Personalized credit management system and method using dynamic causal inference and time-series forecasting

Abstract

According to one embodiment of the present disclosure, a personalized credit management method performed in an electronic device comprising a processor comprises: (a) receiving a user’s financial goal and time-series financial data over a plurality of points in time of the user; (b) generating a dynamic causal graph representing causal relationships between financial variables based on the received time-series financial data and updating the dynamic causal graph periodically or based on detected events; (c) generating sequential improvement path scenarios composed of a plurality of actions based on the set financial goal and the dynamic causal graph; (d) predicting a financial state trajectory at a future point in time for each of the generated improvement path scenarios to identify an optimal improvement path in which the probability of achieving the financial goal satisfies a preset criterion; (e) generating a natural language explanation report including the identified optimal improvement path and causal grounds inferred from the dynamic causal graph in which each action of the path contributes to the achievement of the goal; and (f) providing the natural language explanation report to the user.

Description

Technology Field

[0001] The present disclosure relates to Personalized Credit Management technology, and in particular to an electronic device and method that automatically generate and explain an optimal course of action for achieving a user's financial goals through analysis based on Dynamic Causal Inference and Time-Series Forecasting. More specifically, it applies to a technology that analyzes time-series financial data to learn dynamic causal relationships between financial variables and utilizes this to provide personalized credit improvement strategies and natural language-based explanatory reports. Background Technology

[0002] Existing credit evaluation models have operated by calculating a user's credit score, credit rating, and loan approval status based on numerical input information. These models have traditionally relied on statistical approaches, such as regression analysis, and have had limitations in interpreting complex financial characteristics using simplified formulas.

[0003] Recently, with the advancement of AI technology, various financial institutions are utilizing machine learning or AI-based models to conduct loan evaluations. While these AI-based models offer excellent predictive and classification capabilities by extensively utilizing both structured and unstructured data, they present a challenge in that the prediction results are difficult to interpret intuitively due to their complex algorithmic structures.

[0004] In particular, general users find it difficult to grasp a clear explanation of why a particular evaluation result was reached and to identify directions for improvement. To address this, Explainable AI (XAI) technology is being partially introduced; however, in most cases, it is limited to analyzing the causes of loan results and has limitations in that it fails to provide specific actionable advice that can actually improve loan conditions.

[0005] Therefore, there is a growing need for an automated analysis and explanation system that goes beyond merely explaining the results of credit information and can present specific paths and measures for users to improve their credit status.

[0007] Meanwhile, prior art patents (e.g., KR 10-2025-0067371, etc.) contribute to some extent to providing explanations regarding credit evaluation results, but there are structural and technical limitations as follows to achieve the purpose of the present disclosure.

[0008] First, since prior art relies on statistical correlation to observe only changes in outcomes resulting from changes in specific input values, it is difficult to identify the causal structure regarding the causes behind outcomes such as loan approval or changes in credit scores. In other words, there is a possibility of relying on spurious correlations, where observed patterns in the data appear to be causal even when they are not. Consequently, it is difficult to provide a fundamental causal explanation for questions such as "why a specific change affects the probability of approval."

[0009] Second, prior art remains limited to static analysis based on information from a single point in time, making it difficult to incorporate dynamic changes in the external financial environment—such as interest rate fluctuations, regulatory shifts, and economic factors—into its models. Due to these limitations, it cannot adequately reflect the actual financial environment where relationships between financial variables change over time, which may consequently reduce the accuracy and effectiveness of the explanation.

[0010] Third, prior art tends to present only point-wise counterfactuals. For example, it offers only individual advice, such as "slightly changing a specific value increases the likelihood of approval," and fails to present a strategic path outlining the sequential actions users should take to achieve long-term goals. In other words, it is not suitable for actual financial decision-making processes that require multi-step action plans.

[0011] Fourth, prior art models have a fixed structure, so they lack the ability to update knowledge or rearrange causal structures on their own, even when new financial environment changes or unexpected data patterns occur. As a result, the models are likely to gradually diverge from reality over time, and the reliability of their explanations may also decrease.

[0012] To address these issues, the present disclosure utilizes a dynamic causal graph based on causation to quantitatively model the causal structure among financial variables and is configured to periodically update said structure in response to changes in the financial environment. Furthermore, it generates multi-step improvement paths for achieving long-term financial goals and evaluates the causal contribution of each path to goal achievement, thereby providing a strategic roadmap that users can actually follow. Moreover, based on an adaptive model equipped with self-adaptive capabilities, the present disclosure can maintain the stability and reliability of the explanation by automatically updating the causal graph even when new financial data or external changes occur. Prior art literature

[0013] KR 10-2025-0029843 A: Method and apparatus for predicting a user's financial activity (2025.03.05) KR 10-2835558 B1: Method for generating user-customized alarm information for financial transactions and providing an alarm service based on user setting information (2025.07.14) The problem to be solved

[0014] The purpose of the present disclosure is to provide an electronic device and a method capable of analyzing a user's time-series financial data and suggesting a path for improving credit information by reflecting causal relationships between financial variables.

[0015] More specifically, the present disclosure aims to provide an electronic device and a method that intuitively explain, in natural language, the optimal credit improvement strategy that a user can follow by generating behavioral scenarios for achieving financial goals and predicting the effectiveness of each scenario.

[0016] In addition, the present disclosure aims to provide an electronic device that enables a personalized credit management service by transmitting the natural language description report to a user. means of solving the problem

[0017] According to one embodiment of the present disclosure for solving the above problem, a personalized credit management method performed in an electronic device including a processor comprises: (a) receiving a user’s financial goals and time-series financial data over multiple time points of the user; (b) generating a dynamic causal graph representing causal relationships between financial variables based on the received time-series financial data, and updating the dynamic causal graph periodically or based on detected events; (c) generating sequential improvement path scenarios composed of multiple actions based on the set financial goals and the dynamic causal graph; (d) predicting a financial state trajectory at a future time point for each of the generated improvement path scenarios to identify an optimal improvement path in which the probability of achieving the financial goals satisfies a preset criterion; (e) generating a natural language description report including the identified optimal improvement path and the causal basis for each action of the path contributing to the achievement of the goals, inferred from the dynamic causal graph. A personalized credit management method is provided, characterized by including the step of (f) providing the above natural language description report to a user.

[0018] According to one embodiment of the present disclosure, the dynamic causal graph of step (b) can quantitatively model the causal effect of an intervention on a specific financial variable on another variable.

[0019] According to one embodiment of the present disclosure, the trajectory prediction of step (d) can be performed by combining the dynamic causal graph and the time series prediction model.

[0020] According to one embodiment of the present disclosure, the natural language description report of step (e) may further include reliability information indicating the uncertainty of the predicted trajectory.

[0021] According to one embodiment of the present disclosure, the natural language description report may be generated by inputting a prompt including the causal graph and trajectory prediction results into a large-scale language model (LLM).

[0022] According to one embodiment of the present disclosure, the financial goal includes at least one of a numerical-based goal or a categorical goal, and different improvement path generation logic may be applied depending on the goal type.

[0023] According to another embodiment of the present disclosure, a computer-readable recording medium is provided for storing a program for executing each of the above methods.

[0024] According to another form of the present disclosure, a personalized credit management system is provided, comprising: a memory; and a processor connected to the memory and configured to (a) receive a user’s financial goals and time-series financial data over multiple points in time of the user, (b) generate a dynamic causal graph representing causal relationships between financial variables based on the received time-series financial data and update the causal graph periodically or based on detected events, (c) generate sequential improvement path scenarios composed of multiple actions based on the set financial goals and the dynamic causal graph, (d) predict a financial state trajectory at a future point in time for each of the generated improvement path scenarios to identify an optimal improvement path in which the probability of achieving the financial goals meets a criterion, and (e) generate a natural language description report including causal grounds inferred from the dynamic causal graph in which each action included in the identified optimal improvement path contributes to the achievement of the goals. Effects of the invention

[0025] According to embodiments of the present disclosure, a personalized credit management system and method can be provided that dynamically model the causal relationships between financial variables based on a user's time-series financial data and calculate an optimal improvement path to achieve a set financial goal.

[0026] In particular, the present disclosure goes beyond existing statistics-based explanation methods that merely suggest "what value should be changed (What)" and can provide causal grounds for "why the corresponding action is effective in achieving the goal (Why)" in the form of a natural language report.

[0027] As a result, users can receive causally justified and actionable financial strategies rather than fragmentary advice, significantly enhancing the credibility and transparency of the explanations.

[0028] Furthermore, the Dynamic Causal Graph of the present disclosure can be updated in real time in response to changes in the external financial environment, such as interest rates, regulations, and market conditions, thereby ensuring adaptability and robustness that existing static models cannot provide. This structure enables the generation of stable and up-to-date credit and loan advice, even in real-world situations where the financial environment is constantly changing.

[0029] Furthermore, the system of the present disclosure provides a self-evolving architecture that automatically updates its causal structure by reflecting the user's actual behavior, feedback, new data, etc.

[0030] Accordingly, the performance of the model continuously improves over the long term, and it can provide highly personalized financial strategies that accurately reflect the individual characteristics and financial patterns of users.

[0031] Therefore, the present disclosure possesses high technical value as a next-generation credit management technology by providing technical effects not found in prior art, such as deepened explainability, enhanced personalized feasibility, and the ability to actively respond to environmental changes. Brief explanation of the drawing

[0032] FIG. 1 is a block diagram of an electronic device according to various embodiments of the present disclosure. FIG. 2 is a flowchart illustrating an embodiment of a personalized credit management method according to the present disclosure. FIG. 3 is a drawing showing an electronic device displaying information including a plurality of characteristics according to various embodiments of the present disclosure. FIG. 4 is a flowchart of the operation of an electronic device for providing a response description to a user according to various embodiments of the present disclosure. FIG. 5 is a drawing showing an electronic device that provides user input and response descriptions according to various embodiments of the present disclosure. FIG. 6 is a flowchart of the operation of an electronic device for transmitting a report according to various embodiments of the present disclosure. FIG. 7 is a drawing showing an electronic device displaying a report according to various embodiments of the present disclosure. Specific details for implementing the invention

[0033] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to the exemplary drawings. In assigning reference numerals to the components of each drawing, the same components may have the same reference numeral as much as possible, even if they are shown in different drawings. Furthermore, in describing the embodiments, if it is determined that a detailed description of related known components or functions may obscure the essence of the technical concept, such detailed description may be omitted. Where terms such as "comprising," "having," or "consisting of" are used in this specification, other parts may be added unless "only" is used. Where a component is expressed in the singular, it may include a plural unless otherwise specified.

[0034] Additionally, terms such as first, second, A, B, (a), (b), etc., may be used to describe the components of the present disclosure. These terms are used merely to distinguish the components from other components, and the nature, order, sequence, or number of the components are not limited by such terms.

[0035] In describing the positional relationship of components, where it is stated that two or more components are "connected," "combined," or "joined," it should be understood that while the two or more components may be directly "connected," "combined," or "joined," they may also be "connected," "combined," or "joined" with other components "intervened." Here, the other components may be included in one or more of the two or more components that are "connected," "combined," or "joined" with one another.

[0036] In describing the temporal flow relationship regarding components, methods of operation, or methods of production, for example, when the temporal or sequential relationship is described using "after," "following," "next," or "before," it may include cases where the relationship is not continuous unless "immediately" or "directly" is used.

[0037] Meanwhile, where numerical values ​​or corresponding information regarding a component (e.g., levels, etc.) are mentioned, even without separate explicit notation, the numerical values ​​or corresponding information may be interpreted as including a range of error that may occur due to various factors (e.g., process factors, internal or external shocks, noise, etc.).

[0038] In this disclosure, steps and actions may have the same meaning. Additionally, the order of execution between steps or actions may be changed. Furthermore, at least some of the steps or actions may be omitted within the overall operation flow.

[0039] In the present disclosure, a credit may include information related to credit (e.g., credit evaluation), results related to credit (e.g., credit rating), information related to a loan (e.g., loan amount), and results related to a loan (e.g., whether the loan was approved).

[0040] Hereinafter, various embodiments of the present disclosure will be described in detail with reference to the attached drawings.

[0042] FIG. 1 is a block diagram of an electronic device (100) in which a personalized credit management system according to various embodiments of the present disclosure is implemented.

[0043] Referring to FIG. 1, the electronic device (100) may include a processor (110), a communication circuit (120), a memory (130), and a display (140). The electronic device (100) may be a smartphone, a tablet, a PC, or a server device, but is not limited thereto.

[0044] The processor (110) performs the main computational functions of the electronic device and can implement the personalized credit management method of the present disclosure by executing a program or module stored in memory (130).

[0045] For example, the processor (110) can perform the following functions (see FIG. 2):

[0046] 1. Receive time-series financial data and financial objectives (S210)

[0047] Financial data (CB, MyData, non-financial data, etc.) of the user is received through the communication circuit (120) and stored in memory (130). Financial goals set by the user are received.

[0048] 2. Generation and Update of Dynamic Causal Graph (DCKG) (S220)

[0049] It learns the causal relationships between financial variables from time-series financial data and generates causal graphs. It periodically updates the causal structure or causal strength based on the external financial environment (interest rates, policies, market events). It can calculate intervention effects based on causal inference (do-operator).

[0050] 3. Create Improvement Path Scenario (S230)

[0051] Multiple Sequential Improvement Paths are constructed based on financial goals and the DCKG. The causal impact of each combination of actions on the achievement of long-term goals is considered.

[0052] 4. Prediction of future financial state trajectory and identification of optimal path (S240)

[0053] It predicts the future financial state of each path by combining time series forecasting models (LSTM, Transformer, etc.) with DCKG. It selects the optimal improvement path where the probability of achieving the goal is above a threshold.

[0054] 5. Generation of Causal-Based Natural Language Explanation Reports (S250)

[0055] It infers the causal reasons why each action in the optimal path contributes to goal achievement. It generates LLM-based natural language reports and can include uncertainty information.

[0056] 6. Report provision and user feedback reflection (S260)

[0057] Results are provided to the user through a display (140) or a communication circuit (120). The DCKG is dynamically updated to reflect user behavior or changes in the financial environment.

[0058] The memory (130) stores program code, learned models, data, etc., and may include the following modules:

[0059] - Time Series Data Processing Module: Normalizes, refines, and converts user financial data into a time series structure

[0060] - DCKG Engine Module: Performs causal structure learning, intervention effect calculation, and dynamic update functions

[0061] - Scenario Generation Module: Generation of goal-based action combinations and multi-step path configuration

[0062] - Time series forecasting models (LSTM / Transformer, etc.): Perform financial state trajectory prediction

[0063] - Optimal Path Selection Module: Identifying the optimal path based on the probability of achieving the goal

[0064] - LLM-based Explanation Generation Module: Generates natural language reports including causal explanations and reliability

[0065] - Explanation and Report Provision Module

[0066] The communication circuit (120) performs the following:

[0067] - Receive user financial data from external servers / institutions (CB, MyData, etc.)

[0068] - Send natural language reports and route guidance messages to user terminals

[0069] - Receive information on changes in the financial environment (interest rates, policies, etc.)

[0070] The display (140) provides visual information to the user, such as reports, predicted trajectories, and behavioral paths. It may include touch interface functions.

[0071] That is, the electronic device (100) of FIG. 1 indicates that it is not a simple counterfact-based explanation system, but an integrated architecture that executes all of the dynamic causal inference + time series forecasting + optimal path search + LLM-based report generation intended by the present invention.

[0073] FIG. 2 is a flowchart of a personalized credit management method according to one embodiment of the present disclosure.

[0074] The present method may be executed by a processor of an electronic device and may include the following steps.

[0075] S210: Step for setting user's financial goals and receiving time-series financial data

[0076] Receives financial goals and time-series financial data from the user across multiple points in time.

[0077] Financial data may include various unstructured and non-financial data such as CB data, MyData-based account / card / loan data, e-commerce data, and telecommunications data.

[0078] Financial goals can be numerical or categorical goals, such as "getting a mortgage within one year" or "recovering a credit score to 800."

[0079] S220: Dynamic Causal Graph (DCKG) Generation and Personalization Step

[0080] Based on received time-series financial data and the current financial environment (e.g., interest rates, policies, economic indicators, etc.), a Dynamic Causal Graph (DCKG) representing the causal relationships between financial variables is generated and periodically updated.

[0081] This process may include intervention-based causal inference (do-operator), time-series causal pattern analysis, and adjustment of the causal influence of external environmental factors.

[0082] In addition, DCKG can be personalized and activated according to user characteristics.

[0083] S230: Improvement Path Scenario Generation Step

[0084] Based on the established financial goals and the aforementioned dynamic causal graph, multiple improvement path scenarios consisting of a sequence of actions are generated.

[0085] For example, a multi-step strategy such as "repaying 20% ​​of debt over 3 months → strengthening income verification for 6 months → applying for a loan" may be included.

[0086] Each path can consist of a combination of strategic actions for long-term goal optimization.

[0087] S240: Future financial status prediction and optimal path identification step

[0088] For each generated improvement path scenario, the financial state trajectory at a future point in time is predicted, and the impact of each path on achieving the goal is evaluated to identify the optimal improvement path where the probability of achieving the financial goal is greater than or equal to a preset standard.

[0089] At this time, time series forecasting models (LSTM, Transformer, etc.) and dynamic causal models (DCKG-based causal simulation) can be used in combination.

[0090] S250: Causal-based natural language explanation report generation step

[0091] The causal rationale for each action included in the identified optimal improvement path contributing to goal achievement is inferred from the DCKG, and a natural language explanation report containing this is generated.

[0092] For example: "During the current period of rising interest rates, the sequence of 'debt reduction → income verification' has a greater causal impact on DSR improvement, so this sequence increases the probability of achieving the goal."

[0093] The generation process may include LLM-based prompt generation and the inclusion of uncertainty information.

[0094] S260: User provision and feedback incorporation stage

[0095] The generated natural language description report is provided to the user through a display or communication circuit.

[0096] The delivery method can vary, including app screens, notification messages, email, and messenger.

[0097] Furthermore, when actual user behavior, feedback, and changes in the financial environment are collected, the causal relationships in the DCKG are continuously updated, enabling the maintenance of the model's adaptability and explanatory reliability.

[0098] The method of the present embodiment configured as described above can present an advanced personalized financial strategy by going beyond a fragmentary counterfactual explanation that merely changes specific values, and by providing a long-term improvement path that the user can actually follow along with its causal basis.

[0100] FIG. 3 illustrates an example of a user financial characteristic input interface that can be displayed on a display (140) of an electronic device (100) in various embodiments of the present disclosure.

[0101] Referring to FIG. 3, multiple characteristics (310) that can be collected from a user can be extended to include not only personal identity information but also various time-series-based financial data. These characteristics can be used as basic information for generating dynamic causal graphs (S220), generating improvement path scenarios (S230), predicting future trajectories (S240), and generating natural language description reports (S250), which will be described later.

[0102] 1. Personal Information

[0103] Multiple characteristics (310) may include personal information related to user identification and basic credit evaluation. For example, name, age, gender, place of residence, employment information, etc. This information may be used as a basis for determining financial risk ratings or establishing credit strategies.

[0104] 2. Card Usage Information (Time Series Characteristics)

[0105] Multiple characteristics (310) may include card usage history accumulated over the past few months or years. For example,

[0106] - Credit card usage amount over the past 12 months

[0107] - Cash advance amount

[0108] - Cash advance limit

[0109] Items such as may be included.

[0110] These time-series characteristics are utilized for analyzing changes in spending patterns and assessing cash flow stability, and can be used in dynamic causation graphs to model the causal impact of increases or decreases in spending on creditworthiness or loan eligibility.

[0111] 3. Loan Information (Time Series Characteristics)

[0112] Multiple characteristics (310) may include time series data regarding loan balances and volatility. For example,

[0113] - Balance of mortgage loan

[0114] - Total amount of non-bank credit loans

[0115] - Number of non-bank credit lending institutions

[0116] - Loan amount growth rate over the past 3 months

[0117] This may include factors such as the Debt Service Ratio (DSR) and changes in debt structure, which can be used to analyze the causal effects on lending potential.

[0118] 4. Delinquency and repayment history

[0119] Multiple characteristics (310) may include information related to delinquency and repayment patterns. For example,

[0120] - Number of unrefunded overdue cases

[0121] - Monthly repayment history

[0122] This may include the back.

[0123] In the dynamic causal graph of the present disclosure, it can be utilized as a key note constituting a causal path such as "delinquency → change in creditworthiness" or "delinquency → decrease in approval probability."

[0124] 5. Income and Financial Health Information

[0125] Multiple characteristics (310) may include information regarding income and assets. For example,

[0126] - Monthly income

[0127] - Income verification history

[0128] - Asset Information

[0129] This may include, and can be used for causal path modeling, such as "income change → loan limit change," and for exploring improvement paths to achieve goals.

[0130] 6. Credit Score and Financial Behavior Indicators

[0131] Multiple characteristics (310) may include credit scores and financial activity indicators provided by financial institutions or CB institutions. For example,

[0132] - Credit score (KCB, NICE, etc.)

[0133] - Financial activity stability indicators

[0134] This may include the back.

[0135] Time series credit scores are used to predict the trajectory of creditworthiness changes at future points in time and can be utilized as a key variable to calculate the probability of achieving goals.

[0136] 7. Actual values ​​of the characteristics (320)

[0137] Multiple values ​​(320) are actual values ​​corresponding to each of the above characteristics.

[0138] - Text format (name, address, company name, etc.)

[0139] - Numerical form (age, card usage amount, loan balance, number of overdue payments, etc.)

[0140] It can be composed of.

[0141] These values ​​do not include only values ​​from a single point in time, but can be collected over multiple points in time and stored in the form of time-series data to be used as input for dynamic causal graphs and prediction models.

[0142] 8. User Input and Automatic Collection

[0143] The electronic device (100) can display a characteristic input interface (330) on the display (140). The user may input values ​​directly, or check values ​​that are automatically collected and updated through MyData, CB institution API, financial institution linkage, etc.

[0144] The processor (110) can perform various steps (S220 to S260) presented in FIG. 2 based on these inputs and collected data, and the data can be updated by the user as needed or periodically updated by the system.

[0146] Hereinafter, a personalized credit management system and method based on dynamic causal inference and time-series forecasting according to another embodiment of the present disclosure will be described. This embodiment may be implemented independently of the embodiments described above, or may be optionally combined and operated.

[0147] 1. Time Series Data Reception and Financial Goal Setting Module

[0148] The electronic device according to the present embodiment may include a module that receives time-series financial data over multiple points in time of the user, along with the user's long-term and short-term financial goals (e.g., “achieving a credit score of 800 within one year,” “approval of a specific loan within six months”).

[0149] The above time series financial data may include the following.

[0150] - Monthly card usage history

[0151] - Loan repayment history

[0152] - Time of occurrence and period of delinquency

[0153] - Income change records

[0154] - Changes in credit scores over time

[0155] The received data is used as input for causal modeling, improvement path generation, and future financial state prediction performed in subsequent steps.

[0157] 2. Dynamic Causal Knowledge Graph (DCKG) Engine

[0158] The electronic device according to the present embodiment may include an engine that constructs a Dynamic Causal Knowledge Graph (DCKG) based on the time-series financial data and external financial environment information.

[0159] The above DCKG engine performs the following functions.

[0160] - Inference of the causal relationship structure between financial variables

[0161] - Quantification of the causal effect of a specific intervention

[0162] - Dynamic updating of causal structure or causal strength in response to changes in the external environment, such as interest rates, policies, and economic conditions

[0163] for example,

[0164] - “Increased income → Improved DSR → Increased probability of loan approval”

[0165] - “Rising interest rates → Intensified negative impact of short-term delinquency”

[0166] Causal relationships such as this can reflect characteristics that change over time and with changes in the financial environment.

[0168] 3. Optimal Improvement Path Search and Simulation Module

[0169] According to the present embodiment, the electronic device can generate a number of improvement path scenarios based on the set financial goal and the DCKG.

[0170] Each scenario may include a combination of the following sequential actions.

[0171] - Reduction of debt ratio over a certain period

[0172] - Strengthening income verification

[0173] - Early resolution of specific overdue items

[0174] - Adjustment of unnecessary expenditure items

[0175] For each generated scenario, the electronic device combines the time series forecasting model with the DCKG to predict the trajectory of future financial states and evaluates the probability of achieving the goal.

[0176] Among the scenarios where the probability of achieving the goal meets a preset criterion (e.g., 70% or more), the optimal improvement path having the most efficient combination of actions can be selected.

[0178] 4. Causality-based explanation generation module

[0179] The electronic device according to the present embodiment may include a module that generates a natural language-based explanatory report including the optimal improvement path and the causal basis for each action of the path contributing to the achievement of a goal.

[0180] A feature of this embodiment is that it generates a causal-based explanation that clearly indicates “why” the corresponding action contributes to achieving the goal, rather than simply listing conditions or numerical changes.

[0181] To this end, the electronic device may configure a deep prompt including the following elements and provide it to a large-scale language model (LLM).

[0182] - Sequential actions of the optimal improvement path (e.g., Action A → Action B → Action C)

[0183] - Causal contribution of each action

[0184] - Expected future financial state trajectory

[0185] - Probability of achieving goals after behavior change

[0186] - Model uncertainty and reliability information

[0187] Based on the above prompt, LLM can generate a natural language description such as the following, for example.

[0188] It is predicted that resolving current delinquencies within two months will increase your credit score by approximately 18 points, which is a key factor in meeting loan approval criteria.

[0189] The generated explanation report can be provided to the user through a display or communication circuit.

[0191] FIG. 4 is a flowchart of the operation of an electronic device for providing a response description based on a user query according to various embodiments of the present disclosure.

[0192] FIG. 5 illustrates an electronic device in which a natural language-based explanation and question-and-answer interface is provided through a display according to various embodiments of the present disclosure.

[0194] As previously described in FIG. 2, the electronic device (100) can generate a natural language description report containing the optimal improvement path and the causal basis for each action in the path contributing to the achievement of the goal (S250), and then provide it to the user (S260).

[0195] The user may request additional interpretation or details regarding this explanation, and accordingly, the electronic device may perform an extended function of generating and providing a causal-based response explanation in response to the user's question.

[0197] First, referring to FIG. 4, in operation 410 (S410), the processor (110) of the electronic device (100) can provide an input interface (510) for receiving a user query through a display (140).

[0198] The input interface (510) can be implemented as a touch input, keyboard input, or voice input-based UI, and the user can input a natural language query (input prompt 520) (S420).

[0200] Referring to FIG. 5, the input prompt (520) may be in the form of, for example, the following:

[0201] “Why must we reduce debt first?”

[0202] Why is the income verification step important in this path?

[0203] “What becomes more dangerous during a period of rising interest rates?”

[0205] Referring again to FIG. 4, in operation 430 (S430), the processor (110) of the electronic device (100) provides an input prompt (520) to an LLM-based response generation model.

[0206] At this time, the response generation model can generate a response description (530) by combining the following information:

[0207] - Causal structure between the user's time-series financial data and the financial variables represented by DCKG

[0208] - Dynamic causal inference-based optimal improvement path information of the present disclosure

[0209] - Causal basis elements included in the natural language explanation report (500)

[0210] - External financial environment and uncertainty information at that time

[0211] - User's additional question (Input prompt 520)

[0212] That is, the response explanation (530) is not a simple FAQ-level sentence generation, but,

[0213] It can be generated as a “causal-based follow-up explanation” including the causal paths shown by DCKG, priorities of improvement strategies, and expected effects by time of change.

[0214] For example, it could take the form of, “Since the negative causal effect of short-term delinquency on credit scores is amplified during the current period of rising interest rates, the DCKG analysis evaluates the impact of resolving delinquency as greater than that of reducing debt.”

[0215] In operation 440 (S440), the processor (110) provides the user with the generated response description (530). This may be displayed as text on the display (140), transmitted via email or messenger through the communication circuit (120), or optionally provided in the form of voice output.

[0217] Referring to FIG. 5, this response description (530) does not merely describe counterfactual-based change conditions, but,

[0218] - “What causal path operates when a specific action A is performed first”,

[0219] - “Why action B becomes necessary after time t”,

[0220] - “How changes in market conditions adjust the effectiveness of that channel”

[0221] It enables the provision of sophisticated explanations based on dynamic causal inference, such as...

[0222] In addition, in an embodiment according to the present disclosure, when the electronic device (100) continuously receives user queries and actual financial behavior (e.g., whether to execute debt repayment) or changes in the financial environment, such information may be reflected in the DCKG update module and time series prediction model to improve the accuracy of subsequent explanations. That is, the question-and-answer flow itself may also function as part of the system's self-updating process.

[0223] Furthermore, explanations regarding loans and the results of causal analysis may be necessary not only for users but also for external entities such as financial institutions or loan review agencies.

[0224] Accordingly, the above response description (530) or causal-based report may also be provided to entities other than individual users through the communication circuit (120), and related operations may be described in subsequent drawings.

[0226] FIG. 6 is a flowchart of the operation of an electronic device (100) for transmitting a report (700) according to various embodiments of the present disclosure, and FIG. 7 shows an example of an electronic device (100) on which the report (700) is displayed.

[0228] As previously described in FIG. 2, the processor (110) of the electronic device (100) can identify an optimal improvement path based on the user's time-series financial data and a dynamic causal knowledge graph (DCKG) (S240) and generate a natural language description including the causal basis of the path (S250). This embodiment includes a function to generate and transmit a structured report (700) to be delivered to various stakeholders (users, financial institutions, etc.) in parallel with the flow of providing such natural language descriptions.

[0230] Referring to FIG. 6, the processor (110) can configure a prompt for generating a report based on financial objectives, time series data, the causal structure of DCKG, and the results of the optimal improvement path analysis (S610).

[0231] The above prompt may be configured to include some or all of the following elements.

[0232] - Sequential actions and expected effects of the optimal improvement path

[0233] - Causal effect of each action on the probability of achieving the goal (Effect size)

[0234] - The trajectory of future financial states calculated by a time series forecasting model

[0235] - Credit scores before and after the change / Loan-related indicators such as approval probability

[0236] - Details of adjustments for the impact of changes in the external financial environment (interest rates, policies, etc.)

[0237] - Model uncertainty and reliability information

[0238] The processor (110) can input the above prompt into a large-scale language model (LLM) to generate a report (700).

[0239] The above LLM may be the same model used to generate natural language descriptions in FIGS. 2 to 5, and can automatically generate a structured analysis summary based on causal graphs and time series prediction results.

[0241] As illustrated in FIG. 7, the report (700) may include the following items.

[0242] - Values ​​and temporal changes of major financial variables (e.g., credit card usage, loan balance, delinquency history)

[0243] - Summary of DCKG-based causal structure (e.g., “Increased income → Improved DSR → Increased probability of approval”)

[0244] - Loan-related figures before and after the change in improvement path (e.g., approval probability 25% → 75%)

[0245] - Analysis of the causal contribution of each behavior to goal achievement

[0246] - Time-series forecast-based trajectory of credit score or loan likelihood for the next 3 to 12 months

[0247] - Model reliability (uncertainty score) or risk factors

[0248] These reports can be provided not only directly to users but also to financial institutions, credit assessment systems, or external evaluation and consulting systems.

[0249] The processor (110) can transmit the report (700) through the communication circuit (120) in operation 620 (S620). The transmission method can take various forms, such as app notifications, email, message APIs, and integration with internal corporate systems. The receiving electronic device may visually display the report (700) through a display or output it in the form of audio through a speaker or voice interface.

[0250] This embodiment can provide natural language descriptions (Figs. 4, 5) and structured reports (Figs. 6, 7) independently or in parallel.

[0251] for example,

[0252] - General Users: Provides natural language-based explanations

[0253] - Financial Institutions / Partner Organizations: Provides reports focused on analysis summaries

[0254] As such, the format of provision can be varied depending on the user's role and purpose.

[0255] In addition, recipient feedback on transmitted reports or changes in the financial environment can be reflected back into DCKG updates and time-series forecasting model training, which can be utilized to improve the accuracy and adaptability of subsequently generated reports.

[0257] Referring to FIG. 7, the report (700) may include at least some of a plurality of characteristics (310) (e.g., total amount of cash service usage, total number of non-bank credit lending institutions), a value corresponding to at least some of the plurality of characteristics (310) (e.g., 41200, 0, 2, 1), or a probability associated with the loan (e.g., predicted probability before change -25%, predicted probability after change -75%).

[0258] For example, the probability associated with the goddess may include goddess information (e.g., approval probability before change) based on information associated with the goddess (300) and goddess information (e.g., prediction probability after change) based on counterfacts.

[0259] For example, the report (700) may include reliability. The processor (110) may evaluate reliability based on LLM output evaluation scores. For example, the processor (110) may evaluate reliability using SOFTMAX scores, inverse perplexity, loglikelihood, F1 Score, BLEU, ROUGE, METEOR, and multidimensionalibration. However, it is not limited to the examples presented.

[0260] The report (700) can be in natural language form.

[0261] Referring again to FIG. 6, in operation 620 (S620), the processor (110) can transmit a report (700) through the communication circuit (120).

[0262] For example, the processor (110) can transmit a report (700) to an electronic device (100) and another electronic device through a communication circuit (120).

[0263] Another electronic device can receive the report (700) through a communication circuit within the other electronic device and display the report (700) through a display within the other electronic device.

[0264] Another electronic device can receive the report (700) through a communication circuit within the other electronic device and output the report (700) through a speaker within the other electronic device.

[0265] A method for providing goddess information and an electronic device according to embodiments of the present disclosure may be described as follows.

[0266] A method for providing credit information for an electronic device comprising a memory, a communication circuit, a display, and a processor, the method for providing credit information may include the steps of: receiving information associated with a credit including a plurality of characteristics and a plurality of values ​​corresponding to each of the plurality of characteristics; obtaining a counterfact from the information which is information in which at least some of the plurality of values ​​have been changed; generating a prompt from the counterfact and the information; and generating a natural language description including credit information by inputting the prompt into a model.

[0267] The method for providing goddess information may further include the steps of providing a natural language description through a display, providing an input interface through a display, receiving a user's input prompt for the interface, and generating a response description based on inputting the input prompt into a model.

[0268] A method for providing credit information may include, upon inputting a prompt into a model: a step of generating a report including at least some of a plurality of characteristics, a value corresponding to at least some of the plurality of characteristics, or a probability associated with the credit upon obtaining a counterfact; and a step of transmitting the report through a communication circuit.

[0269] Multiple characteristics may include at least one of information related to the amount of cash service usage, the total amount of mortgage loans, the total number of non-bank credit lending institutions, the total amount of non-bank credit loans, the loan amount growth rate, whether the user is subject to credit management, the age of the user, the gender of the user, or the gender of the user.

[0270] The step of obtaining a counterfact from information may further include the step of generating a plurality of counterfacts by changing a plurality of values ​​corresponding to each of a plurality of characteristics, and the step of obtaining a counterfact by selecting at least one of the plurality of counterfacts.

[0271] The step of obtaining counterfacts from information may further include the step of determining the range of change of multiple values ​​corresponding to each of the multiple characteristics.

[0272] By selecting at least one of a plurality of counterfacts, the step of acquiring the counterfact may include the step of selecting a counterfact in which the change in representative values ​​is minimized by reflecting the change amounts of a plurality of values ​​and a plurality of weights pre-set for each of the plurality of characteristics.

[0273] The model can be a Large Language Model (LM) with policy-based filters configured.

[0274] The method for providing credit information may further include a step of filtering counterfacts.

[0275] The step of generating a prompt from counterfacts and information may further include the step of generating a prompt from filtered counterfacts and information.

[0276] The step of filtering counterfacts may include removing at least one counterfact depending on the direction of increase or decrease of multiple values, the characteristics of multiple values, or the allowable range of multiple values.

[0277] The electronic device may include memory, communication circuits, a display, and a processor.

[0278] The processor can receive information including multiple characteristics and multiple values ​​corresponding to each of the multiple characteristics.

[0279] The processor can obtain a counterfact that is data in which at least some of the multiple values ​​are changed from the information.

[0280] The processor can generate prompts from counterfacts and information.

[0281] By inputting prompts into the model, natural language descriptions containing goddess information can be generated.

[0283] FIG. 7 illustrates an example in which a report (700) generated by an electronic device (100) is displayed on a display (140) according to various embodiments of the present disclosure.

[0284] As previously described, the processor (110) of the electronic device (100) can generate a structured report (700) separate from the natural language description by combining a dynamic causal knowledge graph (DCKG), time series prediction results, and counterfactual-based analysis results. This embodiment further includes an operation to generate and transmit a structured report that can be delivered to a financial institution or a third party in parallel with the provision of a natural language-based description (S250, S260).

[0285] Referring again to FIG. 6, the processor (110) can configure a prompt for generating a report based on counterfacts, information (300) associated with credit, time-series financial data, and the causal structure of DCKG in operation 610 (S610). The prompt may include at least one of the following elements.

[0286] - Current state of multiple characteristics (310) and values ​​(320) corresponding to each characteristic

[0287] - Change values ​​and direction of change based on counterfact analysis results

[0288] - Causal paths inferred from dynamic causal graphs

[0289] - Loan-related probabilities before and after the change calculated by the time series forecasting model (e.g., approval probability)

[0290] - Sensitivity or causal contribution (effect size) of key variables

[0291] - Reliability information indicating the uncertainty of the prediction results

[0292] The processor (110) can input the above prompt into an LLM-based model to generate a report (700), and the model may be the same model as the model that generates natural language descriptions described in FIGS. 2 to 5.

[0293] Referring to FIG. 7, the report (700) may include at least one of the following items.

[0294] - Some of the multiple characteristics (310) (e.g., amount of cash service usage, number of non-bank credit lending institutions, etc.)

[0295] - Values ​​(320) corresponding to the above characteristics (e.g., 41200, 0, 2, 1, etc.)

[0296] - Probability information associated with the deposit (e.g., approval probability 25% before change, predicted probability 75% after change)

[0297] - Results of the causal impact assessment of the changed value

[0298] - Forecast of future financial conditions based on time series predictions

[0299] - Model reliability (e.g., Softmax score, inverse perflexity, Log-likelihood, BLEU, ROUGE, METEOR, multidimensionalibration, etc.)

[0300] The above report (700) may be in a natural language form or in a structured document form including a table, chart, or summary text.

[0301] Referring again to FIG. 6, in operation 620 (S620), the processor (110) can transmit the report (700) to an external electronic device through a communication circuit (120). The receiving electronic device can visually display the report through its own display or output it in the form of audio through a speaker. Accordingly, the report generation function of the present disclosure can be used to provide credit analysis results to third parties, such as financial institutions, cooperating institutions, and review agencies, as well as individual users.

[0302] Below, a method and system for providing credit information according to various embodiments of the present disclosure are summarized.

[0303] The method for providing goddess information of the present disclosure may include the following.

[0304] 1. A step of receiving credit-related information including multiple characteristics and multiple values ​​corresponding thereto.

[0305] 2. A step of obtaining a reflection fact in which at least some values ​​have been changed from the above information.

[0306] 3. Step to generate a prompt from counterfactual and goddess-related information

[0307] 4. Inputting the above prompt into an LLM-based model to generate a natural language description containing credit information

[0308] This method may additionally include the following.

[0309] - Step of providing natural language explanations through the display

[0310] - The step of providing an input interface and receiving user input prompts

[0311] - Step of generating a causal-based response explanation using an input prompt

[0312] - Step of generating a report containing counterfacts, values, characteristics, or credit probabilities

[0313] - The step of transmitting the above report through a communication circuit

[0314] The process of acquiring counterfacts may include the following.

[0315] - Create multiple counterfacts by changing values ​​corresponding to multiple attributes

[0316] - Selecting the counterfact that minimizes the change in weighted representative values

[0317] - Removal of unchangeable characteristics or counterfacts outside the permissible range

[0318] The model can be an LLM with policy-based filters set.

[0319] The electronic device of the present disclosure includes a memory, a communication circuit, a display, and a processor, and the processor may be configured to perform functions such as receiving goddess-related information, obtaining counterfacts, generating prompts, and generating natural language descriptions.

[0321] Hereinafter, a personalized loan management system and method based on Dynamic Causal Inference and Time-Series Forecasting, which can be implemented independently or optionally combined with the above embodiments, are described. This embodiment may additionally include components to complement the limitations of the counterfact-based loan explanation method presented in the prior embodiments and to support strategic decision-making for the achievement of a user's long-term financial goals.

[0323] 1. Dynamic Causal Knowledge Graph (DCKG)-based modeling

[0324] According to the present embodiment, the electronic device can construct a dynamic causal knowledge graph (DCKG) representing causal relationships between financial variables by utilizing not only individual financial information of users but also time-series financial data collected from multiple users (e.g., monthly card usage amount, changes in loan balance, delinquency history, etc.).

[0325] The above DCKG may include the following features:

[0326] - Causal relationships are expressed as connections (Edges) in the form of "Cause → Effect," and the strength (weight) of the connection changes over time according to changes in the external financial environment (e.g., interest rate hikes, regulatory changes, economic cycles).

[0327] - Expresses the fundamental operating principles between financial variables by modeling intervention-based causal effects (P(outcome | do(intervention))) rather than simple conditional probability (P(outcome | condition)).

[0328] - Automatically extracts statistical and semantic causal structures using neuro-symbolic causal pattern analysis techniques and maintains a dynamic structure that is automatically updated when new data is input.

[0329] The causal-based approach of the present embodiment can overcome the limitations of prior art centered on statistical correlations and improve the reliability of explanation and policy consistency.

[0331] 2. Exploring the Optimal Improvement Path Based on Time Series Forecasting

[0332] This embodiment can explore an Optimal Improvement Path that maximizes the possibility of achieving goals based on financial goals set by the user (e.g., "achieve a credit score of 750 within 6 months," "secure a mortgage loan limit within 1 year").

[0333] The electronic device can perform the following:

[0334] - Generate multiple sequential improvement path scenarios by combining various action candidates (e.g., reducing debt ratio, strengthening income verification, clearing delinquent items, etc.) based on DCKG

[0335] - Predict the trajectory of future financial states using time series forecasting models (LSTM, Transformer, etc.) for each scenario

[0336] - Calculate the probability of achieving financial goals based on the predicted trajectory, and select the optimal path among scenarios that meet preset criteria (e.g., 70% or higher).

[0337] This configuration has the advantage of being able to provide a strategic and step-by-step action roadmap that takes the flow of time into account, going beyond the level of "minimal change advice at a single point in time" offered by existing technology.

[0339] 3. Self-organizing and Metacognitive-based Adaptive Architecture

[0340] The electronic device of the present embodiment may include an adaptive architecture that automatically reconstructs the knowledge structure (DCKG) when there is a change in the financial environment or new user data is input, so as not to degrade performance.

[0341] Specifically, the electronic device can perform the following:

[0342] - Utilizing meta-learning and continual learning to maintain existing causal knowledge when learning new information and prevent catastrophic forgetting

[0343] - Quantify the uncertainty of the prediction model to provide an explanation including confidence information, such as "This path is capable of achieving the goal with approximately 75% confidence."

[0344] - Ensures the stability and robustness of explanations and predictions by automatically reflecting the impact of causal structures in the event of environmental changes (e.g., sudden spikes in interest rates).

[0345] As such, this embodiment can implement an active and intelligent financial decision-making system that does not remain in a fixed static model but interacts with the environment and updates its knowledge autonomously.

[0347] As such, the present embodiment can provide long-term, strategic, and highly reliable personalized credit management functions that existing filed technologies and prior patents have not been able to solve, by including in-depth causal analysis through a dynamic causal knowledge graph, time-series-based optimal improvement path search, and a self-organizing and metacognitive-based adaptive structure.

[0349] The devices, methods, configurations, glyphs, and operations described in this disclosure may be implemented in digital electronic circuits, or computer software, firmware, or hardware comprising structures disclosed in this disclosure and structural equivalents, or combinations of one or more of these. The glyphs described in this disclosure may be implemented as one or more computer programs, for example, as one or more modules of computer program instructions encoded on a computer storage medium to control execution by a data processing device or operation by a data processing device. Program instructions may be encoded in artificially generated propagated signals, for example, mechanically generated electrical, optical, or electromagnetic signals generated to encode information for transmission to a suitable receiver device for execution by a data processing device. The computer storage medium may be or may include a computer-readable storage device, a computer-readable storage substrate, a random or serial access memory array or device, or a combination of one or more of these. Although the computer storage medium is not a propagated signal, the computer storage medium may be a source or destination of computer program instructions encoded in an artificially generated propagated signal. Additionally, a computer storage medium may be or include one or more individual physical components or media (e.g., multiple CDs, disks, or other storage devices). The operations described in this disclosure may be implemented as operations performed by a data processing device on data stored in one or more computer-readable storage devices or data received from other sources.

[0350] The foregoing description is merely an illustrative explanation of the technical concept of the present disclosure, and those skilled in the art to which the present disclosure pertains may make various modifications and variations within the scope of the essential characteristics of the technical concept. Furthermore, since these embodiments are intended to explain rather than limit the technical concept of the present disclosure, the scope of the technical concept is not limited by these embodiments.

Claims

Claim 1 A personalized credit management method performed on an electronic device including a processor, comprising: (a) receiving a user’s financial goals and time-series financial data over multiple points in time of the user; (b) generating a dynamic causal graph representing causal relationships between financial variables based on the received time-series financial data, and updating the dynamic causal graph periodically or based on detected events; (c) generating sequential improvement path scenarios composed of multiple actions based on the set financial goals and the dynamic causal graph; (d) predicting a financial state trajectory at a future point in time for each of the generated improvement path scenarios to identify an optimal improvement path in which the probability of achieving the financial goals satisfies a preset criterion; (e) generating a natural language explanation report including the identified optimal improvement path and the causal basis for each action of the path contributing to the achievement of the goal, inferred from the dynamic causal graph; and (f) providing the natural language explanation report to the user. Claim 2 A personalized credit management method according to claim 1, wherein the dynamic causal graph of step (b) quantitatively models the causal effect of an intervention on a specific financial variable on another variable. Claim 3 A personalized credit management method according to claim 1, wherein the trajectory prediction of step (d) is performed by combining the dynamic causal graph and the time series prediction model. Claim 4 A personalized credit management method according to claim 1, wherein the natural language description report of step (e) further includes reliability information indicating the uncertainty of the predicted trajectory. Claim 5 A personalized credit management method according to claim 1, characterized in that the natural language description report is generated by inputting a prompt including the causal graph and trajectory prediction results into a large-scale language model (LLM). Claim 6 A personalized credit management method according to claim 1, wherein the financial goal includes at least one of a numerical-based goal or a categorical goal, and a different improvement path generation logic is applied depending on the type of the financial goal. Claim 7 A computer-readable recording medium storing a program for executing the method according to paragraph 1. Claim 8 A personalized credit management system characterized by comprising: a memory; and a processor connected to the memory and configured to: (a) receive a user’s financial goals and time-series financial data over multiple points in time of the user; (b) generate a dynamic causal graph representing causal relationships between financial variables based on the received time-series financial data, and update the dynamic causal graph periodically or based on detected events; (c) generate sequential improvement path scenarios composed of multiple actions based on the set financial goals and the dynamic causal graph; (d) predict the financial state trajectory at a future point in time for each of the generated improvement path scenarios to identify an optimal improvement path in which the probability of achieving the financial goals meets criteria; and (e) generate a natural language description report including the causal basis in which each action included in the identified optimal improvement path contributes to the achievement of the goals, inferred from the dynamic causal graph.

Images