Method and apparatus for bond recovery prediction based on an artificial intelligence model

Abstract

A server providing an AI model-based debt recovery prediction service can acquire source data related to debt from a user terminal, including promissory note information, information regarding creditors and debtors, consultation data, and community-based information. The server can perform standardization and normalization processing on the source data, including natural language processing-based text normalization, rule-based parsing, outlier removal, categorical encoding, and the generation of derived variables. The server can predict the recovery rate of the debt through an AI agent that takes the standardized and normalized data as input. Based on the predicted recovery rate, the server can calculate a rating value indicating the recoverability and risk level of the debt.

Description

Technology Field

[0001] This specification relates to a technology for predicting a bond recovery rate, calculating a bond rating based on the predicted value, and generating a digital bond object using the rating and bond attribute information. Background Technology

[0002] Non-financial or peer-to-peer monetary transactions often occur outside the credit rating systems of financial institutions, making it difficult to objectively assess the recoverability or risk of debts. In this environment, various unstructured factors—such as the drafting of promissory notes, the debtor's attitude, the intimacy of the relationship, and age and occupation information—are known to influence recovery potential; however, there is a problem in that these factors are not systematically recorded or quantified, resulting in limited utilization. Furthermore, even if consultation records or community-based information accumulate during the creditor's response process following a default, it is difficult to integrate and analyze this data into a standardized structure, leading to a strong tendency to rely on human-centered, empirical judgment.

[0003] Meanwhile, while existing financial sector management systems for non-performing or delinquent loans utilize models based on standardized credit information or repayment history, it is difficult to directly apply these financial models to peer-to-peer financial transactions due to the insufficient availability of such information. Furthermore, existing classification-based prediction techniques often merely categorize loans into normal, delinquent, or non-performing, which limits their ability to precisely assess actual recovery potential or predict recovery rates based on loan amounts.

[0004] Recently, attempts have been made to comprehensively analyze various elements, including unstructured data, using machine learning or deep learning. However, text-based unstructured data, such as consultation records, text conversations, and community responses, is difficult to use directly as input for models because its quality is inconsistent and it is difficult to standardize. Additionally, derived information, such as anomalous relationship data, repeat transaction status, and family status, is not directly included in the source data and requires a separate preprocessing step; however, since the methods for generating these derived variables are not standardized, there is a possibility of deviation in the analysis results.

[0005] Furthermore, to utilize predicted recovery rates in the actual debt management process, it is necessary to structure debt-specific characteristic information, predicted recovery rates, and risk levels, and provide them in a format that is easy to use by external systems or collection-related platforms. However, existing technologies have lacked the functionality to generate prediction results as standardized digital objects or grade them for reusable management, which has resulted in difficulties in utilizing them for creditor decision-making or collaboration with collection agencies.

[0006] As such, there is an increasing demand for technology to process various structured and unstructured data generated from peer-to-peer financial transactions or non-financial bonds, accurately predict recovery rates based on this data, and provide the prediction results in the form of structured bond ratings or digital assets. Prior art literature

[65535] Korean Published Patent No. 10-2023-0092270 The problem to be solved

[0007] The purpose of this specification to solve the aforementioned problems is to standardize and normalize various source data to transform them into a form usable by an artificial intelligence agent, thereby enabling more accurate prediction of the debt recovery rate based on this, in situations where it is difficult to objectively assess the likelihood of debt recovery due to the structural incompleteness and analytical difficulty of structured and unstructured data generated during inter-personal or non-financial financial transactions.

[0008] Another objective of the present specification to solve the above-mentioned problems is to support creditors in rationally establishing debt management strategies by providing rating information indicating the risk or recoverability of debts based on predicted recovery rates, and furthermore, to increase the efficiency and transparency of debt management by creating the corresponding rating and debt attribute information as structured digital objects that can be utilized in external systems or debt collection platforms. means of solving the problem

[0009] A method for providing an artificial intelligence model-based debt recovery prediction service according to one embodiment of the present specification for achieving the above objective may include: a step of obtaining source data related to debt, including promissory note information, information related to creditors and debtors, consultation data, and community-based information, from a user terminal; a step of performing standardization and normalization processing on the source data, including natural language processing-based text normalization, rule-based parsing, outlier removal, categorical encoding, and generation of derived variables; a step of predicting the recovery rate of the debt through an artificial intelligence agent that takes the standardized and normalized data as input; and a step of calculating a grade value representing the recoverability and risk level of the debt based on the predicted recovery rate.

[0010] Here, the method may include the step of creating a digital bond object that structures attribute information of the bond, including the grade value, the standardized and normalized data, and identification information of the bond.

[0011] Here, the method may include a step of performing natural language processing-based text normalization and rule-based parsing on counseling data or community-based data.

[0012] Here, the method may include a step of converting the amount, interest rate, maturity, elapsed period, marketing consent, and recurring transaction status included in the source data related to the bond into structured data.

[0013] Here, the method may include the step of collecting text data containing keywords related to financial transactions, delayed repayment, avoidance of collection, legal response, or debt repayment from posts, inquiries, or comments generated by a user presumed to be a debtor on an online community, bulletin board, or anonymous counseling platform, and analyzing the posting time, keyword frequency, avoidance expression patterns, and linguistic features related to the intention to avoid legal risk of said text data to generate an external community-based derived variable indicating the debtor's possibility of avoiding recovery.

[0014] Here, the method may include the step of collecting information on the types of communication channels used as means of contact between the debtor and the creditor, contact attempt history by channel, channel change history, response status and response time by channel, and analyzing the said information to generate communication path-based derived variables indicating the diversity of contact paths, contact avoidance patterns, or contactability.

[0015] Here, the method may include the step of collecting information from online communities, small groups, local organizations, or club platforms indicating the position, operational role, or position of responsibility held by the debtor, and calculating the debtor's reputation maintenance pressure or social responsibility burden based on the said information to generate a social responsibility-based derivative variable indicating the possibility of recovery.

[0016] Here, the method may include the step of generating a training dataset containing the time series history of rating values ​​by bond, changes in recovery rates over time, the time series of debtor behavior signals, and fluctuation information of external community-based derived variables, training an artificial intelligence agent to learn the fluctuation pattern of rating values ​​using the training dataset, and predicting the possibility that the rating value of the bond will rise or fall over a predetermined period based on the fluctuation pattern.

[0017] Here, the method may include a step of recommending the timing of sale or the necessity of sale of the bond when the rating value is determined to be relatively high or immediately before a downward transition based on the possibility that the rating value of the bond may decline in the future or increase in volatility.

[0018] Here, the method may include the step of obtaining investment portfolio information including stock items held or traded by the debtor, trading frequency, volatility by item, composition ratio of stock groups, and whether it is dividend-centric; analyzing the ratio of high-volatility assets, short-term trading frequency, whether it is centered on blue-chip or dividend stocks, and risk diversification based on the above information to calculate a risk propensity index representing the debtor's investment propensity; and adjusting the debtor's repayment propensity or recovery possibility according to the above risk propensity index.

[0019] An apparatus for providing an artificial intelligence model-based debt recovery prediction service according to another embodiment of the present specification for achieving the above objective comprises a processor and a memory storing at least one instruction executed by the processor, wherein the at least one instruction may be configured to acquire source data related to debt, including promissory note information, information related to creditors and debtors, consultation data, and community-based information from a user terminal, perform standardization and normalization processing on the source data, including natural language processing-based text normalization, rule-based parsing, outlier removal, categorical encoding, and generation of derived variables, predict the recovery rate of the debt through an artificial intelligence agent that takes the standardized and normalized data as input, and calculate a grade value indicating the recoverability and risk level of the debt based on the predicted recovery rate.

[0020] Here, the at least one command may be configured to generate a digital bond object that structures attribute information of the bond, including the grade value, the standardized and normalized data, and identification information of the bond.

[0021] Here, the at least one command may be configured to perform natural language processing-based text normalization and rule-based parsing on counseling data or community-based data.

[0022] Here, at least one command may be set to convert the amount, interest rate, maturity, elapsed period, marketing consent, and recurring transaction status included in the source data related to the bond into structured data.

[0023] Here, the at least one command may be configured to collect posts, inquiries, or comments generated by a user presumed to be a debtor from an online community, bulletin board, or anonymous counseling platform, and to generate external community-based derived variables by analyzing the posting time, keyword frequency, avoidance expression patterns, and linguistic features related to legal risk avoidance of the text data.

[0024] Here, the at least one command may be configured to collect information on the type of communication channel used as a means of contact between the debtor and the creditor, contact attempt history by channel, channel change history, response status by channel, and response time, and to analyze the information to generate a communication path-based derived variable indicating the diversity of contact paths, contact avoidance patterns, or contactability.

[0025] Here, the above at least one command may be configured to collect information on the position, operational role, or position of responsibility held by the debtor from an online community, small group, local group, or club platform, and to generate a derived variable representing reputation maintenance pressure or social responsibility burden based on said information.

[0026] Here, the at least one command may be configured to generate a training dataset including time series history of rating values ​​by bond, changes in recovery rates, time series of behavioral signals, and fluctuation information of external community-based derived variables, learn a pattern of fluctuation of rating values ​​using the training dataset, and predict the possibility that the rating value of the bond will rise or fall during a predetermined period based on the pattern of fluctuation.

[0027] Here, at least one command may be configured to recommend the timing of sale or the necessity of sale of the bond when the rating value is determined to be in a relatively high state or immediately before a downward transition.

[0028] Here, the at least one command may be configured to calculate a risk propensity index indicating investment propensity by analyzing investment portfolio information including stock items held or traded by the debtor, trading frequency, volatility, stock group composition ratio, and whether it is dividend-centric, and to adjust the repayment propensity or recovery possibility based on the risk propensity index. Effects of the invention

[0029] According to one embodiment of the present specification, by standardizing various forms of data generated from interpersonal or non-financial financial transactions and utilizing them as input to an artificial intelligence agent, it is possible to predict the debt recovery rate more precisely compared to existing methods. In particular, as unstructured data such as consultation records, text-based responses, and creditor-debtor relationships are quantified and reflected in the model, the semantic differences of unstructured elements can be considered in the process of predicting the recovery rate. This is useful in that it can reflect expanded data characteristics compared to conventional technology that relied only on structured information such as amount, interest rate, and maturity.

[0030] According to one embodiment of the present specification, rating information indicating the risk or recoverability of a debt can be generated based on a predicted recovery rate, which has the effect of helping a creditor determine the relative priority of each debt or efficiently allocate collection resources. Since the predicted recovery rate and rating are provided in a structured form, the creditor can reduce reliance on subjective judgment or experience and establish debt management strategies based on more objective criteria.

[0031] According to one embodiment of the present specification, since a digital bond object is created based on bond-specific attribute information and prediction results, it has the effect of enabling data exchange in a consistent structure even during the integration process with external collection systems, cooperating bond management companies, digital platforms, etc. This can mitigate information asymmetry or management inefficiency that occurred in conventional environments where it was difficult to reflect changes in bond information in real time, and can contribute to forming a digital-based bond management and valuation ecosystem. Brief explanation of the drawing

[0032] FIG. 1 is a system diagram including an artificial intelligence model-based debt recovery prediction service providing server according to one embodiment of the present specification. FIG. 2 is a block diagram showing the configuration of a server providing an artificial intelligence model-based debt recovery prediction service according to one embodiment of the present specification. FIG. 3 is a flowchart illustrating a method for providing an artificial intelligence model-based debt recovery prediction service according to one embodiment of the present specification. Specific details for implementing the invention

[0033] As the present specification is susceptible to various modifications and may have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present specification to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the present specification. Similar reference numerals have been used for similar components in the description of each drawing.

[0034] Terms such as first, second, A, B, etc., may be used to describe various components, but said components should not be limited by said terms. These terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the rights of this specification, the first component may be named the second component, and similarly, the second component may be named the first component. The term "and / or" includes a combination of a plurality of related described items or any of a plurality of related described items.

[0035] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0036] The terms used in this application are used merely to describe specific embodiments and are not intended to limit this specification. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to indicate the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0037] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which this specification pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0038] Hereinafter, preferred embodiments of the present specification will be described in more detail with reference to the attached drawings. To facilitate overall understanding in describing the present specification, the same reference numerals are used for identical components in the drawings, and redundant descriptions of identical components are omitted.

[0039] FIG. 1 is a system diagram including an artificial intelligence model-based debt recovery prediction service providing server according to one embodiment of the present specification.

[0040] Referring to FIG. 1, the method for providing an artificial intelligence model-based debt recovery prediction service according to an embodiment of the present specification may be performed on a computing device that is equipped with storage space and connected to the Internet, such as a PC (Personal Computer), and is not easily portable, or on a portable terminal such as a smartphone. In this case, the method for providing an artificial intelligence model-based debt recovery prediction service may be executed after an application in which the method is implemented is downloaded from an App Store, etc., and installed on the portable terminal.

[0041] In addition, the method for providing a debt recovery prediction service based on the artificial intelligence model described above may be performed by inserting it into a computing device such as a PC while it is recorded on a recording medium such as a CD (Compact Disc) or USB (Universal Serial Bus) memory and executing it through an access operation of said computing device, or it may be performed by storing it from said recording medium into the storage space of said computing device and then executing it through an access operation of said computing device.

[0042] Meanwhile, if the above computing device or portable terminal can access a server connected to the Internet, the method for providing an artificial intelligence model-based debt recovery prediction service may also be executed on the server upon a request from the computing device or portable terminal.

[0043] In the following, a computing device, portable terminal, or server, etc., on which the above-mentioned method for providing a debt recovery prediction service based on an artificial intelligence model is executed may be collectively referred to as an artificial intelligence model-based debt recovery prediction service providing device.

[0044] The above artificial intelligence model-based debt recovery prediction service provider may have the same configuration as the artificial intelligence model-based debt recovery prediction service provider exemplified in FIG. 2, and the above artificial intelligence model-based debt recovery prediction service provider may not be limited to the artificial intelligence model-based debt recovery prediction service provider illustrated in FIG. 1.

[0045] A system according to one embodiment may include a user terminal (110), a user terminal (120), a user terminal (130), and an artificial intelligence model-based debt recovery prediction service providing server (140) (hereinafter, server (140)).

[0046] The user terminal (110), user terminal (120), and user terminal (130) may be, but are not limited to, a vehicle, a mobility device, a smartphone, a tablet PC, a PC, a mobile phone, a PDA (personal digital assistant), a laptop, a media player, a micro server, a GPS (global positioning system) device, and other mobile or non-mobile computing devices. Additionally, the user terminal (110), user terminal (120), and user terminal (130) may be wearable devices equipped with communication functions and data processing functions. However, they are not limited to.

[0047] The server (140) may be implemented as a computer device or a plurality of computer devices that communicate with the user terminal (110), user terminal (120), and user terminal (130) through a network to provide commands, code, files, content, services, etc.

[0048] For example, the server (140) may provide a file for installing an application to a user terminal (110), a user terminal (120), and a user terminal (130) connected via a network. In this case, the user terminal (110), the user terminal (120), and the user terminal (130) may install the application using the file provided by the server (140).

[0049] Additionally, user terminals (110), user terminals (120), and user terminals (130) can connect to a server (140) under the control of an operating system (OS) and at least one program (e.g., a browser or an installed application) to receive services or content provided by the server (140).

[0050] As another example, the server (140) may establish a communication session for data transmission and reception and route data transmission and reception between user terminal (110), user terminal (120), and user terminal (130) through the established communication session.

[0051] User terminals (110), user terminals (120), user terminals (130), and an AI model-based debt recovery prediction service provider server (140) can communicate using a network. For example, the network includes a Local Area Network (LAN), a Wide Area Network (WAN), a Value Added Network (VAN), a mobile radio communication network, a satellite communication network, and combinations thereof, and is a data communication network in a comprehensive sense that enables each network constituent entity shown in FIG. 1 to communicate smoothly with one another, and may include wired internet, wireless internet, and mobile wireless communication networks. In addition, wireless communication may include, for example, wireless LAN (Wi-Fi), Bluetooth, Bluetooth Low Energy, LoRaWAN, Zigbee, WFD (Wi-Fi Direct), UWB (ultra wideband), infrared communication (IrDA, infrared Data Association), NFC (Near Field Communication), etc., but is not limited thereto.

[0052] FIG. 2 is a block diagram showing the configuration of a server providing an artificial intelligence model-based debt recovery prediction service according to one embodiment of the present specification.

[0053] Referring to FIG. 2, a server (200) (hereinafter, server (200)) providing an artificial intelligence model-based debt recovery prediction service may be configured to include a hardware processor (220), a communication unit (210), and a storage unit (230). Only components related to the embodiment are shown in the server (200) of FIG. 2. Therefore, a person skilled in the art will understand that other general-purpose components may be included in addition to the components shown in FIG. 2.

[0054] The communication unit (210) may include hardware such as a modem, network interface card (NIC), wireless communication module, or Bluetooth / Wi-Fi module for transmitting and receiving data to and from a user terminal, an external data provision system, an online community server, or a financial / investment information provision system via a wired / wireless network. Through this, the communication unit (210) can transmit source data related to bonds, user input data, text data provided from an external platform, behavioral signal data, or investment portfolio data to the server (200) in real-time or asynchronously.

[0055] The processor (220) may be implemented as a multi-core structure combining one or more central processing units (CPUs), graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), or at least one of these, and may execute commands stored in the storage unit (230) to preprocess source data related to bonds and perform data normalization logic such as natural language processing-based analysis, rule-based parsing, outlier removal, and generation of derived variables. Additionally, the processor (220) may execute an artificial intelligence agent to calculate recovery rates or grade values ​​based on standardized data, learn grade fluctuation patterns through a time-series learning model, and perform calculations to predict the possibility of future grade increases or decreases. If necessary, the processor (220) may utilize a GPU or TPU to perform text analysis, deep learning-based recovery rate prediction models, time-series prediction models, investment propensity analysis models, etc., at high speed in a parallel processing manner.

[0056] The storage unit (230) includes non-volatile memory or volatile memory accessible by the processor (220), and may include a hard disk drive (HDD), a solid-state drive (SSD), flash memory, storage with a RAID structure, an NVMe-based high-speed storage device, etc. The storage unit (230) may be used to store source data related to bonds, consultation data, community-based text, contact path data, social responsibility-based data, investment portfolio data, training datasets, model parameters, recovery rate prediction models, and rating fluctuation pattern models. Additionally, the storage unit (230) may provide a software environment necessary for the server (200) to actually operate, including an operating system (OS), middleware, a database system (DBMS), API modules, and security modules.

[0057] According to one embodiment of the present specification, the server (200) can operate multiple processor nodes simultaneously by configuring a load balancer and a distributed server environment, and can be implemented in a cloud computing environment (AWS, GCP, Azure) or an on-premises data center environment. The server (200) can perform distributed processing of source data, artificial intelligence analysis, creation of digital bond objects, API response processing, etc., through a separate data processing server, functional microservice modules, or a container-based architecture (Docker, Kubernetes). Since such a physical or logical configuration is a concrete hardware-based system that physically performs actual data input / output, network communication, parallel computation, model training, and storage / retrieval functions rather than simple algorithm execution, it can clearly guarantee industrial applicability and technical effects as a computer-implemented invention.

[0058] Additionally, the server (200) may include a security processor or be linked with an external security module (HSM) to store and process user identity information, bond data, investment data, consultation data, etc., in an encrypted form, and may include an electronic signature module, an access control module, a session management module, etc., to maintain data integrity. The server (200) may include an API server for transmitting bond recovery prediction results and digital bond objects to an external platform or partner system, and through this, it may be linked with other financial systems, collection agencies, and bond sale platforms.

[0059] The processor (220) may be implemented using at least one of 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 other electrical units for performing functions.

[0060] The storage unit (230) may be implemented as a non-transient computer-readable medium and may include both volatile memory (RAM) and non-volatile memory (ROM, flash memory, hard disk, etc.). The storage unit (230) may store the operating system, application programs, communication protocol stack, database, and learning model parameters of the artificial intelligence module of the server device. Additionally, the storage unit (230) may store log data, communication history, information received from an external terminal, etc., which can be utilized for subsequent operations.

[0061] The storage unit (230) may include RAM (random access memory), such as DRAM (dynamic random access memory) or SRAM (static random access memory), ROM (read-only memory), EEPROM (electrically erasable programmable read-only memory), CD-ROM, Blu-ray or other optical disc storage, HDD (hard disk drive), SSD (solid state drive), or flash memory. The storage unit (230) may also be referred to as memory.

[0062] The storage unit (230) may store at least one command executed through the processor (220). The at least one command may be configured to obtain source data related to debt, including promissory note information, information related to creditors and debtors, consultation data, and community-based information, from a user terminal, perform standardization and normalization processing on the source data, including natural language processing-based text normalization, rule-based parsing, outlier removal, categorical encoding, and generation of derived variables, predict the recovery rate of the debt through an artificial intelligence agent that takes the standardized and normalized data as input, and calculate a grade value indicating the recoverability and risk level of the debt based on the predicted recovery rate.

[0063] Here, the at least one command may be configured to generate a digital bond object that structures attribute information of the bond, including the grade value, the standardized and normalized data, and identification information of the bond.

[0064] Here, the at least one command may be configured to perform natural language processing-based text normalization and rule-based parsing on counseling data or community-based data.

[0065] Here, at least one command may be set to convert the amount, interest rate, maturity, elapsed period, marketing consent, and recurring transaction status included in the source data related to the bond into structured data.

[0066] Here, the at least one command may be configured to collect posts, inquiries, or comments generated by a user presumed to be a debtor from an online community, bulletin board, or anonymous counseling platform, and to generate external community-based derived variables by analyzing the posting time, keyword frequency, avoidance expression patterns, and linguistic features related to legal risk avoidance of the text data.

[0067] Here, the at least one command may be configured to collect information on the type of communication channel used as a means of contact between the debtor and the creditor, contact attempt history by channel, channel change history, response status by channel, and response time, and to analyze the information to generate a communication path-based derived variable indicating the diversity of contact paths, contact avoidance patterns, or contactability.

[0068] Here, the above at least one command may be configured to collect information on the position, operational role, or position of responsibility held by the debtor from an online community, small group, local group, or club platform, and to generate a derived variable representing reputation maintenance pressure or social responsibility burden based on said information.

[0069] Here, the at least one command may be configured to generate a training dataset including time series history of rating values ​​by bond, changes in recovery rates, time series of behavioral signals, and fluctuation information of external community-based derived variables, learn a pattern of fluctuation of rating values ​​using the training dataset, and predict the possibility that the rating value of the bond will rise or fall during a predetermined period based on the pattern of fluctuation.

[0070] Here, at least one command may be configured to recommend the timing of sale or the necessity of sale of the bond when the rating value is determined to be in a relatively high state or immediately before a downward transition.

[0071] Here, the at least one command may be configured to calculate a risk propensity index indicating investment propensity by analyzing investment portfolio information including stock items held or traded by the debtor, trading frequency, volatility, stock group composition ratio, and whether it is dividend-centric, and to adjust the repayment propensity or recovery possibility based on the risk propensity index.

[0072] FIG. 3 is a flowchart illustrating a method for providing an artificial intelligence model-based debt recovery prediction service according to one embodiment of the present specification.

[0073] Referring to FIG. 3, the server can obtain source data related to debt (S300). For example, the server can obtain source data related to debt from a user terminal, including promissory note information, information related to creditors and debtors, consultation data, and community-based information.

[0074] According to one embodiment of the present specification, a server can obtain source data related to debt from a user terminal, including promissory note information, information related to creditors and debtors, text data generated during the consultation process, and conversation or post information based on an anonymous community.

[0075] The server can perform standardization and normalization (S310). For example, the server can perform standardization and normalization processing on the source data, including natural language processing-based text normalization, rule-based parsing, outlier removal, categorical encoding, and generation of derived variables.

[0076] Considering that source data may have different formats and structures, the server can operate a multi-stage data normalization module to process various forms of input, such as unstructured text, semi-structured documents, and structured numerical information, according to the same analysis criteria. For example, since a debtor's consultation messages may contain colloquialisms, typos, and duplicate phrases, the server can apply natural language processing-based text normalization techniques to correct sentence structures and separate semantic units. Additionally, structured information such as debt amounts, interest rates, maturities, and elapsed periods can be reconstructed by applying rule-based parsing techniques, and outliers can be removed by identifying missing values, extreme values, and format errors during this process. If necessary, the server can transform source data into a structured form that the model can utilize by performing categorical variable encoding, generation of time and date-based derived variables, analysis of sentiment change patterns in consultation messages, and extraction of risk signals based on keyword frequencies in community posts.

[0077] The server can predict the recovery rate of the bond (S320). For example, the server can predict the recovery rate of the bond through an artificial intelligence agent that takes the standardized and normalized data as input.

[0078] The server can predict the recovery rate of debt by providing standardized and normalized data as described above as input to an artificial intelligence agent. The AI ​​agent can be implemented using machine learning or deep learning-based models and can utilize a multi-input structure that processes structured and unstructured data simultaneously. For example, numerical variables such as a debtor's repayment history, repeat transaction status, and regional information can be processed using a taper data model, while unstructured text, such as consultation text or community posts, can be converted into vector form and combined using a natural language processing-based embedding model. Using this complex data structure, the server can calculate the likelihood of debt recovery in the form of probability values ​​or scores, which can be utilized to predict the probability of repayment within a specific period.

[0079] The server can calculate the recoverability and risk of the receivable (S330). For example, the server can calculate a grade value representing the recoverability and risk of the receivable based on the predicted recovery rate.

[0080] The server can calculate a rating value indicating the recoverability and risk level of a receivable by considering the predicted recovery rate and risk factors derived from source data. The rating value can be determined by comprehensively reflecting factors such as repayment uncertainty, debtor behavior patterns, community-based negative signals, and transaction conditions, and can be designed in various forms depending on the service purpose, such as A–E grades or a continuous value format. Based on this rating value, the server can provide information that can be utilized for establishing receivable management strategies, adjusting collection intensity, and deciding whether to sell the receivable; furthermore, the rating can be updated through iterative forecasting for the same receivable.

[0081] According to one embodiment of the present specification, the server can provide a more stable and consistent assessment of recovery potential compared to existing methods that relied on manual, subjective debt judgments by integrally performing functions such as data acquisition, normalization processing, recovery rate prediction, and grade calculation. Furthermore, since the server can automatically process high-dimensional inputs including various external data and unstructured information, it can improve the proximity to the actual recovery rate by reflecting complex debtor behavioral signals or online activity patterns in the analysis.

[0082] According to one embodiment of the present specification, a server may create a digital bond object by combining a calculated grade value, standardized and normalized data, and unique identification information for identifying a bond. The digital bond object may be configured in a structured data form so that it can be referenced in a consistent format during inter-system linkage or subsequent analysis processes, and the server may record it in a storage unit or transmit it through communication with an external system.

[0083] The server can extend the components of a digital bond object with various attributes depending on the purpose. For example, a digital bond object can be created to include basic bond information (amount, interest rate, maturity, elapsed time), recovery rate prediction results, risk rating values, variables derived from debtor behavior, signals based on external communities, and variables based on creditor-debtor communication patterns. Through this, the digital bond object can be utilized not merely as a simple bond record, but as a unit of analysis necessary for determining recoverability.

[0084] The generated digital bond objects can be used as foundational data for the entire subsequent process performed on the server. For example, they can be utilized as a common reference format in various functions, such as predicting future rating changes, recommending sale timing, adjusting collection strategies, analyzing large-scale bond portfolios, and integrating with bond sale platforms. Furthermore, based on these digital bond objects, the server can facilitate the automation of bond management tasks or API integration with external systems.

[0085] According to the present embodiment, the digital bond object operates not as a concept used within the service, but as an entity constituting the data flow of the entire actual system, and can be stored in the server's storage or an external database. Since such a structured object can accumulate and update various analysis snapshots of the same bond in a relatively single format, it can also be utilized for tracking changes over time or training predictive models. Additionally, depending on the case, it may be possible to record the digital bond object in a tradable form by linking with an immutability-based storage system such as a blockchain.

[0086] According to one embodiment of the present specification, a server may perform natural language processing-based text normalization and rule-based parsing to convert unstructured text contained in counseling data or community-based data into a form that can be utilized by a model. Counseling data may consist of conversations with debtors, counseling records, message exchange contents, etc., and due to its high degree of unstructuredness, preprocessing steps such as sentence-unit segmentation, spell checking, stop word removal, and tokenization may be required. Through these normalization processes, the server can reduce structural noise in the text and separate core semantic units to provide data in a form that allows a future prediction model to learn stably.

[0087] Furthermore, the server can identify specific patterns within the text, such as date expressions, financial transaction expressions, changes in a debtor's stance, and expressions of repayment promises or evasion, through rule-based parsing. For example, phrases like "I will send it next week," "I am in a difficult situation right now," or "I plan to deposit soon" can be interpreted as important signals related to repayment behavior, and the server can convert these into structured features according to specific parsing rules. Since community-based data is also unstructured, sentence patterns, sentiment signals, and risk keywords in posts or comments can be normalized and included in the analysis.

[0088] Such text normalization and parsing processes function not merely as simple string conversions, but as a preprocessing step to identify the debtor's intentions, circumstances, and risk signals. Through this, the server can support the subsequent recovery rate prediction model in accurately reflecting semantic variations occurring in the text. Therefore, the text-based normalization operation according to this embodiment can contribute to improving the accuracy of debt recovery rate prediction.

[0089] According to one embodiment of the present specification, a server may convert structured information, such as amount, interest rate, maturity, elapsed period, marketing consent, and recurrence transaction status, contained in source data related to bonds into structured data for use in a prediction model. Bond amount, interest rate, or maturity information may be obtained through document images, message-based input, or online forms, and there may be cases where the format or unit of the data is inconsistent. The server may convert these values ​​into a unified numerical format and reconstruct them into normalized features by applying standard units.

[0090] The elapsed period can be calculated based on the difference between the date the debt was incurred or the date the promissory note was drawn up and the current point in time, and the server can parse the variety of date formats to generate time-based values ​​in daily or weekly units. Whether marketing consent or recurring transactions have occurred can be extracted from previous interactions with the debtor or transaction history within the platform, and can be converted into binary or categorical values ​​to be used as input values ​​for the model.

[0091] In addition, the server may perform procedures for correcting missing values, verifying ranges, or checking data to ensure the consistency and reliability of structured information. For example, if negative monetary values, abnormally high interest rates, or unrealistic maturity values ​​are entered, the server may perform processing such as removing the values ​​or converting them to replacement values ​​according to predefined criteria. This structured data transformation process supports the recovery rate prediction model in reliably interpreting numerical-based features and can contribute to improving the quality of model input.

[0092] According to one embodiment of the present specification, a server may obtain external signals related to debt recovery by collecting posts, inquiries, or comments generated by a user presumed to be a debtor from an online community, bulletin board, or anonymous consultation platform. Such text data may include expressions implying financial transactions, delayed repayment, avoidance of collection, consideration of legal action, financial hardship, family conflict, or other debt situations, and the server may apply keyword-based filtering or natural language processing-based search techniques to extract the core meaning within these posts. Furthermore, since users are likely to describe their situations more frankly based on the anonymity inherent in the platform, community-based data can be utilized as useful external material reflecting the debtor's actual behavioral intentions or psychological state.

[0093] The server can generate derived variables to assess the likelihood of a debtor avoiding recovery by analyzing the collected text, including the time of publication, keyword frequency, and patterns of evasive expressions. For example, expressions revealing evasion or anxiety, such as "going silent," "changing phone number," "it is difficult to reach you right now," or "worried about legal issues," can be interpreted as signals of a declining recovery rate. If such expressions are repeated over a certain period or their frequency increases, the server can generate derived variables that set the likelihood of recovery avoidance high.

[0094] Information on the time of posting can be used to determine whether there has been a recent surge in activity, whether it is concentrated in specific time periods, or whether the same patterns are repeated. The server can also incorporate this into the analysis by considering the possibility that changes in text patterns over time may be associated with changes in the debtor's behavior. For example, if negative expressions suddenly increase after a certain point in time or similar posts are repeatedly generated within a short period, this may indicate intensified economic pressure or an increased intention to avoid recovery.

[0095] According to one embodiment of the present specification, a server may collect various information related to communication channels to determine the possibility of contact or the tendency to avoid contact between a debtor and a creditor. Communication channels may consist of telephones, text messages, messenger applications, emails, and Direct Messages (DMs) on social platforms, and the server may analyze the debtor's contact behavior based on the type of channel actually used, the history of contact attempts, whether a response was made by channel, response delay times, and channel switching patterns. For example, if a debtor does not respond on a specific channel but responds intermittently on an alternative channel, this may be associated with selective avoidance or debt avoidance behavior.

[0096] The server can calculate the diversity of contact paths based on collected communication channel information, which can indicate whether the creditor's contact attempts were made continuously across multiple channels or concentrated on a specific channel. Additionally, response time analysis can be utilized to quantitatively assess the debtor's reaction speed; for instance, a sudden tendency toward delayed responses can be interpreted as a sign of increased likelihood of debt avoidance. By reflecting these channel-based quantitative indicators and changes in time patterns, the server can generate communication path-based derived variables that indicate contact avoidance patterns, the likelihood of contact, or the likelihood of maintaining contact.

[0097] In particular, since a debtor’s actions of changing or blocking existing contact channels can act as a leading signal of a decline in recovery rates, the server can analyze the frequency of channel changes, the timing of the changes, and response patterns following the changes to reflect them as risk signals. For instance, a pattern where communication is maintained solely via text messages or messenger after a phone number change may be associated with an intention to evade collection or avoid the exposure of personal information.

[0098] According to one embodiment of the present specification, considering that a debtor's social role or position of responsibility may affect the possibility of recovery, a server may collect information related to the position, operational role, or position of responsibility within an organization currently performed by a debtor from online communities, small groups, local organizations, or hobby / activity-based club platforms. This information may consist of data indicating whether a specific role is performed, such as whether the debtor is a member of the group's management team, the role of a bulletin board administrator, records of participation in event operations, or whether the debtor is a team leader. The server may further analyze whether the role is publicly verifiable, the size of the community in which the role is performed, the intensity or difficulty of the responsibility, and how long the role has been maintained.

[0099] The server determines that the aforementioned information may influence the debtor's reputation management pressure or the burden of social responsibility; therefore, it can assess whether a debtor is likely to increase their willingness to repay in order to avoid a negative reputation when performing operational or managerial roles within a large group with high levels of role disclosure. Furthermore, since maintaining trust and a sense of obligation within social networks may be relatively strong when performing multiple roles simultaneously within a short period or holding a role with significant operational responsibility for an extended period, the server can generate social responsibility-based derived variables by reflecting these factors.

[0100] For example, if a debtor has a record of serving on the management team of a local self-help group or performing an administrator role for a certain period in an online community with a large number of members, the server can quantify the burden of social responsibility based on the duration of the role, the size of the membership, and the level of publicity. Conversely, roles performed short-term or temporarily in small, highly anonymous groups may have low reputational risk and thus a low direct correlation with the likelihood of repayment; in such cases, the server can set the social responsibility-based derived variables to a low value.

[0101] In this way, the server can estimate the social behavioral tendencies of debtors by analyzing the nature of roles and levels of responsibility within the organization from various angles. By providing this as an auxiliary input to the recovery rate prediction model, it can incorporate the psychological behavioral factors of debtors into the prediction process, which were difficult to assess using existing structured data alone. This contributes to a more sophisticated identification of the likelihood of repayment avoidance and can help improve the performance of the overall prediction model.

[0102] According to one embodiment of the present specification, a server may generate a training dataset that analyzes time-series changes in rating values ​​to more precisely evaluate the recoverability and risk of a debt. The server may collect multiple time-series information, including rating values ​​calculated at regular intervals, the difference between a previous prediction point and the actual recovery situation, behavioral signals of the debtor, or temporal patterns of increase or decrease of external community-based derived variables, and configure this information into a training dataset for explaining or predicting rating changes. This time-series information may be linked to the timing of repayment delays, the timing of increased contact avoidance patterns, or the timing of a surge in negative keywords, and through this, the server can identify the correlation between changes in the debtor's behavior and rating changes.

[0103] The server can train an artificial intelligence agent to learn the fluctuation patterns of rating values ​​using the generated training dataset. In this process, Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTMs), Gated Recurrent Units (GRUs), or deep learning structures suitable for time-series volatility analysis may be used, and the server can adjust parameters within the model to identify correlations over time, periods of rapid fluctuation, and gradual downward or upward trends. During the model's training process, the server learns whether a surge in specific behavioral signals acts as a leading indicator of a rating downgrade, or whether a relationship exists between an increase in retrieval avoidance expressions and rating volatility, thereby enabling the server to interpret rating fluctuation patterns.

[0104] According to one embodiment of the present specification, a server can predict the direction of rating changes over a predetermined period based on learned fluctuation patterns. For example, it can determine whether there is a high probability of a rating upgrade over a future period, whether there is a risk of a downgrade, or whether there is a possibility that the rating value will become unstable due to increased volatility. Unlike conventional methods that make decisions based solely on the current rating, these prediction results can be utilized to detect changes in risk over time at an early stage and to adjust bond management strategies in advance.

[0105] According to one embodiment of the present specification, a server may determine the timing of the sale of a bond or the necessity of selling it based on the prediction results of a rating change pattern. The server may interpret as a risk signal if it is predicted that the likelihood of the rating value rising or stabilizing over a certain period decreases, while conversely, the likelihood of a downward reversal increases; such prediction results may be derived from the output value of an artificial intelligence agent that analyzes the rating change pattern. For example, if volatility increases sharply and the uncertainty of the rating prediction rises, or if external behavioral signals and community-based signals simultaneously fluctuate in a negative direction, the server may determine that the current rating value is unlikely to be maintained in the future.

[0106] The server can suggest to the user a time to consider selling if, through the analysis of rating fluctuation patterns, it determines that a specific point in time is maintaining a relatively high rating value—that is, if it is judged to be a phase prior to an expected future decline. In this process, the server does not make a judgment based solely on the absolute value of the rating, but can also incorporate complex factors such as the slope of the rating's rise or fall, the extent of volatility expansion, and correlations with external signals. For instance, even if a slight upward trend has been maintained over the past few days, if a signal indicating a downward reversal is detected, it can be judged as the end of the uptrend and presented as a candidate for a selling point.

[0107] The server can provide recommendation information to user terminals or linked debt management systems to perform the function of recommending sale timing; this can lead not only to simple notifications but also to specific follow-up tasks, such as registering items on sale review lists and classifying priority recovery targets. Additionally, the server may decide to postpone the sale if it determines that the rating fluctuation pattern is in a stable upward or static state, thereby preventing the inefficient execution of debt management strategies through such dual judgment. The server can repeatedly determine sale timing by continuously updating predicted values ​​of rating fluctuation patterns to reflect volatility in real time, which supports the immediate incorporation of rapidly changing behavioral signals or risk factors based on external communities into the model.

[0108] The sales timing recommendation function according to the present embodiment does not stop at merely predicting the recovery rate, but considers various time series characteristics such as periods of increased volatility, periods of downward reversal, and the risk of a sharp decline after a short-term rebound to provide information that can be directly utilized in actual decision-making, thereby contributing to increasing the efficiency of bond management.

[0109] According to one embodiment of the present specification, the server may utilize investment-related information, such as stock items held or traded by the debtor, stock group composition, trading frequency, volatility level, and whether the investment is dividend-oriented, taking into account that the debtor's financial propensity or risk preference may also affect the possibility of recovery. The server may obtain investment portfolio information from an external financial information provision system, an asset inquiry function of a user terminal, or an affiliated investment management platform, and may calculate a risk propensity index representing the debtor's investment propensity based thereon.

[0110] The server can determine whether a debtor is stability-oriented, neutral, or high-risk seeker by analyzing various factors, such as the proportion of blue-chip stocks, high-volatility stocks, frequency of short-term trading, holding of leveraged stocks, and concentration on dividend-focused ETFs. For example, frequent trading aimed at short-term profits, an increased proportion of high-risk and high-volatility stocks, or investment patterns centered on leveraged products may indicate a debtor's risk preference sensitive to fluctuations in asset value and suggest instability in repayment ability due to changes in economic conditions. Conversely, a portfolio centered on dividend stocks or stable ETFs may imply a possibility of having relatively stable consumption and financial habits.

[0111] According to one embodiment of the present specification, the server may reflect the calculated risk propensity index in the adjustment of recoverability. For example, if the risk propensity index is high and the debtor's financial condition is highly likely to be significantly affected by market volatility, the server may apply a risk correction factor to lower the actual recoverability, even if the predicted recovery rate is the same. Conversely, if a conservative and stability-oriented investment pattern appears, the server may determine that the likelihood of repayment is relatively high and adjust the recoverability upward. Through this adjustment process, the server can provide a more precise assessment of the recovery rate that combines the debtor's behavioral and financial characteristics.

[0112] In addition, the server can track trends in investment propensity over time; for example, patterns of sudden shifts from high-risk stocks to stable stocks, or conversely, shifts from a stable portfolio to short-term high-risk stocks, can reflect potential changes in the debtor's financial situation. The server can enhance the overall adaptability of the model by continuously reflecting the impact of these changes on recovery rate prediction results.

[0113] As such, the investment propensity analysis according to the present embodiment can reflect the debtor's behavior-based financial propensity, which was difficult to verify using only existing consultation data or basic financial data, thereby contributing to improving the accuracy and effectiveness of recovery rate prediction.

[0114] The operation according to the embodiments of this specification may be implemented as a computer-readable program or code on a computer-readable recording medium. A computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored. Additionally, a computer-readable recording medium may be distributed across networked computer systems, allowing computer-readable programs or code to be stored and executed in a distributed manner.

[0115] When the embodiment is implemented in software, the above-described technique may be implemented as a module (process, function, etc.) that performs the above-described function. The module may be stored in memory and executed by a processor. The memory may be located inside or outside the processor and may be connected to the processor by various well-known means.

[0116] In addition, computer-readable recording media may include hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Program instructions may include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

[0117] Some aspects of this specification have been described in the context of a device, but may also be described according to a corresponding method, wherein a block or device corresponds to a method step or a feature of a method step. Similarly, aspects described in the context of a method may also be described according to a corresponding block or item or a feature of a corresponding device. Some or all of the method steps may be performed by (or using) a hardware device, such as, for example, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, one or more of the most important method steps may be performed by such a device.

[0118] In the embodiments, a programmable logic device (e.g., a field-programmable gate array) may be used to perform some or all of the functions of the methods described herein. In the embodiments, the field-programmable gate array may operate with a microprocessor to perform one of the methods described herein. Generally, it is preferable that the methods be performed by some hardware device.

[0119] Although the foregoing has been described with reference to preferred embodiments of this specification, those skilled in the art will understand that various modifications and changes can be made to this specification without departing from the spirit and scope of the specification as set forth in the following claims.

Claims

Claim 1 A method for providing an AI model-based debt recovery prediction service performed by at least one server, comprising: a step of obtaining source data related to debt, including promissory note information, information related to creditors and debtors, consultation data, and community-based information from a user terminal; a step of collecting text data containing keywords related to monetary transactions, delayed repayment, avoidance of collection, legal response, or debt repayment among posts, inquiries, and comments generated by a user presumed to be a debtor from an online community, bulletin board, or anonymous consultation platform; a step of generating an external community-based derived variable indicating the debtor's possibility of avoiding recovery by analyzing the text data regarding the time of posting, frequency of keyword appearance, avoidance expression patterns, and linguistic features related to the intention to avoid legal risk; a step of collecting information on the types of communication channels used as means of contact between the debtor and the creditor, contact attempt history by channel, channel change history, response status by channel, and response time; a step of generating a communication path-based derived variable indicating the diversity of contact paths, contact avoidance patterns, or contact possibility by analyzing the collected communication channel-related information; a step of collecting information indicating the position, operational role, or position of responsibility within an organization held by the debtor from an online community, small group, local group, or club platform; and the position or operation A step of generating a social responsibility-based derivative variable indicating the possibility of recovery by calculating the debtor's reputation maintenance pressure or social responsibility burden based on the number of roles, the intensity of responsibility, the number of members, and whether the corresponding roles are disclosed; a step of performing standardization and normalization processing, including natural language processing-based text normalization, rule-based parsing, outlier removal, and categorical encoding, on debt-related data including the generated external community-based derivative variable, communication path-based derivative variable, and social responsibility-based derivative variable in the source data;A method for providing an AI model-based bond recovery prediction service, comprising: a step of predicting the recovery rate of the bond through an AI agent that takes the standardized and normalized data as input; a step of calculating a rating value representing the recoverability and risk of the bond based on the predicted recovery rate; a step of generating a training dataset including a time series history of the rating value for each bond, changes in the recovery rate over time, a time series of debtor behavior signals, and fluctuation information of external community-based derived variables; a step of training the AI ​​agent to learn the fluctuation pattern of the rating value using the training dataset and calculating model parameters representing the fluctuation pattern; a step of predicting the likelihood that the rating value of the bond will rise or fall over a predetermined period based on the calculated fluctuation pattern; and a step of automatically recommending the timing of sale or the necessity of sale of the bond when the rating value is determined to be in a relatively high state or on the verge of a downward transition based on the prediction result of the fluctuation pattern indicating the likelihood that the rating value of the bond will fall in the future or that volatility will increase, if the rating value is determined to be in a relatively high state or a state immediately before a downward transition. Claim 2 A method for providing an artificial intelligence model-based bond recovery prediction service, further comprising the step of generating a digital bond object that structures attribute information of the bond, including the grade value, the standardized and normalized data, and identification information of the bond, in accordance with claim 1. Claim 3 A method for providing an artificial intelligence model-based debt recovery prediction service, wherein the step of performing the standardization and normalization processing according to claim 1 includes the step of performing natural language processing-based text normalization and rule-based parsing on consultation data or community-based data. Claim 4 A method for providing an artificial intelligence model-based debt recovery prediction service, wherein the step of performing the standardization and normalization processing according to claim 1 includes converting the amount, interest rate, maturity, elapsed period, whether marketing consent and whether recurring transactions are included in the source data related to the debt into structured data. Claim 5 delete

Images