Server for classifying payment history data and operating method thereof

Abstract

A method for classifying payment history data performed by a server according to an embodiment of the present invention may comprise the steps of: redundantly providing payment history data to a plurality of inspection terminals; generating, by each of the plurality of inspection terminals, classification data obtained by matching the payment history data based on a categorization system by using a first classification model; receiving the classification data from each of the plurality of inspection terminals; and determining classification data matching the payment history data by comparing the classification data generated by each of the plurality of inspection terminals.

Description

Server and its operation method for classifying payment history data

[0001] The embodiments disclosed in this document relate to a server for classifying payment history data and an operating method thereof, which generates classification data matching payment history data using an artificial intelligence classification model based on an LLM (Large-Scale Language Model), and trains a new artificial intelligence model using the generated classification data to provide an integrated model for classifying payment history.

[0002]

[0003] Thanks to advancements in IT and the proliferation of smartphones, various payment services are being developed that break away from traditional payment methods. When making a payment through these payment services, payment history data containing various information, such as price, merchant, payment date, and payment method, can be generated.

[0004] Payment service providers can provide individual consumption data by analyzing individual consumer consumption trends based on payment history data, and further provide customized financial services for each consumer.

[0005] However, existing payment data provision services have made it difficult to efficiently utilize payment history data due to inaccuracies in the processing or processing of payment history data.

[0006] In addition, existing payment history classification services classified large amounts of data based on human resources, and thus faced various problems such as the payment data processing speed not keeping up with the accumulation speed of payment history data, inaccuracy in data classification due to human error, and difficulty in responding to changing market environments.

[0007] Therefore, an improved classification method for processing payment history data may be required.

[0008]

[0009] One purpose of the embodiments disclosed in this document is to provide a method for efficiently classifying payment history data and generating classification data matched to the payment history data.

[0010] One purpose of the embodiments disclosed in this document is to provide a method for cross-validating classification data generated from each inspection terminal to determine classification data matched to payment history data.

[0011] One purpose of the embodiments disclosed in this document is to provide a method for learning a first classification model based on payment history data and classification data matched to the payment history data, and for generating a lightweight second classification model based on the learned first classification model.

[0012] The technical problems of the embodiments disclosed in this document are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the descriptions below.

[0013]

[0014] A method for classifying payment history data performed by a server according to one embodiment of the present invention may include the steps of: providing payment history data to a plurality of inspection terminals in duplicate; allowing the plurality of inspection terminals to each use a first classification model to generate classification data matching the payment history data to a classification system; receiving the classification data from each of the plurality of inspection terminals; and comparing the classification data generated from each of the plurality of inspection terminals to determine classification data matching the payment history data.

[0015] In addition, the classification system includes a plurality of categorization levels having a hierarchical structure, and the plurality of categorization levels are stored in the form of nodes configured with a parent-child relationship, and the plurality of classifications included in each of the plurality of categorization levels can be expressed as terms corresponding to the reasons for generating the payment history data.

[0016] In addition, the method for classifying the payment history data further includes a step of adding a new category to a classification step of a specific level included in the classification system, and the classification step of the specific level to which the new category is added can maintain a parent-child relationship within the classification system.

[0017] In addition, the method for classifying the payment history data further includes a step of providing an interface for each of the plurality of inspection terminals to match the payment history data to the classification system, wherein the interface visually provides reference data and the classification system to each of the plurality of inspection terminals, and when the payment history data is matched to the classification system in each of the plurality of inspection terminals, the classification included in the classification data to which the payment history data is matched can be highlighted.

[0018] In addition, the step of providing the payment history data to a plurality of inspection terminals in a redundant manner includes: a step of constructing an inspection system including a distributed architecture that expands or reduces the plurality of inspection terminals; a step of predicting a workload that occurs when the inspection system inspects the payment history data; a step of determining a plurality of inspection terminals that can accommodate the workload; and a step of redundantly allocating the payment history data to each of the plurality of confirmed inspection terminals, wherein the payment history data redundantly allocated to each of the plurality of confirmed inspection terminals may include payment history data that is redundant at a preset ratio according to preset conditions.

[0019] In addition, the step of generating the classification data may include a step of determining whether classification data matching the payment history data can be generated using the first classification model, and if the classification data cannot be generated, a step of storing the payment history data for which the classification data could not be generated in a standby database; a step of modifying the classification system so that classification data for the payment history data can be generated; and a step of generating classification data matching the payment history data based on the modified classification system.

[0020] In addition, the step of generating the classification data may include a step of obtaining reference data that provides additional information about the payment history data using the payment history data; a step of extracting keywords related to the payment history data from the reference data; and a step of generating classification data that matches the payment history data to the classification system based on the similarity between the keywords and keywords included in the classification system.

[0021] In addition, the reference data may include at least one of business information, location information, image information, and product / service information, and the step of extracting a keyword related to the payment history data from the reference data may include the steps of: obtaining image information related to the payment history data; obtaining at least one keyword describing an object included in the image information; and extracting a keyword related to the payment history data from at least one keyword describing the object.

[0022] In addition, the step of determining the classification data may include a step of comparing whether the classification data generated from each of the plurality of inspection terminals matches; a step of providing a notification if a mismatch in the classification data is found; and a step of confirming the classification data if the classification data matches.

[0023] In addition, when an inconsistency in the classification data is found, the method may further include a step of providing the payment history data, classification data, and reference data in which the inconsistency is found to the management terminal; a step of receiving, from the management terminal, corrected classification data matching the payment history data in which the inconsistency is found; and a step of confirming the classification data based on the corrected classification data.

[0024] In addition, the method for classifying the payment history data may further include a step of training the first classification model using the payment history data and the classification data; and a step of generating a lightweight second classification model using the trained first classification model.

[0025]

[0026] A server and an operating method according to one embodiment disclosed in this document can provide classification data to a user by matching payment history data from an upper classification level to a lower classification level through a first classification model.

[0027] The server and operating method according to one embodiment disclosed in this document can provide personalized consumption information to a user by providing the user with more accurately and specifically classified classification data through a classification data model.

[0028] In addition, the server and the operating method according to one embodiment disclosed in this document can provide a more improved payment history data classification service to a user by retraining a first classification model and generating a lightweight second classification model based on the trained first classification model.

[0029] The effects that can be obtained from the present invention 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 art to which the present invention belongs from the description below.

[0030]

[0031] FIG. 1 is a block diagram of a server system according to one embodiment of the present invention.

[0032] FIG. 2 is a drawing for explaining the operation of a server according to one embodiment of the present invention.

[0033] FIG. 3 is a drawing for explaining the operation of a server according to another embodiment of the present invention.

[0034] FIG. 4 is a drawing for explaining a classification system according to one embodiment of the present invention.

[0035] FIG. 5 and FIG. 6 are drawings for explaining an additional method of classification according to one embodiment of the present invention.

[0036] FIG. 7 and FIG. 8 are drawings for explaining an interface providing method according to one embodiment of the present invention.

[0037] FIG. 9 is a diagram for explaining a method for providing payment history data according to one embodiment of the present invention.

[0038] Figure 10 is for explaining a method for generating classification data according to one embodiment of the present invention.

[0039] FIG. 11 and FIG. 12 are drawings for explaining a method for generating classification data according to another embodiment of the present invention.

[0040] FIG. 13 is a diagram for explaining a method for determining classification data according to one embodiment of the present invention.

[0041] FIG. 14 is a diagram for explaining a method for generating a second classification model according to one embodiment of the present invention.

[0042]

[0043] Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. When designating components in each drawing, it should be noted that, where possible, identical components will be given the same reference numerals, even if they appear in different drawings. Furthermore, when describing embodiments of the present invention, detailed descriptions of related known structures or functions will be omitted if they are deemed to hinder understanding of the embodiments of the present invention.

[0044] In describing components of embodiments of the present invention, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only intended to distinguish the components from other components, and the nature, order, or sequence of the components are not limited by these terms. Furthermore, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning they have in the context of the relevant technology, and shall not be interpreted in an idealized or overly formal sense unless explicitly defined in this application.

[0045]

[0046] FIG. 1 is a block diagram of a server system according to one embodiment of the present invention.

[0047] Referring to FIG. 1, the server system (10) may include a server (100), a plurality of inspection terminals (200-1, 200-2 to 200-n), and a management terminal (300), and may be connected to an external network (20) via wired or wireless communication.

[0048] The entities included in the server system (10) are not limited to the example shown in FIG. 1, and the server (100) and the administrator terminal (300) may be multiple, as may the inspection terminals (200-1, 200-2 to 200-n, etc.).

[0049] An external network (20) may, for example, refer to a public communication network such as the Internet. According to another embodiment, the external network (20) may include a local area communication network such as Ethernet or an intranet.

[0050] The external network (20) may be a network capable of communicating with a Value Added Network (VAN) server. By communicating with the VAN server, the external network (20) allows the server system (10) to access payment history data between the financial institution and the merchant. The payment history data may include comprehensive payment information.

[0051] For example, payment history data may include information about the merchant where the payment occurred, information about the time the payment was authorized, information about the amount of the payment authorized, and information about the card from which the payment authorization was requested.

[0052] Payment history data stored in the VAN server can be transmitted to the server system (10) via an external network (20).

[0053] The server (100) can collect payment history data from an external network (20). According to one embodiment, the server (100) can transmit the collected payment history data to a plurality of inspection terminals (200-1, 200-2 to 200-n) and receive classification data matching the payment history data to a classification system from the plurality of inspection terminals (200-1, 200-2 to 200-n). The server (100) can compare the received classification data to determine classification data matching the payment history data.

[0054] In another embodiment, the server (100) may generate classification data by matching the collected payment history data to a categorization system using the first classification model. When the server (100) generates the classification data, the server (100) may request the generation of inspection data indicating the accuracy of the classification from a plurality of inspection terminals (200-1, 200-2 to 200-n), and may determine classification data matching the payment history data based on the received inspection data.

[0055] The first classification model may be an artificial intelligence model, including a Large Language Model (LLM). The LLM may be a pre-trained model based on a large amount of data.

[0056] The first classification model may include multiple AI model architectures for extracting text and matching it to a classification system. For example, the first classification model may include a Transformer model, including an encoder, decoder, and self-attention layer. The Transformer model is an AI model advantageous for natural language processing and may be advantageous for processing payment history data provided in text format.

[0057] The classification system that the first classification model uses to match payment history data may include multiple categorization levels with a hierarchical structure, and these categorization levels may be stored as nodes with parent-child relationships. Furthermore, the multiple categories contained within each categorization level may be expressed using terms that correspond to the reasons for generating the payment history data.

[0058] Alternatively, the classification system that the first classification model uses to match payment history data can be structured hierarchically, from upper to lower categories. Upper and lower categories can be stored as nodes with parent-child relationships. Furthermore, upper categories can be expressed using words representing higher-level concepts that correspond to the reasons for which the payment history data was generated.

[0059] Accordingly, a first classification model that matches payment history data with a classification system may include a model that structures payment history data so that keywords for classification can be extracted from the payment history data, a model that extracts keywords from the structured data, and a model that matches the extracted keywords with each of the classifications included in the classification system.

[0060] According to another embodiment, the server (100) can modify and update a classification system for classifying payment history data.

[0061] If the classification system does not have an appropriate classification, it may be difficult to match payment history data to the classification system using the first classification model, and it may be difficult to generate classified data. The server (100) may generate a new classification based on the received payment history data and add the new classification to a specific level of classification. In this case, the new classification may be expressed in terms corresponding to the reason for generating the received payment history data, and the classification stage at the specific level to which the new classification is added may maintain a hierarchical structure within the classification system.

[0062] The server (100) may provide an interface for matching payment history data to a classification system on multiple inspection terminals (200-1, 200-2, and 200-n). The interface may be visually provided through each inspection terminal (e.g., 200-1) and may include, for example, a UI (User Interface) provided to a user of each inspection terminal (e.g., 200-1). For example, the interface may visually provide payment history data and a classification system, and may visually provide reference data that provides additional information about the payment history data.

[0063] When the server (100) generates classification data that matches payment history data to a categorization system, the server (100) can use the payment history data to obtain reference data that provides additional information about the payment history data. The reference data may include at least one of the business information of the merchant where the payment occurred, the merchant's location information, image information, and product / service information. The image information may be payment history, an image of the payment merchant, or an image of a product / service sold by the payment merchant.

[0064] Reference data may be generated based on a separate artificial intelligence model architecture distinct from the first classification model. More specifically, the server (100) may obtain reference data related to payment history data from an external network (20) based on keywords extracted from the payment history data. The reference data may have any data format, such as images, text, or videos. The artificial intelligence model generating the reference data may extract keywords from images, videos, or text, assign semantic weights to the extracted keywords, and provide them as reference data. For example, the artificial intelligence model generating the reference data may include a Large Multimodal Model (LMM) capable of processing various types of data formats.

[0065] The server (100) can receive classification data from each of the first to nth inspection terminals (200-1, 200-2 to 200-n). The server (100) can compare the classification data generated from each of the first to nth inspection terminals (200-1, 200-2 to 200-n) to determine classification data that matches the payment history data.

[0066] If the classification data matches, the server (100) can confirm the classification data as classification data for the payment history data. If an inconsistency in the classification data is found, the server (100) can provide a notification to the inspection terminals where the inconsistency is found. In addition, the notification can be provided together with the management terminal (300). If an inconsistency in the classification data is found, the server (100) can provide the payment history data, classification data, and reference data related to the payment history where the inconsistency is found to the management terminal (300). Then, the server can receive modified classification data matching the payment history data where the inconsistency is found from the management terminal (300). The server (100) can confirm the classification data for the payment history data based on the modified classification data received from the management terminal (300).

[0067] In another embodiment, if a mismatch in classification data is found, the management terminal can transmit the payment history data in which the mismatch is found to the server or inspection terminals to regenerate the classification data. The server or inspection terminals can regenerate the classification data using the first classification model. The management terminal can compare the multiple regenerated classification data to generate revised classification data. Furthermore, the management terminal can compare the multiple inspection data received from the multiple inspection terminals and generate revised classification data.

[0068] Thereafter, the server can confirm the classification data based on the modified classification data. Furthermore, the server can determine (or confirm) the classification data matching the payment history data based on the received review data.

[0069] The server (100) and the first to nth inspection terminals (200-1, 200-2 to 200-n) can form a distributed architecture for generating classification data.

[0070] A distributed architecture may refer to an architecture that can selectively expand or reduce processors participating in a task for efficient processing of data in a system (e.g., a server system (10)) including multiple processors (e.g., a server (100) and multiple inspection terminals (200-1, 200-2 to 200-n)).

[0071] More specifically, when the server (100) according to one embodiment of the present invention repeatedly provides payment history data to a plurality of inspection terminals (200-1, 200-2 to 200-n) for classification data generation, the server (100) may selectively determine a plurality of inspection terminals to accommodate the workload based on the predicted workload. For example, if the predicted workload is high, the server (100) may additionally allocate inspection terminals for classification data generation. If the predicted workload is low, the server (100) may reduce the number of assigned inspection terminals.

[0072] Additionally, the server (100) can redundantly allocate payment history data to each of the multiple confirmed inspection terminals. The redundantly allocated payment history data can be maintained at a preset ratio for each inspection terminal. For example, the preset ratio may be 30% of the payment history data allocated to each inspection terminal. The preset ratio may be determined based on factors such as workload and the accuracy required for the classified data.

[0073] For example, if the workload is excessive and all inspection terminals are participating in the process, the preset ratio can be reduced to increase the amount of payment history data processed. If the accuracy required for classification data is high, the preset ratio can be increased to allow comparison of classification data generated from multiple inspection terminals.

[0074] Classification data for duplicated payment history data may be data that is cross-verified through the server (100).

[0075] In another embodiment, the server (100) and the first to nth inspection terminals (200-1, 200-2 to 200-n) may configure a distributed architecture for generating inspection data. As described above, when the server (100) generates classification data that matches payment history data to a classification system using the first classification model, the server (100) may request the generation of inspection data from a plurality of inspection terminals (200-1, 200-2 to 200-n).

[0076] When the server (100) provides redundant inspection data generation to multiple inspection terminals (200-1, 200-2, and 200-n), the server (100) can selectively determine multiple inspection terminals to accommodate the workload based on the predicted workload. For example, if the predicted workload is high, the server (100) can additionally allocate inspection terminals for inspection data generation. If the predicted workload is low, the server (100) can reduce the number of assigned inspection terminals.

[0077] The server (100) can learn a first classification model using payment history data and classification data determined to match the payment history data. Learning using the payment history data and the determined classification data can be performed using supervised learning.

[0078] The server (100) can generate a second classification model based on the learned first classification model. The second classification model may be a classification model that is a lightweight version of the first classification model.

[0079] Model lightweighting can be achieved by optimizing the architecture of an AI model to create a more efficient model.

[0080] Model lightweighting can include various methods, such as optimizing the number of computational layers included in an AI model, optimizing the weights used in the computational layers, optimizing the connections between computational layers, network slimming, or weight sharing.

[0081] The first to nth inspection terminals (200-1, 200-2 to 200-n) may be various types of devices capable of processing or generating data based on data received from the server (100) or visually outputting a received interface.

[0082] The first to nth inspection terminals (200-1, 200-2 to 200-n) may include a display for outputting an interface, or may be connected to a separate display device. In addition, the first to nth inspection terminals (200-1, 200-2 to 200-n) may additionally include an input device for receiving external input.

[0083] According to one embodiment, the first to nth inspection terminals (200-1, 200-2 to 200-n) may each receive payment history data from the server (100) and match the received payment history data to a classification system using the first classification model.

[0084] In other words, even if the server (100) does not match the payment history data to the classification system, classification data for the payment data can be generated using the first classification model in each inspection terminal (e.g., 200-1). However, the first classification model operated in each inspection terminal (e.g., 200-1) may be substantially identical to the first classification model described above in the server (100).

[0085] Modification and update of the classification system or acquisition of reference data can also be performed at the first to nth inspection terminals (200-1, 200-2 to 200-n) on behalf of the server (100).

[0086] Likewise, learning of the first classification model using payment history data and determined classification data can be performed in the first to nth inspection terminals (200-1, 200-2 to 200-n), respectively, and generation of the second classification model, which is a lightweight version of the first classification model, can also be performed in the first to nth inspection terminals (200-1, 200-2 to 200-n), respectively.

[0087] The management terminal (300) may be an upper terminal for the first to nth inspection terminals (200-1, 200-2 to 200-n).

[0088] According to one embodiment, when a mismatch in classification data is found, the management terminal (300) may receive payment history data, classification data, and reference data, etc. in which a mismatch is found, along with a notification from the server (100), and transmit corrected classification data to the server (100) based on the received data.

[0089] The revised classification data may be classification data determined by an administrator with access to the management terminal (300) to match the payment history data. The payment history data and the revised classification data matched to the payment history data may then be used to train the first classification model. The administrator may request the generation of classification data from a specific inspection terminal again via the management terminal (300).

[0090] If an inconsistency in the classification data is found, the management terminal (300) can transmit the payment history data in which the inconsistency is found to the server (100) or the first to nth inspection terminals (200-1, 200-2 to 200-n) to regenerate the classification data. The classification data can be performed by the first classification model. Individual or multiple classification data regenerated by the first classification model can be transmitted to the first to nth inspection terminals (200-1, 200-2 to 200-n). In addition, individual or multiple classification data regenerated by the first classification model can be transmitted to the management terminal (300). The management terminal (300) can compare the received plurality of classification data or inspection data to generate corrected classification data. Thereafter, the server can confirm the classification data based on the corrected classification data. In addition, the server can determine (or confirm) classification data matching the payment history data based on the received inspection data.

[0091] According to another embodiment, when an inconsistency in the inspection data is found, the management terminal (300) may receive payment history data, classification data, reference data, etc. in which an inconsistency is found along with a notification from the server (100), and may transmit the modified inspection data to the server (100) based on the received data. The administrator may request a specific inspection terminal to generate inspection data again through the management terminal (300).

[0092]

[0093] FIG. 2 is a drawing for explaining the operation of a server according to one embodiment of the present invention.

[0094] Through Fig. 2, data transmitted and received between an external network (20), a server (100), and a first inspection terminal (200-1) and operations performed in each configuration can be specifically described.

[0095] An embodiment of generating classification data using a first classification model in each inspection terminal (e.g., the first inspection terminal (200-1)) can be described through FIG. 2.

[0096] According to the embodiment of FIG. 2, the server (100) can receive payment history data from an external network (20) (S100).

[0097] Payment history data received by the server (100) from an external network (20) may be de-identified within the server (100) at preset intervals. De-identification processing may refer to a process of making it impossible to infer a specific individual through processing such as deletion, replacement, and categorization of the data. The de-identified payment history data may be stored within the server (100).

[0098] As described in FIG. 1, the external network (20) may be a network capable of communicating with an external server (e.g., a VAN server) related to payment.

[0099] The server (100) can provide payment history data received from an external network to multiple inspection terminals in duplicate (S200).

[0100] For convenience of explanation, in FIG. 2, the server (100) is depicted as providing payment history data to the first inspection terminal (200-1), but in reality, the server (100) may configure a distributed architecture with multiple inspection terminals, and may provide payment history data to multiple inspection terminals in duplicate for efficient processing of payment history data.

[0101] A plurality of inspection terminals (e.g., the first inspection terminal (200-1)) that have received payment history data can each use the first classification model to generate classification data that matches the payment history data to a classification system (S300).

[0102] As described in Figure 1, the classification system may include multiple classification levels having a hierarchical structure, and each classification level may include multiple classifications. Furthermore, the classification levels may be stored in the form of nodes organized in parent-child relationships.

[0103] An exemplary classification scheme will be described in detail through Figure 4.

[0104] The first classification model may be an artificial intelligence model that structures payment history data, extracts keywords from the structured data, and matches the extracted keywords to each category within the classification system. The first classification model may include a natural language processing model. The first classification model may be an artificial intelligence model that includes a Large Language Model (LLM). The LLM may be a pre-trained model based on a large amount of data.

[0105] The first classification model may be stored within each inspection terminal (e.g., 200-1) or stored on an AI server located within the server, within the server system, or externally. If the first classification model is stored on the AI ​​server, each inspection terminal (e.g., 200-1) can access the AI ​​server and execute the first classification model. Since the first classification model is stored on a separate AI server, the resources required for each inspection terminal (e.g., 200-1) to execute the first classification model may be reduced.

[0106] The server (100) can receive classification data from each of the multiple inspection terminals (S400). The classification data may be data matching payment history data to multiple categories included in the classification system. The classification system may include multiple classification stages, and payment history data may be matched to the most relevant category among the categories included in the classification stage.

[0107] The server (100) can compare classification data generated from each of a plurality of inspection terminals to determine classification data matching the payment history data (S500). The server (100) can receive multiple classification data for any payment history data and determine the classification data by verifying whether the received classification data matches.

[0108]

[0109] FIG. 3 is a drawing for explaining the operation of a server according to another embodiment of the present invention.

[0110] Through Fig. 3, data transmitted and received between an external network (20), a server (100), and a first inspection terminal (200-1) and operations performed in each configuration can be specifically described.

[0111] An embodiment of generating classification data using a first classification model in a server (100) can be described through FIG. 3.

[0112] According to the embodiment of FIG. 3, the server (100) can receive payment history data from an external network (20) (S1100).

[0113] The server (100) can use the first classification model to create classification data that matches payment history data to a classification system (S1200).

[0114] The first classification model may be stored within the server (100) or stored in an artificial intelligence server located within or outside the server system. If the first classification model is stored in the artificial intelligence server, the server (100) can access the artificial intelligence server and execute the first classification model. Since the first classification model is stored in a separate artificial intelligence server, the resources required by the server (100) for executing the first classification model may be reduced.

[0115] According to an embodiment, the server (100) can generate multiple classification data for one payment history data using the first classification model.

[0116] Additionally, the server can generate one or more classification data for a single payment history data using the first classification model, and request the generation of inspection data for the generated one or more classification data through multiple inspection terminals. The server or management terminal can receive multiple inspection data from multiple inspection terminals and verify whether the inspections match.

[0117] The server (100) can request inspection data indicating the accuracy of classification for each of a plurality of inspection terminals (S1300).

[0118] For convenience of explanation, in FIG. 3, the server (100) is depicted as requesting the first inspection terminal (200-1) to generate inspection data. However, the actual server (100) may configure a distributed architecture with multiple inspection terminals, and may request the generation of inspection data to multiple inspection terminals in duplicate for efficient generation of inspection data.

[0119] Each of the multiple inspection terminals can generate inspection data for the classification data (S1400).

[0120] According to an embodiment, each inspection terminal may include a pre-trained inspection model and may determine the accuracy of classification data generated by the server (100) based on the inspection model. The inspection model may be trained based on payment history data and determined classification data, for example.

[0121] The server (100) can receive inspection data from each of a plurality of inspection terminals (e.g., the first inspection terminal (200-1)) (S1500).

[0122] The server (100) can compare the received inspection data and determine classification data matching the payment history data (S1600).

[0123] Unlike the embodiment of FIG. 2, the embodiment of FIG. 3 performs classification data generation on the server (100), which may result in a greater computational load on the server (100) compared to the embodiment of FIG. 2. However, the embodiment of FIG. 3 processes payment history data within the first classification model included in the server (100) and is not provided separately to the inspection terminal. Therefore, compared to the embodiment of FIG. 2, the embodiment has the advantage of a lower risk of security issues with payment history data, which is personal information.

[0124]

[0125] FIG. 4 is a drawing for explaining a classification system according to one embodiment of the present invention.

[0126] An exemplary classification system (CS1) is illustrated in FIG. 4. The classification system (CS1) may include a plurality of classification levels (CL1, CL2, CL3, and CL4) having a hierarchical structure. The classification system (CS1) illustrated in FIG. 4 includes four classification levels (CL1, CL2, CL3, and CL4), but the number of classification levels may vary depending on the embodiment. Each of the classification levels (CL1, CL2, CL3, and CL4) has a hierarchical structure and may include a plurality of classifications for each classification level.

[0127] Looking at the highest level, the first classification level (CL1), the first classification level (CL1) can include food expenses, housing expenses, transportation expenses, communication expenses, clothing expenses, and cultural expenses as each category.

[0128] The second classification stage (CL2), which is the second layer, may include classifications that are configured in a parent-child relationship with the classifications included in the first classification stage (CL1). In one embodiment, the second classification stage (CL2) may be configured as a subnode of the 'food expenses' classification included in the first classification stage (CL1). The second classification stage (CL2) may include restaurants, groceries, and beverages as their respective classifications. Although the subnodes of 'food expenses' are illustrated in FIG. 4 as being included in the second classification stage (CL2), classifications corresponding to subnodes of housing expenses, transportation expenses, communication expenses, or clothing expenses (e.g., 'monthly rent', which is a subnode of housing expenses) may also be included in the second classification stage (CL2) while maintaining a parent-child relationship with the upper classification.

[0129] The third classification level (CL3), which is the third layer, may include classifications that are configured in a parent-child relationship with the classifications included in the second classification level (CL2). In one embodiment, the third classification level (CL3) may be configured as a subnode of the 'restaurant' classification included in the second classification level (CL2). The third classification level (CL3) may include Korean food, Japanese food, Chinese food, coffee shop, and bakery as classifications, respectively. Although the subnodes of 'restaurant' are illustrated in FIG. 4 as being included in the third classification level (CL3), classifications corresponding to subnodes of groceries or beverages (e.g., 'mart', which is a subnode of groceries) may also be included in the third classification level (CL3) while maintaining a parent-child relationship with the upper classification.

[0130] The fourth classification level (CL4), which is the fourth layer, may include classifications that are comprised of parent-child relationships with the classifications comprised by the third classification level (CL3). In one embodiment, the fourth classification level (CL4) may be comprised of subnodes of the 'Korean food' classification comprised by the third classification level (CL3). The fourth classification level (CL4) may include, for example, subnodes of rice bowl specialty stores, porridge specialty stores, gamjatang specialty stores, and kimbap specialty stores, as their respective classifications. Although FIG. 4 illustrates that subnodes of 'Korean food' are comprised of the fourth classification level (CL4), in another embodiment, classifications corresponding to subnodes of Japanese food, Chinese food, coffee shops, or bakeries (for example, 'udon specialty stores', which are subnodes of Japanese food) may also be comprised of the fourth classification level (CL4) while maintaining a parent-child relationship with the upper classification.

[0131] A single payment history data can be matched with categories within each of multiple classification stages. In other words, the classification data may be a hierarchical structure that matches multiple categories with payment history data.

[0132] For example, if the payment details data is for a kimbap specialty store, the payment details data may be matched with 'food expenses', 'restaurant', 'Korean food', and 'kimbap specialty store', and the classification data may include 'food expenses', 'restaurant', 'Korean food', and 'kimbap specialty store'.

[0133]

[0134] FIG. 5 and FIG. 6 are drawings for explaining an additional method of classification according to one embodiment of the present invention.

[0135] Referring to FIG. 5, the method for classifying payment history data may further include a step (S220) of adding a new classification to a specific level of classification steps included in the classification system. The step (S220) of adding a new classification may be performed via the server (100), and the server (100) may add a classification based on the payment history data collected, and thereafter transmit the classification system with the added classification to each of the inspection terminals.

[0136] In another embodiment, individual inspection terminals (e.g., 200-1) may add new classifications to a particular level of classification steps.

[0137] A classification step at a particular level where a new classification is added can maintain the node form consisting of existing parent-child relationships within the classification system.

[0138]

[0139] Additional methods of classification are described in detail with reference to Fig. 6.

[0140] Figure 6 illustrates exemplary payment history data (PDD1) and a classification system (CS2) with a new classification added. A server (100) or an inspection terminal (e.g., 200-1) can extract keywords for classification from the payment history data (PDD1). For example, the keyword for classification in the payment history data (PDD1) of Figure 6 may be "beer."

[0141] The server (100) or the inspection terminal (e.g., 200-1) can add classifications to each of the classification stages (CL1, CL2, CL3, or CL4) based on the extracted keywords. Comparing the classification system (CS1) of FIG. 4 with the classification system (CS2) of FIG. 6, the classification system (CS2) of FIG. 6 can add 'pub' to the third classification stage (CL3). Accordingly, the fourth classification stage (CL4) can be configured as a subnode of the 'pub' classification included in the third classification stage (CL3). In one embodiment, the fourth classification stage (CL4) can include beer specialty stores, soju specialty stores, liquor specialty stores, makgeolli specialty stores, wine specialty stores, and sake specialty stores by adding them to each classification.

[0142] Even if new categories (e.g., "Pub" and "Beer Shop") are added, classification levels (e.g., CL3 and CL4) can maintain existing parent-child nodes. More specifically, the newly added "Food," "Restaurant," "Pub," and "Beer Shop" can be included in the first through fourth classification levels (CL1 through CL4), respectively, while still maintaining parent-child nodes among themselves.

[0143]

[0144] FIG. 7 and FIG. 8 are drawings for explaining an interface providing method according to one embodiment of the present invention.

[0145] Referring to FIG. 7, the method for classifying payment history data may further include a step (S240) in which the server (100) provides an interface for each of a plurality of inspection terminals to match payment history data to a classification system.

[0146] The interface may, for example, visually provide reference data and a classification system to each of a plurality of inspection terminals. The interface may include a user interface (UI) provided to users of the inspection terminals.

[0147] When payment history data is matched to a classification scheme, the interface can highlight the classifications to which the payment history data is matched.

[0148]

[0149] An exemplary interface provided through any inspection terminal is illustrated in FIG. 8.

[0150] Exemplary payment history data may be illustrated at the top of Figure 8. The server (100) may obtain reference data that provides additional information about the payment history data. The reference data may be obtained in the form of an image, video, or text, and the server (100) may extract keywords related to the payment history data from the reference data. The exemplary reference data may be illustrated at the bottom of the interface.

[0151] Additionally, the classification system can be visually presented through the interface. The classification data matching the payment history data depicted in Figure 8 may include the categories "Food Expenses," "Restaurant," "Korean Food," and "Bowl Specialty Store." Accordingly, the interface can highlight the categories "Food Expenses," "Restaurant," "Korean Food," and "Bowl Specialty Store" that match the payment history data.

[0152]

[0153] FIG. 9 is a diagram for explaining a method for providing payment history data according to one embodiment of the present invention.

[0154] The step (S200a) of providing payment history data to multiple inspection terminals in duplicate is specifically described through FIG. 9.

[0155] The step (S200a) of providing payment history data to multiple inspection terminals in duplicate may include the step (S210a) of building an inspection system including a distributed architecture that expands or reduces the number of inspection terminals. The distributed architecture may be an architecture that selectively expands or reduces the number of processors participating in a task for efficient data processing in a system including multiple processors.

[0156] The server (100) can interconnect with multiple inspection terminals to generate classification data, and can selectively expand or reduce the number of inspection terminals to accommodate the workload for generating classification data for any payment history data.

[0157] After the inspection system including the distributed architecture is built, the server (100) can predict the workload incurred when the inspection system inspects payment history data (S220a). The server (100) can collect current load information from each of the multiple inspection terminals and predict the workload based on the collected load information and payment history data. For example, if the payment history data has a large amount of information, excessive computation is required to structure the payment history data, or if the payment history data is difficult to classify, the workload can be predicted to be high.

[0158] The server (100) can determine a plurality of inspection terminals capable of handling the workload (S230a). For example, among the plurality of inspection terminals capable of handling the workload, those with available processing resources may be preferentially selected. A terminal with available processing resources may refer to a terminal with a lower current processing load compared to the maximum load it can handle, or a terminal with a lower current processing load compared to other terminals.

[0159] After multiple terminals are confirmed, the server (100) can assign payment history data to each of the multiple confirmed inspection terminals (S240a). The payment history data assigned to each of the multiple confirmed inspection terminals may include a preset ratio of duplicate payment history data based on preset conditions.

[0160] Pre-set conditions may include, for example, the size of the payment history data, the load that each verification terminal can handle, the capacity of the total payment history data being processed by the server system, and the accuracy required for the classification data.

[0161] For example, if the volume of payment history data being processed on a server system is large and the load allocated to additional payment history data is small, the percentage of payment history data that is redundantly allocated to each inspection terminal can be reduced. A lower redundancy ratio allows each inspection terminal to process different payment history data instead of processing the same payment history data repeatedly, resulting in faster data processing.

[0162] Conversely, if the accuracy level required for classification data is high, the rate of duplicate assignments may increase. By increasing the rate of duplicate assignments to each inspection terminal, multiple classification data sets can be generated for payment history data, and the accuracy of these classification data can be improved by comparing them with each other.

[0163]

[0164] Figure 10 is for explaining a method for generating classification data according to one embodiment of the present invention.

[0165] The step (S300a) of creating classification data by matching payment history data to a classification system is specifically described through Fig. 10.

[0166] Each inspection terminal or server may include a step (S310a) of determining whether classification data matching the payment history data can be generated using the first classification model.

[0167] If classification data can be generated (YES path of S310a), each inspection terminal can generate classification data according to the classification system (S320a).

[0168] If classification data cannot be generated (NO path of S310a), the inspection terminal can store payment history data for which classification data could not be generated in a standby database (S330a).

[0169] The standby database can temporarily store payment history data. Payment history data stored in the standby database can be used to create classification data after modifying the classification system. The standby database can include at least a portion of the storage area contained in the server or inspection terminal.

[0170] Each inspection terminal or server can modify the classification system so as to generate classification data for the payment history data (S340a).

[0171] Modifying a classification system may involve adding categories to a specific level of classification within the classification system, or adding new categories at a new level. Modifying a classification system may also involve deleting certain categories from an existing classification system or replacing them with more appropriate terminology.

[0172] Each inspection terminal or server can generate classification data matching the payment history data based on the modified classification system (S350a).

[0173] The modified classification system may include a classification for matching payment history data temporarily stored in the standby database, and each inspection terminal or server may generate classification data by matching payment history data stored in the standby database based on the modified classification system.

[0174]

[0175] FIG. 11 and FIG. 12 are drawings for explaining a method for generating classification data according to another embodiment of the present invention.

[0176] The step (S300b) of creating classification data by matching payment history data to a classification system is specifically described through Fig. 11.

[0177] Each inspection terminal or server can obtain reference data that provides additional information about the payment history data by using the payment history data (S310b).

[0178] Reference data may include, for example, at least one of business information, location information, image information, and product / service information. In some embodiments, the reference data may be data generated from additional data collected or retrieved by the inspection terminal or server based on payment history data.

[0179] The inspection terminal or server may utilize a separate AI model architecture, distinct from the first classification model, to acquire reference data. For example, the inspection terminal or server may utilize a Convolutional Neural Network (CNN) model architecture to generate reference data. The CNN model may be an AI model suitable for image processing.

[0180] The inspection terminal or server can extract keywords related to payment history data from the reference data (S320b).

[0181] The step of extracting keywords may include, for example, a step of obtaining image information related to payment history data, a step of recognizing an object included in the image information to obtain at least one keyword describing the object, and a step of extracting a keyword related to the payment history data from at least one keyword describing the object.

[0182] For example, data related to payment history data can be collected in any data format. If the collected data includes image information, objects contained in the image information can be extracted using an image analysis AI model, such as the CNN described above. The inspection terminal or server can obtain keywords based on the extracted objects and, from the keywords associated with the objects, separately extract keywords related to the payment history data.

[0183] Thereafter, the inspection terminal or server can generate classification data that matches the payment history data to the classification system based on the similarity between keywords related to the payment history data and keywords included in the classification system (S330b).

[0184]

[0185] A method for obtaining reference data is specifically described with reference to Fig. 12.

[0186] Figure 12 illustrates exemplary payment history data (PDD2) and reference data (RD1). An inspection terminal or server can perform a search through an external network based on merchant information (e.g., Egg Bomb Rice Bowl Samcheok Branch) contained in the payment history data (PDD2), and the search results can be collected in the form of reference data (RD1) that provides additional information.

[0187] The inspection terminal or server can perform image processing, video processing, or text processing on the collected reference data to extract keywords.

[0188] Keywords extracted from Fig. 12 may include, for example, keywords for price (10,000 won, 5,900-7,500 won, 8,500-9,500 won, etc.) or keywords for menu (bulgogi, egg, fried food, rice bowl, etc.), and the keywords may be used to create classification data corresponding to payment history data.

[0189] Based on keywords extracted from the reference data, the verification terminal or server can infer that the payment history data (PDD2) is collected from a Korean rice bowl specialty restaurant. Therefore, the verification terminal or server can determine that the payment history data (PDD2) matches the classification data, including "food expenses," "restaurant," "Korean food," and "rice bowl specialty restaurant."

[0190]

[0191] FIG. 13 is a diagram for explaining a method for determining classification data according to one embodiment of the present invention.

[0192] The step (S500a) of comparing classification data through Figure 13 to determine classification data matching payment history data is specifically described.

[0193] The server can compare the classification data generated from each of the multiple inspection terminals for consistency (S510a). By determining whether the classification data generated from each of the multiple inspection terminals matches, the server can improve the accuracy of the classification data.

[0194] If the classification data matches (YES path of S510a), the server can confirm the classification data (S520a).

[0195] If a mismatch in classification data is detected (NO path of S510a), the server may provide a notification (S530a). The server may provide a notification to the inspection terminal that generated the classification data where the mismatch occurred.

[0196] If a mismatch in classification data is detected, the server can notify the management terminal. The management terminal may be a parent terminal for multiple inspection terminals. The management terminal collects information about the inspection terminals where the mismatch occurred and, based on this information, evaluates and manages the classification accuracy of each inspection terminal.

[0197] The server can provide payment history data, classification data, and reference data in which inconsistencies are found to the management terminal (S540a).

[0198] The server can receive correction classification data matching the payment history data in which a mismatch is found from the management terminal (S550a).

[0199] The management terminal may determine a more appropriate classification data among multiple classification data as the modified classification data, or, if none of the multiple classification data is appropriate, may match new appropriate classification data based on the payment history data, classification data, and reference data.

[0200] In another embodiment, if a mismatch in classification data is found, the management terminal can transmit the payment history data containing the mismatch to a server or inspection terminals to regenerate the classification data. The server or inspection terminals can regenerate the classification data using the first classification model. The management terminal can compare the regenerated multiple classification data to generate revised classification data.

[0201] Additionally, the management terminal can compare multiple inspection data received from multiple inspection terminals and generate modified classification data.

[0202] Thereafter, the server can confirm the classification data based on the modified classification data (S560a). Furthermore, the server can determine (or confirm) the classification data matching the payment history data based on the received inspection data.

[0203]

[0204] FIG. 14 is a diagram for explaining a method for generating a second classification model according to one embodiment of the present invention.

[0205] The second classification model may be a lightweight classification model compared to the first classification model.

[0206] After classification data matching the payment history data is determined, the server or inspection terminal can train the first classification model using the payment history data and classification data (S600).

[0207] Training of the first classification model using payment history data and determined classification data can be performed using supervised learning. Supervised learning refers to a learning method that improves the accuracy of an AI model by inputting both the target data and the correct answer into the model.

[0208] For example, the classification accuracy of the first classification model can be improved by inputting payment history data and determined classification data together for the first classification model.

[0209] The server or inspection terminal can create a lightweight second classification model using the learned first classification model (S700).

[0210] Compared to the first classification model, the lightweight second classification model has a reduced number of computational layers and can optimize the weights used in the computational layers. Furthermore, the second classification model can optimize the connections between computational layers.

[0211] The server or verification terminal can generate classification data for new payment history data based on a lightweight second classification model instead of the first classification model. The second classification model can enable faster data processing while reducing the resources required for data processing.

[0212]

[0213] The above description is merely an example of the technical idea of ​​the present disclosure, and those skilled in the art to which the present disclosure pertains will be able to make various modifications and variations without departing from the essential characteristics of the present disclosure.

[0214] Accordingly, the embodiments disclosed in this disclosure are intended to illustrate, rather than limit, the technical concepts of this disclosure, and the scope of the technical concepts of this disclosure is not limited by these embodiments. The scope of protection of this disclosure should be interpreted by the claims below, and all technical concepts within the scope equivalent thereto should be construed as being included within the scope of the rights of this disclosure.

Claims

1. A method for classifying payment history data performed by a server, A step of providing payment history data to multiple inspection terminals in duplicate; A step in which the plurality of inspection terminals each use the first classification model to generate classification data matching the payment history data to a classification system; A step of receiving the classification data from each of the plurality of inspection terminals; and A method comprising a step of comparing the classification data generated from each of the plurality of inspection terminals and determining classification data matching the payment history data.

2. In paragraph 1, The above classification system includes multiple categorization levels having a hierarchical structure, The above multiple classification steps are stored in the form of nodes consisting of parent-child relationships, A method in which each of the plurality of classification steps included in the above-mentioned plurality of classifications is expressed as a term corresponding to the reason for generating the payment history data.

3. In paragraph 1, The method of classifying the above payment history data is as follows: Further comprising a step of adding a new category to a specific level of classification step included in the above classification system, A method for maintaining a parent-child relationship within the classification system at a specific level of classification steps to which the above new classification is added.

4. In paragraph 1, The method of classifying the above payment history data is as follows: Further comprising a step of providing an interface for each of the plurality of inspection terminals to match the payment history data to the classification system, The above interface visually provides reference data and the above classification system to each of the plurality of inspection terminals, A method for highlighting a classification included in the classification data to which the payment history data is matched when the payment history data is matched to the classification system in each of the plurality of inspection terminals.

5. In paragraph 1, The step of providing the above payment details data to multiple inspection terminals in duplicate is: A step of building an inspection system including a distributed architecture that expands or reduces the plurality of inspection terminals; A step of predicting the workload that occurs when the above inspection system inspects the payment history data; A step of determining a plurality of inspection terminals capable of accommodating the above workload; and Including a step of redundantly assigning the above payment details data to each of the plurality of confirmed inspection terminals, A method in which payment history data that is duplicated and allocated to each of the above-determined plurality of inspection terminals includes payment history data that is duplicated at a preset ratio according to preset conditions.

6. In paragraph 1, The step of generating the above classification data is: Including a step of determining whether classification data matching the payment history data can be generated using the first classification model, If the above classification data cannot be generated, A step of storing the payment history data for which the above classification data could not be generated in a standby database; A step of modifying the classification system so as to generate classification data for the payment history data; and A method comprising the step of generating classification data matching the payment history data based on a modified classification system.

7. In paragraph 1, The step of generating the above classification data is: A step of obtaining reference data that provides additional information about the payment history data by using the payment history data; A step of extracting keywords related to the payment history data from the above reference data; and A method comprising a step of generating classification data that matches the payment history data to the classification system based on the similarity between the keyword and the keyword included in the classification system.

8. In paragraph 7, The above reference data is, May include at least one of business information, location information, image information, and product / service information; The step of extracting keywords related to the payment history data from the above reference data is as follows: A step of obtaining image information related to the above payment history data; A step of obtaining at least one keyword describing an object included in the image information; and A method comprising the step of extracting keywords related to the payment history data from at least one keyword describing the object.

9. In paragraph 1, The step of determining the above classification data is: A step of comparing whether the classification data generated from each of the plurality of inspection terminals matches; A step of providing a notification when an inconsistency in the above classification data is found; and A method comprising a step of confirming the classification data when the classification data matches.

10. In paragraph 9, If any inconsistency is found in the above classification data, A step of providing payment history data, classification data and reference data in which inconsistencies are found to the management terminal; A step of receiving correction classification data matching the payment history data in which the above inconsistency is found from the above management terminal; and A method further comprising a step of confirming the classification data based on the modified classification data.

11. In paragraph 1, The method of classifying the above payment history data is as follows: A step of training the first classification model using the payment history data and the classification data; and A method further comprising the step of generating a lightweight second classification model using the learned first classification model.

Images