Method for analyzing life metrics based on financial / non-financial factors and providing personalized consulting, and artificial intelligence system for performing same

Abstract

A method for operating an artificial intelligence system according to one aspect of the present invention comprises the steps of: obtaining a plurality of pieces of training data including at least one financial variable, at least one non-financial variable, and a label; generating a plurality of pieces of primary temporary data on the basis of the acquired plurality of pieces of training data; inputting the plurality of pieces of primary temporary data to a temporary data detection model including a machine learning model; and training a metrics prediction model including a machine learning model using at least one piece of secondary temporary data classified as real data by the temporary data detection model among the plurality of pieces of primary temporary data.

Description

A method for analyzing life indicators and providing personalized consulting based on financial and non-financial factors, and an artificial intelligence system for performing this.

[0001] The technical idea of ​​this disclosure relates to a method for analyzing various life indicators related to a user's life, such as happiness, based on various financial / non-financial factors of the individual, and providing personalized consulting based on the analysis results, and an artificial intelligence system that performs this.

[0002]

[0003] In a capitalist society, quality-of-life indicators, such as happiness, can be significantly influenced by economic factors. However, they can also be judged in a complex manner based on a variety of factors, including individual values ​​and perceived priorities. However, many existing happiness assessment methods tend to focus primarily on financial factors such as an individual's income level and assets, and are limited by a lack of indicators that allow for analysis from various perspectives.

[0004] In addition to these happiness indicators, there are various indicators related to a person's life or living, and these indicators can be utilized in various fields for purposes such as improving the quality of life. However, like happiness indicators, there is a problem in that the analysis is not carried out accurately.

[0005] Meanwhile, deep learning models have become a key technology in the field of artificial intelligence and are widely used in various industries to solve problems such as prediction and classification. While these deep learning models excel at learning large amounts of data and inferring complex patterns, their performance can be degraded by missing data. Various techniques have been proposed to handle this missing data, but most are limited to cases where specific variables are missing.

[0006] As the application of AI technology expands, recent attempts are being made to apply it to areas such as the analysis of social science phenomena and financial and non-financial consulting. In these social science analyses, the analysis of new variables—that is, the emergence of new social phenomena or issues—can be crucial. When an AI model must learn to incorporate new variables, the existing training dataset may lack values ​​for these new variables, resulting in missing data.

[0007]

[0008] The task to be solved by the present invention is to provide a method that can predict indicators related to the quality of life of a user, such as the user's happiness, based on various financial / non-financial factors, and effectively provide feedback or consulting to improve the quality of life by improving the indicators.

[0009] The problem that the present invention seeks to solve is to provide a method for constructing various debt management assessment scenarios for evaluating an individual's debt management ability.

[0010] The problem that the present invention seeks to solve is to provide a method for improving an individual's debt management ability by providing feedback or education based on the results of an evaluation of debt management ability.

[0011] The problem that the present invention seeks to solve is to provide a method that can perform effective learning even when there is insufficient data for learning an artificial intelligence model for predicting indicators.

[0012] The task to be solved by the present invention is to provide a method for performing learning of an artificial intelligence model that reflects new variables when a learning dataset with new variables added is acquired.

[0013] The task to be solved by the present invention is to provide a method for predicting and interpreting various indicators related to life based on a learned artificial intelligence model.

[0014]

[0015] In order to achieve the above object, an operating method of an artificial intelligence system according to an aspect of the technical idea of ​​the present disclosure includes the steps of: acquiring a plurality of learning data each including at least one financial variable, at least one non-financial variable, and a label; generating a plurality of primary pseudo data based on the acquired plurality of learning data; inputting the plurality of primary pseudo data into a pseudo data detection model trained to classify input data as real data or pseudo data; and training an artificial intelligence model using at least one secondary pseudo data classified as real data by the pseudo data detection model among the plurality of primary pseudo data.

[0016] According to one embodiment, the step of generating the plurality of primary data may include the step of setting labels of the primary data to be generated based on the plurality of learning data; the step of filtering learning data having a label whose difference from the set label is equal to or less than a first value among the plurality of learning data; and the step of generating the plurality of primary data based on the filtered learning data.

[0017] According to one embodiment, the step of setting the label of the primary data to be generated may include the step of setting the label of the primary data to be generated based on a ratio of labels of the plurality of learning data.

[0018] According to one embodiment, the step of generating the plurality of primary data based on the filtered learning data may include, when there are a plurality of filtered learning data, the step of selecting any two learning data among the plurality of filtered learning data; and the step of generating the primary data based on a weighted average of the two selected learning data.

[0019] In one embodiment, the step of generating the plurality of primary data based on the filtered learning data includes the step of filtering learning data having a label whose difference from the set label is equal to or less than a second value, if the number of filtered learning data is less than two; and the step of generating the plurality of primary data based on the filtered learning data, wherein the second value may have a value greater than the first value.

[0020] According to one embodiment, the method may further include the steps of collecting input data including financial data and non-financial data about a user; obtaining real data based on some of a plurality of variables included in the collected input data; and obtaining a life indicator prediction result for the user from the real data using the learned artificial intelligence model.

[0021] According to one embodiment, the step of obtaining the life indicator prediction result may include the step of assigning the user to one of a plurality of predefined groups based on some of the plurality of variables included in the input data; and the step of obtaining the life indicator prediction result for the user from the real data using an artificial intelligence model corresponding to the group to which the user is assigned, among a plurality of artificial intelligence models learned for each group.

[0022] According to one embodiment, the method may further include a step of performing an interpretation on the learned artificial intelligence model.

[0023] According to one embodiment, the step of performing an interpretation on the artificial intelligence model may include the steps of calculating feature importance for each of the plurality of variables; generating a partial dependence plot (PDP) for the plurality of variables; and extracting decision rules of a plurality of decision trees included in the artificial intelligence model.

[0024] In one embodiment, the artificial intelligence model may include a random forest model composed of a plurality of decision trees.

[0025] A method of operating an artificial intelligence system according to one aspect of the technical idea of ​​the present disclosure includes the steps of collecting input data including financial data and non-financial data about a user; obtaining real data based on some of a plurality of variables included in the collected input data; and obtaining a life indicator prediction result for the user from the real data using an artificial intelligence model, wherein the artificial intelligence model may include a random forest model composed of a plurality of decision trees.

[0026] An artificial intelligence system including at least one computing device according to one aspect of the technical idea of ​​the present disclosure includes: a pseudo-data generation unit that generates a plurality of primary pseudo-data based on a plurality of learning data, wherein the plurality of learning data includes at least one financial variable, at least one non-financial variable, and a label; a pseudo-data detection model trained to detect and classify whether each of the plurality of primary pseudo-data is real data or pseudo-data; and an artificial intelligence model trained to predict a life index related to quality of life from input data using at least one secondary pseudo-data misclassified as real data by the pseudo-data detection model among the plurality of primary pseudo-data, and the plurality of learning data.

[0027] According to one embodiment, the data generation unit may set a label of primary data to be generated based on the plurality of learning data, filter out learning data having a label whose difference from the set label is less than or equal to a first value among the plurality of learning data, and generate the plurality of primary data based on the filtered learning data.

[0028] According to one embodiment, when there are multiple filtered learning data, the data generation unit may select any two learning data among the multiple filtered learning data, and generate primary data based on a weighted average of the two selected learning data.

[0029] According to one embodiment, when there are less than two pieces of filtered learning data, the data generation unit filters learning data having a label whose difference from the set label is equal to or less than a second value, and generates the plurality of primary data based on the filtered learning data, wherein the second value may be greater than the first value.

[0030] An artificial intelligence system including at least one computing device according to one aspect of the technical idea of ​​the present disclosure includes a data collection and preprocessing unit that collects input data including financial data and non-financial data about a user, and obtains real data based on some of a plurality of variables included in the collected input data; and a prediction unit that obtains a life indicator prediction result for the user from the real data using an artificial intelligence model, wherein the artificial intelligence model may include a random forest model composed of a plurality of decision trees.

[0031]

[0032] According to an embodiment of the present disclosure, more accurate indicator prediction and analysis for users can be made by utilizing an artificial intelligence model learned using various financial / non-financial variables.

[0033] In addition, according to an embodiment of the present disclosure, when a variable corresponding to an indicator to be analyzed among the financial / non-financial variables is selected, a new artificial intelligence model can be easily learned and constructed by setting the selected variable as a label.

[0034] In addition, according to an embodiment of the present disclosure, by assigning a user to a group based on variables collected about the user and predicting the user's indicator using an artificial intelligence model corresponding to the assigned group, the problem of decreased prediction accuracy due to differences in characteristics by group can be effectively resolved.

[0035] In addition, according to an embodiment of the present disclosure, a plurality of fake data can be generated from learning data, and an artificial intelligence model can be trained by utilizing fake data having characteristics similar to real data among the generated fake data, so that even if sufficient learning data is not secured, accuracy can be improved through sufficient training of the artificial intelligence model.

[0036] Furthermore, according to embodiments of the present disclosure, when a new learning dataset containing new variables is acquired, the AI ​​system can easily build an AI model that reflects the influence of the new variables by performing new variable processing on the existing learning dataset and then utilizing the new learning dataset together with the new variable for AI model training. Through this method, the AI ​​system can effectively implement and provide an AI model that evolves with the addition of variables.

[0037] In addition, according to an embodiment of the present disclosure, the artificial intelligence system can construct various types of evaluation scenarios by obtaining environment setting information of evaluation scenarios to be applied to a debt management ability evaluation simulation from a user of the terminal, and thus can evaluate the user's debt management ability level more accurately and from various perspectives through various evaluation scenarios.

[0038] In addition, according to an embodiment of the present disclosure, the artificial intelligence system can effectively analyze various debt management data that can be acquired as various evaluation scenarios progress, using a deep learning model, etc., and evaluate debt management capabilities, thereby improving the reliability of the evaluation results and effectively providing more optimized feedback information to the user.

[0039] The effects according to the technical idea of ​​the present disclosure are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present invention belongs from the description below.

[0040]

[0041] To more fully understand the drawings cited in this disclosure, a brief description of each drawing is provided.

[0042] Figure 1 is an example diagram of an environment in which an artificial intelligence system according to an exemplary embodiment of the present disclosure is applied.

[0043] Figure 2 is a block diagram showing an example configuration of the artificial intelligence system illustrated in Figure 1.

[0044] FIG. 3 is a diagram showing the connection relationship and data flow between components included in an artificial intelligence system according to an exemplary embodiment of the present disclosure.

[0045] FIG. 4 is a diagram for explaining in more detail the learning operation of the model learning unit according to an exemplary embodiment of the present disclosure.

[0046] FIGS. 5 to 7 are drawings for explanation related to an artificial intelligence model according to an exemplary embodiment of the present disclosure.

[0047] FIG. 8 is a flowchart for explaining a learning method of an artificial intelligence model according to an embodiment of the present disclosure.

[0048] FIG. 9 is a flowchart for explaining a learning method of an artificial intelligence model according to one embodiment of the present disclosure.

[0049] Fig. 10 is an example diagram for explaining the artificial intelligence model learning operation of the artificial intelligence system according to the learning method of Fig. 9.

[0050] FIG. 11 is a flowchart illustrating a method for an artificial intelligence system according to an exemplary embodiment of the present disclosure to obtain and provide prediction results and interpretation results from input data using an artificial intelligence model.

[0051] Figures 12 to 16 are drawings for explanation related to interpretation data generated and provided through the artificial intelligence system of Figure 3.

[0052] FIG. 17 is a block diagram of components related to the debt management capability evaluation operation of an artificial intelligence system according to an exemplary embodiment of the present disclosure.

[0053] Figure 18 is a diagram showing an example of a configuration of the debt management data collection unit illustrated in Figure 17.

[0054] Figures 19 to 21 are examples showing a simulation interface provided by the debt management capability evaluation simulation control unit illustrated in Figure 18.

[0055] FIG. 22 is an exemplary diagram illustrating a process for evaluating debt management capabilities and providing feedback information according to an exemplary embodiment of the present disclosure.

[0056] FIG. 23 is a flowchart illustrating a debt management capability evaluation method according to an exemplary embodiment of the present disclosure.

[0057] Figure 24 is a flowchart illustrating a specific embodiment of the debt management data acquisition step illustrated in Figure 23.

[0058] Figure 25 is a flowchart illustrating a specific embodiment of the debt management capability evaluation and feedback information provision steps illustrated in Figure 23.

[0059] FIG. 26 is a schematic hardware configuration block diagram of a computing device constituting an artificial intelligence system according to an exemplary embodiment of the present disclosure.

[0060]

[0061] The exemplary embodiments according to the technical idea of ​​the present disclosure are provided to more completely explain the technical idea of ​​the present disclosure to those skilled in the art. The embodiments below may be modified in various different forms, and the scope of the technical idea of ​​the present disclosure is not limited to the embodiments below. Rather, these embodiments are provided to more faithfully and completely convey the technical idea of ​​the present disclosure to those skilled in the art.

[0062] Although terms such as first, second, etc. are used in this disclosure to describe various members, regions, layers, portions, and / or components, it is to be understood that these members, parts, regions, layers, portions, and / or components should not be limited by these terms. These terms do not imply a specific order, hierarchy, or order, and are only used to distinguish one member, region, portion, or component from another member, region, portion, or component. Accordingly, a first member, region, portion, or component described below may refer to a second member, region, portion, or component without departing from the teachings of the technical spirit of this disclosure. For example, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component, without departing from the scope of the present disclosure.

[0063] Unless otherwise defined, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by those of ordinary skill in the art to which the concepts of this disclosure pertain. Furthermore, commonly used terms, as defined in dictionaries, should be interpreted to have a meaning consistent with their meaning within the context of the relevant technology, and should not be interpreted in an overly formal sense unless explicitly defined herein.

[0064] In cases where an embodiment is otherwise feasible, a particular process sequence or sequence of steps may be performed in a different order than the one described. For example, two processes or steps described in succession may be performed substantially simultaneously, or in a reverse order from the one described.

[0065] In addition, terms such as “~part”, “~device”, “~sub-subject”, and “~module” described in this specification mean a unit that processes at least one function or operation, which may be implemented by hardware such as a processor, a microprocessor, a microcontroller, a central processing unit (CPU), a graphics processing unit (GPU), an accelerate processor unit (APU), a drive signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a combination of hardware and software, and may also be implemented in a form combined with a memory that stores data necessary for processing at least one function or operation.

[0066] In addition, it should be clarified that the division of components in this specification is merely a division based on the main function of each component. In other words, two or more components described below may be combined into one component, or one component may be further divided into two or more components with more detailed functions. In addition to its own main function, each component described below may additionally perform some or all of the functions performed by other components, and of course, some of the main functions of each component may be exclusively performed by other components.

[0067] The term 'and / or' as used herein includes each and every combination of one or more of the mentioned absences.

[0068] Hereinafter, embodiments according to the technical idea of ​​the present disclosure will be described in detail with reference to the attached drawings.

[0069] Figure 1 is an example diagram of an environment in which an artificial intelligence system according to an exemplary embodiment of the present disclosure is applied.

[0070] Referring to FIG. 1, the artificial intelligence system (10) according to an embodiment of the present disclosure may refer to a combination of software and hardware designed to enable a computer to learn, think, and solve problems like a human. The artificial intelligence system (10) can learn and predict data through machine learning, deep learning, and other methods.

[0071] For example, the artificial intelligence system (10) can collect (acquire) various financial / non-financial data of the user and predict various indicators (life metrics) related to the user's life based on the collected data. In this specification, various embodiments of the present disclosure are described using happiness among the various indicators, but the indicator may include various factors related to the quality of life such as life satisfaction, degree of freedom, psychological well-being, economic stability, social relationships, self-realization, etc., or factors related to social phenomena such as child planning and education expenses. In addition, the artificial intelligence system (10) can provide the interpretation results of the learned indicator prediction model, thereby identifying the main factors related to the corresponding indicator and helping to determine guidelines or policies for improving the corresponding indicator. In addition, the artificial intelligence system (10) can analyze factors (variables) that mainly influence the indicators of a user through interpretation of the indicator prediction results for the user, and the analysis results can be provided to a server (30) that provides services in various fields (finance, psychological counseling, etc.) and used for consulting or advising to improve the indicators of the user.

[0072] Such an artificial intelligence system (10) may be configured to include at least one computing device. For example, each of the at least one computing device may be a hardware-based device including a processor, memory, communication unit, input unit, and / or output unit, and may include a server, a PC, etc. In this case, the components (modules) included in the artificial intelligence system (10) may be implemented as hardware, software, or a combination thereof, and may be implemented by being integrated or divided into the at least one computing device.

[0073] A user can access the artificial intelligence system (10) via a terminal (20) and provide data. Furthermore, the user can receive and confirm prediction results, interpretation results, consulting information, etc. from the artificial intelligence system (10) via the terminal (20). Such terminals (20) may include various electronic devices such as smartphones, tablet PCs, desktop PCs, and laptops.

[0074] The server (30) can be connected to the artificial intelligence system (10) and the terminal (20) via a network. As described above, the server (30) can receive analysis results on factors (variables) affecting the user's indicators from the artificial intelligence system (10) and utilize the analysis results when providing services in specific fields (finance, psychological counseling, etc.). Alternatively, the server (30) can be a server of an institution or company that manages various financial / non-financial data on the user, and in this case, the server (30) can provide at least some of the financial / non-financial data for predicting and analyzing the user's indicators to the artificial intelligence system (10).

[0075] FIG. 2 is a block diagram illustrating an example configuration of the artificial intelligence system illustrated in FIG. 1. FIG. 3 is a diagram illustrating the connection relationships and data flow between components included in an artificial intelligence system according to an exemplary embodiment of the present disclosure. FIG. 4 is a diagram for explaining in more detail the learning operation of a model learning unit according to an exemplary embodiment of the present disclosure.

[0076] Referring to FIG. 2, the artificial intelligence system (10) may include a data collection and preprocessing unit (110), a user group assignment unit (120), a model learning unit (130), a predictor unit (140), an interpreter unit (150), and a database (160). The artificial intelligence system (10) according to an embodiment of the present disclosure is not limited to the configuration of FIG. 2 and may include more or fewer configurations.

[0077] The data collection and preprocessing unit (110) collects input data from a user and, through preprocessing the collected input data, obtains at least one variable that can be input into an artificial intelligence model. Depending on the embodiment, the preprocessing process for the input data may not be performed.

[0078] Specifically, referring to the embodiment illustrated in FIG. 3, the AI ​​system in this embodiment may be a system that predicts indicators related to a user's quality of life (life indicators) from the user's financial / non-financial data and interprets the prediction results. In this case, the data collection and preprocessing unit (110) may include a financial data collection unit (112), a non-financial data collection unit (114), and a data preprocessing unit (116).

[0079] The financial data collection unit (112) can collect financial data including financial variables such as the user's assets, income, debts, monthly spending amount, and savings amount. For example, the financial data collection unit (112) can obtain the financial data by providing an interface that prompts the user to input the financial variables through the terminal (20). Alternatively, the financial data collection unit (112) can receive financial data about the user from a server (30) of a financial institution such as a bank, securities company, or insurance company, or other various institutions / companies that manage financial data.

[0080] The non-financial data collection unit (114) can collect non-financial data including various non-financial variables such as the user's unique information, marital status / child status, occupation, industry, health status, etc. For example, the non-financial data collection unit (114) can obtain the non-financial data by providing an interface (such as a survey interface) for collecting the non-financial variables through the terminal (20). The interface may be provided so that the user can directly input the values ​​of the non-financial variables, but depending on the embodiment, it may be configured to select or input answers to various questions, like a survey, and the non-financial variables may be collected by processing and / or manipulating the input answers. Alternatively, the non-financial data collection unit (114) may receive the non-financial data from the server (30) of various institutions / companies that manage the non-financial data.

[0081] The data preprocessing unit (116) can obtain data to be provided to each of the user group assignment unit (120), the model learning unit (130), the prediction unit (140), and the analysis unit (150) through preprocessing of the acquired input data (financial / non-financial data). For example, the preprocessing may include processes such as extraction of some of the plurality of variables included in the acquired input data, combination of some of the plurality of variables, substitution (such as replacing text with numbers), and processing.

[0082] For example, the data preprocessing unit (116) may obtain group assignment data for assigning a user to one of a plurality of pre-defined groups through preprocessing of the acquired input data. The group assignment data may include variables related to the classification criteria of the pre-defined groups. For example, if the classification criteria of the groups is 'industry (occupation)', the group assignment data may include variables related to the user's 'industry' among the plurality of variables.

[0083] In addition, the data preprocessing unit (116) can obtain data (hereinafter referred to as “real data”) for predicting and analyzing indicators (such as happiness) through preprocessing of the acquired input data. The real data may be composed of variables input to an artificial intelligence model (162) to be described later among a plurality of variables. For example, variables in a form other than numbers may not be easy to learn or utilize in the artificial intelligence model (162), and in the case of variables directly related to indicators, there is a risk that the label will be explained only by the corresponding variable, and some variables may have very low relevance to the indicator. Therefore, the data preprocessing unit (116) can obtain real data by removing the corresponding variables among the plurality of variables.

[0084] According to an embodiment, the real data may include, among the multiple variables included in the input data, variables having values ​​in quantitative form, such as numbers, degrees, levels, etc., and variables having values ​​in text form but that can be converted into quantitative form, but this is not necessarily the case.

[0085] The above real data may be provided to the prediction unit (140) and the analysis unit (150). According to an embodiment, the above real data may be provided to the model learning unit (130) and utilized in the learning of the artificial data detection model and artificial intelligence model described below.

[0086] Meanwhile, the artificial intelligence model (162) may be a deep learning-based model that outputs prediction results for the aforementioned life indicators based on input real data. For example, if the aforementioned life indicator is "happiness," the artificial intelligence model (162) may output prediction results for happiness based on input real data. The artificial intelligence model will be described in more detail later with reference to FIGS. 5 to 7.

[0087] The user group assignment unit (120) can assign (classify) the user to one of a plurality of predefined groups (G1 to Gn; n is a natural number greater than or equal to 2) based on the group assignment data for the user provided from the data collection and preprocessing unit (110). As described below, the artificial intelligence model (162) can be implemented for each of a plurality of groups, or can be implemented to receive group information when making a prediction. This is so that the artificial intelligence model (162) can make a more accurate prediction, since the range of a specific variable or the variable affecting the prediction result may be different for each group. For the sake of ease of understanding, an example will be given. In the case of the 'high school student' group, the deviation in financial data may not be large and its importance may also be low. However, in the case of the 'self-employed' group, the deviation in financial data may be large and its importance may also be high. Therefore, if the artificial intelligence model (162) learns and makes a prediction operation without distinguishing between groups and the group information is not reflected, the prediction accuracy may be very low.

[0088] For example, if a group-specific artificial intelligence model (162) is stored in a database (160) and the user is assigned to the second group (G2), the model learning unit (130), the prediction unit (140), and the analysis unit (150) can receive an artificial intelligence model (162) corresponding to the second group (G2) from the database (160) when learning, predicting, and interpreting the model using the user's real data.

[0089] The model learning unit (130) corresponds to a configuration that performs a learning operation on an artificial intelligence model (162) using input data (the actual data or data separately generated and input for learning purposes; collectively referred to as learning data). According to an embodiment, the model learning unit (130) may receive a label (or ground truth) of the learning data together with the learning data. For example, the label may correspond to an indicator value (happiness value) corresponding to the learning data. The model learning unit (130) may compare an indicator prediction result obtained from an artificial intelligence model based on the learning data with the label, and may learn the artificial intelligence model (update parameters (weights, etc.) of the artificial intelligence model) so that the indicator prediction result becomes identical to the label based on the comparison result. The model learning unit (130) can update the artificial intelligence model (162) by relearning the artificial intelligence model (162) whenever a predetermined number of data (real data) collected as the artificial intelligence system (10) provides services to the user is acquired.

[0090] In some embodiments, the model learning unit (130) may set one of the various variables included in the training data as a label and use the remaining variables to train an artificial intelligence model corresponding to the set label. In this case, the artificial intelligence model may be implemented as a continuous or binary model, depending on the distribution of the set values ​​of the labels included in the plurality of training data.

[0091] That is, the model learning unit (130) can learn and build artificial intelligence models capable of predicting / analyzing various elements from the acquired learning data. Each variable included in the learning data can be utilized as a setting variable (input value) in a specific artificial intelligence model, and as a response variable (label) in another specific artificial intelligence model.

[0092] Meanwhile, the accuracy of an artificial intelligence model may generally increase as the number of learning cycles increases. However, it may be difficult for the model learning unit (130) to easily acquire a large number of learning data (data sets) due to issues such as time and cost. In order to effectively learn an artificial intelligence model (162) using a minimum amount of learning data, the model learning unit (130) according to an embodiment of the present disclosure may be implemented to generate a large number of pseudo-data based on the learning data and to train the artificial intelligence model (162) by utilizing some of the generated pseudo-data.

[0093] In this regard, referring to FIG. 4, the model learning unit (130) may include a data generation unit (131) and a data detection model (133).

[0094] The pseudo data generation unit (131) can generate at least one pseudo data based on the learning data input through the data preprocessing unit (116), etc. According to the embodiment of FIG. 3, the learning data can include at least one financial variable and at least one non-financial variable. The generated pseudo data will be referred to as primary pseudo data to distinguish it from pseudo data misclassified as real data by the pseudo data detection model (133), and the misclassified pseudo data will be referred to as secondary pseudo data.

[0095] For example, the pseudo-data generation unit (131) may be implemented to generate primary pseudo-data based on a crossover operation of a genetic algorithm. For example, the pseudo-data generation unit (131) may set labels of primary pseudo-data to be generated based on a plurality of learning data, filter learning data having a label whose difference from the label of the set primary pseudo-data is less than or equal to a predetermined value among the plurality of learning data, and generate a plurality of primary pseudo-data based on the filtered learning data. Specifically, the pseudo-data generation unit (131) may (i) set a label value (y) of the primary pseudo-data to be generated. As described above, the label may mean an indicator value (happiness value). The label value (y) may be set based on a ratio of labels of the plurality of learning data, but is not limited thereto, and may be set in various ways. The pseudo data generation unit (131) may (ii) set the label selection range (e) to an initial value (e.g., '1'), and (iii) filter out learning data whose labels are greater than or equal to ye and less than or equal to y+e among the plurality of learning data. If there are two or more filtered learning data, the pseudo data generation unit (131) may randomly select two learning data among the filtered learning data, and generate the first pseudo data through a weighted average of the selected learning data (repeating (iii) to generate the plurality of first pseudo data). The standard for the weighted average may correspond to the difference between the set label (y) and the label of the selected learning data. On the other hand, if there are less than two filtered learning data, the pseudo data generation unit (131) may increase the label selection range (e) (e.g., increase by '1') and re-perform the process of (iii). This method can generate a variety of primary data even when the number of previously input learning data (real data) is not sufficiently large, so that effective learning of the indicator prediction model (162) can be achieved using a minimum amount of learning data.

[0096] The pseudo data detection model (133) can detect and classify whether the input primary pseudo data generated by the pseudo data generation unit (131) is real data (training data) or pseudo data when the input primary pseudo data is input. For example, the pseudo data detection model (133) may correspond to a random forest model based on a discriminator of a Generative Adversarial Network (GAN), but is not limited thereto.

[0097] The detection result of the pseudo-data detection model (133) may correspond to either a first detection result indicating that the primary pseudo-data is pseudo-data, or a second detection result indicating that the primary pseudo-data is real data. The first detection result may correspond to a result in which the pseudo-data detection model (133) accurately detects that the primary pseudo-data is pseudo-data (true detection; 135), and the second detection result may correspond to a result in which the pseudo-data detection model (133) incorrectly detects that the primary pseudo-data is real data (false detection; 137).

[0098] Meanwhile, the model learning unit (130) can perform learning on the pseudo-data detection model (133) using the learning data and the first pseudo-data. For example, the model learning unit (130) can train the pseudo-data detection model (133) to detect that the learning data is real data, and to detect that the first pseudo-data is pseudo-data.

[0099] According to an embodiment, the model learning unit (130) may set the ratio of the output (real data / pseudo data) of the pseudo data detection model (133) when learning the pseudo data detection model (133). For example, the ratio of real data to pseudo data may be set to 1:k, and as k increases, the likelihood that the input primary pseudo data will be detected as pseudo data relatively increases, and as k decreases, the likelihood that the input primary pseudo data will be detected as real data relatively increases. That is, as k increases, the detection accuracy of the pseudo data detection model (133) increases, and there is an advantage that the similarity between the secondary pseudo data and real data (data input by actual users) increases, but the time and cost required to sufficiently secure the secondary pseudo data may increase. On the other hand, as k decreases, the detection accuracy of the pseudo data detection model (133) decreases, and there is a disadvantage that the similarity between the secondary pseudo data and real data decreases, but the time and cost required to sufficiently secure the secondary pseudo data can be reduced. Depending on these trade-offs, the value of k can be set appropriately.

[0100] The model learning unit (130) can perform learning on the artificial intelligence model (162) using the learning data and the first artificial data (secondary artificial data; 137) that is misdetected as real data by the artificial data detection model (133). The prediction unit (140) can predict the life indicator for the user using the artificial intelligence model (162) learned by the model learning unit (130). The interpretation unit (150) can analyze the user's input data, output data (prediction result), and artificial intelligence model, and generate and provide an interpretation result for the output data based on the analysis result. According to an embodiment, the interpretation unit (150) can generate and provide an interpretation result that estimates factors that are considered important in relation to the indicator for users (or a user group corresponding to the artificial intelligence model) through analysis of the learned artificial intelligence model itself. In addition, according to an embodiment, the interpretation unit (150) can convert the interpretation result into a natural language form and provide it through an artificial intelligence language model such as an LLM (large language model). Learning, prediction, and interpretation of the artificial intelligence model (162) will be described in more detail with reference to FIGS. 5 to 7.

[0101] The database (160) can store and manage various data related to the operation of the artificial intelligence system (10). For example, the database (160) can store the aforementioned group-specific artificial intelligence model (162), and can store and manage input data, output data (prediction results), and interpretation results of various users. Depending on the embodiment, the database (160) may be implemented as a separate configuration (data center, cloud, etc.) connected to the artificial intelligence system (10) via a network.

[0102] FIGS. 5 to 7 are drawings for explanation related to an artificial intelligence model according to an exemplary embodiment of the present disclosure.

[0103] An artificial intelligence model according to an embodiment of the present disclosure can be implemented in the form of a random forest model that ensembles decision trees in a bagging manner.

[0104] Referring to FIG. 5, a decision tree is a supervised learning model that sequentially applies decision rules created using given input data to predict output values, and the decision rules can be expressed in a tree structure. Such a decision tree can be composed of multiple nodes representing specific conditions and branches connecting one node to another. The node located at the top of the multiple nodes corresponds to the root node, and the node located at the bottom that does not branch any further corresponds to the leaf node. Since the decision tree can clearly identify the meaning of each node and branch, and the path from the root node to the leaf node, the analysis unit (150) can later extract the decision rule corresponding to the path.

[0105] For convenience of explanation, the decision tree illustrated in Fig. 5 is assumed to have three variables of input data (x1, x2, x3), a total of seven nodes, and a label y with two values ​​of 0 and 1. In this case, the tuple of each node represents (the number of samples (data) with y=0, the number of samples with y=1). When the input data X is (8, 6, 9), the input data X reaches the leaf node node (6) through node (1), node (2), and node (5). In the case of node (6), since there are 40 samples with y=0 and 2 samples with y=1, the label y of the input data X can be classified as 0. That is, the decision tree can perform decision making on the input data X based on the label distribution of the data for the leaf nodes. Additionally, based on the decision path (node ​​(1), node (2), node (5), and node (6)) of the input data X, the factor by which the value of the label y for the input data X is determined can be interpreted.

[0106] Learning of these decision trees can be defined as finding a rule that minimizes impurity and branching each node until a predetermined termination condition is reached. The impurity may correspond to the variance of the labels of the samples included in the node. In other words, if a node includes samples with various labels, the impurity can be understood as being large. However, if learning is performed without limiting the depth of the decision tree, the number of branches increases excessively, making the model more complex. As the model becomes more complex, accuracy may increase, but the load increases, and the interpretation of the prediction results may be difficult. To prevent this, the termination condition may further include information about the maximum depth of the decision tree.

[0107] Meanwhile, decision trees have the advantage of being robust to outliers (abnormal values) and missing values. To minimize decision tree impurity, the branch point criterion for each variable is typically set to the median. Therefore, even if outliers are present, the model is extremely unlikely to be biased by them. Furthermore, even in the case of missing values, the model can be flexibly adapted to handle them by converting the missing values ​​into values ​​that would not normally occur and inputting them into the model. This type of missing value can be effectively utilized in situations where new variables are added to the input data.

[0108] To explain this in more detail, while the artificial intelligence system (10) performs indicator prediction and interpretation for users by utilizing the artificial intelligence model, a new variable that may affect the indicator may be added due to changes in the questionnaire items, changes in the social environment, or the occurrence of a new event (e.g., the outbreak of 'COVID-19'). In this case, the artificial intelligence system (10) may perform learning of the artificial intelligence model by missing-processing the new variable for the previously acquired input data (e.g., setting the value of the new variable to an impossible value, such as -999 or -1) until the number of input data with the new variable added exceeds a preset number. Thereafter, by performing learning of the artificial intelligence model every time a predetermined number of input data with the new variable added is collected, the influence of the missing data can be gradually reduced. On the other hand, when the number of input data with the new variable added exceeds a preset number, the artificial intelligence model can be newly learned using the input data.

[0109] Meanwhile, as the number of variables included in the input data increases, the number of variables not utilized in the trained model also increases. This can lead to the problem of only utilizing a small number of variables highly correlated with the label. To overcome this, the AI ​​model according to the embodiment of the present disclosure can utilize a random forest model that combines decision trees using a bagging method.

[0110] Referring to Figure 6, bagging (bootstrap aggregating) is one of the ensemble learning techniques, which is a method of independently training multiple models and then synthesizing the results through averaging or voting to improve prediction accuracy.

[0111] The model learning unit (130) can randomly sample a portion of the entire learning data set using a bootstrap technique. At this time, the sampling is performed a number of times corresponding to the number of models (e.g., n; n is a natural number greater than or equal to 2), and each sampling result can be input for learning the corresponding model.

[0112] Referring to FIG. 7, the prediction unit (140) can input real data collected and preprocessed for the user into each of a plurality of decision trees included in the artificial intelligence model. When prediction results (RESULT-1 to N) for the real data input from each of the plurality of decision trees are output, the prediction unit (140) can output the final prediction result (FINAL RESULT) through a synthesis (average, voting, etc.) of the output multiple prediction results (RESULT-1 to N). For example, the prediction result (FINAL RESULT) can be output as any one of a plurality of predefined values ​​(for example, any one of the values ​​0 to 7). When the artificial intelligence model is a happiness prediction model, it can be understood that the greater the value of the prediction result (FINAL RESULT), the higher the happiness.

[0113] Meanwhile, the interpretation unit (150) can generate interpretation results regarding the correlation and importance between the labels and variables of the learned artificial intelligence model through analysis of the learned artificial intelligence model. As a specific example, the interpretation unit (150) can perform interpretation of the indicator prediction model through i) calculating feature importance for each variable, ii) generating a PDP (Partial Dependence Plot), and iii) extracting rules from a decision tree.

[0114] In addition, the analysis unit (150) can perform an interpretation of the prediction result for the user using the interpretation result of the artificial intelligence model and the user's input data (or actual data). Specifically, the analysis unit (150) can perform an interpretation of the prediction result and generate an interpretation result by using the interpretation result of the artificial intelligence model according to the above-described i) to iii) and iv) extraction of the decision path of the user's actual data.

[0115] First, the calculation of feature importance is explained as follows.

[0116] In order to interpret the prediction results, it may be essential to identify the variable with high feature importance among the multiple variables input to the artificial intelligence model. The higher the feature importance of a variable, the more likely it is that the impurity of the decision tree will be reduced by that variable. If an arbitrary node T is T l and T r The amount of impurity reduction when branched can be measured as in mathematical equation 1 below.

[0117]

[0118]

[0119]

[0120] Here, N is the total number of samples (training data), N T , N Tl , N Tr are nodes T and T respectively l , T rThe number of samples included in , I represents an impurity measurement function (e.g., Gini index, entropy index, sum of squared errors, etc.). The feature importance of a specific variable can be calculated by dividing the impurity reduction amount due to the variable by the sum of the impurity reduction amounts for all nodes. Through the above-described method, the analysis unit (150) can analyze important variables having a relatively high correlation with the label among multiple variables included in the real data. For example, the analysis unit (150) can generate and provide an analysis result listing (table, graph, etc.) a predetermined number of variables in order of high feature importance based on the analysis result.

[0121] Next, the PDP is a graph that represents the change in a label according to the change in the value of one variable when the values ​​of all variables except one variable are fixed. The analysis unit (150) generates and provides the PDP, thereby allowing the relationship between variables and labels to be visually confirmed.

[0122] The analysis unit (150) can extract decision rules for each of the multiple decision trees constituting the random forest model. The decision rules may correspond to data describing the flow from the root node to each leaf node for each of the multiple decision trees. Depending on the embodiment, the extracted decision rules may be converted into a script for use as input to an artificial intelligence language model and provided in the form of a text file or the like.

[0123] In addition, the analysis unit (150) can obtain a prediction result for the user's real data from the prediction unit (140), and extract a decision path of the prediction result based on the obtained prediction result and the real data. The analysis unit (150) can generate an interpretation result on how the indicator for the real data was predicted based on the extracted decision path. The generated interpretation result can be provided to the user through the terminal (20) or transmitted to the server (30) and utilized in various ways. According to an embodiment, the interpretation unit (150) converts the interpretation result into a natural language form using an artificial intelligence language model linked to the artificial intelligence system (10) and provides it to the user's terminal (20), and the user, etc. can input additional questions, etc. based on the provided interpretation result to obtain additional information related to the interpretation result from the artificial intelligence language model. This will be described in more detail later with reference to FIGS. 11 to 16.

[0124] FIG. 8 is a flowchart for explaining a learning method of an artificial intelligence model according to an embodiment of the present disclosure.

[0125] Referring to FIG. 8, the artificial intelligence system (10) can acquire real data by extracting some variables through preprocessing of input data (S800) and generate primary data from the acquired real data (S810). Although not illustrated, as described above, the artificial intelligence system (10) can also generate primary data from separately provided learning data (dataset) instead of the real data.

[0126] The artificial intelligence system (10) can input the generated primary data into a data detection model to obtain secondary data that has been misclassified as real data (S820).

[0127] The artificial intelligence system (10) can learn an artificial intelligence model using the acquired secondary data (S830). According to an embodiment, the artificial intelligence system (10) can learn an artificial intelligence model using the secondary data and the user's actual data.

[0128] FIG. 9 is a flowchart illustrating a learning method for an artificial intelligence model according to one embodiment of the present disclosure. FIG. 10 is an exemplary diagram illustrating an artificial intelligence model learning operation of an artificial intelligence system according to the learning method of FIG. 9.

[0129] Referring to FIGS. 9 and 10, the artificial intelligence system (10) can acquire a preset number or more of new learning data with new variables added (S900).

[0130] The data collection and preprocessing unit (110) of the artificial intelligence system (10) can acquire new learning data with new variables and variable values ​​added. The new variables may correspond to variables not reflected in the learning of the artificial intelligence model (162) included in the artificial intelligence system (10). The new learning data may be data generated and provided for the purpose of learning the artificial intelligence model (162), or may be data acquired from a user's terminal (20), etc., according to the service provision of the artificial intelligence system (10).

[0131] For example, if the artificial intelligence system (10) corresponds to a system that predicts life indicators as described above in FIG. 3, while the artificial intelligence system (10) performs indicator prediction and interpretation for users, new variables that may affect the indicators may be added depending on changes in questionnaire items, changes in the social environment, or occurrence of new events (e.g., the outbreak of 'COVID-19').

[0132] Meanwhile, if a sufficient number of new learning data with new variables and variable values ​​added is not acquired, the accuracy may be very low even if the artificial intelligence model (162) is trained by reflecting the new learning data set, so the model training unit (130) may not perform training on the artificial intelligence model (162) until a preset number or more of new learning data is acquired. In this case, according to an embodiment, the artificial intelligence system (10) may notify that the new variable has not been reflected when providing output data to a user or the like.

[0133] The artificial intelligence system (10) can process values ​​for new variables of existing learning data based on the acquired new learning data (S310).

[0134] The model learning unit (130) can learn the artificial intelligence model (162) by reflecting new variables when the number of new learning data included in the new learning dataset is greater than or equal to a preset number. However, when the artificial intelligence model (162) is retrained using only the new learning dataset, the problem of a decrease in the prediction accuracy for not only new variables but also existing variables due to the small amount of learning data may occur.

[0135] To solve this problem, the model learning unit (130) can learn the artificial intelligence model (162) by utilizing not only the new learning dataset but also the existing learning dataset (161) stored in the database (160). However, since the plurality of existing learning data constituting the existing learning dataset (161) do not include values ​​for new variables, a process of processing the new variable values ​​of the existing learning dataset may be required to learn the artificial intelligence model (162). To this end, the model learning unit (130) may include a new variable processing module (132).

[0136] According to one embodiment, the new variable processing module (132) may perform training of the artificial intelligence model (162) by processing the new variable as missing for each existing training data (e.g., setting the value of the new variable to a value outside the range of the variable (e.g., 99999, -1, etc.)) until the number of new training data with the new variable added exceeds a preset number. Thereafter, by performing training of the artificial intelligence model every time a predetermined number of input data with the new variable added is collected, the influence of the missing data can be gradually reduced. On the other hand, when the number of new training data exceeds a preset number, the artificial intelligence model can be newly trained using the new training data.

[0137] According to one embodiment, the new variable processing module (132) may estimate the value of a new variable for each of the existing learning data based on the patterns of variable values ​​included in each of the plurality of new learning data. Specifically, the new variable processing module (132) may analyze the pattern (relationship) between the value of the new variable and the values ​​of the existing variables from each of the new learning data, and may estimate the new variable value of the existing learning data by applying the analyzed pattern to the existing learning data. According to an embodiment, when estimating the new variable values ​​of the existing learning data, the new variable processing module (132) may estimate the new variable values ​​by reflecting statistical characteristics (e.g., normal distribution, statistical model, etc.) together with the variable value pattern of the new learning dataset.

[0138] According to one embodiment, the new variable processing module (132) may apply an appropriate processing technique based on the characteristics of the new variable among the processing techniques for the new variable values ​​described above. For example, if the new variable is not correlated with other existing variables, the new variable processing module (132) may process the new variable values ​​of the existing learning dataset as missing. On the other hand, if the new variable is correlated with other existing variables, the new variable processing module (132) may estimate the new variable values ​​of the existing learning dataset based on the variable value pattern of the new learning data.

[0139] The artificial intelligence system (10) can perform learning for an artificial intelligence model using existing learning data and new learning data in which values ​​for new variables have been processed (S320).

[0140] The learning control module (134) of the model learning unit (130) can control learning for the artificial intelligence model (162) using an existing learning dataset and a new learning dataset in which values ​​for new variables have been processed.

[0141] For example, if the artificial intelligence model (162) is implemented as the above-described random forest model, the new variable processing module (132) can provide an existing learning dataset with new variable values ​​processed by using different new variable processing techniques for each decision tree included in the random forest model. For example, the new variable processing module (132) can process missing new variables of an existing learning dataset provided as a first decision tree model, estimate new variables of an existing learning dataset provided as a second decision tree model according to a first pattern of variable values, and estimate new variables of an existing learning dataset provided as a third decision tree model according to a second pattern of variable values.

[0142] Depending on the embodiment, the new variable processing techniques applied to each decision tree model may be determined based on the characteristics of the new variable. For example, if the new variable has no correlation with existing variables, the new variable processing module (132) may process the new variable of the existing learning data provided to each decision tree model as missing. Alternatively, if the new variable has a correlation with existing variables, the new variable processing module (132) may estimate the new variable of the existing learning data provided to each decision tree model as the same value or different values ​​based on the patterns of the variable values.

[0143] In this case, existing learning datasets containing new variables processed in various ways are provided to decision tree models included in the random forest model, and the learning results of each decision tree model are synthesized to provide more reliable and accurate prediction results.

[0144] According to an embodiment, the AI ​​system (10) may perform additional training on the AI ​​model whenever a predetermined number of new training data with new variables is acquired (S930). The additional training operation for the AI ​​model may be substantially identical to steps S910 to S920. As the additional training is repeated, the influence of existing training datasets with missing or estimated values ​​for new variables is reduced, thereby improving the prediction accuracy of the AI ​​model.

[0145] According to an embodiment of the present disclosure, when a new learning dataset including a new variable is acquired, the artificial intelligence system (10) performs new variable processing on an existing learning dataset and then utilizes the new learning dataset together with the new variable for training the artificial intelligence model (162), thereby quickly providing an artificial intelligence model (162) that reflects the influence of the new variable. Through this method, the artificial intelligence system (10) can effectively implement and provide an artificial intelligence model that evolves with the addition of variables.

[0146] FIG. 11 is a flowchart illustrating a method for an artificial intelligence system according to an exemplary embodiment of the present disclosure to obtain and provide prediction results and interpretation results from input data using an artificial intelligence model.

[0147] Referring to FIG. 11, the artificial intelligence system (10) can obtain real data by extracting some variables through preprocessing of input data (S1100), and input the obtained real data into an indicator prediction model to obtain an indicator prediction result (S1110).

[0148] The artificial intelligence system (10) can interpret the indicator prediction results (S1120) and provide the interpretation results to users, etc. (S1130).

[0149] According to an embodiment, the artificial intelligence system (10) may provide interpretation results to users, etc. in natural language form through linkage with an artificial intelligence language model including a large language model (LLM), such as chatGPT. This will be described in more detail below with reference to FIGS. 12 to 16.

[0150] Figures 12 to 16 are drawings for explanation related to interpretation data generated and provided through the artificial intelligence system of Figure 3.

[0151] Referring to FIG. 12, the interpretation unit (150) of the artificial intelligence system may include a variable definition prompt (152), a rule prompt (154), a command prompt (156), and an interpretation result output module (158).

[0152] The variable definition prompt (152) is a prompt that includes text defining each variable included in the user's real data (or input data), enabling the generative artificial intelligence model to understand the meaning of each variable within the real data. Furthermore, the variable definition prompt (152) may include information regarding the value range of each variable.

[0153] The rule prompt (154) is a prompt that includes rules for indicator values ​​according to combinations of variables and value ranges, and the rules can be extracted from the learned artificial intelligence model (162). For example, if the artificial intelligence model (162) corresponds to the random forest model described above, the rule prompt (154) can extract the rules from the decision paths of the decision tree models included in the random forest model. The rule prompt (154) can be changed as the learning of the artificial intelligence model (162) continues.

[0154] The command prompt (156) may include a command that causes the generative artificial intelligence model to generate interpretation data based on the actual data, indicator prediction results, variable definition prompt (152), and rule prompt (156). Depending on the embodiment, the command prompt (156) may further include conditions regarding the method, format, type, etc. of generating the interpretation data.

[0155] The interpretation result output module (158) can provide interpretation data provided from the generative artificial intelligence model operating system (1000) to a terminal (20), etc., or output it through an output means (display, etc.). For example, the interpretation data may include various data such as an indicator prediction result and cause, a major variable affecting the indicator, consulting information for improving the indicator, and an indicator change expected in the future according to the consulting information. Referring to FIGS. 13 to 16, when the indicator predicted by the indicator prediction system corresponds to 'life satisfaction,' the interpretation data may include the prediction result and cause of the user's 'life satisfaction' as shown in FIG. 13, and the major variables affecting the user's 'life satisfaction' as shown in FIG. 14. In addition, the interpretation data may include consulting information for improving the user's 'life satisfaction' as shown in FIG. 15, and an expected change in the user's 'life satisfaction' as shown in FIG. 16.

[0156] FIG. 17 is a block diagram of components related to the debt management capability evaluation operation of an artificial intelligence system according to an exemplary embodiment of the present disclosure. FIG. 18 is a diagram illustrating an example of a configuration of the debt management data collection unit illustrated in FIG. 17. FIGS. 19 to 21 are exemplary diagrams illustrating a simulation interface provided by the debt management capability evaluation simulation control unit illustrated in FIG. 18. FIG. 22 is an exemplary diagram illustrating a debt management capability evaluation and feedback information provision process according to an exemplary embodiment of the present disclosure.

[0157] According to an embodiment, the artificial intelligence system (10) described above in FIG. 1 can perform a debt management ability evaluation operation to improve a user's debt management ability and knowledge by evaluating the user's debt management ability and providing customized feedback based on the evaluation results. The artificial intelligence system (10) can obtain debt management data from users while executing a debt management ability evaluation simulation based on a predetermined evaluation scenario, and can evaluate the level of debt management ability (knowledge) using the obtained debt management data. In addition, the artificial intelligence system (10) can provide education, such as feedback information, to improve the users' debt management ability (knowledge) level based on the evaluation results.

[0158] A user can access an artificial intelligence system (10) via a terminal (20) to request execution of a debt management capability assessment simulation, and input debt management data based on an interface (screen, etc.) provided as the simulation is executed. Furthermore, the user can receive and confirm debt management capability assessment results and feedback information via the terminal (20). Such terminal (20) may include various electronic devices such as smartphones, tablet PCs, desktop PCs, and laptops.

[0159] The server (30) may be connected to the artificial intelligence system (10) and the terminal (20) via a network. For example, the server (30) may be a database server that stores various data related to the debt management capability evaluation process of the artificial intelligence system (10). According to an embodiment, the server (30) may correspond to a server (language model operation server) that operates a large language model (LLM) such as GPT. The artificial intelligence system (10) may transmit data such as the debt management data and evaluation results described above to the server (30), and the server (30) may provide feedback information including an interpretation message for the evaluation results to the artificial intelligence system (10).

[0160] Based on this, referring to FIG. 17, the artificial intelligence system (10) may include a debt management data collection unit (1710), a debt management capability evaluation unit (1720), a feedback information provision unit (1730), and a database (1740). The artificial intelligence system (10) according to the embodiment of the present disclosure is not limited to the configuration of FIG. 2, and may include more or fewer configurations.

[0161] The debt management data collection unit (1710) can collect data (debt management data) for evaluating a user's debt management capabilities. For example, the debt management data collection unit (1710) can run a simulation to evaluate a user's debt management capabilities and collect the debt management data through the simulation.

[0162] In this regard, referring to FIG. 18, the debt management data collection unit (1710) may include a transmission / reception unit (1712) and a debt management capability evaluation simulation control unit (1714).

[0163] The transmitter and receiver (1712) can transmit and receive various data according to the execution of the debt management capability evaluation simulation with the user's terminal (20).

[0164] The debt management capability evaluation simulation control unit (1714) can receive a request for execution of a debt management capability evaluation simulation from a terminal (20) through a transmission / reception unit (1712), and execute the debt management capability evaluation simulation in response to the received execution request.

[0165] According to one embodiment, the debt management capability evaluation simulation control unit (1714) may receive, from the terminal (20) via the transmission / reception unit (1712), environment setting information for generating an evaluation scenario to be applied to the executed debt management capability evaluation simulation. To this end, the debt management capability evaluation simulation control unit (1714) may provide, to the terminal (20) via the transmission / reception unit (1712), an interface (screen, etc.) for allowing a user to input the environment setting information via the terminal (20).

[0166] A specific embodiment of the above environment setting information and the interface provided for inputting the above environment setting information will be described below with reference to FIGS. 19 to 20c.

[0167] Referring to FIGS. 19 to 20c, the debt management capability evaluation simulation control unit (1714) may provide interfaces for inputting environment setting information to the terminal (20). For example, the interface may include a first interface (1900) for inputting basic environment information including the evaluation scenario's progress period, currency unit, initial assets, income and expenditure, and a second interface (2000) for inputting information related to the debt status (debt information), but the form of providing the interfaces is not limited thereto. In this case, the environment setting information may include the progress period, currency unit, initial assets, income status, expenditure status, and debt status.

[0168] The progress period may be related to the number of rounds in the evaluation scenario. For example, a debt management capability assessment simulation may proceed with each round on a monthly basis according to the evaluation scenario, and the number of rounds may be determined based on the entered progress period. According to the first interface (1900) of Figure 19, the progress period may be set to a yearly basis, but this is not necessarily the case.

[0169] The currency unit can refer to the currency unit displayed in the simulation. Initial assets can refer to the amount held by the user at the start of the evaluation scenario.

[0170] Income status can refer to the amount a user earns monthly or in a specific month in an evaluation scenario. This income status can include fixed income, which occurs monthly, and variable income, which occurs in a specific month or any month. The example in Figure 19 illustrates monthly fixed income as an example of fixed income, and bonuses as an example of variable income. However, the types of fixed and variable income are not limited to these.

[0171] Meanwhile, bonuses (fluid income) can be selectively activated by user input, and upon activation, additional information input fields for the bonus can also be activated. For example, additional information for the bonus can include payment type, amount, and payment pattern. For payment type, this field contains information on the month in which the bonus is paid. For example, either "paid in the first month of a set year" or "paid in a random month a set number of times each year" can be selected and set, but is not limited thereto. The amount field can refer to the bonus payment amount. For payment pattern, this field is activated when "paid in the first month of a set year" is selected for the payment type, allowing the user to set the year in which the bonus is paid.

[0172] Expenditure status can refer to the amount a user spends each month or in a specific month in an evaluation scenario. This expenditure status can include fixed expenses occurring monthly and variable expenses occurring in a specific month or any month. The example in Figure 19 illustrates monthly living expenses as an example of fixed expenses, and condolence events (types and patterns of condolence events, etc.) as an example of variable expenses. However, the types of fixed and variable expenses are not limited to these.

[0173] Meanwhile, information on the type and pattern of condolence payments (flexible expenditure status) may be included. For example, the type of condolence payment can be either "randomly," which generates condolence payments randomly, or "monthly," which generates condolence payments on a fixed monthly basis. Condolence expenses can be automatically calculated as a percentage of the entered monthly living expenses, but this is not always the case. The pattern of condolence payments is activated when "randomly" is selected and can include a value for the number of condolence payments.

[0174] Debt status may correspond to an item defining the debt held by a user within an assessment scenario. Regarding the debt status setting, the second interface (2000) of FIG. 20a may include a debt list (2010) containing at least one entered debt, and a debt input interface (2020) for entering debt-related information.

[0175] The debt list (2010) may display information on the debt type, debt name, debt amount, interest rate, holding year, and / or repayment target year for each of at least one debt entered as the debt status.

[0176] The debt input interface (2020) may change input items depending on the selected debt type. Referring to FIG. 20b, if the debt type is "short-term debt (e.g., living expenses loan)," input items provided through the debt input interface (2022) may include the debt name, monthly debt amount, overdue interest rate, and holding years. The monthly debt amount refers to the amount of debt incurred each month based on the debt, the overdue interest rate refers to the interest rate added to the overdue amount if the debt is not repaid, and the holding years may refer to information regarding how many years the debt will be held in the future.

[0177] Meanwhile, referring to FIG. 20c, if the debt type is "long-term debt (e.g., credit loan, etc.)", the input items provided through the debt input interface (2024) may include the debt name, amount of debt held, annual interest rate, and target repayment year. Among these, the target repayment year may refer to the target year for repaying all of the debt.

[0178] The environment setting information input through the above-described interfaces (1900, 2000) can be provided to the debt management capability evaluation simulation control unit (1714) through the transceiver unit (1712) from the terminal (20). The debt management capability evaluation simulation control unit (1714) can create an evaluation scenario based on the environment setting information and execute a debt management capability evaluation simulation according to the created evaluation scenario.

[0179] The debt management capability evaluation simulation control unit (1714) can execute a debt management capability evaluation simulation according to a generated evaluation scenario, provide a debt fluctuation situation according to the progress of the round, and provide a simulation interface (screen, etc.) to the terminal (20) for obtaining repayment data from the user for each round.

[0180] In this regard, referring to the example diagram of FIG. 21, the simulation interface (2100) may include a progress information provision area (2110) indicating the progress of the simulation, a repayment data input area (2120) for the current round (month), and a debt change status information provision area (2130) indicating the debt change status according to the progress of the simulation.

[0181] For example, the progress information provision area (2110) may provide information on the simulation elapsed (running) time, the number of completed rounds, the total number of rounds, income status (fixed income and floating income) based on previously acquired environment setting information, the amount available (cash) in the current round, the total amount of debt existing in the current round, and the amount of expenditure (fixed expenditure and floating expenditure) in the current round. In some embodiments, the expenditure amount in the current round may be adjusted by user input.

[0182] For example, the repayment data input area (2120) may provide information on the amount of income (fixed and floating income) for the current round and the debt status (type of debt, amount of debt, etc.) for the current round. Furthermore, the repayment data input area (2120) may include input fields for entering the type and amount of debt to be repaid in the current round. As the user enters the type and amount of debt to be repaid in the input fields, repayment data for the current round may be obtained. In other words, the repayment data obtained for each round may include the type and amount of debt to be repaid in that round.

[0183] For example, the debt change status information provision area (2130) may include graphs showing the balance change status of each debt according to the simulation progress, thereby providing the user with information on the debt change status.

[0184] The debt management capability evaluation simulation control unit (1714) can obtain round-by-round repayment data through the interface (2100) described above and analyze the debt repayment status for each round based on the obtained debt management data. For example, the debt repayment status for each round can include information such as the type and amount of remaining debt upon completion of the relevant round, and the cumulative repayment amount. In addition, the debt management capability evaluation simulation control unit (1714) can analyze the final debt repayment status upon completion of the final round. For example, the final debt repayment status can include information such as the type, number, amount, and cumulative repayment amount of remaining debt.

[0185] By executing this debt management capability evaluation simulation, the debt management data collection unit (1710) can obtain (collect) debt management data including round-by-round repayment data, round-by-round debt repayment status, and final debt repayment status.

[0186] Referring again to FIG. 17, the debt management ability evaluation unit (1720) can evaluate the user's debt management ability (knowledge) level based on debt management data collected as the debt management ability evaluation simulation progresses. The feedback information provision unit (1730) can obtain feedback information based on the evaluation results of the debt management ability evaluation unit (1720) and provide the obtained feedback information to the terminal (20).

[0187] Referring to the embodiment of FIG. 22, the debt management ability evaluation unit (1720) may include a deep learning-based evaluation model (1722) trained to evaluate a user's debt management ability level based on collected debt management data. For example, the evaluation model (1722) may be trained to output a user's debt management ability level as a value such as a score or grade based on the input debt management data, but this is not necessarily the case.

[0188] The feedback information provision unit (1730) can generate feedback information based on debt management data and evaluation results and provide it to the terminal (20).

[0189] For example, the feedback information provision unit (1730) may extract feedback information corresponding to the evaluation result for the user from among a plurality of pieces of feedback information that are matched according to the score or grade of the evaluation result and provide the extracted feedback information to the terminal (20).

[0190] For example, the feedback information providing unit (1730) may provide feedback information by comparing the debt management data and evaluation results of the user with the debt management data and evaluation results of multiple users stored in a database (1740). Specifically, the feedback information providing unit (1730) may extract debt management data of users with similar evaluation results from the database (1740), compare the extracted debt management data with the debt management data of the user, determine relatively vulnerable items (vulnerable items), and provide feedback information including guides or educational materials for improving capabilities for the determined vulnerable items.

[0191] For example, the feedback information provider (1730) may generate portfolio information for an optimal debt repayment plan for the user based on at least some of the debt management data, evaluation results, and the aforementioned environment setting information, and provide feedback information including the generated portfolio information. Alternatively, the feedback information provider (1730) may analyze the differences between the generated portfolio information and the debt management data, and provide feedback information based on the analyzed differences. To this end, the feedback information provider (1730) may include algorithms, software, and / or deep learning models for generating and analyzing the portfolio information.

[0192] For example, the feedback information providing unit (1730) may transmit the user's debt management data and evaluation result together with a previously generated and stored interpretation request prompt to the server (30). At this time, the server (30) may correspond to a language model operating server that operates a large language model (LLM). The interpretation request prompt may include text requesting the generation of an interpretation message for the debt management data and evaluation result transmitted together. In addition, the interpretation request prompt may further include content regarding an interpretation method for the debt management data and evaluation result, content regarding the meaning of the values ​​of each item in the debt management data and the values ​​of the evaluation result, etc. The server (30) may generate an interpretation message regarding the user's debt management ability evaluation result based on the received debt management data, evaluation result, and interpretation request prompt, and transmit the interpretation message to the feedback information providing unit (1730). The feedback information providing unit (1730) may provide feedback information including the interpretation message to the terminal (20). For example, the above interpretation message may include text containing various information, such as analysis information of the evaluation results or a guide for improving the user's debt management ability, such as an explanation of the debt management ability evaluation results, a guide for improving ability in vulnerable items, and portfolio information for an optimal debt repayment plan.

[0193] Referring still to FIG. 17, the database (1740) can store and manage various data related to the operation of the artificial intelligence system (10). For example, the database (1740) can store at least some of the environment setting information, debt management data, evaluation results, and feedback information acquired for the user. In the present disclosure, the database (1740) is illustrated as being included in the artificial intelligence system (10), but depending on the embodiment, it may be implemented as a separate component (data center, cloud, etc.) connected to the artificial intelligence system (10) via a network.

[0194] Figure 23 is a flowchart illustrating a debt management capability assessment method according to an exemplary embodiment of the present disclosure. Figure 24 is a flowchart illustrating a specific embodiment of the debt management data acquisition step illustrated in Figure 23. Figure 25 is a flowchart illustrating a specific embodiment of the debt management capability assessment and feedback information provision step illustrated in Figure 23.

[0195] Referring to FIG. 23, the artificial intelligence system (10) receives a request to execute a debt management capability evaluation simulation from a connected terminal (20) (S2300), and in response to the received execution request, requests and obtains environment setting information for creating an evaluation scenario (S2310).

[0196] The artificial intelligence system (10) can generate an evaluation scenario corresponding to the acquired environment setting information (S2320) and execute a debt management capability evaluation simulation to which the generated evaluation scenario is applied, thereby obtaining debt management data from the user (S2330).

[0197] More specifically, with respect to step S2330 through Fig. 24, the artificial intelligence system (10) can obtain repayment data including the type of debt to be repaid and the repayment amount for each of a plurality of rounds according to the progress period information in the environment setting information (S2332).

[0198] The artificial intelligence system (10) can analyze the debt repayment status by round based on the acquired repayment data (S2334). As described above, the debt repayment status by round may include information such as the type and amount of remaining debt at the end of each round, and the accumulated repayment amount.

[0199] The artificial intelligence system (10) can analyze the final debt repayment status upon completion of the final round according to the evaluation scenario (S2336). For example, the final debt repayment status may include information such as the type, number, amount, and cumulative repayment amount of remaining debt.

[0200] The artificial intelligence system (10) can obtain debt management data including the round-by-round repayment data, round-by-round debt repayment status, and final debt repayment status by providing the user with a debt management capability evaluation simulation to which the above-mentioned evaluation scenario is applied (S2338).

[0201] Referring again to FIG. 23, the artificial intelligence system (10) can evaluate the user's debt management ability based on the acquired debt management data and provide feedback information based on the evaluation results (S2340).

[0202] Referring to FIG. 25, a specific embodiment related to step S2340 will be described. The artificial intelligence system (10) can obtain a debt management capability evaluation result based on debt management data through a deep learning-based evaluation model (S2342).

[0203] The artificial intelligence system (10) can transmit debt management data, debt management capability evaluation results, and interpretation request prompts to the language model operation server (S2344).

[0204] The artificial intelligence system (10) can receive an interpretation message regarding the evaluation result from the language model operation server (S2346) and provide the received interpretation message to the terminal (20) (S2348).

[0205] FIG. 26 is a schematic hardware configuration block diagram of a computing device constituting an artificial intelligence system according to an exemplary embodiment of the present disclosure.

[0206] The hardware configuration of the computing device (2600) illustrated in FIG. 26 may correspond to the hardware configuration of each of at least one computing device constituting the artificial intelligence system (10) illustrated in FIG. 1.

[0207] Referring to FIG. 26, a computing device (2600) may include a communication unit (2610), an input unit (2620), an output unit (2630), a control unit (2640), and a memory (2650). The control configuration illustrated in FIG. 26 is an example for convenience of explanation, and the computing device (2600) may include more or fewer configurations than the configuration illustrated in FIG. 26.

[0208] The communication unit (2610) may include one or more communication modules that enable communication with other terminals or servers by connecting the computing device (2600) to a network. For example, the communication module may include a mobile communication module such as LTE or 5G, a wireless communication module such as Wi-Fi, and / or other various wired or wireless communication modules.

[0209] The input unit (2620) is a configuration for obtaining information such as user input, video, and audio, and may include various input means such as various mechanical / electronic input means, cameras, and microphones. The output unit (2630) is for providing information to users by generating output related to visual, auditory, or tactile senses, and may include a display, speaker, vibration module, and the like.

[0210] The control unit (2640) can control the overall operation of the computing device (2600). The control unit (2640) can process signals, data, information, etc. input or output through the above-described components, or provide predetermined information or functions according to various applications or algorithms stored in the memory (2650). For example, the control unit (2640) can control the overall process for learning the artificial intelligence model disclosed in this specification, prediction and interpretation using the artificial intelligence model, and the overall process related to the debt management capability evaluation method.

[0211] The control unit (2640) may include at least one processor and / or at least one programmable circuit. For example, the control unit (2640) may be implemented as hardware such as a CPU, an application processor (AP), an MCU, a GPU, an NPU, an integrated circuit, an ASIC, an FPGA, etc.

[0212] The memory (2650) can store programs and data required for the operation of the computing device (2600). In addition, the memory (2650) can store data generated or acquired through the control unit (2640). The memory (2650) can be configured as a storage medium such as a ROM, a RAM, a flash memory, an SSD, an HDD, or a combination of storage media.

[0213] The embodiments of the present disclosure described above can be implemented as computer-readable code on a program-recorded medium. Computer-readable media include all types of recording devices that store data that can be read by a computer system. Examples of computer-readable media include hard disk drives (HDDs), solid-state disk drives (SSDs), silicon disk drives (SDDs), read-only memory (ROM), random-access memory (RAM), CD-ROMs, magnetic tapes, floppy disks, and optical data storage devices.

[0214] The description of the above embodiments is merely an example with reference to the drawings for a more thorough understanding of the present disclosure, and should not be construed as limiting the technical idea of ​​the present disclosure.

[0215] In addition, it will be apparent to a person skilled in the art to which the present disclosure pertains that various changes and modifications are possible within a scope that does not depart from the basic principles of the present disclosure.

Claims

1. A method of operating an artificial intelligence system including at least one computing device, A step of obtaining a plurality of training data each including at least one financial variable, at least one non-financial variable, and a label; A step of generating a plurality of primary data based on a plurality of acquired learning data; A step of inputting the plurality of primary fake data into a fake data detection model trained to classify the input data as real data or fake data; and A step of learning an artificial intelligence model using at least one secondary pseudo-data classified as real data by the pseudo-data detection model among the plurality of primary pseudo-data, method.

2. In paragraph 1, The step of generating the above plurality of primary data is: A step of setting a label of the primary data to be generated based on the plurality of learning data; A step of filtering out learning data having a label whose difference from a set label is less than or equal to a first value among the plurality of learning data; and A step of generating the plurality of primary data based on the filtered learning data, method.

3. In paragraph 2, The step of setting the label of the primary data to be generated above is: A step of setting the label of the primary data to be generated based on the ratio of the labels of the plurality of learning data, method.

4. In paragraph 2, The step of generating the plurality of primary data based on the filtered learning data is as follows: When there are multiple filtered learning data, a step of selecting any two learning data among the multiple filtered learning data; and A step of generating primary data based on a weighted average of two selected training data, method.

5. In paragraph 2, The step of generating the plurality of primary data based on the filtered learning data is as follows: If the number of the filtered learning data is less than two, a step of filtering learning data having a label whose difference from the set label is less than or equal to a second value; and A step of generating the plurality of primary data based on the filtered learning data is included, The second value is greater than the first value, method.

6. In paragraph 1, A step of collecting input data including financial and non-financial data about the user; A step of obtaining real data based on some of the multiple variables included in the collected input data; and A step of obtaining a life indicator prediction result for the user from the real data using the learned artificial intelligence model is further included. method.

7. In paragraph 6, The step of obtaining the above life indicator prediction results is: A step of assigning the user to one of a plurality of predefined groups based on some of the plurality of variables included in the input data; A step of obtaining a life indicator prediction result for the user from the real data by using an artificial intelligence model corresponding to the group to which the user is assigned among a plurality of artificial intelligence models learned by group, method.

8. In paragraph 6, Further comprising a step of performing interpretation on the learned artificial intelligence model, method.

9. In paragraph 8, The steps for performing interpretation of the above artificial intelligence model are: A step of calculating feature importance for each of the above multiple variables; A step of generating a PDP (partial dependence plot) for the above multiple variables; and A step of extracting decision rules of multiple decision trees included in the above artificial intelligence model, method.

10. In paragraph 1, The above artificial intelligence model is, Contains a random forest model composed of multiple decision trees, method.

11. In a method of operating an artificial intelligence system including at least one computing device, A step of collecting input data including financial and non-financial data about the user; A step of obtaining real data based on some of the multiple variables included in the collected input data; and A step of obtaining a life indicator prediction result for the user from the real data using an artificial intelligence model, The above artificial intelligence model is, Contains a random forest model composed of multiple decision trees, method.

12. In paragraph 11, The step of obtaining the above life indicator prediction results is: A step of assigning the user to one of a plurality of predefined groups based on some of the plurality of variables included in the input data; A step of obtaining a life indicator prediction result for the user from the real data by using an artificial intelligence model corresponding to the group to which the user is assigned among a plurality of artificial intelligence models implemented for each group, method.

13. In paragraph 11, Further comprising a step of performing an interpretation of the life indicator prediction result based on the artificial intelligence model, the life indicator prediction result, and the input data, The steps for performing the above interpretation are: A step of calculating feature importance for each of the above multiple variables; A step of generating a PDP (partial dependence plot) for the above multiple variables; and A step of extracting decision rules of multiple decision trees included in the above artificial intelligence model and decision paths of the real data. method.

14. In an artificial intelligence system including at least one computing device, A data generation unit for generating a plurality of primary data based on a plurality of learning data, wherein the plurality of learning data includes at least one financial variable, at least one non-financial variable, and a label; A pseudo-data detection model trained to detect and classify whether each of the plurality of primary pseudo-data is real data or pseudo-data; An artificial intelligence model that learns to predict life indicators related to quality of life from input data using at least one secondary pseudo-data misclassified as real data by the pseudo-data detection model among the plurality of primary pseudo-data, and the plurality of learning data. Artificial intelligence system.

15. In paragraph 14, The above data generation unit is, Based on the above plurality of learning data, the label of the primary data to be generated is set, Among the above plurality of learning data, filtering learning data having a label whose difference from the set label is less than or equal to a first value, Generating the plurality of primary data based on the filtered learning data, Artificial intelligence system.

16. In paragraph 15, The above data generation unit is, If there are multiple filtered learning data, select two random learning data from among the multiple filtered learning data, Generate primary data based on the weighted average of two selected training data. Artificial intelligence system.

17. In paragraph 15, The above data generation unit is, If the number of the above filtered learning data is less than two, filtering the learning data having a label whose difference from the set label is less than or equal to the second value, Generate the plurality of primary data based on the filtered learning data, The second value is greater than the first value, Artificial intelligence system.

18. In an artificial intelligence system including at least one computing device, A data collection and preprocessing unit that collects input data including financial and non-financial data about users and obtains real data based on some of the multiple variables included in the collected input data; and It includes a prediction unit that obtains a life indicator prediction result for the user from the real data using an artificial intelligence model, The above artificial intelligence model is, Contains a random forest model composed of multiple decision trees, Artificial intelligence system.

19. In paragraph 18, Further comprising a group assignment unit that assigns the user to one of a plurality of predefined groups based on some of the plurality of variables included in the input data, The above prediction part, Among the multiple artificial intelligence models implemented for each group, an artificial intelligence model corresponding to the group to which the user is assigned is used to obtain a life indicator prediction result for the user from the real data. Artificial intelligence system.

20. In paragraph 18, Further comprising an analysis unit that performs an interpretation of the life indicator prediction result based on the artificial intelligence model, the life indicator prediction result, and the input data, The above interpretation section, Calculate the feature importance for each of the above multiple variables, Generate a PDP (partial dependence plot) for the above multiple variables, Extracting the decision rules of multiple decision trees included in the above artificial intelligence model and the decision path of the above real data, Artificial intelligence system.

Images