Target prediction method and system

Abstract

A target prediction method for predicting the future prospects of a target performed by a computing device according to an embodiment of the present invention, characterized in that, after collecting relevant structured and unstructured data when a user requests target prediction, the relationship between the target and variables influencing the target is analyzed at a semantic level, and then the future target prospects are calculated.

Description

Technical Field

[0001] The present invention is a method and system for predicting a target based on relationship information between a target influence variable and a target at a semantic level.

Background Art

[0002] Recently, with the emergence of pre-trained language models (LLMs) for large-scale general domain data, various tasks that were conventionally processed manually have been replaced by artificial intelligence infrastructure.

[0003] In particular, the technology of using large language models to perform future prediction tasks is an interesting and rapidly developing technical field, which is called futurecasting.

[0004] Futurecasting means using sophisticated algorithms to predict future trends, events, or actions, which can be applied to various fields, from predicting weather patterns to predicting market trends and even social or political changes.

[0005] Machine learning, which is a subset of artificial intelligence, can play a very important role in this process. In this futurecasting, several main aspects and technologies that can be replaced by machine learning models are as follows.

[0006] It is possible to perform the task of predicting future events using past data. A machine learning model can identify patterns in a large dataset and use them for prediction. For example, in the financial field, such a model can predict the trends of the stock market, and in the medical field, it can predict the occurrence of diseases.

[0007] Time series analysis is possible in forecasting fields such as meteorology, economics, and resource management. Machine learning models can analyze data points collected at consecutive time intervals to predict future points in a series.

[0008] In future casting, natural language processing (NLP) can be used to analyze news, social media, and other text data to measure public opinion or predict political or social trends.

[0009] However, the accuracy of future casting models can be limited by the quality and quantity of data, and they face particular difficulties in predicting numerical values ​​using both structured and unstructured data simultaneously. [Overview of the project] [Problems that the invention aims to solve]

[0010] This invention aims to propose a target prediction method and system that utilizes language models to accurately predict the short-term and / or medium- to long-term prospects of a target based on data on various variables that can influence the target.

[0011] In detail, the target prediction method and system of the present invention can detect target influence variables related to the target at a semantic level, accurately extract features that influence the target based on the detected target influence variables at the semantic level, and predict the prospects for the target.

[0012] Furthermore, the present invention can provide a method and system for accurately predicting a target based on structured and unstructured data for the target and target influencing variables.

[0013] Furthermore, the target prediction method and system of the present invention aims to predict the outlook for targets in the medium to long term and beyond, based on various text data such as news and reports that serve as the basis for predicting targets and target influencing variables.

[0014] Furthermore, the present invention aims to propose a target prediction method and system that clearly presents the basis for predicting the prospects of a target based on target influencing variables and their associated characteristics.

[0015] In detail, the target prediction method and system of the present invention can analyze target influence variables that affect the target at a semantic level and provide justification for the projected target based on the causal relationship between the characteristics of the target influence variables and the target. [Means for solving the problem]

[0016] A target prediction method for predicting the future prospects of a target performed by a processor of a computing device according to an embodiment of the present invention, comprising the steps of: receiving a target prediction request from a user; determining target prediction elements to be predicted from the received target prediction request; searching for a target prospect report for the determined target prediction elements and generating semantic-level relationship information between target-target influence variables based on the searched target prospect report; filtering out unstructured data, including structured data of features related to the target influence variables and unstructured data of text documents related to the target influence variables, based on the target influence variables of the generated relationship information; calculating future target prospects based on the filtered structured and unstructured data; generating the basis for the calculated target prospects as relationship information at the feature level; and providing the user with the calculated target prospects and the relationship information at the feature level.

[0017] In this case, the step of receiving a target prediction request from the user may include the steps of providing the user with a chat interface and receiving text containing the target prediction request, and analyzing the received text in a context-based manner to detect the context indicating the target prediction request.

[0018] Furthermore, the step of determining the target prediction elements may further include NamedEntityRecognition of the text containing the target prediction request to determine the target keywords, total forecast period, and forecast unit period as the target prediction elements.

[0019] In this case, the step of determining the target prediction element may further include, if multiple target keywords of a higher concept and multiple target keywords of a lower concept are recognized for the target of the target prediction element, the step of listing the recognized multiple target keywords and providing them for the user to select.

[0020] Furthermore, the step of generating the relational information may include the step of defining target influence variables that affect the target at a semantic level, and the step of generating a causal relationship graph as relational information, in which the names of each defined target influence variable are used as node names.

[0021] In this case, the step of generating the relational information may further include the step of indicating the sequence of events between the target influencing variables represented by each node in the causal relationship graph using arrows.

[0022] Furthermore, the step of filtering out unstructured data from the structured data of the features and text documents related to the target influence variables may include the steps of classifying the features stored in the data store as target influence variables defined at the semantic level, and combining the structured data of the features classified as target influence variables to generate a structured dataset.

[0023] Furthermore, the step of filtering out unstructured data from the structured data of the features and text documents related to the target influence variables may include the step of inputting the text documents into a document classification prompt template and determining, via a language model, whether or not the text documents are documents that influence the target.

[0024] In this case, the step of calculating the future target outlook may include: detecting a target outlook report that predicts the target outlook from the text documents; performing sentiment analysis on the sentences predicting the target in the target outlook report via a language model to classify the target outlook values ​​as positive, neutral, or negative for each target outlook report; quantifying the level of the classified sentiment tone and returning it to predictive scoring data; and arranging the predictive scoring data of the target outlook report in chronological order to generate quantitative data.

[0025] Furthermore, the step of calculating the future target outlook may include the steps of combining the standard data and the quantified data to generate an integrated structured dataset, and inputting the generated integrated structured dataset into a predictive model to output target outlook values.

[0026] Further, the step of calculating the future target outlook may further include the step of adjusting the target outlook value based on the relationship information between the target-target influence variables.

[0027] Also, the step of generating the relationship information at the feature level may include the step of generating, as relationship information, the features of the target influence variables that are the basis for predicting the target outlook value.

[0028] Also, the step of generating the relationship information at the feature level may include the step of generating the relationship information including the numerical values of the features that affected the target outlook value predicted at a certain point.

[0029] Further, when a prediction environment change input is received from the user, it may further include the step of performing a simulation according to the received prediction environment change input.

[0030] At this time, the step of performing a simulation according to the received prediction environment change input includes, when there is a change in the features of the target influence variable, the step of changing the standard data of the features according to the changed target influence variable, and re-executing the process of interpreting the target outlook value and the basis based on the changed standard data and non-standard data, and outputting the target outlook value and the basis information according to the virtual simulation.

[0031] Further, the step of performing a simulation according to the received prediction environment change input may further include, when a specific event occurrence is received from the user as a prediction environment change input, the step of detecting a case similar to the occurrence of the specific event and calculating a target outlook value based on the detected case.

[0032] A server computing system that receives a target prediction request from a user computing device according to an embodiment of the present invention and executes a target prediction task, comprising: a data store for storing target prediction-related data; a memory for storing instructions and data for performing the target prediction task; and at least one processor that performs the target prediction task according to the instructions and data in the memory, wherein the at least one processor receives a target prediction request from a user, determines the target prediction elements to be predicted in the received target prediction request, searches for a target outlook report relating to the target of the determined target prediction elements, generates semantic-level relationship information between target-target influence variables based on the retrieved target outlook report, filters out structured data of features related to the target influence variables and unstructured data of text documents related to the target influence variables based on the target influence variables of the generated relationship information, calculates a future target outlook based on the filtered structured and unstructured data, generates the basis for the calculated target outlook as relationship information at the feature level, and provides the user with the calculated target outlook and the relationship information at the feature level. [Effects of the Invention]

[0033] The target prediction method and system according to the present invention can accurately predict target prospects by predicting target prospects based on prediction basis data obtained by precisely performing the data preparation necessary for target prediction.

[0034] Specifically, the target prediction method and system according to the present invention can define target influence variables that affect the target at a semantic level, and then precisely filter the structured and unstructured data related to the target influence variables.

[0035] Furthermore, the target prediction method and system according to the present invention can integrate the filtered structured data and unstructured data in this manner and accurately predict target prospect values ​​through a time series prediction model.

[0036] Specifically, the target prediction method and system according to the present invention can quantify unstructured data, accurately filter out only the features related to the target influencing variable from structured data, and then generate an integrated structured dataset, thereby generating a structured dataset necessary for time series target prediction.

[0037] Furthermore, the target prediction method and system according to the present invention can enhance the reliability of the target outlook by interpreting the basis for the predicted target outlook value and providing supporting data that forms the basis of the prediction.

[0038] Furthermore, the target prediction method and system according to the present invention have the advantage of being able to respond to a variety of user requirements by performing simulations in response to changes in the user's target prediction environment. [Brief explanation of the drawing]

[0039] [Figure 1] This figure shows an example block diagram of a computing system that performs a target prediction method according to an embodiment of the present invention. [Figure 2] This figure shows an example block diagram of a computing device, which is one of the components of a computing system that performs a target prediction method according to an embodiment of the present invention. [Figure 3] This figure shows an example of a block diagram of another aspect of a computing device, which is one of the components of a computing system that performs a target prediction method according to an embodiment of the present invention. [Figure 4] This is a flowchart illustrating a method for predicting the prospects of a target through a machine learning model according to an embodiment of the present invention. [Figure 5]This invention presents a meta-architecture for performing a method for predicting the prospects of a target according to embodiments of the present invention. [Figure 6] This is an example of the process of performing a target prediction task according to an embodiment of the present invention and determining causal relationship information with the target predictor variable. [Figure 7] This is an example of the process of performing data preparation based on relationship information between the target and the target influence variable according to an embodiment of the present invention. [Figure 8] This is an example of the process of converting unstructured data into quantitative data according to an embodiment of the present invention. [Figure 9] This is an example of a process for calculating target prospects by integrating standardized data and quantitative data according to embodiments of the present invention. [Figure 10] This is an example of the process for deriving the basis for target prospects according to embodiments of the present invention, and the process for predicting additional target prospects through user simulation. [Figure 11] This is an example chart of the target outlook predicted by embodiments of the present invention. [Figure 12] This is an example of a causal relationship graph presented as supporting evidence for the target outlook predicted by embodiments of the present invention. [Figure 13] This is another example of a causal relationship graph presented as supporting evidence for the target outlook predicted by embodiments of the present invention. [Modes for carrying out the invention]

[0040] The present invention is capable of various modifications and may have multiple embodiments, but specific embodiments will be illustrated in the drawings and described in detail in the detailed description. The effects and features of the present invention, as well as the methods for achieving them, should become clear when referring to the embodiments described in detail below in conjunction with the drawings. However, the present invention is not limited to the embodiments disclosed below and can be embodied in a variety of forms. In the following embodiments, terms such as "first," "second," etc., are used not in a restrictive sense but to distinguish one component from another. Also, singular expressions include plural expressions unless the context clearly indicates otherwise. Furthermore, terms such as "includes" or "has" mean that the features or components described in the specification exist, and do not preclude the possibility that one or more other features or components may be added.

[0041] Figure 1 shows an example block diagram of a computing system that performs a target prediction method according to an embodiment of the present invention.

[0042] Referring to Figure 1, a computing system 1000 for target prediction according to one embodiment of the present invention includes a user computing device 110, a training computing system 150, and a server computing system 130, each device and system being communicated via a network 170.

[0043] According to embodiments of the present invention, 1) a user computing device 110 can perform a target prediction method by utilizing local and / or external machine learning models 120, or by utilizing a machine learning model 140 provided by a server.

[0044] Furthermore, according to another embodiment of the present invention, 2) a server computing system 130 communicating with a user computing device 110 can provide the user computing device 110 with a target prediction service via an application and / or on the web in response to a user request through the user computing device 110.

[0045] Furthermore, according to another embodiment of the present invention, 3) the user computing device 110 and the server computing system 130 can perform at least a part of the method for performing target prediction in conjunction with each other to provide a target prediction service to the user.

[0046] Furthermore, according to various embodiments of the present invention, the user computing device 110 and / or the server computing system 130 can learn machine learning models 120 / 140 for target prediction through interaction with a training computing system 150 that is communicatively connected via a network 170. In this case, the training computing system 150 may be separate from the server computing system 130 or may be part of the server computing system 130.

[0047] In some embodiments, the training computing system 150 may be part of the server computing system 130, or part of the user computing device 110.

[0048] The following description assumes that the user computing device 110 connects to the server computing system 130 to perform a target prediction task, the server computing system 130 collects and analyzes the data necessary for target prediction either directly or using a language model from another server, and the user computing device 110 performs target outlook prediction based on the collected and analyzed data. However, if any part of the process described as being performed in the server computing system 130 is performed in the user computing device 110, it is understood that this is naturally included in the description of the present invention.

[0049] The user computing device 110 may include all other types of computing devices, such as smartphones, mobile phones, digital broadcasting devices, PDAs (personal digital assistants), PMPs (portable multimedia players), desktops, wearable devices, embedded computing devices, and / or tablet PCs.

[0050] Such a user computing device 110 includes at least one processor 111 and memory 112. Here, the processor 111 may consist of at least one or more electrically connected processors such as a central processing unit (CPU), graphics processing unit (GPU), ASICs (application-specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field-programmable gate arrays), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0051] The memory 112 may include one or more non-temporary / temporary computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, and combinations thereof, and may include web storage of a server that performs memory storage functions over the internet. Such memory 112 can store the data and instructions necessary for the operation of the application for the at least one or more processors 111 to perform target prediction.

[0052] In one embodiment, the user computing device 110 can store at least one or more machine learning models 120. For example, the user computing device 110 may consist of a variety of machine learning models, such as multiple neural networks (e.g., deep neural networks) that make predictions about a target based on routine / quantitative data, or other types of machine learning models including nonlinear and / or linear models, or a combination thereof.

[0053] For example, predictive models can include linear regression, decision trees, random forests, gradient-boosted pre-trained language models, and / or deep learning models. Neural networks can include at least one of the following: feed-forward neural networks, cyclic neural networks (e.g., short-long-memory cyclic neural networks), convolutional neural networks, and / or other forms of neural networks.

[0054] Furthermore, the user computing device 110 can store the model used in each process and the prompt templates that form the basis of the input to the model, in order to perform at least a portion of the processes performed for target prediction through a large-scale language model (LLM).

[0055] For example, the user computing device 110 can store: 1) prompts for generating queries from user input; 2) prompts for determining causal relationships between targets and target influencing variables; 3) prompts for identifying raw data related to the determined causal relationships; and 4) prompt templates for quantifying unstructured data.

[0056] In other words, in one embodiment, the user computing device 110 can request some execution steps in the target prediction task from a language model on an external server via prompts or the like, and perform target prediction based on the received data.

[0057] In another embodiment, a target prediction task requested through a user computing device 110 can be performed by a server computing system 130 performing target predictions through at least one or more machine learning models 140 and machine learning models on other servers, and then providing the predicted data to the user computing device 110.

[0058] Such a user computing device 110 may include at least one input component 121 that detects user input. For example, the user input component 121 may include a touch sensor (e.g., a touchscreen and / or touchpad) that detects touch from the user's input medium (e.g., a finger or stylus), an image sensor that detects the user's motion input, a microphone that detects the user's voice input, a button, a mouse and / or keyboard, etc. Furthermore, if the user input component 121 receives input to an external controller (e.g., a mouse, keyboard, etc.) via an interface, it may also include an interface and an external controller.

[0059] The server computing system 130 includes at least one processor 131 and memory 132. Here, the processor 131 can consist of at least one or more electrically connected processors from among central processing units (CPUs), graphics processing units (GPUs), ASICs (application-specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field-programmable gate arrays), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0060] The memory 132 may include one or more non-temporary / temporary computer-readable storage media and combinations thereof, such as RAM, ROM, EEPROM, EPROM, flash memory devices, and magnetic disks. Such a memory 132 can store data and instructions necessary for the processor 131 to work through the language model of the server computing system 130 and / or the language model of an external server, such as prompt templates, machine learning models 140 for future casting, and so on.

[0061] For example, the server computing system 130 may include a neural network and / or other multilayer nonlinear model as a machine learning model 140 for future casting. Exemplary neural networks may include feedforward neural networks, deep neural networks, circular neural networks, and convolutional neural networks.

[0062] In one embodiment, the server computing system 130 can be embodied by including at least one computing device. For example, the server computing system 130 can embody multiple computing devices to operate according to a sequential computing architecture, a parallel computing architecture, or a combination thereof. The server computing system 130 can also include multiple computing devices connected by a network.

[0063] In one embodiment, the server computing system 130 may further include a data store computing system 1000 (hereinafter referred to as "data store"), which is storage for continuously storing and managing raw data that forms the basis of future casting for a target. Such a data store can include various forms of data storage, ranging from file systems to cloud storage.

[0064] For example, a data store may include at least one database from among relational databases that use a structured query language (SQL) to define and manipulate data, NoSQL databases designed for flexibility and scalability to process unstructured and semi-structured data, data warehouses used for reporting and data analysis that centralize large amounts of data from multiple sources and are optimized for querying and analysis, data warehouses that store large amounts of raw data in basic forms such as structured data, semi-structured data, and unstructured data, and local storage devices or NAS (Network Attached Storage) that store data in files in a format that can generally be accessed by computer operating systems.

[0065] The training computing system 150 includes at least one processor 151 and a memory 152. Here, the processor 151 can consist of at least one or more electrically connected processors from among central processing units (CPUs), graphics processing units (GPUs), ASICs (application-specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field-programmable gate arrays), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions. The memory 152 can include one or more non-temporary / temporary computer-readable storage media and combinations thereof, such as RAM, ROM, EEPROM, EPROM, flash memory devices, and magnetic disks. Such a memory 152 can store the data and instructions necessary for the processor 151 to train the future casting model.

[0066] For example, the training computing system 150 may include a model trainer 160 that trains machine learning models stored in the user computing device 110 and / or the server computing system 130 using various training or learning techniques, such as error backpropagation.

[0067] For example, the model trainer 160 can backpropagate the updating of one or more parameters of a machine learning model for future casting based on a defined loss function.

[0068] In some concrete examples, error backpropagation may include truncated backpropagation through time. The model trainer 160 can perform numerous generalization techniques (e.g., weight reduction, dropout, knowledge distillation, etc.) to improve the generalization ability of the fusion casting model being trained.

[0069] The model trainer 160 includes computer logic used to provide the desired functionality. The model trainer 160 can be embodied by hardware, firmware, and / or software that control a general-purpose processor. For example, in one embodiment, the model trainer 160 includes a program file stored in a storage device, which is loaded into memory and can be executed by one or more processors. In other embodiments, the model trainer 160 includes one or more sets of computer-executable instructions stored in a tangible computer-readable storage medium such as a RAM hard disk or an optical or magnetic medium.

[0070] Network 170 includes, but is not limited to, 3GPP (3rd Generation Partnership Project) networks, LTE (Long Term Evolution) networks, WiMAX (World Interoperability for Microwave Access) networks, the Internet, LAN (Local Area Network), Wireless LAN, WAN (Wide Area Network), PAN (Personal Area Network), Bluetooth® networks, satellite broadcasting networks, analog broadcasting networks, and / or DMB (Digital Multimedia Broadcasting) networks.

[0071] In general, communication over network 170 can be conducted using any type of wired and / or wireless connection, via a variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), and / or protection schemes (e.g., VPN, Secure HTTP, SSL).

[0072] Figure 2 shows an example of a block diagram of a computing device, which is one of the components of a computing system 1000 that performs a target prediction method according to an embodiment of the present invention.

[0073] Referring to Figure 2, the computing device 100 included in the user computing device 110, server computing system 130, and training computing system 150 contains multiple applications (e.g., application 1 to application N). Each application may include a machine learning library. For example, an application may include a futurecasting application, a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, a chatbot application, another futurecasting application, and so on.

[0074] In one embodiment, the computing device 100 may include a model trainer 160 for training a future casting model, and can store and operate the future casting model to perform a target prediction task on input data.

[0075] Each application of the computing device 100 can communicate with several other components of the computing device, such as one or more sensors, a context manager, a device status component, and / or additional components. In one embodiment, each application can communicate with each device component using an API (e.g., a public API). In one embodiment, the API used by each application may be specific to that application.

[0076] Figure 3 shows an example block diagram of another aspect of a computing device, which is one of the components of a computing system 1000 that performs a target prediction method according to an embodiment of the present invention.

[0077] Referring to Figure 3, the computing device 200 includes multiple applications (e.g., Application 1 to Application N). Each application can communicate with the central intelligence layer. For example, applications may include image processing applications, text messaging applications, email applications, dictation applications, virtual keyboard applications, browser applications, and so on. In one embodiment, each application can communicate with the central intelligence layer (and the models stored therein) using an API (e.g., a common API across all applications).

[0078] The central intelligence layer can include prompts using multiple machine learning models and / or language models. For example, as shown in Figure 3, at least a portion of each machine learning model can be provided to each application and managed by the central intelligence layer. In other embodiments, two or more applications may share a single machine learning model. For example, in some embodiments, the central intelligence layer may provide a single model to all applications. In some embodiments, the central intelligence layer may be contained within the operating system of the computing device 200, or may be implemented differently.

[0079] The central intelligence layer can communicate with the central device data layer. The central device data layer may be a centralized data storage for the computing device 200. As shown in Figure 3, the central device data layer can communicate with many other components of the computing device, such as one or more sensors, a context manager, a device state component, and / or additional components. In some embodiments, the central device data layer can communicate with each device component using an API (e.g., a private API).

[0080] The technologies described herein may refer to servers, databases, software applications, and other computer-based systems, as well as the actions performed and the information transmitted to or from those systems. The inherent flexibility of computer-based systems will be recognized as allowing for a wide range of possible configurations, combinations, and divisions of work, and inter- and inter-component functionality. For example, the processes described herein may be implemented using a single device or component, or multiple devices or components operating in combination. Databases and applications may be implemented by a single system or by systems distributed across multiple systems. Distributed components may operate sequentially or in parallel.

[0081] The following describes a target prediction method and system in which such a computing system 1000 collects raw data using a language model, analyzes the collected raw data to predict the prospects of a target, and provides causal relationship information that forms the basis of the prospect prediction, with reference to Figures 4 to 13.

[0082] First, the computing system 1000 can receive a target prediction request from the user computing device 110 and perform a target prediction task in response to the received target prediction request. (S101)

[0083] In one embodiment, the user computing device 110 can receive text-based target prediction requests from the user via a chat interface and send the text containing the target prediction requests to the server computing system 130 to perform the target prediction task on the server computing system 130.

[0084] The server computing system 130 can perform a target prediction task by detecting pre-stored phrases for target prediction requests from text entered through the chat interface, or by detecting the context of the target prediction request by analyzing the text in a context-based manner.

[0085] The server computing system 130 can then recognize the text containing the target prediction request and determine the target prediction elements for target prediction.

[0086] Here, the target prediction element includes the target to be predicted, and may further include at least one of the prediction length and the prediction unit time.

[0087] In this embodiment, the target includes information on a numerical value that changes over time, and predicting the target includes predicting and calculating the target value in the future, with the prediction unit period as the period, up to the total forecast period.

[0088] Specifically, the server computing system 130 analyzes the text of the target prediction request and inserts it into a query generation prompt template that determines the target prediction elements, inputs it into the language model, and returns at least one of the target prediction elements as output to the language model, thereby determining the target prediction elements.

[0089] For example, a query generation prompt template can be configured to take "text of target prediction request" as input in the interactive prediction request field, recognize values ​​corresponding to the target, total outlook period, and unit period based on object name recognition (NamedEntityRecognition, NER), and return the query's target, total outlook period, and unit period as output values.

[0090] As a more specific example, when a user inputs the target prediction request text "Predict what will happen to lithium prices for the next 12 months, on a monthly basis," the server computing system 130 can determine the target prediction elements in the following way: <Input: Interactive prediction request "Predict what will happen to lithium prices for the next 12 months, on a monthly basis," Operation: Through NER, recognize the values ​​corresponding to the target, total forecast period, and unit period for the input text, generate and return a query like the following, Output: Query - {Target:, Unit Period:, Total Forecast Period:}>> as a prompt to the language model, and output the target prediction elements as {Target: Lithium market price, Unit Period: Monthly, Total Forecast Period: 12 months}.

[0091] In this case, if the target prediction elements are not specified or are abstract, the server computing system 130 can provide another future casting interface for inputting target prediction elements for target prediction, and can send the target prediction elements input through the provided future casting interface to the server computing system 130 to execute the target prediction task. That is, if the target is classified by category from a higher-level concept to multiple lower-level concepts, the server computing system 130 can list target keywords that map to the higher and lower-level concepts and provide them for the user to select.

[0092] For example, the future casting interface provides target keywords derived through object name recognition, sequentially from higher-level concepts to lower-level concepts, allowing the user to select and more accurately determine the target they are trying to predict.

[0093] Once the target prediction elements are determined, the server computing system 130 can determine the relationship information between the target and the target influence variables. (S103)

[0094] First, the server computing system 130 can collect target analysis data for the target. This can be done by filtering data in the data store within the server computing system 130 or by crawling data that exists on the internet.

[0095] For example, the server computing system 130 can perform a keyword search based on keywords representing a determined target to detect target analysis data. Here, target analysis data may be target analysis reports related to the target.

[0096] Specifically, the server computing system 130 can request, based on a pre-configured target analysis report collection prompt template in the language model, to search for analytical materials relevant to the target using the target's keyword base and return them through an analysis report.

[0097] More specifically, the server computing system 130 can obtain a target analysis report as output by using a target analysis report collection prompt template, with the following settings: <<Input: Target - Lithium market price, Operation: Search for and return analysis reports with titles relevant to the target through keyword search>>.

[0098] Furthermore, the server computing system 130 can detect target influence variables that affect the target from the collected target analysis data, and analyze and generate relationship information between the target influence variables and the target.

[0099] In this embodiment, the relational information may include information about target influencing variables that affect the future prediction of the target, and information about the relationship between the target influencing variables and the target.

[0100] More specifically, information regarding target influence variables refers to information that defines the target influence variables at a semantic level, while information regarding the relationship between targets and target influence variables can refer to the causal relationships and weights of influence between targets and target influence variables, as well as the relative influences and weights of those influences.

[0101] In the following, we will refer to the relationship information between the target and the target influencing variable as causal relationship information.

[0102] In this embodiment, the server computing system 130 can generate causal relationship information by analyzing a semantic causal graph at the semantic level as target-target influence variable relationship information based on the collected target analysis data.

[0103] To this end, in one embodiment, the server computing system 130 can perform topic-relevant term recognition on the target analysis data to detect and annotate target influence variables related to the target.

[0104] The server computing system 130 can then input the data into a causal graph generation model, which has been trained to generate a causal relationship graph between targets and target influencing variables based on target analysis data annotated with targets and target influencing variables, thereby generating a causal relationship graph at a semantic level.

[0105] Here, the causal relationship graph between the target and the target influence variable can include information defining the target and the target influence variable at the semantic level, along with the node name, in each node.

[0106] For example, information used to determine the target and target influencing variables at a semantic level may include additional annotations such as the name, keywords, source, domain, region, location, and characteristics of the element.

[0107] Furthermore, the causal relationship graph between target and target influence variables can include information via arrows about the causal relationship regarding whether the mutual influence between each node (target and target influence variable) precedes or follows.

[0108] In one embodiment, the server computing system 130 can collect target analysis data in a context-based manner and output causal relationship information of target-target influencing variables based on the collected target analysis data, all through a RAG (Retrieval Augmented Generation) model.

[0109] Through the process of generating causal information according to such embodiments, target influencing variables can be clearly identified, defined at a semantic level by concepts, categories, subjects, and / or specific criteria, and the context and domain related to the target influencing variables at a semantic level can be accurately determined.

[0110] Then, by annotating the target influencing variables with this defined information and using it later for semantic-level data preparation, it becomes possible to accurately identify the raw data necessary for target prediction.

[0111] Once the causal relationship information between the target and target influencing variables is determined, the server computing system 130 can perform data preparation based on the determined causal relationship information (S105).

[0112] First, the server computing system 130 can collect raw data related to the target of causal information and the target influence variables for prospect prediction of the target.

[0113] In this embodiment, the server computing system 130 can collect unstructured data (e.g., news articles and analysis reports consisting of text) and structured data related to the target and target influencing variables through keyword searches or the like that represent the target and target influencing variables, and store the collected raw data in a data store.

[0114] The server computing system 130 can then determine whether the raw data stored in the data store is related to the target influencing variable at a semantic level (DocumentIdentification) and extract the relevant data. At this time, the raw data can be filtered based on whether or not it matches the semantic definition included in the aforementioned target-target influencing variable, thereby obtaining the basic data necessary for prediction of the target.

[0115] For example, the server computing system 130 takes a document to be judged as input and outputs the semantic level target influencing variable and its relationship as an operation, thereby extracting predictive base data from the raw data that is semantically related to the target and target influencing variable.

[0116] To identify data related to target influence variables that affect such targets, historical data analysis knowledge and domain expertise in the target-related fields are crucial.

[0117] To complement this, the server computing system 130 can derive events relevant to the target and events that are not relevant through a language model.

[0118] For example, the server computing system 130 can instruct a language model through related / unrelated event creation prompts that include phrases instructing it to act as a domain expert on the target, to return a number of related events that affect the target at a semantic level, and a number of unrelated events that have no effect or affect it below a certain threshold.

[0119] Specifically, the relevant / unrelated event creation prompt can include information defining each target influence variable at a semantic level, instructing the language model to distinguish between relevant and unrelated events that affect the target from the underlying prediction data.

[0120] The server computing system 130 then creates document identification prompts through the returned related / unrelated events to classify and identify predictive basis data from the raw data, and based on the created document identification prompts, requests the language model to classify the raw data, thereby accurately extracting predictive basis data related to the target and target influence variables.

[0121] Furthermore, unstructured data related to the target's outlook can be detected from predictive background data related to the target and / or target influencing variables. In other words, the server computing system 130 can classify documents related to the target and / or target influencing variables from the raw data stored in the data store, and detect relevant events and / or texts that affect the target from the documents.

[0122] For example, a document classification prompt may consist of: 1) instructions for the target to act as an expert; 2) inputting at least one document contained in the raw data to be identified as input data; 3) instructions for selecting one of either related event options related to the target's predictions or unrelated event options that do not affect the target; and 5) instructions for adding related events that affect the target to the related event options, or unrelated events that do not affect the target to the unrelated event options.

[0123] As a specific example, if the element to be predicted is "lithium production", then Server Computing System 130 will: <<1) Become a lithium expert. 2) Input: [Documents] 3) Classify the [Documents] related to increases or decreases in lithium production. You have two options for your answer: - Option 1: Highly relevant (list of relevant events), - Option 2: Not relevant (list of unrelated events), 4) First, describe how the information provided in relation to lithium production increases or decreases and why. Then, place the option number on the last line. The prompt consisting of >> will allow you to identify whether the documents in the raw data are related to "lithium production".

[0124] In other words, the server computing system 130 can collect raw data related to the target and / or target influencing variables, classify predictive basis data related to the target and / or target influencing variables from the raw data, determine relevant events and sentences that affect the target outlook from the classified predictive basis data, and filter sentences and related events related to the target outlook from the raw data as unstructured data.

[0125] Next, the server computing system 130 can use a language model to identify and classify whether each feature stored in the data store belongs to a relevant target semantic variable, and generate a structured dataset consisting of standardized data for the relevant features. Here, a feature refers to an attribute of data stored in a standardized data format, representing a variety of elements that can influence the target's perspective, and may include, for example, CSV, Excel file, and / or databasetable.

[0126] For example, if the target is lithium price, then the target influencing variables are variables that have a causal relationship with lithium price, such as "spodumene, lithium mines, lithium salt lakes, lithium carbonate, lithium hydroxide, and lithium batteries." The characteristic feature is that these are structured data belonging to the target influencing variables that influence the outlook for the target, such as "Australia's spodumene production, Australia's spodumene exports, Chile's lithium hydroxide production, Chile's lithium hydroxide exports, China's spodumene imports, China's lithium carbonate imports, China's lithium carbonate production, China's lithium carbonate sales, lithium battery efficiency (km / wh), China's electric vehicle sales, and China's electric vehicle subsidy program."

[0127] In other words, in the embodiment, the target influencing variable may be a specific concept, topic, or category that influences the target outlook, and the feature may mean an attribute of structured data in data storage related to the target influencing variable.

[0128] The server computing system 130 can then filter the data store features to identify relevant features related to the target influencing variable, and integrate the filtered features to generate a structured dataset (StructuredData).

[0129] To explain the process of generating a structured dataset in detail, first, the server computing system 130 can list the features that can be used in the data store as feature names. Then, it can list the description for each feature.

[0130] The server computing system 130 can then filter the listed features based on their relationship to target influence variables, which are those that can influence the target, based on their semantic level definition.

[0131] To this end, the server computing system 130 can utilize machine learning models and language models to classify the relationships between features and target influencing variables.

[0132] In this embodiment, the server computing system 130 lists the feature names and descriptions of the data store, inputs the keywords of the target influencing variables of the causal relationship information into a word embedding model, and maps the features classified to each target influencing variable by detecting the feature names related to the keywords of each target influencing variable according to their feature relevance. Here, word embedding refers to a model trained to classify features relevant to semantic target influencing variables based on their feature names and descriptions.

[0133] The server computing system 130 can then retrieve tabular data corresponding to the names of the classified features from the data store, process the retrieved tabular data through data organization and preprocessing, and arrange it in a structured format to make it suitable for input into target predictive modeling, thereby generating it in a time-series structured data format (e.g., CSV, Excel).

[0134] In this way, the server computing system 130 can collect accurate raw data that forms the basis of target prediction based on causal relationship information between target and target influencing variables, and can precisely filter the collected raw data into structured and unstructured data necessary for target prediction and utilize it as input data for target prediction modeling.

[0135] Next, the server computing system 130 can quantify unstructured data (Text Processing for Forecasting) to generate quantitative data (S107).

[0136] First, the server computing system 130 can generate predictive scoring data by scoring the target prospect values ​​for each target prospect report among documents classified as unstructured data.

[0137] In detail, in the embodiment, the server computing system 130 inputs each target prospect report into a language model, performs sentiment analysis on the relevant sentences classified as predictive target prospects, classifies the target prospects into positive, neutral, and negative, and operates according to a target prospect scoring prompt that returns a numerical value for the tone level, thereby generating predictive scoring data as quantitative data arranged in chronological order.

[0138] Specifically, the target outlook scoring prompt can be configured to, upon inputting a target outlook report (or related text pre-extracted from the target outlook report), classify the opinion on the target outlook from the input text into positive, neutral, or negative, and select a tone for the outlook opinion from the input text within a predetermined level range.

[0139] Furthermore, the server computing system 130 can generate an event list based on relevant events that affect the prospects of the target detected from the document during unstructured data filtering.

[0140] For example, the server computing system 130 can generate quantitative data in the form of an event list that quantifies the date of occurrence of events affecting the target outlook, related characteristics, the value of the related characteristics, and the impact and influence on the target outlook.

[0141] Furthermore, the server computing system 130 can encode each document classified as unstructured data into a latent vector through the language model's encoder and return embedded metrics. Specifically, the server computing system 130 can use the language model to encode documents into latent vectors and obtain an embedded matrix that captures the semantic essence of each document.

[0142] In detail, the server computing system 130 can input unstructured data, such as news articles, into a language model encoder to generate a document embedding matrix for modeling themes that are prevalent in each document. The document embedding thus generated can be used with algorithms such as LDA (LatentDirichletAllocation) to identify broad themes within the documents, thereby highlighting themes (variables, features) that can influence the target's future prospects.

[0143] Thereafter, the server computing system 130 can predict the target outlook based on the generated structured dataset and quantitative data (S109).

[0144] In detail, the server computing system 130 can calculate target predicted values ​​for each forecast unit period during the total outlook period, based on the quantitative data and structured datasets, etc.

[0145] Therefore, the server computing system 130 can combine (concatenate) a structured dataset generated based on standardized data with a quantitative dataset generated based on unstructured data to generate an integrated structured dataset.

[0146] Specifically, the server computing system 130 can first categorize data according to its influence on the target, assign weights to the data, and then combine them. For example, the server computing system 130 can classify the features included in the structured dataset into macro variables if they influence the target above a certain threshold, and into micro variables if they influence it below a certain threshold. Then, after matching the classified macro variables with the quantitative data over time, the server computing system 130 can integrate them into a single macro time-series structured dataset, and can also integrate the data classified as micro variables into a single micro time-series structured dataset.

[0147] In other words, in this embodiment, the event list and predictive scoring data can be matched and combined according to the time-series flow of the structured dataset to generate an integrated structured dataset that includes all information from both structured and unstructured data.

[0148] The server computing system 130 can then input the generated integrated structured dataset into a predictive model to calculate target predicted values ​​for each predictive unit period over the total prospect period. Here, the predictive model may include linear regression, decision trees, random forests, gradient boosting, deep learning models, and / or pre-trained language models.

[0149] In one embodiment, the server computing system 130 can be further inputted with causal information at the semantic level into a prediction model, which can then be guided to predict the target prospect based on the causal information.

[0150] Furthermore, in this embodiment, the server computing system 130 can input the embedded metrics into a second prediction model that predicts target prospects based on the embedded metrics, thereby reflecting unstructured target prediction information that is not present in the structured data into the predicted values.

[0151] In a specific embodiment, the server computing system 130 can input an integrated structured dataset into a first prediction model to first calculate a first target outlook value.

[0152] The server computing system 130 can then adjust the first target prospect value based on a semantic causal relationship graph to calculate a second target prospect value that reflects the causal relationship information between the target influence variable and the target.

[0153] Finally, the server computing system 130 can calibrate the calculated second target prospect value based on unstructured target prediction information to ultimately calculate the final target prospect value.

[0154] Furthermore, the server computing system 130 can perform a rationale interpretation for the target outlook based on causal relationship information and structured datasets, and generate rationale information (S111).

[0155] In detail, referring to Figure 10, the server computing system 130 can interpret the rationale for the final target prospect value and output rationale information based on semantic-level causal information and structured datasets.

[0156] Specifically, the server computing system 130 can generate historical causal relationship graphs at the feature level based on existing target values ​​from the past relative to the present, structured datasets, and semantic causal relationship graphs from a structured dataset.

[0157] Furthermore, the server computing system 130 can generate future causal relationship graphs at the feature level based on a data-driven causal discovery model (Data-driven Causal Discovery) that has learned from past causal relationship graphs, using the present as a baseline, future final target prospect values, a structured dataset, and semantic causal relationship graphs.

[0158] Furthermore, the server computing system 130 can map the target prospect value to provide a future causal relationship graph, which can then be used as supporting information to explain how various characteristics influence the target prospect value and to what extent, thus determining how the target prospect value was calculated.

[0159] For example, as shown in Figure 11, the server computing system 130 can provide a target outlook graph, which represents the target outlook values ​​calculated for each forecast unit period, through the user computing device 110 during the total outlook period.

[0160] Furthermore, as shown in Figure 12, the server computing system 130 can provide the user computing device 110 with a causal relationship graph at the feature level, which interprets the basis for the prediction of the target prospect value, as supporting information.

[0161] In particular, as shown in Figure 13, the server computing system 130 can display specific numerical values ​​of the features that influenced the predicted target outlook at a particular prediction point in time, further improving user confidence in the target outlook.

[0162] Finally, after providing the calculated final target prospect value and supporting information, the server computing system 130 receives input from the user for a change in the predicted environment. In response to the input change in the predicted environment, it can perform a what-if simulation again and provide the target prospect value and supporting information for the changed environment (S113).

[0163] Specifically, referring to Figure 10, the user can input changes to the forecasting environment by modifying the characteristics of target influencing variables that affect the target outlook value, or by inputting the occurrence of a specific event, through the user computing device 110.

[0164] In this embodiment, if there is a change in the target influencing variables, the server computing system 130 modifies the integrated structured dataset according to the changed target influencing variables, then performs the process of interpreting the target prospect values ​​and rationale again, and if necessary, outputs the target prospect values ​​and rationale information obtained from the simulation and provides them to the user computing device 110.

[0165] In another embodiment, the server computing system 130 can receive input for predicted environmental changes due to the occurrence of a specific event. In such a case, if the server computing system 130 can quantitatively reflect the occurrence of a specific event from the event list, it can calculate the changed quantitative data, modify the integrated structured dataset again based on this, and then re-execute the process of interpreting the target prospect value and its rationale, and if necessary, output the target prospect value and rationale information obtained from the simulation and provide it to the user computing device 110.

[0166] Furthermore, while the detailed description of the present invention has been provided with reference to preferred embodiments, a person skilled in the art or with ordinary knowledge in the art will understand that the present invention can be modified and altered in various ways without departing from the spirit and technical scope of the invention as described in the claims below. Therefore, the technical scope of the present invention is not limited to what is described in the detailed description of the specification, but must be defined by the claims. [Industrial applicability]

[0167] The present invention is a method and system for predicting future prospects for a target by analyzing structured and unstructured data at a semantic level, and therefore has industrial applicability.

Claims

1. A target prediction method that predicts the future prospects of a target performed by a computing device, The steps include receiving a target prediction request from the user, The steps include determining the target prediction element to be predicted from the received target prediction request, The steps include: searching for target prospect reports for the target of the determined target prediction elements, and generating semantic-level relationship information between target-target influence variables based on the searched target prospect reports; The steps include filtering out unstructured data from feature data related to the target influence variable and unstructured data from text documents related to the target influence variable based on the target influence variable of the generated relational information, The steps include: calculating future target prospects based on the filtered standardized and unstandardized data; The steps include generating the basis for the calculated target outlook as relational information at the feature level, The step includes providing the user with relationship information between the calculated target outlook and the feature level, Target prediction method.

2. The step of receiving a target prediction request from the user is: The steps include providing the user with a chat interface and receiving text including the target prediction request, The process includes the step of analyzing the received text in a context-based manner to detect a context indicating a target prediction request. The target prediction method according to claim 1.

3. The step of determining the target prediction element is: The process further includes the step of performing NamedEntityRecognition on the text containing the target prediction request to determine the target keywords, total forecast period, and prediction unit period as the target prediction elements. The target prediction method according to claim 2.

4. The step of determining the target prediction element is: If multiple target keywords, both higher-level and lower-level, are recognized for the target of the target prediction element, the method further includes the step of listing the recognized multiple target keywords and providing them for the user to select. The target prediction method according to claim 3.

5. The step of generating the aforementioned relational information is: The steps include defining target influence variables that affect the aforementioned target at a semantic level, The step includes generating a causal relationship graph as relational information, where the names of each defined target influence variable are used as node names. The target prediction method according to claim 1.

6. The step of generating the aforementioned relational information is: The step further includes indicating the temporal relationships between the target influencing variables represented by each node in the aforementioned causal relationship graph using arrows, The target prediction method according to claim 5.

7. The step of filtering out unstructured data from the structured data of the feature and the text documents related to the target influencing variable is: The steps include classifying the features stored in the data store as target influence variables defined at the semantic level, The process includes the step of combining standardized data of features classified as the target influence variables to generate a structured dataset. The target prediction method according to claim 1.

8. The step of filtering out unstructured data from the structured data of the aforementioned feature and the text documents related to the target influencing variable is: The process includes the step of inputting the text document into a document classification prompt template and determining, via a language model, whether or not the text document is a document that affects the target. The target prediction method according to claim 1.

9. The step of calculating the aforementioned future target outlook is: The steps include: detecting a target outlook report from the aforementioned text documents that predicts the prospects of the target; The steps include: performing sentiment analysis (sentimental analysis) on the sentences predicting the target in the target outlook report using a language model, and classifying the target outlook values ​​for each target outlook report as positive, neutral, or negative; The steps include quantifying the level of the classified tone of sensibilities and returning it to the predictive scoring data, The process includes the step of arranging the predictive scoring data from the aforementioned target outlook report in chronological order and generating it as quantifiable data. The target prediction method according to claim 8.

10. The step of calculating the aforementioned future target outlook is: The steps include: combining the standard data and the quantified data to generate an integrated structured dataset; The process includes the step of inputting the generated integrated structured dataset into a predictive model and outputting target prospect values. The target prediction method according to claim 9.

11. The step of calculating the aforementioned future target outlook is: The step further includes adjusting the target prospect value based on the relationship information between the target and target influence variables, The target prediction method according to claim 10.

12. The step of generating relational information at the feature level is: This includes a step of generating relational information on the characteristics of target influencing variables that serve as the basis for predicting the aforementioned target prospect value. The target prediction method according to claim 1.

13. The step of generating relational information at the feature level is: The step of generating the relational information includes the numerical values ​​of features that influenced the predicted target prospect value at a given point in time, The target prediction method according to claim 12.

14. When a predicted environmental change input is received from the user, the method further includes the step of performing a simulation in response to the received predicted environmental change input. The target prediction method according to claim 1.

15. The step of performing a simulation in response to the received predicted environmental change input is: If there is a change in the characteristics of the target influence variable, the step is to modify the characteristic template data according to the changed target influence variable. The process includes re-executing the process of interpreting target prospect values ​​and rationale based on the modified standard and non-standard data, and outputting target prospect values ​​and rationale information corresponding to the virtual simulation. The target prediction method according to claim 14.

16. The step of performing a simulation in response to the received predicted environmental change input is: When the occurrence of a specific event is received from the user as a predicted environmental change input, the system further includes the steps of detecting cases similar to the occurrence of the specific event and calculating a target prospect value based on the detected cases. The target prediction method according to claim 1.

17. A server computing system that receives a target prediction request from a user computing device and performs a target prediction task, A data store that stores target prediction-related data, A memory for storing instruction words and data for performing the aforementioned target prediction task, Includes at least one processor that performs the target prediction task according to the instruction words and data in the memory, The at least one of the aforementioned processors is We receive a target prediction request from the user. Determine the target prediction element to be predicted in the received target prediction request, The system searches for target prospect reports for the targets of the determined target prediction elements, and generates semantic-level relationship information between target-target influence variables based on the retrieved target prospect reports. Based on the target influence variables of the generated relational information, the structured data of features related to the target influence variables and the unstructured data of text documents related to the target influence variables are filtered. Based on the filtered standardized and unstandardized data, future target prospects are calculated. The basis for the calculated target outlook is generated as relational information at the feature level, The relationship information between the calculated target outlook and the feature level is provided to the user. Target prediction system.

Images