results and feedback displayed to the user

[0332] These are the specific steps of the system's processing. The data input and output at each step, as well as the processing details, are clearly defined, creating a system that improves the accuracy of predictions as a whole.

[0333] (Application example 2)

[0334] 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."

[0335] Conventional stock price and exchange rate prediction systems using generative artificial intelligence models make predictions without taking into account the user's emotional state, and therefore are unable to alleviate the user's stress and anxiety. In addition, because the prediction results are not adjusted to take the user's emotions into account, there are issues with the accuracy of predictions and the reliability of investment advice.

[0336] The identification processing by the identification 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 combining a generative artificial intelligence model with time-series data and corresponding natural language data and making predictions using the generative artificial intelligence model; means for improving prediction accuracy using rigorous gradient descent; means for combining with an emotion engine that recognizes the user's emotions and adjusting the prediction results; and means for providing investment advice and risk management based on the user's emotional state. This makes it possible to provide the user with highly accurate predictions that take the user's emotional state into consideration, and reliable investment advice and risk management.

[0337] A "generative artificial intelligence model" is an artificial intelligence technology that uses past data and learning algorithms to generate new data and perform predictions and analysis.

[0338] "Time series data" is a series of data points collected over time, such as stock prices or currency fluctuations.

[0339] "Natural language data" refers to information written in languages ​​that humans use on a daily basis, including news articles and analytical reports.

[0340] "Gradient descent" is an algorithm used to optimize the parameters of a predictive model, which involves repeatedly processing a set of data to minimize error.

[0341] An "emotion engine" is software that analyzes a user's emotional state and recognizes emotions from voice tone, facial expressions, text input, etc.

[0342] "Investment advice" refers to advice that supports users in making appropriate investment decisions, including suggestions based on predictive data and the user's emotional state.

[0343] "Risk management" is the process of minimizing the risks associated with investment activities and proposing optimal investment strategies for users.

[0344] A "Transformer model" is an advanced machine learning model used in natural language processing, which analyzes large amounts of text data to understand meaning and make predictions.

[0345] "Prediction results" are information that indicates future data points and trends calculated by a generative artificial intelligence model.

[0346] "Retraining" is an additional learning process that reviews the parameter settings of the initial model and improves the accuracy of the model based on actual data.

[0347] A "prompt" refers to specific input data or questions used when analyzing natural language data or making predictions.

[0348] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. A detailed configuration for configuring the system of the present invention will be described below.

[0349] System configuration

[0350] It consists of three main elements: the server, the smartphone device, and the user.

[0351] server

[0352] The server uses the following hardware and software:

[0353] Hardware: High-performance CPU, GPU, storage, network interface

[0354] Software: TensorFlow, Transformers (Hugging Face), API, Database Management System

[0355] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. The collected data is then formatted and combined with natural language data such as news articles and analytical reports to create an integrated dataset. This dataset is then used to train a generative artificial intelligence model that combines a transformer model and an LSTM layer to make predictions.

[0356] The server also uses an emotion engine to analyze the user's emotions in real time, which analyzes voice tone, facial expressions, input text messages, etc. to recognize the user's emotions.

[0357] Terminal

[0358] The device used is a smartphone operated by the user. The application on the smartphone provides the following functions:

[0359] Input interface for historical exchange data and corresponding news articles

[0360] An interface for recognizing emotions from speech tone and facial expressions

[0361] Displaying prediction results returned from the server

[0362] Through the application, users input past data and news articles for the period they want to predict. The emotion engine obtains the user's current emotional data and sends it to the server. The server then uses the generative AI model and emotion engine to make predictions and returns the results to the device.

[0363] User

[0364] Users operate their smartphones to input data, have the system recognize their emotions, and then check the prediction results to make trading decisions.

[0365] Specific examples

[0366] The user inputs the past week's USD / JPY exchange rate data and news articles into the terminal. The emotion engine recognizes the user's current emotions from their voice tone and facial expressions, and when they request a USD / JPY forecast for the next week, the terminal sends the data to the server. The server makes a prediction and returns the results to the terminal, adjusting the reliability of the prediction based on the user's emotions. The terminal then displays the received prediction results to the user, who can then begin trading based on that information. Appropriate feedback is also displayed.

[0367] An example prompt is:

[0368] "Enter USD / JPY exchange rate data and news articles from the past week. Then, please tell us your current emotional state by voice."

[0369] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0370] Step 1:

[0371] The server collects historical exchange data and news articles from APIs and databases. It sends requests to collect exchange data and news articles and retrieves the respective data. The input is raw data obtained from the API, and the output is formatted time series data and news article data.

[0372] Step 2:

[0373] The server formats the collected data, associates the time series data with the natural language data, and creates an integrated dataset. The formatted data is then combined and saved as a single dataset. The input is the formatted time series data and news article data, and the output is an integrated dataset.

[0374] Step 3:

[0375] The server combines a Transformer model and an LSTM layer to build a generative AI model, and then trains the model using this integrated dataset. The input is the integrated dataset, and the output is the trained generative AI model. The model processes the data sequentially and improves its prediction accuracy through learning.

[0376] Step 4:

[0377] The server uses an emotion engine to recognize emotions from the user's voice tone and facial expressions. It analyzes voice and image data for emotion recognition and identifies the user's emotional state. The input is the user's voice and image data, and the output is the recognized emotional data.

[0378] Step 5:

[0379] The terminal provides an interface for receiving past exchange data and news articles entered by the user. The user enters data for the desired forecast period, and also collects voice tone and facial expression data. The input is exchange data, news articles, voice and image data from the user, and the output is the collected data.

[0380] Step 6:

[0381] The server receives data sent from the device and inputs it into a generative artificial intelligence model to predict the next exchange rate and stock price. The data is input into the model and a prediction result is generated. The input is data from the device and the output is the prediction result.

[0382] Step 7:

[0383] The server dynamically adjusts the reliability of predictions using the user's emotional data recognized by the emotion engine. The prediction results are revised based on the emotional data to improve reliability. The input is the recognized emotional data, and the output is the adjusted prediction results.

[0384] Step 8:

[0385] The server returns the final prediction result to the terminal, which transmits the prediction result and provides it to the user. The input is the adjusted prediction result, and the output is the data for the user to display.

[0386] Step 9:

[0387] The terminal receives the prediction results from the server and displays them to the user. The received data is displayed on the screen for the user to confirm. The input is the prediction result data from the server, and the output is the displayed prediction result.

[0388] Step 10:

[0389] The user checks the prediction results and emotional feedback displayed on the device and makes a trading decision. Based on the prediction results and feedback, the user makes an appropriate trade. The input is the prediction results and feedback displayed on the device, and the output is the user's trading decision.

[0390] 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.

[0391] 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.

[0392] 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.

[0393] [Second embodiment]

[0394] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0395] 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.

[0396] 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).

[0397] 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.

[0398] 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.

[0399] 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).

[0400] 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.

[0401] 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.

[0402] 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.

[0403] 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.

[0404] 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.

[0405] 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."

[0406] The present invention relates to a system for predicting stock prices and exchange rates using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[0407] System program and processing description

[0408] server

[0409] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The server then formats the data and creates an integrated dataset that combines the time-series data with corresponding news and related document information. This integrated dataset is used as input data for making predictions using a transformer model (e.g., a model for automatic learning) and a time-series prediction model (e.g., LSTM or RNN).

[0410] Next, the server builds a generative artificial intelligence model. Specifically, it loads a "Transformer model" and adds an "LSTM" layer to it to define it as an integrated neural network. It then trains this model using the integrated dataset. For training, it prepares batches of data using a data loader, inputs this data into the model, and generates predictions. It calculates an error based on the difference between the prediction results and the actual data, backpropagates this error to calculate the gradient, and updates the model parameters. This procedure is repeated over multiple epochs (learning cycles).

[0411] The server evaluates the model during the training process. For evaluation, it uses validation data and test data and calculates the difference between the prediction and the actual data as an indicator (e.g., MSE). If the model meets certain performance criteria, the server saves it and provides it to external parties as an API.

[0412] Terminal

[0413] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to forecast. This data is then sent to the server, which makes a forecast request.

[0414] The server uses the received data to input data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period. The prediction results are sent back to the terminal, which displays them to the user, who can use them as a reference when making trading and investment decisions.

[0415] Specific examples

[0416] User

[0417] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the terminal. Next, the user requests a "USD / JPY forecast for the next week." The terminal then sends this data to the server and makes a prediction request.

[0418] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. The forecast results are then sent back to the terminal. The terminal then displays the received forecast results to the user, who can then begin trading based on that information.

[0419] The above is a detailed description of an embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[0420] The processing flow will be explained below.

[0421] Step 1:

[0422] The server collects time-series data on stock prices and exchange rates, and retrieves past data from APIs and databases. For example, it retrieves and stores the past year's USD / JPY exchange rate data.

[0423] Step 2:

[0424] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0425] Step 3:

[0426] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[0427] Step 4:

[0428] The server loads the Transformer model and adds and integrates an LSTM layer to build a generative artificial intelligence model.

[0429] Step 5:

[0430] The server splits the combined dataset into three parts: training, validation, and testing. For example, 80% is for training, 10% is for validation, and 10% is for testing.

[0431] Step 6:

[0432] The server prepares the batch data using a data loader and is configured to load the training data in fixed batch sizes.

[0433] Step 7:

[0434] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated for multiple epochs.

[0435] Step 8:

[0436] The server evaluates the performance of the trained model using the validation data, calculating the difference between the predictions and the actual data as a metric (e.g., MSE) to confirm the model's performance.

[0437] Step 9:

[0438] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[0439] Step 10:

[0440] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[0441] Step 11:

[0442] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[0443] Step 12:

[0444] The terminal transmits the data entered by the user to the server and makes a prediction request.

[0445] Step 13:

[0446] The server inputs the received data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period.

[0447] Step 14:

[0448] The server returns the prediction result to the terminal.

[0449] Step 15:

[0450] The terminal displays the prediction results received from the server to the user, who can then make investment and trading decisions based on the prediction results.

[0451] Example 1

[0452] 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."

[0453] Conventional financial market forecasting systems have relied on a single data source, resulting in low forecast accuracy. Furthermore, the lack of a method for effectively linking natural language data and time series data limits the accuracy of forecasts. Furthermore, the training and evaluation methods for generative AI models have not been fully optimized, preventing the full potential of forecasting performance.

[0454] 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.

[0455] In this invention, the server includes a means for collecting time-series data of stock prices and exchange rates and natural language data, a means for creating an integrated data set by associating the collected time-series data with the natural language data, and a means for constructing a generative artificial intelligence model that combines a Transformer model and a time-series prediction model, thereby enabling highly accurate predictions.

[0456] "Time series data of stock prices and exchange rates" refers to numerical information on stock prices and exchange rates recorded at specific time intervals (e.g., daily, hourly, etc.).

[0457] "Natural language data" is text data written in the language that people use on a daily basis, such as news articles, analytical reports, and social media posts.

[0458] An "integrated dataset" is a single dataset that links time series data and natural language data based on date and time and relevance, and is used to train artificial intelligence models.

[0459] A "generative artificial intelligence model" is an artificial intelligence model designed to make predictions or generate data from input data, and specifically includes transformer models and time series prediction models.

[0460] A "Transformer model" is a type of machine learning for data, and is a neural network model used in particular for natural language processing.

[0461] A "time series prediction model" is an algorithm or model that uses time series data as input to predict future values, and specifically includes LSTM and RNN.

[0462] "Calculating gradients by backpropagating errors" is the process of calculating the gradients of the parameters of an artificial intelligence model using the difference (error) between the predicted value and the actual value, and updating the model parameters.

[0463] "Evaluating a model" is the process of measuring the performance of a trained generative artificial intelligence model using validation data and evaluating its predictive accuracy.

[0464] "Providing the prediction results to the user" means displaying the prediction values ​​generated by the generative artificial intelligence model to the user so that the user can make decisions based on that information.

[0465] The present invention relates to a system for predicting stock prices and exchange rates, and is a system that achieves highly accurate predictions using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[0466] server

[0467] The server operates as follows: First, it uses the APIs and databases of financial data providers to collect time-series data on stock prices and exchange rates. It also uses news APIs to obtain natural language data such as related news articles and analytical reports. Specific data collection uses financial data APIs (e.g., Yahoo! Finance API, Alpha Vantage API) and news APIs.

[0468] The server then processes the collected time series data and natural language data, aligning them by date and time, to create an integrated dataset that links the time series data with the corresponding news articles. This integrated dataset is then used as input data for making predictions using transformer models (e.g., BERT models) and time series prediction models (e.g., LSTM, RNN).

[0469] The server then builds a hybrid neural network that combines a Transformer model with an LSTM layer and trains the model using the integrated dataset. In this invention, we use the Hugging Face Transformers library as the Transformer model. The training process involves using PyTorch to prepare the data loader and generate predictions. The error is calculated from the difference between the prediction result and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. This procedure is repeated multiple times.

[0470] During the training process, the server also evaluates the model. For evaluation, it uses a validation dataset and calculates the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error). If the model meets certain performance criteria, it saves it and provides it to external parties as an API.

[0471] Terminal

[0472] The terminal provides an interface that accepts input from the user. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. This data is sent to the server, which issues a prediction request. The server then uses this data to make a prediction using a generative artificial intelligence model and sends the prediction results back to the terminal. The terminal then displays the prediction results, which the user can use as a reference when making trading and investment decisions.

[0473] User

[0474] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into a terminal. Then, the user requests a "prediction of the USD / JPY for the next week." The terminal sends this data to the server and makes a prediction request. The server inputs the received data into a generative AI model, which predicts the USD / JPY exchange rate for the next week. The prediction results are sent back to the terminal, which displays them to the user. The user then makes trading decisions based on this information. For example, the following prompt sentences can be input into the generative AI model:

[0475] What is your forecast for the dollar-yen exchange rate for the next week?

[0476] Data and news articles from the past week are below.

[0477] Exchange data: [List of data]

[0478] News articles: [list of articles]

[0479] The above is a detailed description of an embodiment of the present invention. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[0480] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0481] Step 1: Data collection

[0482] server

[0483] The server collects stock and exchange time series data from financial data providers' APIs and databases, and also uses news APIs to retrieve natural language data such as related news articles and analytical reports.

[0484] input

[0485] Financial data API, news API.

[0486] output

[0487] Time series data and news article data.

[0488] Specific actions

[0489] The server calls a financial data API to retrieve stock prices and exchange rates for the past few months, and similarly calls a news API to retrieve relevant news articles.

[0490] Step 2: Data Refinement and Integration

[0491] server

[0492] The server then formats the collected time series data and natural language data by aligning the data by date and time, and creating an integrated dataset that links a series of time series data with the corresponding news articles.

[0493] input

[0494] Time series data, news article data.

[0495] output

[0496] Integrated dataset.

[0497] Specific actions

[0498] The server uses the Pandas library to convert the time series data and news article data into data frames, then merges them based on date and time to create a unified dataset.

[0499] Step 3: Model Building

[0500] server

[0501] The server builds a hybrid neural network that combines a Transformer model and an LSTM layer, using a common natural language processing library and a time series forecasting model.

[0502] input

[0503] Natural language processing library, time series forecasting library.

[0504] output

[0505] Hybrid neural network model.

[0506] Specific actions

[0507] The server loads a Transformer model and adds an LSTM layer to it to define a hybrid neural network.

[0508] Step 4: Model training

[0509] server

[0510] The server uses the combined dataset to train the hybrid neural network model, using a data loader to prepare batches of data, which are then fed into the model to generate predictions.

[0511] input

[0512] Integrated datasets, hybrid neural network models.

[0513] output

[0514] The trained neural network model.

[0515] Specific actions

[0516] The server uses PyTorch to prepare a data loader and input batches of data into the model. The error is calculated from the difference between the predicted results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters.

[0517] Step 5: Model evaluation

[0518] server

[0519] The server evaluates the model during the training process, using a validation dataset to calculate the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error).

[0520] input

[0521] Validation dataset, trained neural network model.

[0522] output

[0523] Evaluation results (MSE).

[0524] Specific actions

[0525] The server uses the validation dataset to calculate the difference between the prediction and the actual value, and calculates the MSE to evaluate the model's predictive accuracy.

[0526] Step 6: Processing a prediction request

[0527] Terminal

[0528] The terminal receives input from the user about past exchange data and news articles, and sends this to the server. The server then uses this information to make predictions for the next period and sends the results back to the terminal.

[0529] input

[0530] User input data: historical exchange data, news articles.

[0531] output

[0532] Prediction results.

[0533] Specific actions

[0534] The terminal receives data from the user and sends it to the server via an HTTP request. The server inputs the received data into the model, performs exchange rate predictions for the next period, and sends the results back to the terminal.

[0535] Step 7: View the prediction results

[0536] Terminal

[0537] The terminal displays the prediction results returned from the server to the user, who then makes trading and investment decisions based on these prediction results.

[0538] input

[0539] Prediction results from the server.

[0540] output

[0541] The prediction results displayed on the screen.

[0542] Specific actions

[0543] The terminal displays the prediction results received from the server on the screen so that the user can easily check them. The user can then start trading based on this information.

[0544] (Application example 1)

[0545] 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."

[0546] Conventional systems do not provide sufficient information for investment decisions, particularly in terms of real-time forecasts and related information in financial markets. Furthermore, they are unable to provide personalized information tailored to individual user needs, creating a need for a means to efficiently provide information useful for users' investment decisions.

[0547] 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.

[0548] In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data and making predictions using the generative artificial intelligence model, a means for improving prediction accuracy using rigorous gradient descent, a means for providing information related to financial investments, and a means for providing a personalized notification function based on user interests, thereby enabling users to receive highly accurate forecast information in real time and efficiently obtain related information useful for their investment decisions.

[0549] A "generative artificial intelligence model" is an artificial intelligence model that combines time series data and natural language data to make predictions using a neural network.

[0550] "Time series data" refers to a continuous record of data collected at regular intervals, such as stock prices or currency exchange data.

[0551] "Natural language data" refers to text data written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[0552] "Exact gradient descent" is a computational technique that iteratively updates the model parameters to minimize error during the neural network training process.

[0553] "Information related to financial investment" refers to information to support investment decisions, including stock price forecasts, related news articles, and analytical reports.

[0554] "Personalized notification function" is a function that automatically delivers specific information based on the user's individual interests and concerns.

[0555] A "Transformer model" is a neural network model widely used in natural language processing, and is a model for understanding and generating language data.

[0556] A "time series prediction model" is a neural network model that uses time series data as input to predict future data points.

[0557] A "content distribution service" is a service that provides related information to users in real time.

[0558] This invention relates to a system that uses generative artificial intelligence models to predict stock prices and exchange rates, and provides information related to financial investments in real time. The system includes two main components, a server and a terminal, and provides an interface that is easy for users to use.

[0559] server

[0560] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The collected data is then formatted through preprocessing to create an integrated dataset that combines the time-series data with corresponding news and related document information.

[0561] Next, the server builds a generative artificial intelligence model. Specifically, it loads a Transformer model and adds an LSTM layer to it to define it as an integrated neural network. This model is trained using the integrated dataset. A data loader is used to prepare batch data, which is then input into the model to generate predictions. An error is calculated based on the difference between the generated prediction results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. The model is evaluated using validation and test data, and if it meets certain performance criteria, the server saves the model and provides it externally as an API.

[0562] Terminal

[0563] The device provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict into the device. This data is sent to the server, and a prediction request is made. The server uses the received data to input data into a generative artificial intelligence model, which predicts exchange rates and stock prices for the next period. The prediction results are sent back to the device, which displays them to the user. A personalized notification function based on the user's interests is also provided.

[0564] User

[0565] For example, a user inputs the exchange rate data for a specific currency from the past week and related news articles into their device. Next, the user requests a "currency forecast for the next week." The device sends this data to the server and makes a prediction request. The server inputs the received data into a generative artificial intelligence model and calculates a forecast for the next week's exchange rate. The prediction results are then sent back to the device, which displays them to the user. Personalized notifications are also provided based on the user's interests.

[0566] Usage examples and prompt statements

[0567] As a concrete example, if a user wants to get a forecast for the next week's exchange rate for a specific currency (e.g., the yen), the prompt text is as follows:

[0568] curl -X GET "https: / / api.example.com / stocks?ticker=USDJPY"

[0569] curl -X GET "https: / / api.example.com / news?ticker=USDJPY"

[0570] This prompt shows an API request to gather exchange and news data for a specific currency. This data can be used by the application to make predictions and provide the user with a currency forecast for the coming week.

[0571] The above is a specific embodiment for carrying out the present invention of this system.

[0572] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0573] Step 1:

[0574] The terminal receives input from the user, including past exchange data and corresponding news articles. This allows the terminal to obtain data for the period for which the user wishes to make a prediction. The input data is then sent to the server.

[0575] Step 2:

[0576] The server performs data processing based on the received exchange data and news articles. This process combines time series data and natural language data to create an integrated dataset. The processed data is then created.

[0577] Step 3:

[0578] The server uses the formatted data to input data into a generative artificial intelligence model. Specifically, predictions are made using a Transformer model and an LSTM layer. Data is input into the model, and prediction results are output.

[0579] Step 4:

[0580] The server calculates the difference between the predicted results and the actual data, and then retrains the generative AI model using rigorous gradient descent. This process improves the accuracy of predictions, and a retrained model is created.

[0581] Step 5:

[0582] The server then uses the trained model to make another prediction and returns the prediction results to the device, where they are displayed.

[0583] Step 6:

[0584] The device displays the received prediction results to the user, who can then make investment decisions based on this information.The device also provides personalized notifications based on the user's interests.

[0585] Step 7:

[0586] The user decides on the next investment action based on the displayed prediction results and notified information, and the user's investment action is executed.

[0587] 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.

[0588] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. Detailed embodiments for implementing this system are described below.

[0589] System program and processing description

[0590] server

[0591] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. For example, it retrieves and stores data on the dollar-yen exchange rate for the past year. It also collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0592] The server then transforms the time series data and natural language data, matches them with corresponding news articles and analytical reports to create an integrated dataset, and loads a Transformer model, adding an LSTM layer to integrate them to build a generative artificial intelligence model.

[0593] The server uses this combined dataset to train the model, preparing batches of data using a data loader. The model generates predictions for each batch of data, calculates the error based on the difference between the predictions and the actual data, and updates the model parameters using gradient descent. This procedure is repeated over multiple epochs (learning cycles).

[0594] During the training process, the server evaluates the model's performance using validation data and finally using test data. Models that achieve satisfactory performance are saved and made public as an API.

[0595] The server also incorporates an emotion engine that recognizes the user's emotions by analyzing the user's facial expressions, voice tone, and text messages, and evaluates the user's emotions in real time.

[0596] Terminal

[0597] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict. In addition, the emotion engine obtains the user's current emotion data. This data is sent to the server, and a prediction request is made.

[0598] The server uses the received data to input it into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the prediction based on the user's emotional data recognized by the emotion engine, further improving the accuracy of the prediction.

[0599] The server sends the prediction results back to the device, which then displays them to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with peace of mind.

[0600] Specific examples

[0601] User

[0602] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine then recognizes the user's current emotion from their tone of voice and facial expression. Next, the user requests a "USD / JPY forecast for the next week." The device then sends this data to the server and makes a prediction request.

[0603] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The server then sends the forecast results back to the terminal. The terminal then displays the received forecast results to the user, who then begins trading based on that information.

[0604] Furthermore, the device provides appropriate feedback based on the user's emotional data, making it easier for the user to understand the prediction results and use the device with confidence.

[0605] The above is a detailed embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide more reliable information to users. Users can use these prediction results to make more appropriate trading and investment decisions.

[0606] The processing flow will be explained below.

[0607] Step 1:

[0608] The server collects time series data on stock prices and exchange rates, and retrieves and stores the past year's USD / JPY exchange rate data via API and database.

[0609] Step 2:

[0610] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0611] Step 3:

[0612] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[0613] Step 4:

[0614] The server loads the Transformer model and adds an LSTM layer to integrate it to build a generative artificial intelligence model. Specifically, the output of the Transformer model is input to the LSTM, and the output is used as a prediction value to define a generative artificial intelligence model.

[0615] Step 5:

[0616] The server splits the combined dataset into three parts: training, validation, and testing: 80% of the dataset is stored for training, 10% for validation, and 10% for testing.

[0617] Step 6:

[0618] The server prepares the batch data using a data loader. It uses PyTorch's DataLoader class and configures it to load training data in fixed batch sizes.

[0619] Step 7:

[0620] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated over multiple epochs (learning cycles).

[0621] Step 8:

[0622] The server evaluates the performance of the trained model using the validation data. It calculates the difference between the predictions and the actual data as an evaluation metric (e.g., MSE) to confirm the model's performance.

[0623] Step 9:

[0624] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[0625] Step 10:

[0626] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[0627] Step 11:

[0628] The server incorporates an emotion engine, which analyzes the user's tone of voice, facial expressions, and input text messages to evaluate the user's emotions in real time.

[0629] Step 12:

[0630] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[0631] Step 13:

[0632] The terminal receives data input by the user and transmits it to the server. In addition, the terminal acquires the user's emotion data in real time and transmits it to the server.

[0633] Step 14:

[0634] The server inputs the received time series data and natural language data into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the predictions based on the user's emotional data recognized by the emotion engine.

[0635] Step 15:

[0636] The server returns the prediction result and adjusted reliability information to the terminal.

[0637] Step 16:

[0638] The device displays the prediction results and reliability information received from the server to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with confidence.

[0639] These are the specific processing steps of the system that combines the emotion engine. Users can use the prediction results to make appropriate trading and investment decisions that take emotion into account.

[0640] Example 2

[0641] 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."

[0642] Current stock price and exchange rate prediction systems use time series data and limited natural language data, which limits their prediction accuracy. Furthermore, because they do not take into account the user's emotional state, users may lack confidence in or understand the predictions. This makes it difficult for users to make appropriate investment and trading decisions based on the prediction results.

[0643] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data, a means for making a prediction using the generative artificial intelligence model, a means for improving prediction accuracy using strict gradient descent, a means for collecting user emotion data and analyzing it using an emotion engine, a means for dynamically adjusting the reliability of the prediction based on the analyzed emotion data, and a means for displaying the prediction result to the user. This enables highly accurate prediction using the generative artificial intelligence model, and by taking the user emotion data into consideration, the reliability of the prediction is further increased, allowing the user to use the prediction result with peace of mind.

[0644] A "generative artificial intelligence model" refers to an advanced machine learning algorithm that takes time series data or natural language data as input and performs predictive or generative tasks.

[0645] "Time series data" is a continuous set of data that changes over time, and includes numerical information collected at specific time intervals, such as stock prices or foreign exchange prices.

[0646] "Natural language data" refers to text information written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[0647] "Gradient descent" refers to an algorithm that calculates the gradient of the error function and gradually updates the parameters based on that in order to optimize the parameters of a machine learning model.

[0648] An "emotion engine" is a technology for analyzing a user's emotional state, and takes input data such as facial expressions, voice tone, and text messages.

[0649] "Dynamic adjustment" means that the system automatically changes its settings and behavior in response to changing data and conditions in real time.

[0650] "Prediction reliability" is an indicator that shows the accuracy and credibility of the prediction results output by a generative artificial intelligence model.

[0651] A "Transformer model" is a type of neural network that can perform natural language processing tasks using large datasets, particularly those that utilize attention mechanisms.

[0652] An "LSTM layer" is a recurrent neural network layer with long short-term memory units, which can memorize long-term dependencies.

[0653] A "data loader" is a program or library for efficiently processing batches of data during model training.

[0654] An "epoch" is a unit of machine learning training that refers to passing the entire dataset through a model once.

[0655] This invention combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. An embodiment of this system will be described below.

[0656] server

[0657] The server runs on high-performance hardware and programming languages ​​such as Python. The server first collects time-series data on stock prices and exchange rates through an API. For example, it uses the Yahoo Finance API to obtain data on the dollar-yen exchange rate for the past year and stores it in a database. It also collects natural language data such as news articles and analytical reports using news feeds and RSS readers. This data is stored as text files.

[0658] Next, the server preprocesses the time series data and natural language data to create an integrated dataset. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data is also cleaned by removing unnecessary characters and tags, and then tokenized and vectorized. This creates an integrated dataset that combines time series data and natural language data.

[0659] Using the combined dataset, the server builds and trains a generative artificial intelligence model. It uses a model that combines a transformer model and an LSTM layer, and uses the TensorFlow library for training. It prepares batches of data using a data loader, and separates the data into training, validation, and test data. It calculates the gradient of the error function and updates the model parameters using gradient descent.

[0660] Furthermore, the server collects user emotional data and analyzes it using an emotion engine. It uses a camera and microphone to capture the user's facial expressions and voice tone, and analyzes emotions in real time. Based on this emotional data, the reliability of predictions is dynamically adjusted.

[0661] Terminal

[0662] The terminal provides a user interface and collects prediction requests and emotion data. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. The emotion engine analyzes the user's facial expressions and tone of voice to collect emotion data in real time. This data is then sent to the server, which then makes a prediction request to the server.

[0663] User

[0664] The user inputs, for example, the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine recognizes the user's current emotion from their tone of voice and facial expression, and transmits it to the server. The user inputs the following:

[0665] Based on the past week's USD / JPY exchange rate data and related news articles, please predict next week's USD / JPY exchange rate.

[0666] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The forecast results are sent back to the device, which displays them to the user. The user can then make investment and trading decisions based on the forecast results.

[0667] The above is an embodiment of the present invention. This system combines a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide users with reliable information. Users can use these prediction results to make more appropriate trading and investment decisions.

[0668] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0669] Step 1: Data collection

[0670] The server first uses an API (general term) to collect time-series data on stock prices and exchange rates. Specifically, to obtain, for example, the dollar-yen exchange rate data for the past year, the appropriate API endpoint is called and the data is downloaded in CSV format. This data is then stored in the server's database. Furthermore, news articles for a specified period are collected using a news feed or RSS reader and saved as text files.

[0671] Input: API endpoint, RSS feed URL

[0672] Output: Time series data (CSV), natural language data (text file)

[0673] Step 2: Data preprocessing and integration

[0674] The server formats the collected time series data and natural language data. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data undergoes cleaning (removal of unnecessary tags and characters), tokenization, and vectorization. The time series data and natural language data are then matched based on date to create an integrated dataset.

[0675] Input: time series data, natural language data

[0676] Output: Unified dataset

[0677] Step 3: Building the model

[0678] The server uses the combined dataset to build a generative artificial intelligence model, which combines a transformer model with an LSTM layer, and uses the TensorFlow library in Python to define, compile, and initialize the model.

[0679] Input: Unified dataset

[0680] Output: An initialized generative artificial intelligence model

[0681] Step 4: Train the model

[0682] The server uses a data loader to split the training data into batches and train the model. The model generates predictions on the training data and calculates the error based on the difference between the predictions and the actual data. To minimize the error, the model's parameters are updated using gradient descent. This process is repeated for multiple epochs.

[0683] Input: Initialized generative artificial intelligence model, training data

[0684] Output: A trained generative artificial intelligence model

[0685] Step 5: Evaluate and save the model

[0686] The server evaluates the model's performance using the validation and test data. It calculates evaluation metrics (e.g., RMSE and MAE) and checks whether the model meets the specified performance criteria. If it does, it saves the model to storage and exposes it as an API.

[0687] Input: A trained generative AI model, validation data, and test data

[0688] Output: Saved generative AI model, exposed API

[0689] Step 6: Collect and analyze emotion data

[0690] The device uses a camera and microphone to collect the user's emotional data. This allows the device to capture the user's facial expressions and vocal tone in real time and transmit them to the server. The server then uses an emotion engine to analyze this data and evaluate the user's emotional state.

[0691] Input: Facial expression data, voice tone data

[0692] Output: Parsed emotion data

[0693] Step 7: Request and make predictions

[0694] The device collects past data and news articles for the period the user wants to predict, and sends them along with emotion data to the server. The server then inputs this data into a generative AI model to predict stock prices and exchange rates for the next period. The reliability of the prediction is dynamically adjusted based on the user's emotion data.

[0695] Input: Historical data, news articles, sentiment data

[0696] Output: Prediction results

[0697] Step 8: Displaying prediction results and providing feedback

[0698] The server sends the prediction results to the device, which then displays them to the user. Furthermore, feedback is generated based on the user's emotional data, providing additional information to deepen the user's understanding of the prediction results. This allows the user to use the prediction results with confidence.

[0699] Input: Prediction results, analyzed emotion data

[0700] Output: Prediction results and feedback displayed to the user

[0701] These are the specific steps of the system's processing. The data input and output at each step, as well as the processing details, are clearly defined, creating a system that improves the accuracy of predictions as a whole.

[0702] (Application example 2)

[0703] 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."

[0704] Conventional stock price and exchange rate prediction systems using generative artificial intelligence models make predictions without taking into account the user's emotional state, and therefore are unable to alleviate the user's stress and anxiety. In addition, because the prediction results are not adjusted to take the user's emotions into account, there are issues with the accuracy of predictions and the reliability of investment advice.

[0705] The identification processing by the identification 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 combining a generative artificial intelligence model with time-series data and corresponding natural language data and making predictions using the generative artificial intelligence model; means for improving prediction accuracy using rigorous gradient descent; means for combining with an emotion engine that recognizes the user's emotions and adjusting the prediction results; and means for providing investment advice and risk management based on the user's emotional state. This makes it possible to provide the user with highly accurate predictions that take the user's emotional state into consideration, and reliable investment advice and risk management.

[0706] A "generative artificial intelligence model" is an artificial intelligence technology that uses past data and learning algorithms to generate new data and perform predictions and analysis.

[0707] "Time series data" is a series of data points collected over time, such as stock prices or currency fluctuations.

[0708] "Natural language data" refers to information written in languages ​​that humans use on a daily basis, including news articles and analytical reports.

[0709] "Gradient descent" is an algorithm used to optimize the parameters of a predictive model, which involves repeatedly processing a set of data to minimize error.

[0710] An "emotion engine" is software that analyzes a user's emotional state and recognizes emotions from voice tone, facial expressions, text input, etc.

[0711] "Investment advice" refers to advice that supports users in making appropriate investment decisions, including suggestions based on predictive data and the user's emotional state.

[0712] "Risk management" is the process of minimizing the risks associated with investment activities and proposing optimal investment strategies for users.

[0713] A "Transformer model" is an advanced machine learning model used in natural language processing, which analyzes large amounts of text data to understand meaning and make predictions.

[0714] "Prediction results" are information that indicates future data points and trends calculated by a generative artificial intelligence model.

[0715] "Retraining" is an additional learning process that reviews the parameter settings of the initial model and improves the accuracy of the model based on actual data.

[0716] A "prompt" refers to specific input data or questions used when analyzing natural language data or making predictions.

[0717] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. A detailed configuration for configuring the system of the present invention will be described below.

[0718] System configuration

[0719] It consists of three main elements: the server, the smartphone device, and the user.

[0720] server

[0721] The server uses the following hardware and software:

[0722] Hardware: High-performance CPU, GPU, storage, network interface

[0723] Software: TensorFlow, Transformers (Hugging Face), API, Database Management System

[0724] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. The collected data is then formatted and combined with natural language data such as news articles and analytical reports to create an integrated dataset. This dataset is then used to train a generative artificial intelligence model that combines a transformer model and an LSTM layer to make predictions.

[0725] The server also uses an emotion engine to analyze the user's emotions in real time, which analyzes voice tone, facial expressions, input text messages, etc. to recognize the user's emotions.

[0726] Terminal

[0727] The device used is a smartphone operated by the user. The application on the smartphone provides the following functions:

[0728] Input interface for historical exchange data and corresponding news articles

[0729] An interface for recognizing emotions from speech tone and facial expressions

[0730] Displaying prediction results returned from the server

[0731] Through the application, users input past data and news articles for the period they want to predict. The emotion engine obtains the user's current emotional data and sends it to the server. The server then uses the generative AI model and emotion engine to make predictions and returns the results to the device.

[0732] User

[0733] Users operate their smartphones to input data, have the system recognize their emotions, and then check the prediction results to make trading decisions.

[0734] Specific examples

[0735] The user inputs the past week's USD / JPY exchange rate data and news articles into the terminal. The emotion engine recognizes the user's current emotions from their voice tone and facial expressions, and when they request a USD / JPY forecast for the next week, the terminal sends the data to the server. The server makes a prediction and returns the results to the terminal, adjusting the reliability of the prediction based on the user's emotions. The terminal then displays the received prediction results to the user, who can then begin trading based on that information. Appropriate feedback is also displayed.

[0736] An example prompt is:

[0737] "Enter USD / JPY exchange rate data and news articles from the past week. Then, please tell us your current emotional state by voice."

[0738] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0739] Step 1:

[0740] The server collects historical exchange data and news articles from APIs and databases. It sends requests to collect exchange data and news articles and retrieves the respective data. The input is raw data obtained from the API, and the output is formatted time series data and news article data.

[0741] Step 2:

[0742] The server formats the collected data, associates the time series data with the natural language data, and creates an integrated dataset. The formatted data is then combined and saved as a single dataset. The input is the formatted time series data and news article data, and the output is an integrated dataset.

[0743] Step 3:

[0744] The server combines a Transformer model and an LSTM layer to build a generative AI model, and then trains the model using this integrated dataset. The input is the integrated dataset, and the output is the trained generative AI model. The model processes the data sequentially and improves its prediction accuracy through learning.

[0745] Step 4:

[0746] The server uses an emotion engine to recognize emotions from the user's voice tone and facial expressions. It analyzes voice and image data for emotion recognition and identifies the user's emotional state. The input is the user's voice and image data, and the output is the recognized emotional data.

[0747] Step 5:

[0748] The terminal provides an interface for receiving past exchange data and news articles entered by the user. The user enters data for the desired forecast period, and also collects voice tone and facial expression data. The input is exchange data, news articles, voice and image data from the user, and the output is the collected data.

[0749] Step 6:

[0750] The server receives data sent from the device and inputs it into a generative artificial intelligence model to predict the next exchange rate and stock price. The data is input into the model and a prediction result is generated. The input is data from the device and the output is the prediction result.

[0751] Step 7:

[0752] The server dynamically adjusts the reliability of predictions using the user's emotional data recognized by the emotion engine. The prediction results are revised based on the emotional data to improve reliability. The input is the recognized emotional data, and the output is the adjusted prediction results.

[0753] Step 8:

[0754] The server returns the final prediction result to the terminal, which transmits the prediction result and provides it to the user. The input is the adjusted prediction result, and the output is the data for the user to display.

[0755] Step 9:

[0756] The terminal receives the prediction results from the server and displays them to the user. The received data is displayed on the screen for the user to confirm. The input is the prediction result data from the server, and the output is the displayed prediction result.

[0757] Step 10:

[0758] The user checks the prediction results and emotional feedback displayed on the device and makes a trading decision. Based on the prediction results and feedback, the user makes an appropriate trade. The input is the prediction results and feedback displayed on the device, and the output is the user's trading decision.

[0759] 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.

[0760] 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.

[0761] 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.

[0762] [Third embodiment]

[0763] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0764] 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.

[0765] 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).

[0766] 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.

[0767] 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.

[0768] 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).

[0769] 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.

[0770] 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.

[0771] 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.

[0772] 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.

[0773] 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.

[0774] 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."

[0775] The present invention relates to a system for predicting stock prices and exchange rates using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[0776] System program and processing description

[0777] server

[0778] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The server then formats the data and creates an integrated dataset that combines the time-series data with corresponding news and related document information. This integrated dataset is used as input data for making predictions using a transformer model (e.g., a model for automatic learning) and a time-series prediction model (e.g., LSTM or RNN).

[0779] Next, the server builds a generative artificial intelligence model. Specifically, it loads a "Transformer model" and adds an "LSTM" layer to it to define it as an integrated neural network. It then trains this model using the integrated dataset. For training, it prepares batches of data using a data loader, inputs this data into the model, and generates predictions. It calculates an error based on the difference between the prediction results and the actual data, backpropagates this error to calculate the gradient, and updates the model parameters. This procedure is repeated over multiple epochs (learning cycles).

[0780] The server evaluates the model during the training process. For evaluation, it uses validation data and test data and calculates the difference between the prediction and the actual data as an indicator (e.g., MSE). If the model meets certain performance criteria, the server saves it and provides it to external parties as an API.

[0781] Terminal

[0782] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to forecast. This data is then sent to the server, which makes a forecast request.

[0783] The server uses the received data to input data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period. The prediction results are sent back to the terminal, which displays them to the user, who can use them as a reference when making trading and investment decisions.

[0784] Specific examples

[0785] User

[0786] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the terminal. Next, the user requests a "USD / JPY forecast for the next week." The terminal then sends this data to the server and makes a prediction request.

[0787] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. The forecast results are then sent back to the terminal. The terminal then displays the received forecast results to the user, who can then begin trading based on that information.

[0788] The above is a detailed description of an embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[0789] The processing flow will be explained below.

[0790] Step 1:

[0791] The server collects time-series data on stock prices and exchange rates, and retrieves past data from APIs and databases. For example, it retrieves and stores the past year's USD / JPY exchange rate data.

[0792] Step 2:

[0793] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0794] Step 3:

[0795] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[0796] Step 4:

[0797] The server loads the Transformer model and adds and integrates an LSTM layer to build a generative artificial intelligence model.

[0798] Step 5:

[0799] The server splits the combined dataset into three parts: training, validation, and testing. For example, 80% is for training, 10% is for validation, and 10% is for testing.

[0800] Step 6:

[0801] The server prepares the batch data using a data loader and is configured to load the training data in fixed batch sizes.

[0802] Step 7:

[0803] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated for multiple epochs.

[0804] Step 8:

[0805] The server evaluates the performance of the trained model using the validation data, calculating the difference between the predictions and the actual data as a metric (e.g., MSE) to confirm the model's performance.

[0806] Step 9:

[0807] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[0808] Step 10:

[0809] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[0810] Step 11:

[0811] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[0812] Step 12:

[0813] The terminal transmits the data entered by the user to the server and makes a prediction request.

[0814] Step 13:

[0815] The server inputs the received data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period.

[0816] Step 14:

[0817] The server returns the prediction result to the terminal.

[0818] Step 15:

[0819] The terminal displays the prediction results received from the server to the user, who can then make investment and trading decisions based on the prediction results.

[0820] Example 1

[0821] 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."

[0822] Conventional financial market forecasting systems have relied on a single data source, resulting in low forecast accuracy. Furthermore, the lack of a method for effectively linking natural language data and time series data limits the accuracy of forecasts. Furthermore, the training and evaluation methods for generative AI models have not been fully optimized, preventing the full potential of forecasting performance.

[0823] 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.

[0824] In this invention, the server includes a means for collecting time-series data of stock prices and exchange rates and natural language data, a means for creating an integrated data set by associating the collected time-series data with the natural language data, and a means for constructing a generative artificial intelligence model that combines a Transformer model and a time-series prediction model, thereby enabling highly accurate predictions.

[0825] "Time series data of stock prices and exchange rates" refers to numerical information on stock prices and exchange rates recorded at specific time intervals (e.g., daily, hourly, etc.).

[0826] "Natural language data" is text data written in the language that people use on a daily basis, such as news articles, analytical reports, and social media posts.

[0827] An "integrated dataset" is a single dataset that links time series data and natural language data based on date and time and relevance, and is used to train artificial intelligence models.

[0828] A "generative artificial intelligence model" is an artificial intelligence model designed to make predictions or generate data from input data, and specifically includes transformer models and time series prediction models.

[0829] A "Transformer model" is a type of machine learning for data, and is a neural network model used in particular for natural language processing.

[0830] A "time series prediction model" is an algorithm or model that uses time series data as input to predict future values, and specifically includes LSTM and RNN.

[0831] "Calculating gradients by backpropagating errors" is the process of calculating the gradients of the parameters of an artificial intelligence model using the difference (error) between the predicted value and the actual value, and updating the model parameters.

[0832] "Evaluating a model" is the process of measuring the performance of a trained generative artificial intelligence model using validation data and evaluating its predictive accuracy.

[0833] "Providing the prediction results to the user" means displaying the prediction values ​​generated by the generative artificial intelligence model to the user so that the user can make decisions based on that information.

[0834] The present invention relates to a system for predicting stock prices and exchange rates, and is a system that achieves highly accurate predictions using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[0835] server

[0836] The server operates as follows: First, it uses the APIs and databases of financial data providers to collect time-series data on stock prices and exchange rates. It also uses news APIs to obtain natural language data such as related news articles and analytical reports. Specific data collection uses financial data APIs (e.g., Yahoo! Finance API, Alpha Vantage API) and news APIs.

[0837] The server then processes the collected time series data and natural language data, aligning them by date and time, to create an integrated dataset that links the time series data with the corresponding news articles. This integrated dataset is then used as input data for making predictions using transformer models (e.g., BERT models) and time series prediction models (e.g., LSTM, RNN).

[0838] The server then builds a hybrid neural network that combines a Transformer model with an LSTM layer and trains the model using the integrated dataset. In this invention, we use the Hugging Face Transformers library as the Transformer model. The training process involves using PyTorch to prepare the data loader and generate predictions. The error is calculated from the difference between the prediction result and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. This procedure is repeated multiple times.

[0839] During the training process, the server also evaluates the model. For evaluation, it uses a validation dataset and calculates the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error). If the model meets certain performance criteria, it saves it and provides it to external parties as an API.

[0840] Terminal

[0841] The terminal provides an interface that accepts input from the user. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. This data is sent to the server, which issues a prediction request. The server then uses this data to make a prediction using a generative artificial intelligence model and sends the prediction results back to the terminal. The terminal then displays the prediction results, which the user can use as a reference when making trading and investment decisions.

[0842] User

[0843] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into a terminal. Then, the user requests a "prediction of the USD / JPY for the next week." The terminal sends this data to the server and makes a prediction request. The server inputs the received data into a generative AI model, which predicts the USD / JPY exchange rate for the next week. The prediction results are sent back to the terminal, which displays them to the user. The user then makes trading decisions based on this information. For example, the following prompt sentences can be input into the generative AI model:

[0844] What is your forecast for the dollar-yen exchange rate for the next week?

[0845] Data and news articles from the past week are below.

[0846] Exchange data: [List of data]

[0847] News articles: [list of articles]

[0848] The above is a detailed description of an embodiment of the present invention. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[0849] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0850] Step 1: Data collection

[0851] server

[0852] The server collects stock and exchange time series data from financial data providers' APIs and databases, and also uses news APIs to retrieve natural language data such as related news articles and analytical reports.

[0853] input

[0854] Financial data API, news API.

[0855] output

[0856] Time series data and news article data.

[0857] Specific actions

[0858] The server calls a financial data API to retrieve stock prices and exchange rates for the past few months, and similarly calls a news API to retrieve relevant news articles.

[0859] Step 2: Data Refinement and Integration

[0860] server

[0861] The server then formats the collected time series data and natural language data by aligning the data by date and time, and creating an integrated dataset that links a series of time series data with the corresponding news articles.

[0862] input

[0863] Time series data, news article data.

[0864] output

[0865] Integrated dataset.

[0866] Specific actions

[0867] The server uses the Pandas library to convert the time series data and news article data into data frames, then merges them based on date and time to create a unified dataset.

[0868] Step 3: Model Building

[0869] server

[0870] The server builds a hybrid neural network that combines a Transformer model and an LSTM layer, using a common natural language processing library and a time series forecasting model.

[0871] input

[0872] Natural language processing library, time series forecasting library.

[0873] output

[0874] Hybrid neural network model.

[0875] Specific actions

[0876] The server loads a Transformer model and adds an LSTM layer to it to define a hybrid neural network.

[0877] Step 4: Model training

[0878] server

[0879] The server uses the combined dataset to train the hybrid neural network model, using a data loader to prepare batches of data, which are then fed into the model to generate predictions.

[0880] input

[0881] Integrated datasets, hybrid neural network models.

[0882] output

[0883] The trained neural network model.

[0884] Specific actions

[0885] The server uses PyTorch to prepare a data loader and input batches of data into the model. The error is calculated from the difference between the predicted results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters.

[0886] Step 5: Model evaluation

[0887] server

[0888] The server evaluates the model during the training process, using a validation dataset to calculate the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error).

[0889] input

[0890] Validation dataset, trained neural network model.

[0891] output

[0892] Evaluation results (MSE).

[0893] Specific actions

[0894] The server uses the validation dataset to calculate the difference between the prediction and the actual value, and calculates the MSE to evaluate the model's predictive accuracy.

[0895] Step 6: Processing a prediction request

[0896] Terminal

[0897] The terminal receives input from the user about past exchange data and news articles, and sends this to the server. The server then uses this information to make predictions for the next period and sends the results back to the terminal.

[0898] input

[0899] User input data: historical exchange data, news articles.

[0900] output

[0901] Prediction results.

[0902] Specific actions

[0903] The terminal receives data from the user and sends it to the server via an HTTP request. The server inputs the received data into the model, performs exchange rate predictions for the next period, and sends the results back to the terminal.

[0904] Step 7: View the prediction results

[0905] Terminal

[0906] The terminal displays the prediction results returned from the server to the user, who then makes trading and investment decisions based on these prediction results.

[0907] input

[0908] Prediction results from the server.

[0909] output

[0910] The prediction results displayed on the screen.

[0911] Specific actions

[0912] The terminal displays the prediction results received from the server on the screen so that the user can easily check them. The user can then start trading based on this information.

[0913] (Application example 1)

[0914] 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."

[0915] Conventional systems do not provide sufficient information for investment decisions, particularly in terms of real-time forecasts and related information in financial markets. Furthermore, they are unable to provide personalized information tailored to individual user needs, creating a need for a means to efficiently provide information useful for users' investment decisions.

[0916] 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.

[0917] In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data and making predictions using the generative artificial intelligence model, a means for improving prediction accuracy using rigorous gradient descent, a means for providing information related to financial investments, and a means for providing a personalized notification function based on user interests, thereby enabling users to receive highly accurate forecast information in real time and efficiently obtain related information useful for their investment decisions.

[0918] A "generative artificial intelligence model" is an artificial intelligence model that combines time series data and natural language data to make predictions using a neural network.

[0919] "Time series data" refers to a continuous record of data collected at regular intervals, such as stock prices or currency exchange data.

[0920] "Natural language data" refers to text data written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[0921] "Exact gradient descent" is a computational technique that iteratively updates the model parameters to minimize error during the neural network training process.

[0922] "Information related to financial investment" refers to information to support investment decisions, including stock price forecasts, related news articles, and analytical reports.

[0923] "Personalized notification function" is a function that automatically delivers specific information based on the user's individual interests and concerns.

[0924] A "Transformer model" is a neural network model widely used in natural language processing, and is a model for understanding and generating language data.

[0925] A "time series prediction model" is a neural network model that uses time series data as input to predict future data points.

[0926] A "content distribution service" is a service that provides related information to users in real time.

[0927] This invention relates to a system that uses generative artificial intelligence models to predict stock prices and exchange rates, and provides information related to financial investments in real time. The system includes two main components, a server and a terminal, and provides an interface that is easy for users to use.

[0928] server

[0929] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The collected data is then formatted through preprocessing to create an integrated dataset that combines the time-series data with corresponding news and related document information.

[0930] Next, the server builds a generative artificial intelligence model. Specifically, it loads a Transformer model and adds an LSTM layer to it to define it as an integrated neural network. This model is trained using the integrated dataset. A data loader is used to prepare batch data, which is then input into the model to generate predictions. An error is calculated based on the difference between the generated prediction results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. The model is evaluated using validation and test data, and if it meets certain performance criteria, the server saves the model and provides it externally as an API.

[0931] Terminal

[0932] The device provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict into the device. This data is sent to the server, and a prediction request is made. The server uses the received data to input data into a generative artificial intelligence model, which predicts exchange rates and stock prices for the next period. The prediction results are sent back to the device, which displays them to the user. A personalized notification function based on the user's interests is also provided.

[0933] User

[0934] For example, a user inputs the exchange rate data for a specific currency from the past week and related news articles into their device. Next, the user requests a "currency forecast for the next week." The device sends this data to the server and makes a prediction request. The server inputs the received data into a generative artificial intelligence model and calculates a forecast for the next week's exchange rate. The prediction results are then sent back to the device, which displays them to the user. Personalized notifications are also provided based on the user's interests.

[0935] Usage examples and prompt statements

[0936] As a concrete example, if a user wants to get a forecast for the next week's exchange rate for a specific currency (e.g., the yen), the prompt text is as follows:

[0937] curl -X GET "https: / / api.example.com / stocks?ticker=USDJPY"

[0938] curl -X GET "https: / / api.example.com / news?ticker=USDJPY"

[0939] This prompt shows an API request to gather exchange and news data for a specific currency. This data can be used by the application to make predictions and provide the user with a currency forecast for the coming week.

[0940] The above is a specific embodiment for carrying out the present invention of this system.

[0941] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0942] Step 1:

[0943] The terminal receives input from the user, including past exchange data and corresponding news articles. This allows the terminal to obtain data for the period for which the user wishes to make a prediction. The input data is then sent to the server.

[0944] Step 2:

[0945] The server performs data processing based on the received exchange data and news articles. This process combines time series data and natural language data to create an integrated dataset. The processed data is then created.

[0946] Step 3:

[0947] The server uses the formatted data to input data into a generative artificial intelligence model. Specifically, predictions are made using a Transformer model and an LSTM layer. Data is input into the model, and prediction results are output.

[0948] Step 4:

[0949] The server calculates the difference between the predicted results and the actual data, and then retrains the generative AI model using rigorous gradient descent. This process improves the accuracy of predictions, and a retrained model is created.

[0950] Step 5:

[0951] The server then uses the trained model to make another prediction and returns the prediction results to the device, where they are displayed.

[0952] Step 6:

[0953] The device displays the received prediction results to the user, who can then make investment decisions based on this information.The device also provides personalized notifications based on the user's interests.

[0954] Step 7:

[0955] The user decides on the next investment action based on the displayed prediction results and notified information, and the user's investment action is executed.

[0956] 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.

[0957] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. Detailed embodiments for implementing this system are described below.

[0958] System program and processing description

[0959] server

[0960] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. For example, it retrieves and stores data on the dollar-yen exchange rate for the past year. It also collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0961] The server then transforms the time series data and natural language data, matches them with corresponding news articles and analytical reports to create an integrated dataset, and loads a Transformer model, adding an LSTM layer to integrate them to build a generative artificial intelligence model.

[0962] The server uses this combined dataset to train the model, preparing batches of data using a data loader. The model generates predictions for each batch of data, calculates the error based on the difference between the predictions and the actual data, and updates the model parameters using gradient descent. This procedure is repeated over multiple epochs (learning cycles).

[0963] During the training process, the server evaluates the model's performance using validation data and finally using test data. Models that achieve satisfactory performance are saved and made public as an API.

[0964] The server also incorporates an emotion engine that recognizes the user's emotions by analyzing the user's facial expressions, voice tone, and text messages, and evaluates the user's emotions in real time.

[0965] Terminal

[0966] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict. In addition, the emotion engine obtains the user's current emotion data. This data is sent to the server, and a prediction request is made.

[0967] The server uses the received data to input it into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the prediction based on the user's emotional data recognized by the emotion engine, further improving the accuracy of the prediction.

[0968] The server sends the prediction results back to the device, which then displays them to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with peace of mind.

[0969] Specific examples

[0970] User

[0971] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine then recognizes the user's current emotion from their tone of voice and facial expression. Next, the user requests a "USD / JPY forecast for the next week." The device then sends this data to the server and makes a prediction request.

[0972] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The server then sends the forecast results back to the terminal. The terminal then displays the received forecast results to the user, who then begins trading based on that information.

[0973] Furthermore, the device provides appropriate feedback based on the user's emotional data, making it easier for the user to understand the prediction results and use the device with confidence.

[0974] The above is a detailed embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide more reliable information to users. Users can use these prediction results to make more appropriate trading and investment decisions.

[0975] The processing flow will be explained below.

[0976] Step 1:

[0977] The server collects time series data on stock prices and exchange rates, and retrieves and stores the past year's USD / JPY exchange rate data via API and database.

[0978] Step 2:

[0979] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[0980] Step 3:

[0981] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[0982] Step 4:

[0983] The server loads the Transformer model and adds an LSTM layer to integrate it to build a generative artificial intelligence model. Specifically, the output of the Transformer model is input to the LSTM, and the output is used as a prediction value to define a generative artificial intelligence model.

[0984] Step 5:

[0985] The server splits the combined dataset into three parts: training, validation, and testing: 80% of the dataset is stored for training, 10% for validation, and 10% for testing.

[0986] Step 6:

[0987] The server prepares the batch data using a data loader. It uses PyTorch's DataLoader class and configures it to load training data in fixed batch sizes.

[0988] Step 7:

[0989] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated over multiple epochs (learning cycles).

[0990] Step 8:

[0991] The server evaluates the performance of the trained model using the validation data. It calculates the difference between the predictions and the actual data as an evaluation metric (e.g., MSE) to confirm the model's performance.

[0992] Step 9:

[0993] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[0994] Step 10:

[0995] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[0996] Step 11:

[0997] The server incorporates an emotion engine, which analyzes the user's tone of voice, facial expressions, and input text messages to evaluate the user's emotions in real time.

[0998] Step 12:

[0999] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[1000] Step 13:

[1001] The terminal receives data input by the user and transmits it to the server. In addition, the terminal acquires the user's emotion data in real time and transmits it to the server.

[1002] Step 14:

[1003] The server inputs the received time series data and natural language data into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the predictions based on the user's emotional data recognized by the emotion engine.

[1004] Step 15:

[1005] The server returns the prediction result and adjusted reliability information to the terminal.

[1006] Step 16:

[1007] The device displays the prediction results and reliability information received from the server to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with confidence.

[1008] These are the specific processing steps of the system that combines the emotion engine. Users can use the prediction results to make appropriate trading and investment decisions that take emotion into account.

[1009] Example 2

[1010] 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."

[1011] Current stock price and exchange rate prediction systems use time series data and limited natural language data, which limits their prediction accuracy. Furthermore, because they do not take into account the user's emotional state, users may lack confidence in or understand the predictions. This makes it difficult for users to make appropriate investment and trading decisions based on the prediction results.

[1012] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data, a means for making a prediction using the generative artificial intelligence model, a means for improving prediction accuracy using strict gradient descent, a means for collecting user emotion data and analyzing it using an emotion engine, a means for dynamically adjusting the reliability of the prediction based on the analyzed emotion data, and a means for displaying the prediction result to the user. This enables highly accurate prediction using the generative artificial intelligence model, and by taking the user emotion data into consideration, the reliability of the prediction is further increased, allowing the user to use the prediction result with peace of mind.

[1013] A "generative artificial intelligence model" refers to an advanced machine learning algorithm that takes time series data or natural language data as input and performs predictive or generative tasks.

[1014] "Time series data" is a continuous set of data that changes over time, and includes numerical information collected at specific time intervals, such as stock prices or foreign exchange prices.

[1015] "Natural language data" refers to text information written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[1016] "Gradient descent" refers to an algorithm that calculates the gradient of the error function and gradually updates the parameters based on that in order to optimize the parameters of a machine learning model.

[1017] An "emotion engine" is a technology for analyzing a user's emotional state, and takes input data such as facial expressions, voice tone, and text messages.

[1018] "Dynamic adjustment" means that the system automatically changes its settings and behavior in response to changing data and conditions in real time.

[1019] "Prediction reliability" is an indicator that shows the accuracy and credibility of the prediction results output by a generative artificial intelligence model.

[1020] A "Transformer model" is a type of neural network that can perform natural language processing tasks using large datasets, particularly those that utilize attention mechanisms.

[1021] An "LSTM layer" is a recurrent neural network layer with long short-term memory units, which can memorize long-term dependencies.

[1022] A "data loader" is a program or library for efficiently processing batches of data during model training.

[1023] An "epoch" is a unit of machine learning training that refers to passing the entire dataset through a model once.

[1024] This invention combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. An embodiment of this system will be described below.

[1025] server

[1026] The server runs on high-performance hardware and programming languages ​​such as Python. The server first collects time-series data on stock prices and exchange rates through an API. For example, it uses the Yahoo Finance API to obtain data on the dollar-yen exchange rate for the past year and stores it in a database. It also collects natural language data such as news articles and analytical reports using news feeds and RSS readers. This data is stored as text files.

[1027] Next, the server preprocesses the time series data and natural language data to create an integrated dataset. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data is also cleaned by removing unnecessary characters and tags, and then tokenized and vectorized. This creates an integrated dataset that combines time series data and natural language data.

[1028] Using the combined dataset, the server builds and trains a generative artificial intelligence model. It uses a model that combines a transformer model and an LSTM layer, and uses the TensorFlow library for training. It prepares batches of data using a data loader, and separates the data into training, validation, and test data. It calculates the gradient of the error function and updates the model parameters using gradient descent.

[1029] Furthermore, the server collects user emotional data and analyzes it using an emotion engine. It uses a camera and microphone to capture the user's facial expressions and voice tone, and analyzes emotions in real time. Based on this emotional data, the reliability of predictions is dynamically adjusted.

[1030] Terminal

[1031] The terminal provides a user interface and collects prediction requests and emotion data. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. The emotion engine analyzes the user's facial expressions and tone of voice to collect emotion data in real time. This data is then sent to the server, which then makes a prediction request to the server.

[1032] User

[1033] The user inputs, for example, the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine recognizes the user's current emotion from their tone of voice and facial expression, and transmits it to the server. The user inputs the following:

[1034] Based on the past week's USD / JPY exchange rate data and related news articles, please predict next week's USD / JPY exchange rate.

[1035] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The forecast results are sent back to the device, which displays them to the user. The user can then make investment and trading decisions based on the forecast results.

[1036] The above is an embodiment of the present invention. This system combines a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide users with reliable information. Users can use these prediction results to make more appropriate trading and investment decisions.

[1037] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1038] Step 1: Data collection

[1039] The server first uses an API (general term) to collect time-series data on stock prices and exchange rates. Specifically, to obtain, for example, the dollar-yen exchange rate data for the past year, the appropriate API endpoint is called and the data is downloaded in CSV format. This data is then stored in the server's database. Furthermore, news articles for a specified period are collected using a news feed or RSS reader and saved as text files.

[1040] Input: API endpoint, RSS feed URL

[1041] Output: Time series data (CSV), natural language data (text file)

[1042] Step 2: Data preprocessing and integration

[1043] The server formats the collected time series data and natural language data. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data undergoes cleaning (removal of unnecessary tags and characters), tokenization, and vectorization. The time series data and natural language data are then matched based on date to create an integrated dataset.

[1044] Input: time series data, natural language data

[1045] Output: Unified dataset

[1046] Step 3: Building the model

[1047] The server uses the combined dataset to build a generative artificial intelligence model, which combines a transformer model with an LSTM layer, and uses the TensorFlow library in Python to define, compile, and initialize the model.

[1048] Input: Unified dataset

[1049] Output: An initialized generative artificial intelligence model

[1050] Step 4: Train the model

[1051] The server uses a data loader to split the training data into batches and train the model. The model generates predictions on the training data and calculates the error based on the difference between the predictions and the actual data. To minimize the error, the model's parameters are updated using gradient descent. This process is repeated for multiple epochs.

[1052] Input: Initialized generative artificial intelligence model, training data

[1053] Output: A trained generative artificial intelligence model

[1054] Step 5: Evaluate and save the model

[1055] The server evaluates the model's performance using the validation and test data. It calculates evaluation metrics (e.g., RMSE and MAE) and checks whether the model meets the specified performance criteria. If it does, it saves the model to storage and exposes it as an API.

[1056] Input: A trained generative AI model, validation data, and test data

[1057] Output: Saved generative AI model, exposed API

[1058] Step 6: Collect and analyze emotion data

[1059] The device uses a camera and microphone to collect the user's emotional data. This allows the device to capture the user's facial expressions and vocal tone in real time and transmit them to the server. The server then uses an emotion engine to analyze this data and evaluate the user's emotional state.

[1060] Input: Facial expression data, voice tone data

[1061] Output: Parsed emotion data

[1062] Step 7: Request and make predictions

[1063] The device collects past data and news articles for the period the user wants to predict, and sends them along with emotion data to the server. The server then inputs this data into a generative AI model to predict stock prices and exchange rates for the next period. The reliability of the prediction is dynamically adjusted based on the user's emotion data.

[1064] Input: Historical data, news articles, sentiment data

[1065] Output: Prediction results

[1066] Step 8: Displaying prediction results and providing feedback

[1067] The server sends the prediction results to the device, which then displays them to the user. Furthermore, feedback is generated based on the user's emotional data, providing additional information to deepen the user's understanding of the prediction results. This allows the user to use the prediction results with confidence.

[1068] Input: Prediction results, analyzed emotion data

[1069] Output: Prediction results and feedback displayed to the user

[1070] These are the specific steps of the system's processing. The data input and output at each step, as well as the processing details, are clearly defined, creating a system that improves the accuracy of predictions as a whole.

[1071] (Application example 2)

[1072] 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."

[1073] Conventional stock price and exchange rate prediction systems using generative artificial intelligence models make predictions without taking into account the user's emotional state, and therefore are unable to alleviate the user's stress and anxiety. In addition, because the prediction results are not adjusted to take the user's emotions into account, there are issues with the accuracy of predictions and the reliability of investment advice.

[1074] The identification processing by the identification 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 combining a generative artificial intelligence model with time-series data and corresponding natural language data and making predictions using the generative artificial intelligence model; means for improving prediction accuracy using rigorous gradient descent; means for combining with an emotion engine that recognizes the user's emotions and adjusting the prediction results; and means for providing investment advice and risk management based on the user's emotional state. This makes it possible to provide the user with highly accurate predictions that take the user's emotional state into consideration, and reliable investment advice and risk management.

[1075] A "generative artificial intelligence model" is an artificial intelligence technology that uses past data and learning algorithms to generate new data and perform predictions and analysis.

[1076] "Time series data" is a series of data points collected over time, such as stock prices or currency fluctuations.

[1077] "Natural language data" refers to information written in languages ​​that humans use on a daily basis, including news articles and analytical reports.

[1078] "Gradient descent" is an algorithm used to optimize the parameters of a predictive model, which involves repeatedly processing a set of data to minimize error.

[1079] An "emotion engine" is software that analyzes a user's emotional state and recognizes emotions from voice tone, facial expressions, text input, etc.

[1080] "Investment advice" refers to advice that supports users in making appropriate investment decisions, including suggestions based on predictive data and the user's emotional state.

[1081] "Risk management" is the process of minimizing the risks associated with investment activities and proposing optimal investment strategies for users.

[1082] A "Transformer model" is an advanced machine learning model used in natural language processing, which analyzes large amounts of text data to understand meaning and make predictions.

[1083] "Prediction results" are information that indicates future data points and trends calculated by a generative artificial intelligence model.

[1084] "Retraining" is an additional learning process that reviews the parameter settings of the initial model and improves the accuracy of the model based on actual data.

[1085] A "prompt" refers to specific input data or questions used when analyzing natural language data or making predictions.

[1086] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. A detailed configuration for configuring the system of the present invention will be described below.

[1087] System configuration

[1088] It consists of three main elements: the server, the smartphone device, and the user.

[1089] server

[1090] The server uses the following hardware and software:

[1091] Hardware: High-performance CPU, GPU, storage, network interface

[1092] Software: TensorFlow, Transformers (Hugging Face), API, Database Management System

[1093] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. The collected data is then formatted and combined with natural language data such as news articles and analytical reports to create an integrated dataset. This dataset is then used to train a generative artificial intelligence model that combines a transformer model and an LSTM layer to make predictions.

[1094] The server also uses an emotion engine to analyze the user's emotions in real time, which analyzes voice tone, facial expressions, input text messages, etc. to recognize the user's emotions.

[1095] Terminal

[1096] The device used is a smartphone operated by the user. The application on the smartphone provides the following functions:

[1097] Input interface for historical exchange data and corresponding news articles

[1098] An interface for recognizing emotions from speech tone and facial expressions

[1099] Displaying prediction results returned from the server

[1100] Through the application, users input past data and news articles for the period they want to predict. The emotion engine obtains the user's current emotional data and sends it to the server. The server then uses the generative AI model and emotion engine to make predictions and returns the results to the device.

[1101] User

[1102] Users operate their smartphones to input data, have the system recognize their emotions, and then check the prediction results to make trading decisions.

[1103] Specific examples

[1104] The user inputs the past week's USD / JPY exchange rate data and news articles into the terminal. The emotion engine recognizes the user's current emotions from their voice tone and facial expressions, and when they request a USD / JPY forecast for the next week, the terminal sends the data to the server. The server makes a prediction and returns the results to the terminal, adjusting the reliability of the prediction based on the user's emotions. The terminal then displays the received prediction results to the user, who can then begin trading based on that information. Appropriate feedback is also displayed.

[1105] An example prompt is:

[1106] "Enter USD / JPY exchange rate data and news articles from the past week. Then, please tell us your current emotional state by voice."

[1107] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1108] Step 1:

[1109] The server collects historical exchange data and news articles from APIs and databases. It sends requests to collect exchange data and news articles and retrieves the respective data. The input is raw data obtained from the API, and the output is formatted time series data and news article data.

[1110] Step 2:

[1111] The server formats the collected data, associates the time series data with the natural language data, and creates an integrated dataset. The formatted data is then combined and saved as a single dataset. The input is the formatted time series data and news article data, and the output is an integrated dataset.

[1112] Step 3:

[1113] The server combines a Transformer model and an LSTM layer to build a generative AI model, and then trains the model using this integrated dataset. The input is the integrated dataset, and the output is the trained generative AI model. The model processes the data sequentially and improves its prediction accuracy through learning.

[1114] Step 4:

[1115] The server uses an emotion engine to recognize emotions from the user's voice tone and facial expressions. It analyzes voice and image data for emotion recognition and identifies the user's emotional state. The input is the user's voice and image data, and the output is the recognized emotional data.

[1116] Step 5:

[1117] The terminal provides an interface for receiving past exchange data and news articles entered by the user. The user enters data for the desired forecast period, and also collects voice tone and facial expression data. The input is exchange data, news articles, voice and image data from the user, and the output is the collected data.

[1118] Step 6:

[1119] The server receives data sent from the device and inputs it into a generative artificial intelligence model to predict the next exchange rate and stock price. The data is input into the model and a prediction result is generated. The input is data from the device and the output is the prediction result.

[1120] Step 7:

[1121] The server dynamically adjusts the reliability of predictions using the user's emotional data recognized by the emotion engine. The prediction results are revised based on the emotional data to improve reliability. The input is the recognized emotional data, and the output is the adjusted prediction results.

[1122] Step 8:

[1123] The server returns the final prediction result to the terminal, which transmits the prediction result and provides it to the user. The input is the adjusted prediction result, and the output is the data for the user to display.

[1124] Step 9:

[1125] The terminal receives the prediction results from the server and displays them to the user. The received data is displayed on the screen for the user to confirm. The input is the prediction result data from the server, and the output is the displayed prediction result.

[1126] Step 10:

[1127] The user checks the prediction results and emotional feedback displayed on the device and makes a trading decision. Based on the prediction results and feedback, the user makes an appropriate trade. The input is the prediction results and feedback displayed on the device, and the output is the user's trading decision.

[1128] 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.

[1129] 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.

[1130] 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.

[1131] [Fourth embodiment]

[1132] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1133] 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.

[1134] 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).

[1135] 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.

[1136] 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.

[1137] 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).

[1138] 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.

[1139] 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.

[1140] 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.

[1141] 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.

[1142] 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.

[1143] 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.

[1144] 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."

[1145] The present invention relates to a system for predicting stock prices and exchange rates using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[1146] System program and processing description

[1147] server

[1148] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The server then formats the data and creates an integrated dataset that combines the time-series data with corresponding news and related document information. This integrated dataset is used as input data for making predictions using a transformer model (e.g., a model for automatic learning) and a time-series prediction model (e.g., LSTM or RNN).

[1149] Next, the server builds a generative artificial intelligence model. Specifically, it loads a "Transformer model" and adds an "LSTM" layer to it to define it as an integrated neural network. It then trains this model using the integrated dataset. For training, it prepares batches of data using a data loader, inputs this data into the model, and generates predictions. It calculates an error based on the difference between the prediction results and the actual data, backpropagates this error to calculate the gradient, and updates the model parameters. This procedure is repeated over multiple epochs (learning cycles).

[1150] The server evaluates the model during the training process. For evaluation, it uses validation data and test data and calculates the difference between the prediction and the actual data as an indicator (e.g., MSE). If the model meets certain performance criteria, the server saves it and provides it to external parties as an API.

[1151] Terminal

[1152] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to forecast. This data is then sent to the server, which makes a forecast request.

[1153] The server uses the received data to input data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period. The prediction results are sent back to the terminal, which displays them to the user, who can use them as a reference when making trading and investment decisions.

[1154] Specific examples

[1155] User

[1156] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the terminal. Next, the user requests a "USD / JPY forecast for the next week." The terminal then sends this data to the server and makes a prediction request.

[1157] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. The forecast results are then sent back to the terminal. The terminal then displays the received forecast results to the user, who can then begin trading based on that information.

[1158] The above is a detailed description of an embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[1159] The processing flow will be explained below.

[1160] Step 1:

[1161] The server collects time-series data on stock prices and exchange rates, and retrieves past data from APIs and databases. For example, it retrieves and stores the past year's USD / JPY exchange rate data.

[1162] Step 2:

[1163] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[1164] Step 3:

[1165] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[1166] Step 4:

[1167] The server loads the Transformer model and adds and integrates an LSTM layer to build a generative artificial intelligence model.

[1168] Step 5:

[1169] The server splits the combined dataset into three parts: training, validation, and testing. For example, 80% is for training, 10% is for validation, and 10% is for testing.

[1170] Step 6:

[1171] The server prepares the batch data using a data loader and is configured to load the training data in fixed batch sizes.

[1172] Step 7:

[1173] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated for multiple epochs.

[1174] Step 8:

[1175] The server evaluates the performance of the trained model using the validation data, calculating the difference between the predictions and the actual data as a metric (e.g., MSE) to confirm the model's performance.

[1176] Step 9:

[1177] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[1178] Step 10:

[1179] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[1180] Step 11:

[1181] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[1182] Step 12:

[1183] The terminal transmits the data entered by the user to the server and makes a prediction request.

[1184] Step 13:

[1185] The server inputs the received data into a generative artificial intelligence model to predict exchange rates and stock prices for the next period.

[1186] Step 14:

[1187] The server returns the prediction result to the terminal.

[1188] Step 15:

[1189] The terminal displays the prediction results received from the server to the user, who can then make investment and trading decisions based on the prediction results.

[1190] Example 1

[1191] 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."

[1192] Conventional financial market forecasting systems have relied on a single data source, resulting in low forecast accuracy. Furthermore, the lack of a method for effectively linking natural language data and time series data limits the accuracy of forecasts. Furthermore, the training and evaluation methods for generative AI models have not been fully optimized, preventing the full potential of forecasting performance.

[1193] 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.

[1194] In this invention, the server includes a means for collecting time-series data of stock prices and exchange rates and natural language data, a means for creating an integrated data set by associating the collected time-series data with the natural language data, and a means for constructing a generative artificial intelligence model that combines a Transformer model and a time-series prediction model, thereby enabling highly accurate predictions.

[1195] "Time series data of stock prices and exchange rates" refers to numerical information on stock prices and exchange rates recorded at specific time intervals (e.g., daily, hourly, etc.).

[1196] "Natural language data" is text data written in the language that people use on a daily basis, such as news articles, analytical reports, and social media posts.

[1197] An "integrated dataset" is a single dataset that links time series data and natural language data based on date and time and relevance, and is used to train artificial intelligence models.

[1198] A "generative artificial intelligence model" is an artificial intelligence model designed to make predictions or generate data from input data, and specifically includes transformer models and time series prediction models.

[1199] A "Transformer model" is a type of machine learning for data, and is a neural network model used in particular for natural language processing.

[1200] A "time series prediction model" is an algorithm or model that uses time series data as input to predict future values, and specifically includes LSTM and RNN.

[1201] "Calculating gradients by backpropagating errors" is the process of calculating the gradients of the parameters of an artificial intelligence model using the difference (error) between the predicted value and the actual value, and updating the model parameters.

[1202] "Evaluating a model" is the process of measuring the performance of a trained generative artificial intelligence model using validation data and evaluating its predictive accuracy.

[1203] "Providing the prediction results to the user" means displaying the prediction values ​​generated by the generative artificial intelligence model to the user so that the user can make decisions based on that information.

[1204] The present invention relates to a system for predicting stock prices and exchange rates, and is a system that achieves highly accurate predictions using a generative artificial intelligence model. Detailed embodiments for implementing this system will be described below.

[1205] server

[1206] The server operates as follows: First, it uses the APIs and databases of financial data providers to collect time-series data on stock prices and exchange rates. It also uses news APIs to obtain natural language data such as related news articles and analytical reports. Specific data collection uses financial data APIs (e.g., Yahoo! Finance API, Alpha Vantage API) and news APIs.

[1207] The server then processes the collected time series data and natural language data, aligning them by date and time, to create an integrated dataset that links the time series data with the corresponding news articles. This integrated dataset is then used as input data for making predictions using transformer models (e.g., BERT models) and time series prediction models (e.g., LSTM, RNN).

[1208] The server then builds a hybrid neural network that combines a Transformer model with an LSTM layer and trains the model using the integrated dataset. In this invention, we use the Hugging Face Transformers library as the Transformer model. The training process involves using PyTorch to prepare the data loader and generate predictions. The error is calculated from the difference between the prediction result and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. This procedure is repeated multiple times.

[1209] During the training process, the server also evaluates the model. For evaluation, it uses a validation dataset and calculates the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error). If the model meets certain performance criteria, it saves it and provides it to external parties as an API.

[1210] Terminal

[1211] The terminal provides an interface that accepts input from the user. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. This data is sent to the server, which issues a prediction request. The server then uses this data to make a prediction using a generative artificial intelligence model and sends the prediction results back to the terminal. The terminal then displays the prediction results, which the user can use as a reference when making trading and investment decisions.

[1212] User

[1213] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into a terminal. Then, the user requests a "prediction of the USD / JPY for the next week." The terminal sends this data to the server and makes a prediction request. The server inputs the received data into a generative AI model, which predicts the USD / JPY exchange rate for the next week. The prediction results are sent back to the terminal, which displays them to the user. The user then makes trading decisions based on this information. For example, the following prompt sentences can be input into the generative AI model:

[1214] What is your forecast for the dollar-yen exchange rate for the next week?

[1215] Data and news articles from the past week are below.

[1216] Exchange data: [List of data]

[1217] News articles: [list of articles]

[1218] The above is a detailed description of an embodiment of the present invention. This system combines time-series data and natural language data, and uses a generative artificial intelligence model to achieve highly accurate predictions. Users can use these prediction results to make more appropriate trading and investment decisions.

[1219] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1220] Step 1: Data collection

[1221] server

[1222] The server collects stock and exchange time series data from financial data providers' APIs and databases, and also uses news APIs to retrieve natural language data such as related news articles and analytical reports.

[1223] input

[1224] Financial data API, news API.

[1225] output

[1226] Time series data and news article data.

[1227] Specific actions

[1228] The server calls a financial data API to retrieve stock prices and exchange rates for the past few months, and similarly calls a news API to retrieve relevant news articles.

[1229] Step 2: Data Refinement and Integration

[1230] server

[1231] The server then formats the collected time series data and natural language data by aligning the data by date and time, and creating an integrated dataset that links a series of time series data with the corresponding news articles.

[1232] input

[1233] Time series data, news article data.

[1234] output

[1235] Integrated dataset.

[1236] Specific actions

[1237] The server uses the Pandas library to convert the time series data and news article data into data frames, then merges them based on date and time to create a unified dataset.

[1238] Step 3: Model Building

[1239] server

[1240] The server builds a hybrid neural network that combines a Transformer model and an LSTM layer, using a common natural language processing library and a time series forecasting model.

[1241] input

[1242] Natural language processing library, time series forecasting library.

[1243] output

[1244] Hybrid neural network model.

[1245] Specific actions

[1246] The server loads a Transformer model and adds an LSTM layer to it to define a hybrid neural network.

[1247] Step 4: Model training

[1248] server

[1249] The server uses the combined dataset to train the hybrid neural network model, using a data loader to prepare batches of data, which are then fed into the model to generate predictions.

[1250] input

[1251] Integrated datasets, hybrid neural network models.

[1252] output

[1253] The trained neural network model.

[1254] Specific actions

[1255] The server uses PyTorch to prepare a data loader and input batches of data into the model. The error is calculated from the difference between the predicted results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters.

[1256] Step 5: Model evaluation

[1257] server

[1258] The server evaluates the model during the training process, using a validation dataset to calculate the difference between the predicted results and the actual values ​​as MSE (Mean Squared Error).

[1259] input

[1260] Validation dataset, trained neural network model.

[1261] output

[1262] Evaluation results (MSE).

[1263] Specific actions

[1264] The server uses the validation dataset to calculate the difference between the prediction and the actual value, and calculates the MSE to evaluate the model's predictive accuracy.

[1265] Step 6: Processing a prediction request

[1266] Terminal

[1267] The terminal receives input from the user about past exchange data and news articles, and sends this to the server. The server then uses this information to make predictions for the next period and sends the results back to the terminal.

[1268] input

[1269] User input data: historical exchange data, news articles.

[1270] output

[1271] Prediction results.

[1272] Specific actions

[1273] The terminal receives data from the user and sends it to the server via an HTTP request. The server inputs the received data into the model, performs exchange rate predictions for the next period, and sends the results back to the terminal.

[1274] Step 7: View the prediction results

[1275] Terminal

[1276] The terminal displays the prediction results returned from the server to the user, who then makes trading and investment decisions based on these prediction results.

[1277] input

[1278] Prediction results from the server.

[1279] output

[1280] The prediction results displayed on the screen.

[1281] Specific actions

[1282] The terminal displays the prediction results received from the server on the screen so that the user can easily check them. The user can then start trading based on this information.

[1283] (Application example 1)

[1284] 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."

[1285] Conventional systems do not provide sufficient information for investment decisions, particularly in terms of real-time forecasts and related information in financial markets. Furthermore, they are unable to provide personalized information tailored to individual user needs, creating a need for a means to efficiently provide information useful for users' investment decisions.

[1286] 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.

[1287] In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data and making predictions using the generative artificial intelligence model, a means for improving prediction accuracy using rigorous gradient descent, a means for providing information related to financial investments, and a means for providing a personalized notification function based on user interests, thereby enabling users to receive highly accurate forecast information in real time and efficiently obtain related information useful for their investment decisions.

[1288] A "generative artificial intelligence model" is an artificial intelligence model that combines time series data and natural language data to make predictions using a neural network.

[1289] "Time series data" refers to a continuous record of data collected at regular intervals, such as stock prices or currency exchange data.

[1290] "Natural language data" refers to text data written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[1291] "Exact gradient descent" is a computational technique that iteratively updates the model parameters to minimize error during the neural network training process.

[1292] "Information related to financial investment" refers to information to support investment decisions, including stock price forecasts, related news articles, and analytical reports.

[1293] "Personalized notification function" is a function that automatically delivers specific information based on the user's individual interests and concerns.

[1294] A "Transformer model" is a neural network model widely used in natural language processing, and is a model for understanding and generating language data.

[1295] A "time series prediction model" is a neural network model that uses time series data as input to predict future data points.

[1296] A "content distribution service" is a service that provides related information to users in real time.

[1297] This invention relates to a system that uses generative artificial intelligence models to predict stock prices and exchange rates, and provides information related to financial investments in real time. The system includes two main components, a server and a terminal, and provides an interface that is easy for users to use.

[1298] server

[1299] The server first collects time-series data on stock prices and exchange rates, and then acquires news articles and analytical reports as natural language data. This data is collected via APIs and databases. The collected data is then formatted through preprocessing to create an integrated dataset that combines the time-series data with corresponding news and related document information.

[1300] Next, the server builds a generative artificial intelligence model. Specifically, it loads a Transformer model and adds an LSTM layer to it to define it as an integrated neural network. This model is trained using the integrated dataset. A data loader is used to prepare batch data, which is then input into the model to generate predictions. An error is calculated based on the difference between the generated prediction results and the actual data, and this error is backpropagated to calculate the gradient and update the model parameters. The model is evaluated using validation and test data, and if it meets certain performance criteria, the server saves the model and provides it externally as an API.

[1301] Terminal

[1302] The device provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict into the device. This data is sent to the server, and a prediction request is made. The server uses the received data to input data into a generative artificial intelligence model, which predicts exchange rates and stock prices for the next period. The prediction results are sent back to the device, which displays them to the user. A personalized notification function based on the user's interests is also provided.

[1303] User

[1304] For example, a user inputs the exchange rate data for a specific currency from the past week and related news articles into their device. Next, the user requests a "currency forecast for the next week." The device sends this data to the server and makes a prediction request. The server inputs the received data into a generative artificial intelligence model and calculates a forecast for the next week's exchange rate. The prediction results are then sent back to the device, which displays them to the user. Personalized notifications are also provided based on the user's interests.

[1305] Usage examples and prompt statements

[1306] As a concrete example, if a user wants to get a forecast for the next week's exchange rate for a specific currency (e.g., the yen), the prompt text is as follows:

[1307] curl -X GET "https: / / api.example.com / stocks?ticker=USDJPY"

[1308] curl -X GET "https: / / api.example.com / news?ticker=USDJPY"

[1309] This prompt shows an API request to gather exchange and news data for a specific currency. This data can be used by the application to make predictions and provide the user with a currency forecast for the coming week.

[1310] The above is a specific embodiment for carrying out the present invention of this system.

[1311] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1312] Step 1:

[1313] The terminal receives input from the user, including past exchange data and corresponding news articles. This allows the terminal to obtain data for the period for which the user wishes to make a prediction. The input data is then sent to the server.

[1314] Step 2:

[1315] The server performs data processing based on the received exchange data and news articles. This process combines time series data and natural language data to create an integrated dataset. The processed data is then created.

[1316] Step 3:

[1317] The server uses the formatted data to input data into a generative artificial intelligence model. Specifically, predictions are made using a Transformer model and an LSTM layer. Data is input into the model, and prediction results are output.

[1318] Step 4:

[1319] The server calculates the difference between the predicted results and the actual data, and then retrains the generative AI model using rigorous gradient descent. This process improves the accuracy of predictions, and a retrained model is created.

[1320] Step 5:

[1321] The server then uses the trained model to make another prediction and returns the prediction results to the device, where they are displayed.

[1322] Step 6:

[1323] The device displays the received prediction results to the user, who can then make investment decisions based on this information.The device also provides personalized notifications based on the user's interests.

[1324] Step 7:

[1325] The user decides on the next investment action based on the displayed prediction results and notified information, and the user's investment action is executed.

[1326] 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.

[1327] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. Detailed embodiments for implementing this system are described below.

[1328] System program and processing description

[1329] server

[1330] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. For example, it retrieves and stores data on the dollar-yen exchange rate for the past year. It also collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[1331] The server then transforms the time series data and natural language data, matches them with corresponding news articles and analytical reports to create an integrated dataset, and loads a Transformer model, adding an LSTM layer to integrate them to build a generative artificial intelligence model.

[1332] The server uses this combined dataset to train the model, preparing batches of data using a data loader. The model generates predictions for each batch of data, calculates the error based on the difference between the predictions and the actual data, and updates the model parameters using gradient descent. This procedure is repeated over multiple epochs (learning cycles).

[1333] During the training process, the server evaluates the model's performance using validation data and finally using test data. Models that achieve satisfactory performance are saved and made public as an API.

[1334] The server also incorporates an emotion engine that recognizes the user's emotions by analyzing the user's facial expressions, voice tone, and text messages, and evaluates the user's emotions in real time.

[1335] Terminal

[1336] The terminal provides an interface for receiving input from the user. The user inputs historical exchange data and corresponding news articles for the period they wish to predict. In addition, the emotion engine obtains the user's current emotion data. This data is sent to the server, and a prediction request is made.

[1337] The server uses the received data to input it into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the prediction based on the user's emotional data recognized by the emotion engine, further improving the accuracy of the prediction.

[1338] The server sends the prediction results back to the device, which then displays them to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with peace of mind.

[1339] Specific examples

[1340] User

[1341] For example, a user inputs the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine then recognizes the user's current emotion from their tone of voice and facial expression. Next, the user requests a "USD / JPY forecast for the next week." The device then sends this data to the server and makes a prediction request.

[1342] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The server then sends the forecast results back to the terminal. The terminal then displays the received forecast results to the user, who then begins trading based on that information.

[1343] Furthermore, the device provides appropriate feedback based on the user's emotional data, making it easier for the user to understand the prediction results and use the device with confidence.

[1344] The above is a detailed embodiment of the system. This system combines time-series data and natural language data, and uses a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide more reliable information to users. Users can use these prediction results to make more appropriate trading and investment decisions.

[1345] The processing flow will be explained below.

[1346] Step 1:

[1347] The server collects time series data on stock prices and exchange rates, and retrieves and stores the past year's USD / JPY exchange rate data via API and database.

[1348] Step 2:

[1349] The server collects natural language data such as news articles and analytical reports. It uses news feeds and RSS readers to collect news articles for a specified period and stores them as text data.

[1350] Step 3:

[1351] The server formats the collected time series data and natural language data, and matches it with news articles and analytical reports that correspond to the time series data to create an integrated dataset.

[1352] Step 4:

[1353] The server loads the Transformer model and adds an LSTM layer to integrate it to build a generative artificial intelligence model. Specifically, the output of the Transformer model is input to the LSTM, and the output is used as a prediction value to define a generative artificial intelligence model.

[1354] Step 5:

[1355] The server splits the combined dataset into three parts: training, validation, and testing: 80% of the dataset is stored for training, 10% for validation, and 10% for testing.

[1356] Step 6:

[1357] The server prepares the batch data using a data loader. It uses PyTorch's DataLoader class and configures it to load training data in fixed batch sizes.

[1358] Step 7:

[1359] The server starts training the generative artificial intelligence model. For each batch of data, it generates a prediction, calculates the error based on the difference between the prediction and the actual data, and updates the model's parameters using gradient descent. This process is repeated over multiple epochs (learning cycles).

[1360] Step 8:

[1361] The server evaluates the performance of the trained model using the validation data. It calculates the difference between the predictions and the actual data as an evaluation metric (e.g., MSE) to confirm the model's performance.

[1362] Step 9:

[1363] If the model meets the predetermined performance criteria, the server uses the test data as a final evaluation to confirm the final performance.

[1364] Step 10:

[1365] The server saves the model that has achieved satisfactory performance and prepares it for publication as an API. It saves the model file and sets up an API endpoint to respond to external prediction requests.

[1366] Step 11:

[1367] The server incorporates an emotion engine, which analyzes the user's tone of voice, facial expressions, and input text messages to evaluate the user's emotions in real time.

[1368] Step 12:

[1369] Users use their terminal to request the latest exchange rate forecast. They input the past week's exchange rate data and corresponding news articles, and then request a forecast for the next period.

[1370] Step 13:

[1371] The terminal receives data input by the user and transmits it to the server. In addition, the terminal acquires the user's emotion data in real time and transmits it to the server.

[1372] Step 14:

[1373] The server inputs the received time series data and natural language data into a generative AI model to predict exchange rates and stock prices for the next period. It also dynamically adjusts the reliability of the predictions based on the user's emotional data recognized by the emotion engine.

[1374] Step 15:

[1375] The server returns the prediction result and adjusted reliability information to the terminal.

[1376] Step 16:

[1377] The device displays the prediction results and reliability information received from the server to the user. The device generates feedback based on the user's emotional data, ensuring that the user can use the prediction results with confidence.

[1378] These are the specific processing steps of the system that combines the emotion engine. Users can use the prediction results to make appropriate trading and investment decisions that take emotion into account.

[1379] Example 2

[1380] 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."

[1381] Current stock price and exchange rate prediction systems use time series data and limited natural language data, which limits their prediction accuracy. Furthermore, because they do not take into account the user's emotional state, users may lack confidence in or understand the predictions. This makes it difficult for users to make appropriate investment and trading decisions based on the prediction results.

[1382] The identification process by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes a generative artificial intelligence model, a means for combining time-series data with corresponding natural language data, a means for making a prediction using the generative artificial intelligence model, a means for improving prediction accuracy using strict gradient descent, a means for collecting user emotion data and analyzing it using an emotion engine, a means for dynamically adjusting the reliability of the prediction based on the analyzed emotion data, and a means for displaying the prediction result to the user. This enables highly accurate prediction using the generative artificial intelligence model, and by taking the user emotion data into consideration, the reliability of the prediction is further increased, allowing the user to use the prediction result with peace of mind.

[1383] A "generative artificial intelligence model" refers to an advanced machine learning algorithm that takes time series data or natural language data as input and performs predictive or generative tasks.

[1384] "Time series data" is a continuous set of data that changes over time, and includes numerical information collected at specific time intervals, such as stock prices or foreign exchange prices.

[1385] "Natural language data" refers to text information written in languages ​​that humans use on a daily basis, such as news articles and analytical reports.

[1386] "Gradient descent" refers to an algorithm that calculates the gradient of the error function and gradually updates the parameters based on that in order to optimize the parameters of a machine learning model.

[1387] An "emotion engine" is a technology for analyzing a user's emotional state, and takes input data such as facial expressions, voice tone, and text messages.

[1388] "Dynamic adjustment" means that the system automatically changes its settings and behavior in response to changing data and conditions in real time.

[1389] "Prediction reliability" is an indicator that shows the accuracy and credibility of the prediction results output by a generative artificial intelligence model.

[1390] A "Transformer model" is a type of neural network that can perform natural language processing tasks using large datasets, particularly those that utilize attention mechanisms.

[1391] An "LSTM layer" is a recurrent neural network layer with long short-term memory units, which can memorize long-term dependencies.

[1392] A "data loader" is a program or library for efficiently processing batches of data during model training.

[1393] An "epoch" is a unit of machine learning training that refers to passing the entire dataset through a model once.

[1394] This invention combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. An embodiment of this system will be described below.

[1395] server

[1396] The server runs on high-performance hardware and programming languages ​​such as Python. The server first collects time-series data on stock prices and exchange rates through an API. For example, it uses the Yahoo Finance API to obtain data on the dollar-yen exchange rate for the past year and stores it in a database. It also collects natural language data such as news articles and analytical reports using news feeds and RSS readers. This data is stored as text files.

[1397] Next, the server preprocesses the time series data and natural language data to create an integrated dataset. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data is also cleaned by removing unnecessary characters and tags, and then tokenized and vectorized. This creates an integrated dataset that combines time series data and natural language data.

[1398] Using the combined dataset, the server builds and trains a generative artificial intelligence model. It uses a model that combines a transformer model and an LSTM layer, and uses the TensorFlow library for training. It prepares batches of data using a data loader, and separates the data into training, validation, and test data. It calculates the gradient of the error function and updates the model parameters using gradient descent.

[1399] Furthermore, the server collects user emotional data and analyzes it using an emotion engine. It uses a camera and microphone to capture the user's facial expressions and voice tone, and analyzes emotions in real time. Based on this emotional data, the reliability of predictions is dynamically adjusted.

[1400] Terminal

[1401] The terminal provides a user interface and collects prediction requests and emotion data. The user inputs historical exchange rate data for the period they wish to predict and corresponding news articles into the terminal. The emotion engine analyzes the user's facial expressions and tone of voice to collect emotion data in real time. This data is then sent to the server, which then makes a prediction request to the server.

[1402] User

[1403] The user inputs, for example, the past week's USD / JPY exchange rate data and related news articles into the device. The emotion engine recognizes the user's current emotion from their tone of voice and facial expression, and transmits it to the server. The user inputs the following:

[1404] Based on the past week's USD / JPY exchange rate data and related news articles, please predict next week's USD / JPY exchange rate.

[1405] The server inputs the received data into a generative artificial intelligence model to calculate a forecast for the USD / JPY exchange rate for the next week. It also adjusts the reliability of the forecast based on the user's sentiment. The forecast results are sent back to the device, which displays them to the user. The user can then make investment and trading decisions based on the forecast results.

[1406] The above is an embodiment of the present invention. This system combines a generative artificial intelligence model and an emotion engine to improve prediction accuracy and provide users with reliable information. Users can use these prediction results to make more appropriate trading and investment decisions.

[1407] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1408] Step 1: Data collection

[1409] The server first uses an API (general term) to collect time-series data on stock prices and exchange rates. Specifically, to obtain, for example, the dollar-yen exchange rate data for the past year, the appropriate API endpoint is called and the data is downloaded in CSV format. This data is then stored in the server's database. Furthermore, news articles for a specified period are collected using a news feed or RSS reader and saved as text files.

[1410] Input: API endpoint, RSS feed URL

[1411] Output: Time series data (CSV), natural language data (text file)

[1412] Step 2: Data preprocessing and integration

[1413] The server formats the collected time series data and natural language data. The time series data is sorted by date, missing values ​​are imputed, and standardized. The natural language data undergoes cleaning (removal of unnecessary tags and characters), tokenization, and vectorization. The time series data and natural language data are then matched based on date to create an integrated dataset.

[1414] Input: time series data, natural language data

[1415] Output: Unified dataset

[1416] Step 3: Building the model

[1417] The server uses the combined dataset to build a generative artificial intelligence model, which combines a transformer model with an LSTM layer, and uses the TensorFlow library in Python to define, compile, and initialize the model.

[1418] Input: Unified dataset

[1419] Output: An initialized generative artificial intelligence model

[1420] Step 4: Train the model

[1421] The server uses a data loader to split the training data into batches and train the model. The model generates predictions on the training data and calculates the error based on the difference between the predictions and the actual data. To minimize the error, the model's parameters are updated using gradient descent. This process is repeated for multiple epochs.

[1422] Input: Initialized generative artificial intelligence model, training data

[1423] Output: A trained generative artificial intelligence model

[1424] Step 5: Evaluate and save the model

[1425] The server evaluates the model's performance using the validation and test data. It calculates evaluation metrics (e.g., RMSE and MAE) and checks whether the model meets the specified performance criteria. If it does, it saves the model to storage and exposes it as an API.

[1426] Input: A trained generative AI model, validation data, and test data

[1427] Output: Saved generative AI model, exposed API

[1428] Step 6: Collect and analyze emotion data

[1429] The device uses a camera and microphone to collect the user's emotional data. This allows the device to capture the user's facial expressions and vocal tone in real time and transmit them to the server. The server then uses an emotion engine to analyze this data and evaluate the user's emotional state.

[1430] Input: Facial expression data, voice tone data

[1431] Output: Parsed emotion data

[1432] Step 7: Request and make predictions

[1433] The device collects past data and news articles for the period the user wants to predict, and sends them along with emotion data to the server. The server then inputs this data into a generative AI model to predict stock prices and exchange rates for the next period. The reliability of the prediction is dynamically adjusted based on the user's emotion data.

[1434] Input: Historical data, news articles, sentiment data

[1435] Output: Prediction results

[1436] Step 8: Displaying prediction results and providing feedback

[1437] The server sends the prediction results to the device, which then displays them to the user. Furthermore, feedback is generated based on the user's emotional data, providing additional information to deepen the user's understanding of the prediction results. This allows the user to use the prediction results with confidence.

[1438] Input: Prediction results, analyzed emotion data

[1439] Output: Prediction results and feedback displayed to the user

[1440] These are the specific steps of the system's processing. The data input and output at each step, as well as the processing details, are clearly defined, creating a system that improves the accuracy of predictions as a whole.

[1441] (Application example 2)

[1442] 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."

[1443] Conventional stock price and exchange rate prediction systems using generative artificial intelligence models make predictions without taking into account the user's emotional state, and therefore are unable to alleviate the user's stress and anxiety. In addition, because the prediction results are not adjusted to take the user's emotions into account, there are issues with the accuracy of predictions and the reliability of investment advice.

[1444] The identification processing by the identification 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 combining a generative artificial intelligence model with time-series data and corresponding natural language data and making predictions using the generative artificial intelligence model; means for improving prediction accuracy using rigorous gradient descent; means for combining with an emotion engine that recognizes the user's emotions and adjusting the prediction results; and means for providing investment advice and risk management based on the user's emotional state. This makes it possible to provide the user with highly accurate predictions that take the user's emotional state into consideration, and reliable investment advice and risk management.

[1445] A "generative artificial intelligence model" is an artificial intelligence technology that uses past data and learning algorithms to generate new data and perform predictions and analysis.

[1446] "Time series data" is a series of data points collected over time, such as stock prices or currency fluctuations.

[1447] "Natural language data" refers to information written in languages ​​that humans use on a daily basis, including news articles and analytical reports.

[1448] "Gradient descent" is an algorithm used to optimize the parameters of a predictive model, which involves repeatedly processing a set of data to minimize error.

[1449] An "emotion engine" is software that analyzes a user's emotional state and recognizes emotions from voice tone, facial expressions, text input, etc.

[1450] "Investment advice" refers to advice that supports users in making appropriate investment decisions, including suggestions based on predictive data and the user's emotional state.

[1451] "Risk management" is the process of minimizing the risks associated with investment activities and proposing optimal investment strategies for users.

[1452] A "Transformer model" is an advanced machine learning model used in natural language processing, which analyzes large amounts of text data to understand meaning and make predictions.

[1453] "Prediction results" are information that indicates future data points and trends calculated by a generative artificial intelligence model.

[1454] "Retraining" is an additional learning process that reviews the parameter settings of the initial model and improves the accuracy of the model based on actual data.

[1455] A "prompt" refers to specific input data or questions used when analyzing natural language data or making predictions.

[1456] The present invention relates to a system that combines a system that uses a generative artificial intelligence model to predict stock prices and exchange rates with an emotion engine that recognizes user emotions. A detailed configuration for configuring the system of the present invention will be described below.

[1457] System configuration

[1458] It consists of three main elements: the server, the smartphone device, and the user.

[1459] server

[1460] The server uses the following hardware and software:

[1461] Hardware: High-performance CPU, GPU, storage, network interface

[1462] Software: TensorFlow, Transformers (Hugging Face), API, Database Management System

[1463] The server first collects time-series data on stock prices and exchange rates, and then retrieves past data from APIs and databases. The collected data is then formatted and combined with natural language data such as news articles and analytical reports to create an integrated dataset. This dataset is then used to train a generative artificial intelligence model that combines a transformer model and an LSTM layer to make predictions.

[1464] The server also uses an emotion engine to analyze the user's emotions in real time, which analyzes voice tone, facial expressions, input text messages, etc. to recognize the user's emotions.

[1465] Terminal

[1466] The device used is a smartphone operated by the user. The application on the smartphone provides the following functions:

[1467] Input interface for historical exchange data and corresponding news articles

[1468] An interface for recognizing emotions from speech tone and facial expressions

[1469] Displaying prediction results returned from the server

[1470] Through the application, users input past data and news articles for the period they want to predict. The emotion engine obtains the user's current emotional data and sends it to the server. The server then uses the generative AI model and emotion engine to make predictions and returns the results to the device.

[1471] User

[1472] Users operate their smartphones to input data, have the system recognize their emotions, and then check the prediction results to make trading decisions.

[1473] Specific examples

[1474] The user inputs the past week's USD / JPY exchange rate data and news articles into the terminal. The emotion engine recognizes the user's current emotions from their voice tone and facial expressions, and when they request a USD / JPY forecast for the next week, the terminal sends the data to the server. The server makes a prediction and returns the results to the terminal, adjusting the reliability of the prediction based on the user's emotions. The terminal then displays the received prediction results to the user, who can then begin trading based on that information. Appropriate feedback is also displayed.

[1475] An example prompt is:

[1476] "Enter USD / JPY exchange rate data and news articles from the past week. Then, please tell us your current emotional state by voice."

[1477] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1478] Step 1:

[1479] The server collects historical exchange data and news articles from APIs and databases. It sends requests to collect exchange data and news articles and retrieves the respective data. The input is raw data obtained from the API, and the output is formatted time series data and news article data.

[1480] Step 2:

[1481] The server formats the collected data, associates the time series data with the natural language data, and creates an integrated dataset. The formatted data is then combined and saved as a single dataset. The input is the formatted time series data and news article data, and the output is an integrated dataset.

[1482] Step 3:

[1483] The server combines a Transformer model and an LSTM layer to build a generative AI model, and then trains the model using this integrated dataset. The input is the integrated dataset, and the output is the trained generative AI model. The model processes the data sequentially and improves its prediction accuracy through learning.

[1484] Step 4:

[1485] The server uses an emotion engine to recognize emotions from the user's voice tone and facial expressions. It analyzes voice and image data for emotion recognition and identifies the user's emotional state. The input is the user's voice and image data, and the output is the recognized emotional data.

[1486] Step 5:

[1487] The terminal provides an interface for receiving past exchange data and news articles entered by the user. The user enters data for the desired forecast period, and also collects voice tone and facial expression data. The input is exchange data, news articles, voice and image data from the user, and the output is the collected data.

[1488] Step 6:

[1489] The server receives data sent from the device and inputs it into a generative artificial intelligence model to predict the next exchange rate and stock price. The data is input into the model and a prediction result is generated. The input is data from the device and the output is the prediction result.

[1490] Step 7:

[1491] The server dynamically adjusts the reliability of predictions using the user's emotional data recognized by the emotion engine. The prediction results are revised based on the emotional data to improve reliability. The input is the recognized emotional data, and the output is the adjusted prediction results.

[1492] Step 8:

[1493] The server returns the final prediction result to the terminal, which transmits the prediction result and provides it to the user. The input is the adjusted prediction result, and the output is the data for the user to display.

[1494] Step 9:

[1495] The terminal receives the prediction results from the server and displays them to the user. The received data is displayed on the screen for the user to confirm. The input is the prediction result data from the server, and the output is the displayed prediction result.

[1496] Step 10:

[1497] The user checks the prediction results and emotional feedback displayed on the device and makes a trading decision. Based on the prediction results and feedback, the user makes an appropriate trade. The input is the prediction results and feedback displayed on the device, and the output is the user's trading decision.

[1498] 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.

[1499] 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.

[1500] 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.

[1501] 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.

[1502] 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.

[1503] 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.

[1504] 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).

[1505] 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.

[1506] 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."

[1507] 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.

[1508] 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).

[1509] 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.

[1510] 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.

[1511] 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.

[1512] 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.

[1513] The hardware resource for executing a specific process can be any of the following processors: An example of a processor 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 example of a 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.

[1514] 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.

[1515] 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.

[1516] 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.

[1517] 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.

[1518] 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.

[1519] The following is further disclosed regarding the above embodiment.

[1520] (Claim 1)

[1521] A generative artificial intelligence model and

[1522] Combine time series data with corresponding natural language data,

[1523] A means for making predictions using the generative artificial intelligence model;

[1524] A means to improve prediction accuracy using rigorous gradient descent;

[1525] A system including:

[1526] (Claim 2)

[1527] Using the generative artificial intelligence model,

[1528] a means of making predictions using time series data and news and related document information as prompts;

[1529] a gradient descent retraining means for improving the model based on the difference between the predicted results and actual data;

[1530] 10. The system of claim 1, comprising:

[1531] (Claim 3)

[1532] The generative artificial intelligence model is

[1533] A method that combines a transformer model and a time series forecasting model;

[1534] receiving the time series data and natural language data as input;

[1535] A means of predicting the next time series data;

[1536] 10. The system of claim 1, comprising:

[1537] "Example 1"

[1538] (Claim 1)

[1539] A means of collecting time series data on stock prices and exchange rates and natural language data,

[1540] a means for associating the collected time series data with the natural language data to create an integrated dataset;

[1541] A means for building a generative artificial intelligence model that combines a transformer model and a time series prediction model;

[1542] means for training a generative artificial intelligence model using the integrated dataset;

[1543] a means for evaluating and retraining the model based on the difference between the predicted results and the actual data;

[1544] A means of predicting stock prices and exchange rates using trained models,

[1545] a means for providing the prediction result to a user;

[1546] A system including:

[1547] (Claim 2)

[1548] A means for receiving historical exchange data and news articles as input from a user and transmitting them to a server;

[1549] means for calculating a forecast for the next period using a generative artificial intelligence model by a server;

[1550] means for returning the prediction result to the terminal and displaying it to the user;

[1551] 10. The system of claim 1, comprising:

[1552] (Claim 3)

[1553] a means for preparing the integrated dataset as a batch of data and inputting it into a model to generate predictions;

[1554] A means for backpropagating errors in training the generative artificial intelligence model to calculate gradients and update parameters of the model;

[1555] Once a certain threshold is reached, the model can be saved and provided as an API.

[1556] 10. The system of claim 1, comprising:

[1557] "Application Example 1"

[1558] (Claim 1)

[1559] A generative artificial intelligence model and

[1560] Combine time series data with corresponding natural language data,

[1561] A means for making predictions using the generative artificial intelligence model;

[1562] A means to improve prediction accuracy using rigorous gradient descent;

[1563] a means for providing information relating to financial investments;

[1564] means for providing personalized notifications based on user interests;

[1565] A system including:

[1566] (Claim 2)

[1567] Using the generative artificial intelligence model,

[1568] a means of making predictions using time series data and news and related document information as prompts;

[1569] a gradient descent retraining means for improving the model based on the difference between the predicted results and actual data;

[1570] a means for distributing information relating to financial investments in real time;

[1571] 10. The system of claim 1, comprising:

[1572] (Claim 3)

[1573] The generative artificial intelligence model is

[1574] A method that combines a transformer model and a time series forecasting model;

[1575] receiving the time series data and natural language data as input;

[1576] A content distribution service that provides users with a means for predicting the next time series data;

[1577] 10. The system of claim 1, comprising:

[1578] "Example 2: Combining Emotion Engines"

[1579] (Claim 1)

[1580] A generative artificial intelligence model and

[1581] Combine time series data with corresponding natural language data,

[1582] A means for making predictions using the generative artificial intelligence model;

[1583] A means to improve prediction accuracy using rigorous gradient descent;

[1584] A means for collecting user emotion data and analyzing it using an emotion engine;

[1585] means for dynamically adjusting the reliability of the prediction based on the analyzed sentiment data;

[1586] means for displaying the prediction result to a user;

[1587] A system including:

[1588] (Claim 2)

[1589] Using the generative artificial intelligence model,

[1590] a means of making predictions using time series data and news and related document information as prompts;

[1591] a gradient descent retraining means for improving the model based on the difference between the predicted results and actual data;

[1592] means for analyzing user emotion data in real time and reflecting the data in the prediction;

[1593] 10. The system of claim 1, comprising:

[1594] (Claim 3)

[1595] The generative artificial intelligence model is

[1596] A method that combines a transformer model and a time series forecasting model;

[1597] receiving the time series data and natural language data as input;

[1598] A means of predicting the next time series data;

[1599] A means for improving the accuracy of predictions based on the results of user sentiment analysis;

[1600] 10. The system of claim 1, comprising:

[1601] "Application example 2 when combining emotion engines"

[1602] (Claim 1)

[1603] A generative artificial intelligence model and

[1604] Combine time series data with corresponding natural language data,

[1605] A means for making predictions using the generative artificial intelligence model;

[1606] A means to improve prediction accuracy using rigorous gradient descent;

[1607] A means for combining an emotion engine that recognizes the user's emotions and adjusts the prediction results;

[1608] a means for providing investment advice and risk management based on the emotional state of the user;

[1609] A system including:

[1610] (Claim 2)

[1611] Using the generative artificial intelligence model,

[1612] a means of making predictions using time series data and news and related document information as prompts;

[1613] a gradient descent retraining means for improving the model based on the difference between the predicted results and actual data;

[1614] A means for analyzing emotions from a user's tone of voice and facial expressions in real time;

[1615] 10. The system of claim 1, comprising:

[1616] (Claim 3)

[1617] The generative artificial intelligence model is

[1618] A method that combines a transformer model and a time series forecasting model;

[1619] receiving the time series data and natural language data as input;

[1620] A means of predicting the next time series data;

[1621] a means for adjusting a prediction result based on a result of the emotion engine recognizing the user's emotion;

[1622] 10. The system of claim 1, comprising: [Explanation of symbols]

[1623] 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. A generative artificial intelligence model and Combine time series data with corresponding natural language data, A means for making predictions using the generative artificial intelligence model; A means to improve prediction accuracy using rigorous gradient descent; A system including:

2. Using the generative artificial intelligence model, a means of making predictions using time series data and news and related document information as prompts; a gradient descent retraining means for improving the model based on the difference between the predicted results and actual data; The system of claim 1 , comprising:

3. The generative artificial intelligence model is A method that combines a transformer model and a time series forecasting model; receiving the time series data and natural language data as input; A means of predicting the next time series data; The system of claim 1 , comprising:

Images