Method and device for providing customer-tailored discount services using a kiosk terminal
The method uses AI to analyze customer behavior and business data through a kiosk terminal, optimizing discount policies for individual customers, enhancing operational efficiency and profitability by dynamically adjusting discount cycles and rates.
Patent Information
- Application Number
- KR1020250170374
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-11-06
- Estimated Expiration
- 2045-11-12
AI Technical Summary
Conventional discount policies fail to account for individual customer behavioral characteristics, leading to inefficiencies in operational efficiency and reduced margins due to discrepancies between discount timing and customer visit cycles, and lack of personalized strategies.
A method using a kiosk terminal and artificial intelligence to analyze customer behavior data and business operation data, dynamically calculating optimized discount cycles and rates through neural networks to provide customer-tailored discounts.
The method precisely predicts customer purchasing patterns and response characteristics, dynamically adjusting discounts to reflect store conditions, maximizing customer satisfaction and business profitability by optimizing discount cycles, rates, and preventing unnecessary discount abuse.
Smart Images

Figure 112025126393836-PAT00003_ABST
Abstract
Description
Technology Field
[0001] Embodiments of the present disclosure relate to a method and apparatus for providing a customer-tailored discount service using a kiosk terminal. Background Technology
[0002] Retail businesses and franchise stores operate various discount policies and coupon systems to increase sales and encourage repeat visits. However, conventional discount policies typically involved applying a uniform discount rate to specific periods, days of the week, time slots, or the entire customer base. Since this approach fails to reflect individual customer behavioral characteristics—such as purchasing propensity, visit frequency, and price sensitivity—it presented a problem in that the discounts did not effectively reach the customers who actually needed the benefits.
[0003] In particular, as ordering via kiosk terminals in offline stores has become commonplace, large amounts of behavioral data—such as operation logs, screen dwell time, and responses to discount offers—are being collected during the process of customers directly selecting menus and making payments. However, conventional systems have failed to fully utilize the capability to analyze this data and generate personalized discount strategies. Most systems remained at the level of simply issuing coupons based on fixed tiers derived from payment amounts or purchase frequency, or applying the same events to all customers indiscriminately.
[0004] Furthermore, existing discount management servers planned short-term promotions solely to compensate for declining sales of specific products, excess inventory, or sales drops during specific time periods; consequently, there were limited technologies capable of dynamically calculating optimized discount cycles and rates by simultaneously considering customer behavior data and business operation data. As a result, inefficiencies occurred in terms of operational efficiency, such as discrepancies between discount timing and actual customer visit cycles, or reduced margins due to the application of excessive discount rates.
[0005] Furthermore, in a franchise environment where multiple locations operate within the same brand, there was a limitation in that optimization by region and time of day could not be achieved because the central server uniformly distributed the same discount conditions, despite differences in operational status such as customer characteristics, congestion, inventory levels, and event availability across stores.
[0006] Accordingly, a method and device for providing customer-tailored discount services using a kiosk terminal are required. The problem to be solved
[0007] Embodiments of the present disclosure may provide a method and apparatus for providing a customer-tailored discount service using a kiosk terminal.
[0008] The technical problems to be solved in the embodiments are not limited to those mentioned above, and other unmentioned technical problems may be considered by those skilled in the art from the various embodiments described below. means of solving the problem
[0009] A method for a server to provide a customer-tailored discount service through a kiosk terminal using artificial intelligence according to one embodiment comprises: receiving a discount proposal request including customer identification information from the kiosk terminal; obtaining customer history data corresponding to the customer identification information; the customer history data including purchase history data and kiosk operation logs; generating a behavioral pattern vector including customer behavior stability, price sensitivity, and search propensity through a customer behavior analysis model using a first neural network based on the customer history data; generating a business situation vector including a discount cycle correction coefficient, a discount rate correction coefficient, and a maximum allowable discount rate correction coefficient through a situation optimization model using a second neural network based on pre-set business operation information; determining discount information including a discount cycle, a discount rate, and a maximum allowable discount rate through a discount combination decision model using a third neural network based on the behavioral pattern vector, the business situation vector, and pre-set reference parameters; wherein the pre-set reference parameters include a reference discount cycle, a reference discount rate, and a basic maximum allowable discount rate; transmitting a discount proposal message including the discount information to the kiosk terminal; and receiving the discount proposal from the kiosk terminal. It may include a step of receiving a response message to a message. For example, the parameters of the first neural network, the second neural network, and the third neural network, respectively, may be updated by utilizing the response message as training data. Effects of the invention
[0010] According to the embodiments, the present disclosure can quantitatively analyze a customer's actual purchasing patterns and response characteristics by calculating the customer's behavioral stability, price sensitivity, and search propensity through an artificial intelligence model based on customer identification information, purchase history data, and operation log data received from a kiosk terminal. Through this, the acceptability of discount offers and the tendency for revisiting can be precisely predicted for each customer, thereby providing a customized discount policy suitable for the behavioral characteristics of individual customers.
[0011] According to the embodiments, the present disclosure can calculate time-based discount cycle correction factors, discount rate correction factors, and maximum allowable discount rate correction factors for a business using an artificial intelligence model based on business operation information. Through this, since the discount policy is dynamically adjusted to reflect the operating status of each store and changes in the business environment by time of day, unnecessary abuse of discounts can be prevented and the balance between sales and margins can be efficiently maintained.
[0012] According to the embodiments, the present disclosure can automatically determine the optimal discount cycle, discount rate, and maximum allowable discount rate to simultaneously maximize customer satisfaction and business profitability by generating a plurality of discount combination candidates using an artificial intelligence model and predicting the expected reward value for each candidate. Accordingly, discount offers respond in real time to both customer behavior and store conditions, and efficiency is significantly improved compared to existing uniform discount policies.
[0013] The effects obtainable from the embodiments are not limited to those mentioned above, and other unmentioned effects can be clearly derived and understood by a person skilled in the art based on the detailed description below. Brief explanation of the drawing
[0014] The accompanying drawings, included as part of the detailed description to aid in understanding the embodiments, provide various embodiments and explain the technical features of the various embodiments together with the detailed description. FIG. 1 is a diagram showing the configuration of an electronic device according to one embodiment. FIG. 2 is a diagram showing the configuration of a program according to one embodiment. FIG. 3 shows a flowchart of a method for a server to provide a customer-tailored discount service to a kiosk terminal according to one embodiment. FIG. 4 is a flowchart illustrating the operation of a customer behavior analysis model using a first neural network according to one embodiment. FIG. 5 is a flowchart illustrating the operation of a situation optimization model using a second neural network according to one embodiment. FIG. 6 is a flowchart illustrating the operation of a discount combination decision model using a third neural network according to one embodiment. FIG. 7 is a block showing the configuration of a server according to one embodiment. Specific details for implementing the invention
[0015] The following embodiments are combinations of the components and features of the embodiments in a predetermined form. Each component or feature may be considered optional unless otherwise explicitly stated. Each component or feature may be implemented in a form not combined with other components or features. Additionally, various embodiments may be constructed by combining some components and / or features. The order of operations described in various embodiments may be changed. Some components or features of one embodiment may be included in another embodiment, or may be replaced with corresponding components or features of another embodiment.
[0016] In the description of the drawings, procedures or steps that could obscure the essence of the various embodiments were not described, nor were procedures or steps that can be understood by a person of ordinary knowledge in the relevant technical field described.
[0017] Throughout the specification, when a part is described as "comprising" or "including" a component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part," "...unit," and "module" as used in the specification refer to a unit that performs at least one function or operation, and this may be implemented in hardware, software, or a combination of hardware and software. Additionally, "one (a or an)," "one," "the," and similar related terms may be used in the context describing various embodiments (particularly in the context of the following claims) in both singular and plural forms, unless otherwise indicated in the specification or clearly contradicted by the context.
[0018] Hereinafter, embodiments according to various examples will be described in detail with reference to the accompanying drawings. The detailed description disclosed below, together with the accompanying drawings, is intended to describe exemplary embodiments of various examples and is not intended to represent the only embodiment.
[0019] In addition, specific terms used in various embodiments are provided to aid in understanding the various embodiments, and the use of such specific terms may be modified in other forms within the scope of not departing from the technical concept of the various embodiments.
[0020] FIG. 1 is a diagram showing the configuration of an electronic device according to one embodiment.
[0021] FIG. 1 is a block diagram of an electronic device (101) in a network environment (100) according to various embodiments. Referring to FIG. 1, in the network environment (100), the electronic device (101) may communicate with an electronic device (102) through a first network (198) (e.g., a short-range wireless communication network) or may communicate with at least one of an electronic device (104) or a server (108) through a second network (199) (e.g., a long-range wireless communication network). According to one embodiment, the electronic device (101) may communicate with the electronic device (104) through a server (108). According to one embodiment, the electronic device (101) may include a processor (120), memory (130), input module (150), sound output module (155), display module (160), audio module (170), sensor module (176), interface (177), connection terminal (178), haptic module (179), camera module (180), power management module (188), battery (189), communication module (190), subscriber identification module (196), or antenna module (197). In some embodiments, at least one of these components (e.g., connection terminal (178)) may be omitted from the electronic device (101), or one or more other components may be added. In some embodiments, some of these components (e.g., sensor module (176), camera module (180), or antenna module (197)) may be integrated into a single component (e.g., display module (160)). The electronic device (101) may be referred to as a client, terminal, or peer.
[0022] The processor (120) can control at least one other component (e.g., hardware or software component) of the electronic device (101) connected to the processor (120) by executing software (e.g., program (140)), for example, and can perform various data processing or operations. According to one embodiment, as at least part of the data processing or operations, the processor (120) can store commands or data received from other components (e.g., sensor module (176) or communication module (190)) in volatile memory (132), process the commands or data stored in volatile memory (132), and store the resulting data in non-volatile memory (134). According to one embodiment, the processor (120) may include a main processor (121) (e.g., central processing unit or application processor) or an auxiliary processor (123) that can operate independently or together with it (e.g., graphics processing unit, neural processing unit (NPU), image signal processor, sensor hub processor, or communication processor). For example, if the electronic device (101) includes a main processor (121) and an auxiliary processor (123), the auxiliary processor (123) may be configured to use lower power than the main processor (121) or to be specialized for a designated function. The auxiliary processor (123) may be implemented separately from the main processor (121) or as part thereof.
[0023] The auxiliary processor (123) may control at least some of the functions or states associated with at least one component of the electronic device (101) (e.g., display module (160), sensor module (176), or communication module (190)) on behalf of the main processor (121) while the main processor (121) is in an inactive (e.g., sleep) state, or together with the main processor (121) while the main processor (121) is in an active (e.g., application execution) state. According to one embodiment, the auxiliary processor (123) (e.g., image signal processor or communication processor) may be implemented as part of another functionally related component (e.g., camera module (180) or communication module (190)). According to one embodiment, the auxiliary processor (123) (e.g., neural network processing unit) may include a hardware structure specialized for processing an artificial intelligence model.
[0024] An artificial intelligence model can be generated through machine learning. Such learning may be performed, for example, on the electronic device (101) itself where the artificial intelligence model is executed, or through a separate server (e.g., server (108)). The learning algorithm may include, for example, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but is not limited to the examples described above. The artificial intelligence model may include a plurality of artificial neural network layers. The artificial neural network may be a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), deep Q-networks, or a combination of two or more of the above, but is not limited to the examples described above. Artificial intelligence models may include software structures, either additionally or as a substitute, in addition to hardware structures.
[0025] The memory (130) can store various data used by at least one component of the electronic device (101) (e.g., processor (120) or sensor module (176)). The data may include, for example, input data or output data for software (e.g., program (140)) and related commands. The memory (130) may include volatile memory (132) or non-volatile memory (134).
[0026] The program (140) may be stored as software in memory (130) and may include, for example, an operating system (142), middleware (144), or an application (146).
[0027] The input module (150) can receive commands or data to be used for a component of the electronic device (101) (e.g., processor (120)) from outside the electronic device (101) (e.g., user). The input module (150) may include, for example, a microphone, a mouse, a keyboard, a key (e.g., a button), or a digital pen (e.g., a stylus pen).
[0028] The sound output module (155) can output a sound signal to the outside of the electronic device (101). The sound output module (155) may include, for example, a speaker or a receiver. The speaker may be used for general purposes, such as multimedia playback or recording playback. The receiver may be used to receive incoming calls. According to one embodiment, the receiver may be implemented separately from the speaker or as part thereof.
[0029] The display module (160) can visually provide information to an external (e.g., user) of the electronic device (101). The display module (160) may include, for example, a display, a holographic device, or a projector and a control circuit for controlling said device. According to one embodiment, the display module (160) may include a touch sensor configured to detect a touch, or a pressure sensor configured to measure the intensity of the force generated by said touch.
[0030] The audio module (170) can convert sound into an electrical signal or, conversely, convert an electrical signal into sound. According to one embodiment, the audio module (170) can acquire sound through the input module (150) or output sound through the sound output module (155) or an external electronic device (e.g., electronic device (102)) (e.g., speaker or headphones) connected directly or wirelessly to the electronic device (101).
[0031] The sensor module (176) can detect the operating state of the electronic device (101) (e.g., power or temperature) or the external environmental state (e.g., user state) and generate an electrical signal or data value corresponding to the detected state. According to one embodiment, the sensor module (176) may include, for example, a gesture sensor, a gyroscope sensor, a barometric pressure sensor, a magnetic sensor, an accelerometer sensor, a grip sensor, a proximity sensor, a color sensor, an IR (infrared) sensor, a biosensing device, a temperature sensor, a humidity sensor, or an illuminance sensor.
[0032] The interface (177) may support one or more specified protocols that can be used for the electronic device (101) to be connected directly or wirelessly to an external electronic device (e.g., electronic device (102)). According to one embodiment, the interface (177) may include, for example, a high definition multimedia interface (HDMI), a universal serial bus (USB) interface, an SD card interface, or an audio interface.
[0033] The connection terminal (178) may include a connector through which the electronic device (101) can be physically connected to an external electronic device (e.g., electronic device (102)). According to one embodiment, the connection terminal (178) may include, for example, an HDMI connector, a USB connector, an SD card connector, or an audio connector (e.g., a headphone connector).
[0034] The haptic module (179) can convert an electrical signal into a mechanical stimulus (e.g., vibration or movement) or an electrical stimulus that the user can perceive through tactile or kinesthetic senses. According to one embodiment, the haptic module (179) may include, for example, a motor, a piezoelectric element, or an electric stimulation device.
[0035] The camera module (180) can capture still images and images. According to one embodiment, the camera module (180) may include one or more lenses, image sensors, image signal processors, or flashes.
[0036] The power management module (188) can manage the power supplied to the electronic device (101). According to one embodiment, the power management module (188) can be implemented, for example, as at least part of a power management integrated circuit (PMIC).
[0037] The battery (189) can supply power to at least one component of the electronic device (101). According to one embodiment, the battery (189) may include, for example, a non-rechargeable primary battery, a rechargeable secondary battery, or a fuel cell.
[0038] The communication module (190) can support the establishment of a direct (e.g., wired) communication channel or a wireless communication channel between an electronic device (101) and an external electronic device (e.g., electronic device (102), electronic device (104), or server (108)), and the performance of communication through the established communication channel. The communication module (190) may include one or more communication processors that operate independently of the processor (120) (e.g., application processor) and support direct (e.g., wired) communication or wireless communication. According to one embodiment, the communication module (190) may include a wireless communication module (192) (e.g., cellular communication module, short-range wireless communication module, or GNSS (global navigation satellite system) communication module) or a wired communication module (194) (e.g., LAN (local area network) communication module, or power line communication module). The corresponding communication module among these communication modules can communicate with an external electronic device (104) through a first network (198) (e.g., a short-range communication network such as Bluetooth, WiFi (wireless fidelity) direct, or IrDA (infrared data association)) or a second network (199) (e.g., a legacy cellular network, a 5G network, a next-generation communication network, the Internet, or a computer network (e.g., a LAN or WAN)). These various types of communication modules may be integrated into a single component (e.g., a single chip) or implemented as multiple separate components (e.g., multiple chips). The wireless communication module (192) can identify or authenticate the electronic device (101) within a communication network such as the first network (198) or the second network (199) using subscriber information (e.g., International Mobile Subscriber Identifier (IMSI)) stored in the subscriber identification module (196).
[0039] The wireless communication module (192) can support 5G networks and next-generation communication technologies following 4G networks, for example, new radio access technology. NR access technology can support high-speed transmission of high-capacity data (enhanced mobile broadband (eMBB)), minimization of terminal power and connection of multiple terminals (massive machine type communications (mMTC)), or high reliability and low latency (ultra-reliable and low-latency communications (URLLC)). The wireless communication module (192) can support a high-frequency band (e.g., mmWave band) to achieve a high data transmission rate, for example. The wireless communication module (192) can support various technologies for securing performance in the high-frequency band, such as beamforming, massive MIMO (multiple-input and multiple-output), full-dimensional MIMO (FD-MIMO), array antenna, analog beam-forming, or large-scale antenna. The wireless communication module (192) can support various requirements specified in the electronic device (101), external electronic device (e.g., electronic device (104)), or network system (e.g., second network (199)). According to one embodiment, the wireless communication module (192) can support a Peak data rate (e.g., 20 Gbps or more) for realizing eMBB, loss coverage (e.g., 164 dB or less) for realizing mMTC, or U-plane latency (e.g., downlink (DL) and uplink (UL) each 0.5 ms or less, or round trip 1 ms or less) for realizing URLLC.
[0040] An antenna module (197) can transmit a signal or power to or from an external source (e.g., an external electronic device). According to one embodiment, the antenna module (197) may include an antenna comprising a radiator made of a conductor or a conductive pattern formed on a substrate (e.g., a PCB). According to one embodiment, the antenna module (197) may include a plurality of antennas (e.g., an array antenna). In this case, at least one antenna suitable for a communication method used in a communication network, such as a first network (198) or a second network (199), may be selected from the plurality of antennas, for example, by a communication module (190). A signal or power may be transmitted or received between the communication module (190) and an external electronic device through the selected at least one antenna. According to some embodiments, in addition to the radiator, other components (e.g., a radio frequency integrated circuit (RFIC)) may be additionally formed as part of the antenna module (197).
[0041] According to various embodiments, the antenna module (197) may form a mmWave antenna module. According to one embodiment, the mmWave antenna module may include a printed circuit board, an RFIC disposed on or adjacent to a first surface (e.g., bottom surface) of the printed circuit board and capable of supporting a specified high frequency band (e.g., mmWave band), and a plurality of antennas (e.g., array antennas) disposed on or adjacent to a second surface (e.g., top surface or side surface) of the printed circuit board and capable of transmitting or receiving a signal of the specified high frequency band.
[0042] At least some of the above components can be connected to each other via a communication method between peripheral devices (e.g., bus, GPIO (general purpose input and output), SPI (serial peripheral interface), or MIPI (mobile industry processor interface)) and exchange signals (e.g., commands or data) with each other.
[0043] According to one embodiment, commands or data may be transmitted or received between the electronic device (101) and an external electronic device (104) through a server (108) connected to a second network (199). Each of the external electronic devices (102, or 104) may be the same or different type of device as the electronic device (101). According to one embodiment, all or part of the operations performed on the electronic device (101) may be performed on one or more of the external electronic devices (102, 104, or 108). For example, if the electronic device (101) needs to perform a function or service automatically or in response to a request from a user or another device, the electronic device (101) may request one or more external electronic devices to perform at least part of the function or service instead of performing the function or service itself or additionally. One or more external electronic devices that receive the above request may execute at least part of the requested function or service, or additional function or service related to the request, and transmit the result of the execution to the electronic device (101). The electronic device (101) may provide the result as is or additionally processed as at least part of the response to the request. For this purpose, for example, cloud computing, distributed computing, mobile edge computing (MEC), or client-server computing technology may be used. The electronic device (101) may provide ultra-low latency services using, for example, distributed computing or mobile edge computing. In another embodiment, the external electronic device (104) may include an Internet of Things (IoT) device. The server (108) may be an intelligent server using machine learning and / or neural networks. According to one embodiment, the external electronic device (104) or the server (108) may be included within a second network (199).The electronic device (101) can be applied to intelligent services (e.g., smart home, smart city, smart car, or healthcare) based on 5G communication technology and IoT-related technology.
[0044] The server (108) is connected to an electronic device (101) and can provide services to the connected electronic device (101). Additionally, the server (108) may proceed with the membership registration process, store and manage various information of users who have registered as members accordingly, and provide various purchase and payment functions related to the service. Furthermore, the server (108) may share execution data of service applications running on each of multiple electronic devices (101) in real time so that services can be shared among users. In terms of hardware, this server (108) may have the same configuration as a conventional web server or service server. However, in terms of software, it may include program modules that perform various functions and are implemented through any language such as C, C++, Java, Python, Golang, Kotlin, etc. Additionally, the server (108) generally refers to a computer system that is connected to an unspecified number of clients and / or other servers through an open computer network such as the Internet, receives requests for task execution from clients or other servers, and derives and provides the results of the tasks, as well as computer software (server program) installed for this purpose. Furthermore, the server (108) should be understood as a broad concept that includes, in addition to the aforementioned server program, a series of application programs running on the server (108) and, in some cases, various databases (DB: Database, hereinafter referred to as "DB") built internally or externally. Accordingly, the server (108) classifies membership registration information and various information and data, stores them in the DB, and manages them; such DB can be implemented internally or externally of the server (108).Additionally, the server (108) can be implemented using various server programs provided for general server hardware according to operating systems such as Windows, Linux, UNIX, and Macintosh. Representative examples include IIS (Internet Information Server) used in Windows environments and CERN, NCSA, APPACH, and TOMCAT used in UNIX environments to implement web services. Additionally, the server (108) may be linked with an authentication system and a payment system for user authentication of the service or purchase payment related to the service.
[0045] The first network (198) and the second network (199) refer to a connection structure capable of exchanging information between each node, such as terminals and servers, or a network connecting a server (108) and electronic devices (101, 104). The first network (198) and the second network (199) include, but are not limited to, the Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), 3G, 4G, LTE, 5G, Wi-Fi, etc. The first network (198) and the second network (199) may be closed types such as LAN, WAN, etc., but it is preferable that they be open types such as the Internet. The Internet refers to a global open computer first network (198) and second network (199) structure that provides protocols such as TCP / IP protocol, TCP, and UDP (user datagram protocol), and various services existing in the upper layer, namely HTTP (HyperText Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), and NIS (Network Information Service).
[0046] A database may have a general data structure implemented in the storage space (hard disk or memory) of a computer system using a database management program (DBMS). A database may have a data storage form that allows for the free retrieval (extraction), deletion, editing, and addition of data. A database may be implemented to suit the purpose of an embodiment of the present disclosure using a relational database management system (RDBMS) such as Oracle, Informix, Sybase, and DB2, an object-oriented database management system (OODBMS) such as Gemston, Orion, and O2, and an XML native database such as Excelon, Tamino, and Sekaiju, and may have appropriate fields or elements to achieve its functions.
[0047] FIG. 2 is a diagram showing the configuration of a program according to one embodiment.
[0048] FIG. 2 is a block diagram (200) illustrating a program (140) according to various embodiments. According to one embodiment, the program (140) may include an operating system (142), middleware (144), or an application (146) executable on the operating system (142) for controlling one or more resources of an electronic device (101). The operating system (142) may include, for example, Android™, iOS™, Windows™, Symbian™, Tizen™, or Bada™. At least some of the programs (140) may be preloaded into the electronic device (101) at manufacturing time, for example, or downloaded or updated from an external electronic device (e.g., electronic device (102 or 104), or server (108)) when used by a user. All or part of the program (140) may include a neural network.
[0049] The operating system (142) can control the management (e.g., allocation or reclamation) of one or more system resources (e.g., processes, memory, or power) of the electronic device (101). The operating system (142) may additionally or substantially include one or more driver programs for driving other hardware devices of the electronic device (101), e.g., an input module (150), an audio output module (155), a display module (160), an audio module (170), a sensor module (176), an interface (177), a haptic module (179), a camera module (180), a power management module (188), a battery (189), a communication module (190), a subscriber identification module (196), or an antenna module (197).
[0050] Middleware (144) may provide various functions to an application (146) so that functions or information provided from one or more resources of an electronic device (101) can be used by the application (146). Middleware (144) may include, for example, an application manager (201), a window manager (203), a multimedia manager (205), a resource manager (207), a power manager (209), a database manager (211), a package manager (213), a connectivity manager (215), a notification manager (217), a location manager (219), a graphics manager (221), a security manager (223), a call manager (225), or a voice recognition manager (227).
[0051] The application manager (201) can, for example, manage the life cycle of the application (146). The window manager (203) can, for example, manage one or more GUI resources used on the screen. The multimedia manager (205) can, for example, identify one or more formats required for the playback of media files and perform encoding or decoding of the corresponding media files among the media files using a codec that matches the selected corresponding format. The resource manager (207) can, for example, manage the source code of the application (146) or the memory space of the memory (130). The power manager (209) can, for example, manage the capacity, temperature, or power of the battery (189) and, using the relevant information, determine or provide relevant information required for the operation of the electronic device (101). According to one embodiment, the power manager (209) can interact with the BIOS (basic input / output system) (not shown) of the electronic device (101).
[0052] The database manager (211) can, for example, create, search, or modify a database to be used by the application (146). The package manager (213) can, for example, manage the installation or update of the application distributed in the form of a package file. The connectivity manager (215) can, for example, manage a wireless or direct connection between the electronic device (101) and an external electronic device. The notification manager (217) can, for example, provide a function to notify the user of the occurrence of a specified event (e.g., an incoming call, a message, or an alarm). The location manager (219) can, for example, manage location information of the electronic device (101). The graphics manager (221) can, for example, manage one or more graphic effects or related user interfaces to be provided to the user.
[0053] The security manager (223) may, for example, provide system security or user authentication. The telephony manager (225) may, for example, manage voice call functions or video call functions provided by the electronic device (101). The voice recognition manager (227) may, for example, transmit user voice data to the server (108) and receive from the server (108) a command corresponding to a function to be performed on the electronic device (101) based on at least part of the voice data, or text data converted based on at least part of the voice data. According to one embodiment, the middleware (144) may dynamically delete some existing components or add new components. According to one embodiment, at least part of the middleware (144) may be included as part of the operating system (142) or implemented as separate software different from the operating system (142).
[0054] The application (146) may include, for example, a home (251), a dialer (253), an SMS / MMS (255), an IM (instant message) (257), a browser (259), a camera (261), an alarm (263), a contact (265), a voice recognition (267), an email (269), a calendar (271), a media player (273), an album (275), a watch (277), a health (279) (e.g., measuring biometric information such as exercise volume or blood sugar), or an environmental information (281) (e.g., measuring atmospheric pressure, humidity, or temperature information). According to one embodiment, the application (146) may further include an information exchange application (not shown) capable of supporting information exchange between the electronic device (101) and an external electronic device. The information exchange application may include, for example, a notification relay application configured to transmit information (e.g., a call, a message, or an alarm) designated to an external electronic device, or a device management application configured to manage the external electronic device. The notification relay application may transmit notification information corresponding to a designated event (e.g., receiving mail) generated in another application of the electronic device (101) (e.g., an email application (269)) to the external electronic device. Additionally or alternatively, the notification relay application may receive notification information from the external electronic device and provide it to the user of the electronic device (101).
[0055] A device management application can control the power (e.g., turn-on or turn-off) or function (e.g., brightness, resolution, or focus) of an external electronic device or a part of its components (e.g., a display module or camera module of the external electronic device) that communicates with the electronic device (101). The device management application can additionally or substantially support the installation, deletion, or updating of applications running on the external electronic device.
[0056] Throughout this specification, the terms neural network, neural network, and network function may be used interchangeably. A neural network may consist of a set of interconnected computational units, which may generally be referred to as “nodes.” These “nodes” may also be referred to as “neurons.” A neural network is composed of at least two nodes. The nodes (or neurons) constituting neural networks may be interconnected by one or more “links.”
[0057] In a neural network, two or more nodes connected via links can form a relative relationship between an input node and an output node. The concepts of input and output nodes are relative; any node in an output node relationship with respect to one node may be in an input node relationship with respect to another node, and vice versa. As previously mentioned, the input node versus output node relationship can be generated based on links. One or more output nodes may be connected to a single input node via links, and vice versa.
[0058] In a relationship between an input node and an output node connected through a single link, the value of the output node can be determined based on data input to the input node. Here, the nodes interconnecting the input node and the output node may have weights. The weights may be variable and may be varied by a user or an algorithm to enable the neural network to perform the desired function. Here, the edges or links interconnecting the input node and the output node have weights that can be variably applied by a user or an algorithm to enable the neural network to perform the desired function. For example, if one or more input nodes are interconnected to a single output node by respective links, the output node may determine its value based on the values input to the input nodes connected to the output node and the weights set on the links corresponding to each input node.
[0059] As described above, a neural network is formed in which two or more nodes are interconnected through one or more links to create input-output node relationships within the network. The characteristics of a neural network can be determined by the number of nodes and links within the network, the relationships between the nodes and links, and the weight values assigned to each link. For example, if two neural networks exist with the same number of nodes and links but different weight values between the links, the two neural networks can be recognized as being different from each other.
[0060] FIG. 3 shows a flowchart of a method for a server to provide a customer-tailored discount service to a kiosk terminal according to one embodiment. The embodiment of FIG. 3 can be combined with various embodiments of the present disclosure.
[0061] Referring to FIG. 3, in step S310, the server may receive a discount offer request including customer identification information from a kiosk terminal.
[0062] The server may be a server that provides customer-tailored discount services to a kiosk terminal using artificial intelligence. Here, the customer-tailored discount service may be a service in which the server predicts the customer's purchase pattern through artificial intelligence based on customer-related information received from the kiosk terminal, generates multiple discount combination candidates using the purchase pattern, selects the most suitable discount combination for the customer, and provides discount information. For example, the server may store discount parameters (standard discount period, standard discount rate, basic maximum discount rate) set by the owner of the business establishment where the kiosk terminal is installed, and may calculate optimized discount conditions based on the customer's purchase characteristics and usage frequency according to the discount parameters. Here, the business establishment may be a place where at least one of product sales or service provision takes place, such as an office, a restaurant, or an academy. For example, the server may manage discount parameters for multiple business establishments and control discount policies by obtaining discount parameters corresponding to each business establishment. For example, the server may include the server (108) of FIG. 1.
[0063] A kiosk terminal may be a device that provides personalized discount services to customers. For example, a kiosk terminal may be an interface device installed within a business premises that customers operate directly to place orders. For instance, a kiosk terminal may communicate with a server via a pre-installed dedicated application or a web-based client. For instance, a kiosk terminal may receive customer identification information (phone number, ID, membership code, QR code, etc.) via a screen and transmit a discount offer request containing the entered identification information to the server. For instance, the kiosk terminal may display a "Member Discount Guide" or "View Personalized Benefits" menu on the initial screen, and if a customer selects the menu, it may guide them to enter identification information through an input field (e.g., entering a phone number or scanning a QR code). In this case, if a QR code is scanned or a phone number is entered, the corresponding information can be converted into customer identification information and transmitted to the server. Additionally, for instance, the kiosk terminal may include additional identification information, such as the customer's login status, recent usage history, or participation in specific events, in the discount offer request and transmit it to the server. For example, a kiosk terminal may include the electronic device (101) of FIG. 1.
[0064] A discount offer request may be a message sent by a kiosk terminal to a server to request a discount offer message for a customer. For example, the discount offer request may include customer identification information, a business identifier, and the time of the request.
[0065] In step S320, the server can obtain customer history data corresponding to customer identification information.
[0066] Customer history data may include purchase history data and kiosk operation logs. For example, the server can pre-store customer history data for each customer at multiple business locations. For instance, the server can manage customer history data by pre-establishing a customer history database for each business location using customer identification information as an index, and by periodically synchronizing transaction logs stored on each business location's transaction server or in a central cloud storage.
[0067] For example, the server maps customer identification information received from a kiosk terminal to an internal customer key (CustKey) and uses the internal customer key to retrieve the corresponding customer's purchase history data and kiosk operation logs from the database. The internal customer key is an internal unique identifier generated by hash mapping customer identification information, and it can serve as a key value to consistently identify the same customer across multiple business locations. The business location identifier is a unique code used to distinguish the business location to which each kiosk terminal belongs, and it can serve as an identifier for managing business-level policies and distinguishing transaction data. In this case, the server uses the business location identifier and the internal customer key as a combined key to filter and extract only the transactions corresponding to the current business location among the transactions generated by the same customer across multiple business locations.
[0068] Purchase history data may be data representing the details of purchases actually made by a customer at the relevant business establishment. For example, purchase history data may include at least one purchase date and time, the purchase amount for said at least one purchase date and time, a value regarding the transaction item group, a value regarding the payment method, and a value regarding whether a discount was applied.
[0069] Transaction group may be a category identification value used to classify individual transaction items transacted by a customer at a relevant business location. For example, transaction group may be distinguished based on the type, nature, or purpose of provision of the transaction items. For instance, in the case of a restaurant, it may be classified according to the type of menu (meals, beverages, desserts, etc.); in the case of an academy, according to the nature of the course (English conversation, advanced math, music lessons, etc.); and in the case of a retail store, according to the product category (food, household goods, clothing, etc.). For instance, transaction group may be expressed in the form of a string code (e.g., “CAT_FOOD”, “CAT_EDU”) or an integer identifier (ID) according to a predefined classification system or operational policy specific to the business location, and each transaction item may be mapped to one or more transaction group values. Values for payment methods may be categorical values indicating the type of payment method, such as cash, card, mobile pay, or points. Values related to whether a discount is applied may include a binary value indicating whether a discount is applied at each purchase, and values for the discount period, discount rate, and maximum allowable discount rate, respectively, when a discount is applied. For example, it may be expressed as 1 when a discount is applied and 0 when it is not applied.
[0070] For example, the server may organize normalized dataframes by transaction unit to standardize the structure of purchase history data. For instance, each transaction record may consist of a set of fields including {transaction number, purchase date and time, transaction item group code, payment type, discount rate, discount period, maximum discount rate, total amount}.
[0071] Additionally, for example, if there are missing items, the server may correct the missing values by interpolating using the average value or exponential moving average (EMA) based on the last 5 transaction data of the same customer.
[0072] Kiosk operation logs may include interface interaction data generated on a transaction-by-transaction basis while a customer proceeds with the ordering process through a kiosk terminal at the business premises. For example, kiosk operation logs may include screen-specific dwell time logs containing the display time and end time of each screen, and touch event stream logs containing the time, coordinates, target widget identifier, and input type of each touch event.
[0073] Screen-specific dwell time logs are data representing the patterns of time users spend on each menu screen per transaction session, and can be used to calculate customer dwell time and conversion frequency on a screen-by-screen basis. For example, a server can calculate screen-specific dwell time by determining the difference between the display start and end times of each screen within the same transaction session.
[0074] The touch event stream log is data that records the sequence of touch inputs performed by customers on on-screen buttons, sliders, product images, etc., in chronological order by transaction session, and can reflect operation patterns such as touch intervals, menu selection order, and cancellation frequency.
[0075] Each touch event may represent a unit input action that occurs during the process of a customer operating the screen of a kiosk terminal. The time of a touch event may be a value recorded in milliseconds at the time when the corresponding touch action occurred. For example, the time of a touch event may be stored in a UTC-based timestamp format, such as “2025-10-29T10:24:35.126”. The coordinates of a touch event are 2D coordinates indicating the location within the screen where the touch occurred, and may be normalized values in the form of (x, y) relative to the top-left corner of the screen. For example, if the screen resolution is 1080*1920, the coordinates of the touch event (540, 960) may represent the center point of the screen. The target widget identifier is a unique identifier of the interface component to which the corresponding touch event acted, and may be a fixed string code assigned to each UI component, such as a button, image, list item, payment icon, back arrow, scroll bar, or product card. For example, identifier values such as “BTN_PAY”, “IMG_MENU_01”, and “CARD_ITEM_A03” may be used. The input type is a value used to distinguish the physical input form of a touch event and can be set to any one of Tap, Long Press, Swipe, Drag, Double Tap, or Pinch.
[0076] In step S330, the server can generate a behavioral pattern vector including the customer's behavioral stability, price sensitivity, and search propensity through a customer behavior analysis model using a first neural network based on customer history data.
[0077] A customer behavior analysis model using a first neural network may be a neural network model that receives chronologically sorted purchase history data and kiosk operation logs as input, learns the continuous changes in customer behavior at each point in time, and converts them into an internal representation. For example, the customer behavior analysis model may utilize a first neural network that includes a time series encoder.
[0078] The behavioral pattern vector is a high-dimensional representation vector that quantifies customers' purchasing and manipulation behaviors in a time-series manner, and may be a representation value in a three-dimensional or multi-dimensional numerical space composed of axes of customers' behavioral stability, price sensitivity, and search propensity. For example, each dimension of the behavioral pattern vector reflects customers' psychological decision-making characteristics and manipulation habits in the form of internal variables, and each value may be a normalized value of a scalar indicator calculated from the regression head of the first neural network.
[0079] For example, the dimension corresponding to behavioral stability indicates how consistent a customer's purchasing behavior is over time, the dimension corresponding to price sensitivity is a numerical value representing the sensitivity of the purchase response to price changes, and the dimension corresponding to search propensity is a value representing the strength of the tendency to explore new transaction items or screens. For example, customer behavioral stability is an indicator calculated based on touch rhythm during the ordering process, response latency, and the repurchase rate for the same transaction item, and can represent the variability of a customer's purchasing intent. For example, customer price sensitivity is a value calculated by analyzing the reaction pattern regarding whether to purchase based on changes in the discount rate, and can be expressed, for instance, as the rate of change in purchase probability when the discount rate increases by 1%. For example, customer search propensity is an indicator calculated based on the number of screens explored within the kiosk terminal, the number of comparisons regarding transaction items, and the shopping cart modification rate, and can represent the tendency to explore new products.
[0080] In step S340, the server can generate a workplace situation vector including a discount period correction factor, a discount rate correction factor, and a maximum allowable discount rate correction factor through a situation optimization model using a second neural network based on pre-configured workplace operation information.
[0081] Preset business operation information may be a data set representing the business and inventory operation status of a business within a specific period. For example, the preset business operation information may include multiple items by time period during a preset period. In this case, the preset period may be set to a period between 3 months and 1 year.
[0082] For example, during the initial setup phase, the server can pre-configure the standard configuration of business site operation information by setting operational metric items (e.g., number of orders, customer density, inventory remaining rate, budget limit, event registration status, etc.) through a kiosk terminal and registering the measurement unit, collection cycle, and reference values (e.g., maximum allowable density, daily budget limit, inventory threshold) for each item in the form of a setting value table.
[0083] For example, business site operation information can be transmitted to the server based on pre-registered operation indicator items and collection cycles on the kiosk terminal.
[0084] For example, multiple operational metric items may include order count, customer density, inventory retention rate, budget exhaustion rate, event frequency, and order failure rate. The order count refers to the total quantity of orders received during a specific time period and can indicate the level of traffic within the business premises. Customer density may be a value calculated as the ratio of the number of customers present within the business premises at any given time to a standard number of seats or standard area. The inventory retention rate may be a value expressing the ratio of the inventory quantity of a specific transaction group to the initial inventory for each transaction group. The budget exhaustion rate may be a value representing the cumulative ratio of the discount amount used to date relative to the discount budget. Event frequency may be a value representing the number of events, such as promotions, seasonal events, or new launch events, that occurred during a specific time period. The order failure rate may be a value representing the percentage of total order attempts that ended in payment failure or cancellation. For example, the order failure rate may be the percentage of total order attempts where a customer abandoned the order, payment was interrupted due to stock shortages, or an order failed due to a network error.
[0085] A situational optimization model using a second neural network may be a time-series forecasting-based deep learning model that receives business operation information sorted in chronological order as input, learns the correlation structure and temporal variation patterns between each operation indicator item, and determines coefficients for dynamically adjusting discount policies.
[0086] A situational optimization model utilizing a second neural network can track the impact of the operational status of a previous time period on discount performance in the next time period through inner product operations to reflect the temporal continuity of business operations, and may be a non-linear regression neural network structure with multiple hidden representations to express non-linear interactions between items. For example, the situational optimization model can be trained to adjust in the direction of increasing the discount cycle to alleviate congestion in sections where the trend of fluctuations in the number of orders and customer density over a certain period shows a correlation higher than a preset threshold, and conversely, to adjust in the direction of increasing the discount rate to promote sales in sections where the order failure rate is higher than a preset first threshold ratio and the inventory remaining rate is higher than a preset second threshold ratio. Additionally, the situational optimization model can consider the ratio of event frequency and budget exhaustion rate together to allow for temporary adjustments to the upper limit of the discount rate during specific event time periods, and to mitigate the discount frequency by adjusting the discount cycle correction factor in the negative direction when the budget exhaustion rate exceeds a threshold.
[0087] The workplace condition vector is a multidimensional numerical vector of three or more dimensions produced as the output of a condition optimization model, which may be a vector that quantifies the current operational status of a workplace. For example, the workplace condition vector may include a discount cycle correction factor, a discount rate correction factor, and a maximum allowable discount rate correction factor, and each item may be a numerical value calculated by normalizing the relative variation of the current measurement value relative to a preset reference value. In other words, the workplace condition vector may not be a simple set of status data, but rather an intrinsic characteristic vector that represents the complex state of workplace load, inventory, and budget as a single integrated expression. The discount cycle correction factor may be a numerical value for adjusting the discount cycle by reflecting fluctuations in workplace congestion and order volume over time. The discount rate correction factor may be a numerical value for adjusting the discount rate based on operational indicator items such as inventory depletion rate, budget depletion rate, and order failure rate. The maximum allowable discount rate correction factor may be a numerical value for adjusting the maximum allowable discount rate, which is the policy-acceptable maximum discount limit, based on operational indicator items.
[0088] In step S350, the server can determine discount information for a customer, including the discount cycle, discount rate, and maximum allowable discount rate, through a discount combination decision model using a third neural network based on a behavior pattern vector, a business location situation vector, and preset reference parameters.
[0089] Discount information may include discount parameters to be proposed to customers. For example, discount parameters may include a discount cycle, a discount rate, and a maximum allowable discount rate. The discount cycle may be the period until the next discount application for the customer. The discount rate may be the applicable discount percentage. The maximum allowable discount rate may be a discount cap that is policy-acceptable. For example, if the discount cycle is 7 days, the discount rate is 1%, and the maximum allowable discount rate is 50%, a discount rate of 1% may be applied once a week to the customer until the maximum allowable discount rate reaches 50%.
[0090] Preset standard parameters may be reference values for discount parameters set by the manager of the business. For example, preset standard parameters may include a standard discount cycle, a standard discount rate, and a standard maximum allowable discount rate. The standard discount cycle may be a discount cycle that applies universally to all customers. The standard discount rate may be a discount rate that applies universally to all customers. The standard maximum allowable discount rate may be a maximum allowable discount rate that applies universally to all customers.
[0091] A discount combination decision model utilizing a third neural network may be a non-linear regression deep learning model designed to calculate the optimal combination between the discount cycle, discount rate, and maximum allowable discount rate based on behavior pattern vectors, business location situation vectors, and pre-set reference parameters. For example, based on a multi-layer perceptron (MLP) structure, the discount combination decision model can integrally learn the interrelationship between behavior pattern vectors representing customer behavior indicators given as input and business location situation vectors representing the integration of business operational indicators through inner product operations. Through this, the discount combination decision model can calculate the expected reward value for each candidate discount combination and determine the discount combination that maximizes the expected reward value as the optimal discount information. During the learning process of the discount combination decision model, actual sales growth rate, customer return rate, margin retention rate, and engagement rate are set as target variables, and by updating weights in a direction that minimizes the prediction error for these target variables, more sophisticated discount policies for each customer can be calculated over time.
[0092] In step S360, the server can send a discount offer message containing discount information to the kiosk terminal.
[0093] The discount offer message may be a message regarding a discount policy proposed by the server through the kiosk terminal. For example, the discount offer message may include application conditions, validity period, and a description of the benefit, along with numerical information corresponding to the discount cycle, discount rate, and maximum allowable discount rate. For example, the discount offer message may be displayed on the kiosk terminal screen in a form such as “3% immediate discount applied upon revisit within the next 7 days, up to a maximum limit of 50%.”
[0094] In step S370, the server can receive a response message for a discount offer message from the kiosk terminal.
[0095] The response message may be a message indicating a response to a discount offer message. For example, the response message may include a value for the response type and a response time. For example, the value for the response type may include a value indicating either Accept or Reject. In this case, if the administrator or customer does not perform input on the kiosk terminal screen within a preset time limit, the value for the response type may be determined as a value indicating Rejection, and the response time may be determined as a preset time limit. The response time may be the elapsed time from the time the discount offer message is displayed on the kiosk terminal until the administrator or customer completes the response input.
[0096] The parameters of the first, second, and third neural networks can be updated by utilizing response messages as training data. This update can be achieved by accumulating recent response data in mini-batches and then adjusting the parameters through backpropagation based on the loss function of each model. For example, the first neural network can be trained to minimize the behavioral stability prediction error, the second neural network to minimize the discount rate correction factor prediction error, and the third neural network to minimize the sales increment error after a discount proposal.
[0097] Additionally, for example, after receiving a response message, the server can convert the response message into training data and update the parameters of each neural network using an Incremental Online Update method. That is, after receiving the response message, the server immediately converts the corresponding response data into training data and can update some of the weights by calculating the loss function for the new data without resetting the parameters of the existing model.
[0098] For example, if the response type is acceptance, the server can reflect that the customer exhibited a consistent purchase intention pattern by upgrading the weights related to behavioral stability and downgrading the weights related to search propensity in the first neural network. Additionally, the server can learn that the current discount strength was appropriate by relaxing the sensitivity weights for the time zone corresponding to the response time in the second neural network and slightly decreasing the discount rate correction factor. Furthermore, the server can reflect that the currently proposed discount combination was suitable by strengthening the reward prediction parameters in the third neural network.
[0099] For example, if the response type is a rejection, the server can reflect the instability of the customer's purchase decision pattern by upgrading the weights related to the search propensity of the first neural network. Additionally, the server can learn to apply a shorter period and a higher discount rate in the next discount proposal by strengthening the sensitivity weights for the time zone corresponding to the response time in the second neural network, upgrading the discount rate correction factor, and downgrading the discount period correction factor. In this case, the values for each of the upgrading and downgrading can be updated through learning. Furthermore, the server can reflect in the policy update that the currently proposed discount combination was ineffective by decreasing the reward prediction parameter of the third neural network.
[0100] FIG. 4 is a flowchart illustrating the operation of a customer behavior analysis model using a first neural network according to one embodiment. The embodiment of FIG. 4 can be combined with various embodiments of the present disclosure.
[0101] Referring to FIG. 4, in step S410, the customer behavior analysis model can calculate a purchase cycle vector that includes the average purchase interval, the variability of the purchase cycle, the average purchase amount, the discounted purchase ratio, and the diversity of the transaction item group for each transaction session based on purchase history data.
[0102] For example, the customer behavior analysis model can calculate a purchase cycle vector for each transaction session through multivariate statistics on purchase history data up to immediately prior to each transaction session. In this process, for instance, the customer behavior analysis model can generate a purchase interval column by sorting the dates of each transaction in the purchase history data in chronological order and calculating the time difference between adjacent transactions. Subsequently, the customer behavior analysis model can calculate the average value of the purchase interval column prior to the transaction session as the average purchase interval (avg_interval) and the variance value of the purchase interval column prior to the transaction session as the purchase cycle variability (var_interval). For example, the customer behavior analysis model can calculate the average value of the purchase amounts prior to the transaction session as the average purchase amount (avg_amount). For example, the customer behavior analysis model can determine the ratio of transactions with a discount status indicated as '1' out of the total number of purchases prior to the transaction session as the discount application ratio (discount_ratio). For example, a customer behavior analysis model can calculate category diversity by normalizing the number of categories included in transactions prior to the transaction session by the total number of transactions prior to the transaction session. The average purchase interval, purchase cycle variability, average purchase amount, discounted purchase ratio, and category diversity calculated in this way can be combined into a 5-dimensional component and expressed as a purchase cycle vector for the transaction session.
[0103] In step S420, the customer behavior analysis model can calculate the response delay vector, operation rhythm vector, purchase intent vector, and discount screen focus vector based on the screen-specific dwell time log and touch event stream log.
[0104] For example, each vector is calculated per transaction session and can then be sorted into a time-ordered sequence according to the chronological order of the transaction session.
[0105] The response delay vector may be a vector that quantifies the time pattern from when the discount offer screen is displayed until the first touch event is received. For example, a customer behavior analysis model can calculate the response delay per transaction session by subtracting the timestamp at which the first touch event occurred from the timestamp at which the discount offer screen was activated. The response delay vector may include average delay (avg_delay), delay variability (var_delay), minimum delay (min_delay), maximum delay (max_delay), and normalized delay ratio (normalized_delay) per transaction session.
[0106] Average latency may be the average value of the response latency for the previous trading sessions. Latency variability may be the variance of the response latency for the previous trading sessions. Minimum latency may be the shortest response latency among the response latencies for the previous trading sessions. Maximum latency may be the longest response latency among the response latency for the previous trading sessions. The normalized latency ratio may be a value between 0 and 1 obtained by linearly normalizing the customer's average latency against the minimum and maximum response latencies of the entire customer group. Specifically, the normalized latency ratio can be calculated as the result of dividing (the customer's average latency - the entire customer group's minimum response latency) by (the entire customer group's maximum response latency - the entire customer group's minimum response latency). A normalized latency ratio closer to 0 may indicate that the customer's response speed is relatively fast compared to the entire customer group, while a ratio closer to 1 may indicate that the customer's response speed is relatively slow.
[0107] The operation rhythm vector may be a vector representing the temporal regularity and consistency between touch events. For example, the operation rhythm vector may include the mean interval (mean_interval), the standard deviation of the interval (std_interval), the skewness of the interval distribution (skew_interval), the touch density per unit time (touch_density), and the input idle ratio (idle_ratio). The mean interval may be the average value of the time intervals between consecutive touch events that occurred in the previous trading sessions. The standard deviation of the interval is a value representing the degree of variability of the time interval between touch events and can be calculated as the standard deviation of the time intervals between touch events for the previous trading sessions. The skewness of the interval distribution is a value representing the asymmetry of the distribution of time intervals between touch events and can be calculated as [(mean interval - median) / standard deviation of the interval]. Here, the median may be the median of the time intervals between touch events in the previous trading sessions. Touch density per unit time is a value representing the frequency of touch events relative to the total session time, and may be the value obtained by dividing the total number of touch events that occurred during all previous transaction sessions by the total session duration of those previous transaction sessions. The input gap ratio is the ratio of periods during the total session time in which no touch input occurs, and may be calculated by dividing the sum of the time periods without touch events during the previous transaction sessions by the total duration of those previous transaction sessions.
[0108] The purchase intent vector may be a vector representing the stability of decision-making through patterns of adding and removing transaction items within the shopping cart. For example, the purchase intent vector may include the add-to-remove ratio (add_remove_ratio), cart edit frequency (edit_frequency), number of times the same transaction item is re-added (readd_occurrence), stable item ratio (stable_item_ratio), and unstable item ratio (unstable_item_ratio). The add-to-remove ratio can be calculated by dividing the total number of transaction items removed across all previous transaction sessions by the total number of transaction items added. Cart edit frequency can be calculated as the sum of the number of add events and the number of remove events that occurred across all previous transaction sessions. The number of times the same item is re-added can be calculated as the cumulative number of events where the same transaction item was removed and then re-added. The stable item ratio can be calculated by dividing the total number of transaction items maintained until the checkout point across previous transaction sessions by the total number of transaction items added at least once during the session. The unstable item ratio can be calculated by dividing the total number of transaction items removed from all previous transaction sessions by the total number of transaction items contained at least once during the session.
[0109] The discount screen focus vector may be a vector reflecting the dwell time on the discount offer screen and the touch focus on the discount offer area. For example, the discount screen focus vector may include screen focus, area focus, focus continuity, and focus intensity. For the current transaction session, the customer behavior analysis model can calculate cumulative statistical values using screen-specific dwell time logs and touch event stream logs from previous transaction sessions, and based on this, calculate the discount screen focus vector for the current session. Screen focus can be calculated by dividing the total sum of dwell times on the discount offer screen in previous transaction sessions by the total sum of the total session dwell times. Area focus can be calculated by dividing the total sum of touch events corresponding to the discount offer area (discount button, coupon guide, etc.) in previous transaction sessions by the total sum of the total number of touch events. Focus continuity can be calculated by dividing the sum of the times of continuous dwell periods within the discount offer screen in previous transaction sessions by the total session dwell time. Concentration intensity can be calculated by dividing the number of touch events within the discount offer screen in the previous trading sessions by the time spent on the discount offer screen.
[0110] Additionally, according to one embodiment, if customer purchase history data and kiosk operation logs are missing or collected abnormally, the server can correct the missing data by calculating the average value for each behavioral vector type using the recent N customer history data (e.g., 50 people) within the same or similar business location. Behavioral vector types may include a purchase cycle vector, a response delay vector, an operation rhythm vector, a purchase intent vector, and a discount screen focus vector.
[0111] Specifically, the server can obtain a purchase cycle vector, a response delay vector, an operation rhythm vector, a purchase intent vector, and a discount screen concentration vector calculated from N customers who have the most recent transaction among a set of customers corresponding to the same store identifier or a set of customers designated with a store identifier similar to the store identifier.
[0112] Subsequently, the server can calculate the average value for each identical item of each vector to generate the average response delay vector, average operation rhythm vector, average purchase intention vector, and average discount screen focus vector, respectively.
[0113] The server can replace each generated average vector with a model input corresponding to the customer's missing behavior vector type.
[0114] Through this, even if individual customer logs are insufficient or outliers exist, a sequence of behavioral features can be reliably constructed by reflecting the general behavioral characteristics of customer groups within the same or similar workplaces.
[0115] In step S430, the customer behavior analysis model can generate a sequence of behavioral feature vectors with preserved time order by merging the purchase cycle vector, response delay vector, operation rhythm vector, purchase intent vector, and discount screen focus vector.
[0116] For example, a customer behavior analysis model can construct a session-unit continuous vector sequence by arranging the purchase cycle vector, response lag vector, operation rhythm vector, purchase intent vector, and discount screen focus vector calculated for each transaction session point in chronological order.
[0117] At this time, each transaction session is considered as a single point in time unit, and the {purchase cycle vector + response delay vector + operation rhythm vector + purchase intent vector + discount screen focus vector} corresponding to that point in time can be combined into a single integrated vector.
[0118] For example, a customer behavior analysis model can form a session-specific behavioral feature vector by sequentially concatenating five sub-vectors for each transaction session, and generate a sequence of behavioral feature vectors that maintain time-series continuity by arranging the formed multiple session-specific behavioral feature vectors in chronological order.
[0119] In step S440, the customer behavior analysis model can generate an integrated representation vector by learning the correlation between time points through a multilayer encoder based on a sequence of behavior feature vectors.
[0120] For example, a customer behavior analysis model can simultaneously learn dependencies between time points and long-term and short-term pattern changes by inputting an input sequence of behavioral feature vectors into a multi-layer encoder composed of multiple encoder layers.
[0121] Each encoder layer of a multilayer encoder can be composed of a self-attention block and a feed-forward block. Each encoder layer can progressively strengthen the interaction of time-series behavioral patterns by calculating the correlation for all time points of the input sequence and correcting the feature representation for each time point by reflecting the calculated correlation information.
[0122] First, the self-attention block can linearly transform each time point vector of the input action feature vector sequence into a query vector, a key vector, and a value vector. Subsequently, the self-attention block can calculate the correlation between time points through the inner product of the query vector and the key vector, and derive attention weights by applying a softmax function. Attention weights can represent the relative importance that the action features of each time point have on the action patterns of other time points. By performing a weighted sum by multiplying the corresponding attention weights by the value vectors, the self-attention block can produce a context vector that reflects the contextual information of surrounding time points for each time point.
[0123] Subsequently, the encoder layer can induce stable learning by performing residual connections between the generated context vector and the input action feature vector sequence, and by performing layer normalization.
[0124] Next, the feedforward block can enhance the abstracted feature representation for each time point by performing a fully connected operation including a non-linear activation function on the normalized time-point vectors.
[0125] The operations of these self-attention blocks and feedforward blocks are repeated in multiple layers, and as you move to higher layers, they can form representations that strengthen long-term dependency and global behavioral context.
[0126] The final output of a multilayer encoder may be a series of high-dimensional vectors that maintain chronological order, and a customer behavior analysis model can generate an integrated representation vector that summarizes the behavioral features of the entire sequence into a single vector by performing global average pooling or attention-based weighted sum operations on the vectors. This integrated representation vector may be a high-dimensional representation in which the temporal variations of response speed, manipulation patterns, purchase decision consistency, and visual attention appearing throughout the customer's past behavioral history are all compressed.
[0127] In step S450, the customer behavior analysis model can calculate behavioral stability, price sensitivity, and search propensity, respectively, through multiple regression heads based on an integrated representation vector.
[0128] Specifically, the customer behavior analysis model branches the integrated representation vector into three independent regression head inputs, which can then be input into a multilayer perceptron-based regression module corresponding to the characteristics of each indicator. Each regression head performs a continuous fully connected operation including a non-linear activation function on the input integrated representation vector, and the output layer can predict each behavior indicator as a single scalar value.
[0129] The behavioral stability regression head can be trained by emphasizing the weights of feature elements corresponding to the response lag vector and the operation rhythm vector within the integrated representation vector. For example, the behavioral stability regression head can regressively predict the reliability of behavioral stability over time, focusing on sub-features that represent the consistency of customer operation patterns, such as the variability of response lag time, the standard deviation of the interval between touches, and the input gap ratio. For example, the behavioral stability regression head can perform non-linear feature transformation by applying a non-linear activation function (ReLU (Rectified Linear Unit) or GELU (Gaussian Error Linear Unit)) in the first hidden layer to the input integrated representation vector, prevent overfitting through normalization (norm) and dropout operations in the second hidden layer, and then calculate behavioral stability in the final output layer. The ReLU function may be a non-linear activation function that outputs 0 when the input value is 0 or less, and outputs the input value itself when it is greater than 0. The GELU function can be a non-linear activation function that multiplies the input value by a Gaussian cumulative distribution function, gradually dampening the output as the input becomes smaller. Behavioral stability is normalized to a value between 0 and 1, and a value closer to 1 may indicate that the customer's response and manipulation behavior are consistent and exhibit a confident behavioral pattern.
[0130] The price sensitivity regression head can be trained based on the weights of feature elements related to the purchase cycle vector and the discount screen concentration vector within the integrated representation vector. For example, the price sensitivity regression head can regressively predict the degree to which customers respond to price fluctuations and discount information based on the average purchase interval of past transaction sessions, the purchase ratio to which discounts are applied, screen concentration, and area concentration. For instance, the price sensitivity regression head can input the input integrated representation vector into a regression module of a multilayer perceptron structure to iteratively perform linear transformations and non-linear activations (GELU or ReLU) in multiple hidden layers. The outputs of the hidden layers undergo batch normalization and dropout operations to minimize weight bias and ensure generalization performance. The final output layer can calculate price sensitivity by applying the sigmoid function as an activation function to normalize the output values into continuous values between 0 and 1. The sigmoid function ( ) can be a non-linear function that outputs a value closer to 1 as the input value increases and a value closer to 0 as the input value decreases. Price sensitivity quantitatively indicates how sensitively customers respond to discount rates, unit price differences, or changes in promotions; a value closer to 1 indicates a highly sensitive customer who reacts immediately to price changes, while a value closer to 0 indicates a low-sensitivity customer who makes purchasing decisions based on factors other than price.
[0131] The search propensity regression head can be trained based on the weights of feature elements related to the manipulation rhythm vector and the purchase intention vector within the integrated representation vector. For example, a customer behavior analysis model can regressively calculate a customer's search propensity by utilizing input elements that reflect exploratory behavior patterns, such as mean interval, standard deviation of intervals, skewness of the interval distribution, touch density per unit time, add-to-remove ratio, and cart edit frequency. For instance, the search propensity regression head receives the integrated representation vector as input and learns weights that emphasize search-related features by performing fully connected operations with non-linear activation (GELU or ReLU) applied in multiple hidden layers. The output of each hidden layer is stabilized through normalization (norm) and dropout processes, and a sigmoid function is applied in the final output layer to produce a search propensity with a range between 0 and 1. Search propensity is a numerical value representing the tendency of customers to explore multiple alternatives or compare prices, configurations, or discount conditions rather than immediately selecting a single item. A value closer to 1 indicates a customer who exhibits an exploratory and non-deterministic consumption pattern, while a value closer to 0 indicates a customer who exhibits deterministic and rapid decision-making.
[0132] Additionally, for example, to prevent overconfidence in prediction probabilities from historical training data, flat scaling-based correction coefficients may be applied to price sensitivity and search propensity. For instance, a flat scaling-based correction coefficient may be a coefficient that readjusts the probability values to match the distribution of actual observed historical data by passing the raw probability values calculated by the model through a correction function in the form of logistic regression. For instance, the actual distribution of each behavioral indicator (price sensitivity and search propensity) can be estimated using customer history data accumulated over the past three months or more, and the correction coefficient can be trained to minimize the difference between that distribution and the mean vector of the model's predicted values. In this case, the correction coefficient may be adjusted to expand the sensitivity of the central region (interval of 0.4 to 0.6) so that the prediction probabilities do not lean toward 0 or 1. In other words, flat scaling-based correction is a process of linearly remapping the probabilistic behavioral indicators output by the model to the empirical probabilities of the actual data; consequently, it can have the effect of normalizing the predicted values of price sensitivity and search propensity so that they are not excessively extreme and are consistent with the average behavioral tendencies of the actual customer group.
[0133] FIG. 5 is a flowchart illustrating the operation of a situation optimization model using a second neural network according to one embodiment. The embodiment of FIG. 5 can be combined with various embodiments of the present disclosure.
[0134] Referring to Fig. 5, in step S510, the situation optimization model can generate an operation sequence by arranging pre-set workplace operation information in daily and hourly units.
[0135] Specifically, the situation optimization model can generate an operation sequence in the form of a multivariate time series input structure by sorting pre-set workplace operation information in chronological order.
[0136] For example, a situational optimization model can generate time-series frames based on multiple operational metrics, such as order throughput, customer density, inventory retention rate, budget exhaustion rate, event frequency, and order failure rate, tailored to the measurement unit and collection cycle of each operational metric.
[0137] Additionally, for example, if data for some operational indicator items is missing within the same time period, i.e., the same time slot, the situation optimization model can correct the missing interval by applying the moving average of the previous time interval or the exponential moving average of the last M measurements. Furthermore, if abnormal outliers (e.g., values exceeding a reference threshold or spikes due to system errors) are detected, the corresponding interval may be replaced using a median correction method. The corrected data can be normalized for each operational indicator item and converted into an operational sequence with the same temporal resolution. For example, the last M measurements may be M measurements prior to the current point in time. For example, M can be preset as a constant value greater than or equal to 1.
[0138] In step S520, the situation optimization model can generate an operation pattern vector sequence by calculating the moving average value, coefficient of variation, and rate of change of the previous N time intervals for each of the multiple operation indicator items within the operation sequence.
[0139] For example, for each of the order throughput, customer density, inventory retention rate, budget exhaustion rate, event occurrence, and order failure rate, the situation optimization model may set a time window including the previous N time intervals and calculate the moving average, coefficient of variation, and rate of change within that time window. For example, the previous N time intervals may be N time intervals prior to the current point in time. For example, N may be a constant value greater than or equal to 1 and may be pre-set.
[0140] The moving average is the arithmetic mean of measurements included in a time window and can be used as an indicator representing the short-term base level of a corresponding operational metric. For example, the moving average of order throughput over a specific time interval is calculated as the average of the number of orders recorded during the preceding N time intervals, and can express the average demand intensity over a certain period.
[0141] The coefficient of variation is calculated as the ratio of the standard deviation of measurements within a time window to the mean, and it serves as an indicator representing the relative variability of each operational metric item as a normalized ratio. For example, a high coefficient of variation for customer density may indicate instability in customer inflow within the same timeframe or the presence of short-term surges or declines caused by specific promotions or events.
[0142] The rate of change can be set as a ratio value calculated based on the relative degree of change between adjacent time intervals using the measurements of consecutive time intervals included in the time window. Specifically, for each time interval within the time window, the situation optimization model can calculate the rate of change for each interval by subtracting the measurement of the preceding time interval from the measurement of the corresponding time interval and dividing the result by the absolute value of the preceding time interval. Subsequently, by calculating the average or weighted average of the rates of change for all intervals within the moving interval, a representative rate of change for the entire moving interval can be determined.
[0143] For example, if the moving interval change rate of the inventory remaining rate is calculated as a negative value, it indicates a trend of continuous decrease in inventory over the last N time intervals, and if the moving interval change rate of the budget exhaustion rate is calculated as a positive value, it may indicate that budget usage is gradually increasing over the same period.
[0144] For example, the situational optimization model can generate a single multidimensional vector by combining the moving average, coefficient of variation, and rate of change calculated for each operational metric item by time interval. In this case, since three types of statistical values are included for each of the six items—order throughput, customer density, inventory remaining rate, budget exhaustion rate, event occurrence, and order failure rate—an 18-dimensional vector with a total of 18 elements for each time interval can be generated as an operational pattern vector.
[0145] For example, a situation optimization model can sequentially connect 18-dimensional vectors for each time interval to generate a sequence of operation pattern vectors aligned along the time axis.
[0146] In step S530, the situation optimization model can learn the temporal correlation between multiple operational indicator items by inputting the operational pattern vector sequence into a hidden layer based on a recurrent neural network.
[0147] For example, a recurrent neural network-based hidden layer can be composed of an LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit) structure, and can sequentially receive operation pattern vectors for each time interval to generate a hidden state vector that reflects state transitions between time intervals.
[0148] For example, when applying an LSTM structure, the input value at each time step can be combined with the hidden state and cell state of the previous time step and pass through an input gate, a forget gate, and an output gate. Through this, the importance of input information over time can be adjusted, and long-term dependency and short-term fluctuation can be learned simultaneously.
[0149] For example, when applying the GRU structure, the input gate and the forget gate are integrated into a single update gate, resulting in relatively high computational efficiency and enabling operation suitable for time-series environments such as real-time data.
[0150] A recurrent neural network-based hidden layer can output a hidden state vector for each time interval through these gate operations, and each hidden state vector may be a high-dimensional representation containing simultaneous correlations between multiple operational indicator items within the time interval.
[0151] As a result, as the time intervals proceed sequentially, the situation optimization model internalizes the pattern flow of workplace operations and can generate a sequence of hidden state vectors for each time point that can be used for weight calculations in the next step.
[0152] In step S540, the situation optimization model can calculate sensitivity weights for each of the multiple operational indicator items by applying a feature gating layer to the output of the hidden layer.
[0153] The output of the hidden layer is a sequence of hidden state vectors by time interval, which may include hidden state vectors by time interval.
[0154] The feature gating layer determines the activation value for each item of the hidden state vector for each time interval and can calculate a gating coefficient for each operational indicator item by considering the interdependence between operational indicator items. In this case, the gating coefficient for each item can be set as a sensitivity weight to reflect the relative importance that the corresponding operational indicator item has on the correction coefficients of the discount policy (e.g., discount rate, discount period, maximum allowable discount rate).
[0155] For example, the feature gating layer determines the activity level of each operational metric item by performing an inner product operation on the hidden state vector of each time interval, and can calculate the importance ratio of all operational metric items to be expressed in a normalized form with a sum of 1 between 0 and 1 by applying a softmax function. In other words, the sensitivity weight of each operational metric item is the ratio obtained by dividing the activity value of that item by the sum of the activity values of all operational metric items, and can be a weight in the form of a normalized probability representing the relative intensity of attention. For example, if the activity value of the customer density item is calculated to be higher than that of other items, the sensitivity weight of the customer density item will have a high value close to 1 as a result of the softmax operation, which may mean that it is evaluated as the item that has the greatest influence on discount rate adjustments in that time interval. Conversely, in time intervals where fluctuations in inventory retention rate or budget exhaustion rate are small, the sensitivity weight of the item is calculated to be relatively low, so the impact on discount policy adjustments may be reflected as small. The sensitivity weights calculated in this way are normalized probability values representing the relative influence of each operational indicator item, and can numerically express how much the corresponding operational indicator item contributes to the calculation of correction factors for the discount rate, discount cycle, and maximum allowable discount rate depending on changes in the workplace status.
[0156] In step S550, the situation optimization model can generate an integrated state vector by multiplying the sensitivity weights by the hidden representation of each operational indicator item and then performing a weighted average operation on all operational indicator items.
[0157] For example, the situational optimization model can calculate a weighted representation vector reflecting the importance of each item by multiplying the sensitivity weight and the gating probability for each operational indicator item in the hidden state vector generated for each time interval. In this case, the gating probability can be calculated as a continuous probability value between 0 and 1 by applying a sigmoid function to the hidden state value of each operational indicator item, and can be used as a stochastic correction factor to compensate for non-linear differences in response sensitivity among each operational indicator item. That is, the gating probability can act as a stochastic damping factor to mitigate the contribution of an item in situations where a specific item is temporarily inactive or has low volatility and has little impact on the discount policy. Conversely, items with a gating probability close to 1 are judged to have high operational volatility at the current time or are items that directly affect the discount policy correction, and their weights may be strengthened. Therefore, the hidden representation of each item reflects both sensitivity weights and gating probabilities, and can be adjusted to emphasize or mitigate the relative impact of the item's operational volatility on the discount policy adjustment.
[0158] For example, a situational optimization model can generate a single integrated representation vector by taking as input a weighted representation vector reflecting the sensitivity weights and gating probabilities of each operational indicator item within the same time interval, calculating relative importance, and performing a weighted average operation based on relative importance. In this case, relative importance may be a ratio value calculated by normalizing the sensitivity weights of each operational indicator item to the weighted sum of all items. That is, the situational optimization model can derive an integrated representation vector for each time interval by multiplying the calculated relative importance values by the weighted representation vectors of each item and summing them for all operational indicator items. The integrated representation vector is a vector that summarizes the operational status of the business site within the corresponding time interval and may be a high-dimensional state representation reflecting the interrelationships between order throughput, customer density, inventory rate, budget exhaustion rate, event occurrence, and order failure rate.
[0159] For example, a situational optimization model can calculate an integrated state vector by combining integrated representation vectors generated from multiple time intervals in chronological order or by applying a time-weighted average method. The integrated state vector may be a multidimensional vector that reflects both the interdependencies between multiple operational indicator items and continuous changes between time intervals. In this case, for example, when applying a time-weighted average method, the situational optimization model can reflect recent situational changes more sensitively by assigning a relatively high weight to the operational state of the most recent time interval. In this case, the time weight can be set using an exponential decay or linear decay method depending on the relative position of the time interval. For example, when the time interval index is denoted as t, the time weight can be defined as an exponential function of the form e^(-λ(Tt)) such that the value is larger for the most recent time interval. Here, T is the total number of time intervals, and λ is the decay rate, which can be set to a value in the range of 0.05 to 0.3. Accordingly, the weight of the most recent segment can be set to account for at least 20% of the total sum of weights, and the cumulative weight of the segment within the previous 3 hours can account for at least 60% of the total. For example, if a pattern is observed where customer density increases rapidly and the budget depletion rate rises during the last 3 hours, the integrated representation vector of those segments can be multiplied by a high time weight and reflected in the final integrated state vector with a weight of at least 60%. On the other hand, for past segments that have elapsed more than 6 hours, the total reflection weight is reduced to less than 10% due to weight attenuation, so that changes in the latest operational conditions can be reflected more sensitively as a result.
[0160] In step S560, the situation optimization model inputs the integrated state vector into a multilayer perceptron-based correction factor calculation layer to calculate a workplace situation vector including a discount period correction factor, a discount rate correction factor, and a maximum allowable discount rate correction factor.
[0161] For example, a situation optimization model can model non-linear interrelationships between operational items by sequentially applying multiple fully connected layers and non-linear activation functions (ReLU or GELU) to an integrated state vector.
[0162] For example, the situation optimization model branches the integrated state vector into three output heads to predict the discount period correction factor, the discount rate correction factor, and the maximum allowable discount rate correction factor, respectively. Each output head consists of a small multilayer perceptron, and at the output stage, a sigmoid function can be applied to normalize the predicted values to a range between 0 and 1. In this case, the discount period correction factor represents the relative ratio of shortening or extending the reference discount period; a value close to 0 indicates a shortening relative to the reference, while a value close to 1 indicates an extension relative to the reference. The discount rate correction factor is an adjustment ratio for the reference discount rate; a value closer to 0 indicates a weakening of the discount strength, while a value closer to 1 indicates a strengthening. The maximum allowable discount rate correction factor is a value that corrects the maximum discount limit allowed by the system and can serve to constrain the discount rate correction factor so that it does not exceed this upper limit.
[0163] Additionally, for example, in the process of calculating correction factors, the situation optimization model can dynamically apply constraints based on the numerical range and reference threshold of each operational indicator.
[0164] For example, if the budget exhaustion rate exceeds a preset first threshold rate (e.g., 80%), the situation optimization model can prevent budget overruns due to additional discounts by limiting the upper limit of the discount rate adjustment factor to a preset first adjustment rate relative to the threshold (e.g., 70%). Conversely, if the budget exhaustion rate remains below a preset second threshold rate (e.g., 30%), the upper limit of the discount rate adjustment factor can be extended to a preset second adjustment rate relative to the threshold (e.g., 110%) to enable enhanced promotions utilizing the remaining budget.
[0165] In addition, during time periods when customer density exceeds a first standard density (e.g., 85% relative to seats), the discount cycle correction factor can be adjusted to a preset first adjustment ratio of the standard cycle (e.g., 1.2 times) to reduce congestion, thereby lowering the discount cycle, and during time periods when customer density is 2nd standard density (e.g., 40%) or lower, the discount cycle correction factor can be shortened to a preset second adjustment ratio of the standard cycle (e.g., 0.8 times) to increase the frequency of discount offers.
[0166] For example, if the inventory remaining rate decreases to a preset first threshold inventory rate (e.g., 20%) or lower, the situation optimization model can prevent excessive depletion of remaining inventory by lowering the discount rate correction factor to a preset first adjustment ratio (e.g., 60%) relative to the base discount rate. Conversely, if the inventory remaining rate is maintained at a preset second threshold inventory rate (e.g., 80%) or higher, the discount rate correction factor can be raised to a preset second adjustment ratio (e.g., 120%) relative to the base discount rate to increase inventory turnover and enhance sales promotion effects.
[0167] For example, if the order failure rate exceeds a preset threshold failure rate (e.g., 10%), the situation optimization model limits the application range of the discount rate adjustment factor to a preset first adjustment ratio relative to the threshold (e.g., 80%), thereby minimizing the inefficient application of the discount policy due to network instability or payment errors. Conversely, if the order failure rate stabilizes at less than half of the threshold failure rate (e.g., 5%), the discount rate adjustment factor is expanded to a preset second adjustment ratio relative to the threshold (e.g., 105%), thereby maintaining an active discount policy in a state of high system reliability.
[0168] For example, if the event frequency exceeds a certain threshold value (e.g., 0.7, i.e., the event is active in more than 70% of the total time periods), the situational optimization model may limit the upper limit of the discount rate adjustment factor to a pre-set first adjustment ratio relative to the threshold (e.g., 50%) to prevent excessive discount accumulation due to the duplicate application of discount policies. Conversely, if the event frequency remains below the threshold value (e.g., 0.3), the situational optimization model may expand the discount rate adjustment factor to a pre-set second adjustment ratio relative to the threshold (e.g., 120%) to continuously maintain a certain level of price incentive effect even when events occur infrequently. In this case, the event frequency can be calculated as the ratio of the event active interval by time period during a pre-set period.
[0169] For example, if two or more items among budget exhaustion rate, customer density, inventory retention rate, order failure rate, and event frequency simultaneously exceed or fall below a threshold, the situation optimization model can calculate a correction factor according to a predefined priority rule. For example, if the budget exhaustion rate exceeds a first threshold rate and the inventory retention rate is below a first threshold rate, the situation optimization model determines that budget preservation is a priority condition and may limit the discount rate correction factor to a pre-set first adjustment rate relative to the threshold (e.g., 50%). Conversely, if customer density is below a second threshold rate and the inventory retention rate is above a second threshold rate, the situation optimization model determines that sales promotion is a priority condition and may extend the discount rate correction factor to a pre-set second adjustment rate relative to the threshold (e.g., 130%).
[0170] Consequently, the standard and adjustment ratios for each item are stored in a predefined configuration table, which is automatically referenced when the situation optimization model receives real-time operational status input, and all correction calculations can be restricted to be performed only within a numerically specified range. Through this, the situation optimization model can quantitatively consider the site's budget, inventory, congestion, and event status to calculate consistent and predictable discount policy correction values at each point in time.
[0171] Additionally, according to one embodiment, if pre-configured business location operation information is unavailable, the server may refer to a pre-established industry average pattern data set based on the same industry group, the same business location type, and the same business time zone. The case where pre-configured business location operation information is unavailable may be when the sales history for the business location is shorter than a pre-configured period or does not exist.
[0172] At this time, the server can determine a reference group that satisfies the most similar conditions based on the industry group, business type, and operating hours metadata of the new business location. For example, industry groups can be classified into "coffee shops," "fast food," "Japanese restaurants," "convenience stores," etc., and business types can be categorized into "dine-in," "takeout / delivery-focused," or "hybrid." Operating hours can be set as "daytime (08:00–18:00)," "nighttime (18:00–02:00)," "24-hour," etc. By combining these three criteria values, the server identifies a reference group having the same industry group, business type, and operating hours as the business location, and can form an industry average pattern data set by statistically aggregating time-based operational indicator data collected from multiple existing business locations belonging to that group. The industry average pattern data set may include average values, standard deviations, and coefficients of variation for the number of orders by time of day, customer density, inventory retention rate, budget exhaustion rate, event frequency, and order failure rate, and each item may be calculated as the time-series average of values measured at multiple business sites with identical conditions.
[0173] For example, the server can extract the average value and coefficient of variation of each operational indicator item from an industry average pattern dataset and generate initial operational information for a business site based on the average value and coefficient of variation of each operational indicator item. Subsequently, the server can use this initial operational information as operational information for the business site.
[0174] For example, actual business operation information collected via kiosk terminals can be transmitted to a server for a certain period (e.g., 2 weeks) after the business has commenced actual operations. The actual business operation information may include actual measurements of the number of orders by time of day, customer density, inventory remaining rate, budget exhaustion rate, event frequency, and order failure rate. Each item can be stored in a normalized form by referring to a predefined table of units, collection cycles, and reference values.
[0175] The server can calculate the item-by-item difference between the actual site operation information and the site's initial operation information, and gradually update the correction coefficient for each operation indicator item based on that difference.
[0176] For example, if the actual number of orders is 20% or more higher than the industry average, the server can correct the underestimation by adjusting the order throughput correction factor upward, and if the inventory retention rate remains consistently lower than the industry average, it can correct the inventory correction factor downward. This correction process can be repeated at regular intervals (e.g., daily or weekly) and can be configured to mitigate short-term fluctuations and reflect only medium-to-long-term trends by applying an Exponential Moving Average (EMA) or an adaptive weight-based update method.
[0177] Through this, even in new business sites with insufficient initial data, stable initialization can be performed based on industry average patterns, and the system can be trained to converge to a corrected state vector that reflects the unique operational characteristics of the business site as operational data accumulates.
[0178] FIG. 6 is a flowchart illustrating the operation of a discount combination decision model using a third neural network according to one embodiment. The embodiment of FIG. 6 can be combined with various embodiments of the present disclosure.
[0179] Referring to FIG. 6, in step S610, the discount combination determination model can produce multiple discount combination candidates by discretizing each item of the preset standard discount parameter.
[0180] For example, the discount combination determination model can set the interval values of the corresponding items by setting the standard discount period, standard discount rate, and standard maximum allowable discount rate, respectively, as central reference values and applying predefined intervals.
[0181] For example, if the standard discount cycle is set to 5 days, the discount combination determination model can set the interval units to a lower interval of 3 days, a standard interval of 5 days, and an upper interval of 7 days centered around the standard value. If the standard discount rate is set to 10 percent, the discount combination determination model can set the lower interval to 5 percent, the standard interval to 10 percent, and the upper interval to 15 percent; and if the standard maximum allowable discount rate is set to 25 percent, the discount combination determination model can set the lower interval to 20 percent, the standard interval to 25 percent, and the upper interval to 30 percent. The interval values for each item set in this way are arranged at regular intervals centered around the standard value, so as to be configured to evenly divide the sensitivity and response range of the discount combination according to parameter changes.
[0182] For example, a discount combination decision model can generate all possible parameter combinations by combining the discrete interval values of each of the three items, and set each combination as an independent discount combination candidate. In this case, each discount combination candidate is represented in the form of a parameter vector consisting of three discrete values—discount period, discount rate, and maximum allowable discount rate—and can be combined with customer behavior information and workplace situation information in a subsequent step to be used as input for compensation prediction evaluation.
[0183] In step S620, the discount combination decision model can generate a candidate-specific context input vector by combining the behavior pattern vector and the workplace situation vector with each discount candidate combination.
[0184] For example, a discount combination decision model can generate a single high-dimensional context input vector along with the parameter values of each discount combination candidate by merging behavior pattern vectors (e.g., behavioral stability, price sensitivity, search propensity) derived from a customer behavior analysis model and business context vectors (e.g., time-based discount cycle correction factor, discount rate correction factor, maximum allowable discount rate correction factor) derived from a context optimization model.
[0185] At this time, the discount combination decision model can minimize input bias by performing a normalization operation considering that the units and ranges of each element are different, and by aligning all input elements on the same numerical scale. For example, since the discount rate is measured in percentage units, it can be normalized to a range of 0.0 to 1.0, and the discount period can be converted into a ratio value by dividing the daily value by the reference period.
[0186] In step S630, the discount combination decision model can perform a reward prediction operation on the context input vector for each candidate to calculate an expected reward value based on the revisit probability, sales growth rate, margin retention rate, and participation rate.
[0187] For example, a discount combination decision model can calculate the expected reward value for each candidate by inputting the context input vector for each candidate into a multilayer perceptron-based reward prediction network and learning the weights for each reward element in the hidden layer.
[0188] For example, a discount combination decision model can sequentially provide each context input vector to the input layer of a multilayer perceptron-based reward prediction network. In the input layer, a fully connected operation is performed to consider the interactions between each element, and in the hidden layer, a non-linear activation function such as ReLU or GELU is applied to learn complex correlations. The output of each hidden layer is stabilized through batch normalization and dropout operations, and weight parameters for each reward element can be simultaneously optimized. Through this process, the discount combination decision model can stably learn the mutual influences between various customer-situation-discount combinations and generate internal representation vectors that express the individual response trends for each reward item.
[0189] Subsequently, the discount combination decision model branches the internal representation vector, which is the output of the hidden layer, into four independent regression heads, allowing for the independent calculation of predicted values for each reward factor.
[0190] The revisit probability regression head can output a revisit probability based on behavioral stability, time-based discount cycle correction factors, and candidate discount cycles. In other words, the revisit probability regression head can evaluate how well the corresponding discount interval aligns with the customer's visit patterns and time-based preferred cycles. Specifically, it can operate with a structure where weights are learned such that the output probability increases as the candidate discount cycle approaches the preferred interval suggested by the time-based discount cycle correction factor, and decreases as the discrepancy widens. The final output can be calculated as a revisit probability normalized to a value between 0 and 1 by applying sigmoid activation. For example, if the time-based discount cycle correction factor is calculated as 1.2, this implies that an interval extended by 20% relative to the standard discount cycle is appropriate, and a high revisit probability can be predicted if the candidate discount cycle is close to this value. Conversely, if the candidate discount cycle is excessively short or long, the output can be adjusted to lower the likelihood of a revisit.
[0191] The sales growth rate regression head can output the sales growth rate based on price sensitivity, time-based discount rate correction factors, and candidate discount rates. In other words, the sales growth rate regression head can evaluate how the actual discount intensity is amplified or mitigated by the time-based correction factors. In time periods where the time-based discount rate correction factor is calculated to be high, the sales growth rate output may be calculated to be relatively high even with the same candidate discount rate; conversely, in time periods where the correction factor is low, the sales growth effect may be calculated to be mitigated even with the same discount rate. Through this, it is possible to quantitatively reflect the sensitivity of the time-based discount effect by distinguishing between time periods where price sensitivity is higher than a specific threshold value and those where it is lower.
[0192] The margin retention rate regression head can output the margin retention rate based on the time-dependent maximum allowable discount rate correction factor, candidate discount rates, and candidate maximum allowable discount rates. In other words, the margin retention rate regression head can evaluate whether the corresponding discount combination falls within the acceptable profit limit range. If the candidate discount rate exceeds the allowable threshold level indicated by the time-dependent maximum allowable discount rate correction factor and the candidate maximum allowable discount rate, the output value decreases sharply, reflecting a section where margin losses surge. Conversely, if the candidate discount rate remains within the allowable limit, the output can be maintained stably to reflect the decline in profit margins gradually.
[0193] The engagement rate regression head can output the engagement rate based on search propensity, time-based discount rate correction factors, time-based discount cycle correction factors, candidate discount rates, and candidate cycles. In other words, the engagement rate regression head can calculate the likelihood that a customer will respond immediately to the frequency and intensity of discount exposures. If the time-based discount rate correction factor is high and the candidate discount rate aligns with it, a high engagement rate output may be produced; conversely, if the time-based discount cycle correction factor is high (i.e., time periods with long discount intervals) and the candidate cycle is set short, the output may be calculated to be mitigated.
[0194] The four predicted values calculated in this way (revisit probability, sales growth rate, margin retention rate, and participation rate) are each expressed as real values between 0 and 1, and can represent the individual utility items of the discount combination candidates.
[0195] For example, a discount combination decision model can derive expected reward values by performing a weighted sum operation by applying preset weight ratios to these four elements. For instance, a weight of 0.3 may be assigned to the probability of revisiting and the sales growth rate, and a weight of 0.2 may be assigned to the margin retention rate and the engagement rate, respectively. In this case, the weighted sum is calculated by reflecting the importance of each reward item, enabling a balanced evaluation among customer retention, sales performance, and profit stability.
[0196] For example, a discount combination decision model can output the expected reward value for each discount combination candidate by normalizing the weighted sum result to a range between 0 and 1. The closer the expected reward value is to 1, the more favorable the corresponding discount combination is for inducing revisits and increasing sales in the current customer and business state, and the closer it is to 0, the more likely it is to be evaluated as a combination with low efficiency.
[0197] In step S640, the discount combination decision model can select the discount candidate combination with the largest expected reward value among multiple discount candidate combinations.
[0198] For example, a discount combination decision model can compare the expected reward values of all candidate discount combinations and rank them in descending order. Subsequently, the discount combination decision model can select the combination with the highest rank.
[0199] For example, the discount combination decision model can compare all candidate combinations based on the expected reward value of each discount candidate combination calculated in step S630, and rank them in order of highest expected reward value. Subsequently, the discount combination decision model can select the discount candidate combination with the highest rank.
[0200] In this case, if multiple discount candidate combinations are ranked highest, the discount combination decision model can make a final selection by applying pre-set auxiliary criteria. For example, the auxiliary criteria can be set to prioritize combinations with lower absolute discount rates, or, in the case of identical discount rates, combinations with relatively higher margin retention rates. Through this, the discount combination decision model can determine a reasonable combination that simultaneously considers profit stability and customer burden reduction, based on consistent rules, even when expected reward values are identical.
[0201] In step S650, the discount combination determination model can determine the discount candidate combination with the largest expected reward value using the discount period, discount rate, and maximum allowable discount rate included in the discount information.
[0202] Additionally, according to one embodiment, the server can relearn the updated parameters of each neural network and apply the updated parameters after learning in real time to subsequent transaction sessions to recalculate the discount cycle, discount rate, and maximum allowable discount rate that reflect the latest response trends of customers.
[0203] For example, to evaluate the performance of the recalculated discount parameter, the server can monitor the return visit rate, sales change rate, and engagement rate for a certain period after the discount offer, and cyclically update the parameter by reflecting these metrics in the next learning cycle.
[0204] Additionally, according to one embodiment, the server can analyze response trends at the group level by aggregating discount offer messages and response messages that occurred within the same workplace during a preset time window (e.g., 30 minutes).
[0205] For example, based on all transaction sessions included in the corresponding time window, the server can aggregate the total number of discount offers received by customers and the number of customers who actually accepted the discount offers, and calculate the group acceptance rate by time period using these figures. In this case, the group acceptance rate can be set as a value representing the proportion of customers who accepted the discount offer within the same time window.
[0206] If the calculated group acceptance rate is above a preset threshold, the server determines that the customer group included in the corresponding time window has responded positively to the discount offer and can automatically increase the discount rate applied to that group by a preset rate. For example, if the base discount rate is 10% and the preset amplification rate is 5%, the server can dynamically control the discount rate within the same group to 15% to strengthen customer engagement within a short period.
[0207] Conversely, if the group acceptance rate remains below a preset threshold, the server can prevent unnecessary budget expenditure by maintaining the discount rate for the same group or lowering it by a certain attenuation rate. This discount rate adjustment is performed on a group basis within the same time window, rather than on a single customer basis, thereby forming a real-time discount policy feedback structure based on customer response trends by time period.
[0208] In addition, the server can record the aforementioned discount rate adjustment event in the session log along with the corresponding time window and customer group identification information. The recorded session log can be reused as training data for the customer behavior analysis model and the discount combination decision model in subsequent learning cycles, and can be reflected in a form that weights the price sensitivity of customers within the same group. Through this, the system can learn short-term responses at the customer group level in real time and update the model to more precisely reflect behavioral patterns at the group level when calculating discount combinations in the future.
[0209] Additionally, according to one embodiment, after the discount cycle, discount rate, and maximum allowable discount rate are determined through a discount combination determination model using a third neural network, the server can recalculate optimized discount parameters for a specific transaction item through a discount adjustment model using a fourth neural network, using an identification value for a specific transaction item and the sales history of the transaction item for a certain period as input.
[0210] For example, the discount adjustment model may query basic attribute information corresponding to the identification value of the transaction item and transaction item group information, and generate an item embedding that expresses the categorical characteristics of the transaction item through a predetermined embedding transformation. Subsequently, the server may collect sales history by transaction item during a pre-set query period and calculate the margin rate of the transaction item by combining the total sales quantity, total revenue, actual sales indicators reflecting refunds and cancellations during the same period, and cost information or a margin table by transaction item. Additionally, the server may calculate the total sales revenue of all transaction items during the same period within the same business location, and set the ratio obtained by dividing the total sales revenue of a specific transaction item by the total sales revenue as the transaction item sales ratio, thereby quantifying the relative contribution that the transaction item accounts for to the total sales of the business location.
[0211] For example, the discount adjustment model can construct a transaction item feature vector representing a specific transaction item by combining the aforementioned item embeddings, the discount period, discount rate, and maximum allowable discount rate calculated through a third neural network, as well as the transaction item sales ratio, margin rate, and selected response rate indicator calculated based on the lookup period. Here, the transaction item feature vector can be defined as a high-dimensional representation combining the categorical characteristics of the transaction item, discount parameters, and recent performance indicators.
[0212] For example, a discount adjustment model can perform similarity-based transfer learning by taking a transaction item feature vector as input and calculating its similarity with multiple transaction item feature vectors stored in a historical training database. Specifically, the discount adjustment model can calculate the similarity between the current transaction item and past transaction items by applying a cosine similarity or embedding distance-based similarity function in the same feature space, and select the similar transaction item with the highest similarity. The selected similar transaction item may contain historical discount parameters along with performance indicators regarding sales growth, margin maintenance, and engagement response observed as a result of their application.
[0213] For example, for selected similar transaction items, the discount adjustment model can calculate the discount parameters of the current transaction item by referencing the historical discount parameters stored in the transaction item and using the similarity values between the current transaction item feature vector and each similar transaction item feature vector as weights, and by integrating the historical discount parameters with the discount period, discount rate, and maximum allowable discount rate calculated through a third neural network.
[0214] In this case, the similarity weight is a value calculated by a cosine similarity or embedding distance-based function, which may be a value that numerically represents the degree of feature matching between two transaction items. The discount adjustment model is configured so that the discount parameter of a transaction item has a greater influence on the result of the current item as similarity increases, while conversely, transaction items with low similarity are configured to have a smaller weight applied.
[0215] In addition, for example, performance-weighted aggregation can be performed instead of a simple similarity average by assigning additional performance weights to parameters of similar transaction items in which sales growth and margin retention rates were simultaneously observed to be favorable above a certain standard in the past, and by applying damping factors to parameters of transaction items with low performance.
[0216] For example, the server sets the results generated through this weighted aggregation process as discount parameters at the transaction item level, and the output discount period, discount rate, and maximum allowable discount rate can be configured as adjusted values that align with the discount parameters of the current time period while reflecting the recent performance of the transaction item and verified results of similar past items.
[0217] FIG. 7 is a block showing the configuration of a server according to one embodiment. The embodiment of FIG. 7 can be combined with various embodiments of the present disclosure.
[0218] As illustrated in FIG. 7, the server (700) may include a processor (710), a communication unit (720), and memory (730). However, not all components illustrated in FIG. 7 are essential components of the server (700). The server (700) may be implemented with more components than those illustrated in FIG. 7, or with fewer components than those illustrated in FIG. 7. For example, according to some embodiments, the server (700) may further include a user input interface (not shown), an output unit (not shown), etc., in addition to the processor (710), the communication unit (720), and memory (730).
[0219] The processor (710) typically controls the overall operation of the server (700). The processor (710) may have one or more processors to control other components included in the server (700). For example, the processor (710) may control the communication unit (720) and the memory (730) overall by executing programs stored in the memory (730). Additionally, the processor (710) may perform the functions of the server (700) described in FIGS. 3 to 6 by executing programs stored in the memory (730).
[0220] The communication unit (720) may include one or more components that enable the server (700) to communicate with another device (not shown) and the server (not shown). The other device (not shown) may be a computing device such as the server (700) or a sensing device, but is not limited thereto. The communication unit (720) may receive user input from another electronic device or receive data stored in an external device from an external device via a network.
[0221] The memory (730) can store a program for processing and controlling the processor (710). For example, the memory (730) can store information input to the server or information received from another device via a network. Additionally, the memory (730) can store data generated by the processor (710). The memory (730) can also store information input to the server (700) or output from the server (700).
[0222] The memory (730) may include at least one type of storage medium among flash memory type, hard disk type, multimedia card micro type, card type memory (e.g., SD or XD memory, etc.), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic disk, and optical disk.
[0223] For example, each of the data collection device and the plurality of sensors may include the processor, memory, and communication unit described above.
[0224] The embodiments described above may be implemented as hardware components, software components, and / or combinations of hardware and software components. For example, the devices, methods, and components described in the embodiments may be implemented using one or more general-purpose or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are also possible.
[0225] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or command the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.
[0226] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiment, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.
[0227] Although the embodiments have been described above with reference to the limited drawings, those skilled in the art can apply various technical modifications and variations based on the above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or if the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.
[0228] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.
Claims
Claim 1 A method for a server to provide a customer-tailored discount service through a kiosk terminal using artificial intelligence comprises: receiving a discount proposal request including customer identification information from the kiosk terminal; acquiring customer history data corresponding to the customer identification information; wherein the customer history data includes purchase history data and kiosk operation logs, and generating a behavioral pattern vector including customer behavior stability, price sensitivity, and search propensity through a customer behavior analysis model using a first neural network based on the customer history data; generating a business situation vector including a discount cycle correction coefficient, a discount rate correction coefficient, and a maximum allowable discount rate correction coefficient through a situation optimization model using a second neural network based on pre-set business operation information; determining discount information including a discount cycle, a discount rate, and a maximum allowable discount rate through a discount combination decision model using a third neural network based on the behavioral pattern vector, the business situation vector, and pre-set reference parameters; wherein the pre-set reference parameters include a reference discount cycle, a reference discount rate, and a basic maximum allowable discount rate, and transmitting a discount proposal message including the discount information to the kiosk terminal.The method comprises the step of receiving a response message regarding the discount offer message from the kiosk terminal, wherein the parameters of the first neural network, the second neural network, and the third neural network are updated using the response message as training data, and the purchase history data includes at least one purchase date and time, and values related to the purchase amount, transaction item group, payment method, and whether a discount is applied for the at least one purchase date and time, and the kiosk operation log includes a screen-specific dwell time log including the display time and end time of each screen, and a touch event stream log including the time, coordinates, target widget identifier, and input type of each touch event, and the customer behavior analysis model calculates a purchase cycle vector including an average purchase interval, variability of the purchase cycle, average purchase amount, discounted purchase ratio, and transaction item group diversity based on the purchase history data, and based on the screen-specific dwell time log and the touch event stream log, a response delay vector quantifying the time pattern from when the discount offer screen is displayed until the first touch event is received, an operation rhythm vector expressing the temporal regularity and consistency between touch events, and a purchase expressing the stability of decision-making through patterns of adding and removing transaction items in the shopping cart. A method comprising: calculating a discount screen concentration vector reflecting a pseudo-vector, dwell time on a discount offer screen, and touch concentration on a discount offer area; merging the purchase cycle vector, the response delay vector, the operation rhythm vector, the purchase pseudo-vector, and the discount screen concentration vector to generate a sequence of behavioral feature vectors in which the time order is preserved; generating an integrated representation vector by learning the correlation between time points through a multilayer encoder based on the sequence of behavioral feature vectors; and calculating the behavioral stability, the price sensitivity, and the search propensity, respectively, through a plurality of regression heads based on the integrated representation vector. Claim 2 delete Claim 3 In claim 1, the pre-set business site operation information includes multiple operation indicator items by time period during a pre-set period, and the multiple operation indicator items include the number of orders, customer density, inventory remaining rate, budget exhaustion rate, event occurrence frequency, and order failure rate; the situation optimization model generates an operation sequence by sorting the pre-set business site operation information by day and hour; generates an operation pattern vector sequence by calculating the moving average value, coefficient of variation, and rate of change of the previous N time intervals for each of the multiple operation indicator items within the operation sequence; learns the temporal correlation between the multiple operation indicator items by inputting the operation pattern vector sequence into a hidden layer based on a recurrent neural network; calculates a sensitivity weight for each of the multiple operation indicator items by applying a feature gating layer to the output of the hidden layer; generates an integrated state vector by multiplying the sensitivity weight by the hidden representation of each operation indicator item and performing a weighted average operation on all operation indicator items; and inputs the integrated state vector into a multi-layer perceptron-based correction coefficient calculation layer to the discount period A method comprising calculating a correction factor, the discount rate correction factor, and the maximum allowable discount rate correction factor for each time period during the preset period, wherein the discount combination determination model discretizes each item of the preset standard discount parameter to produce a plurality of discount combination candidates, combines the behavior pattern vector and the workplace situation vector with each discount candidate combination to generate a context input vector for each candidate, performs a reward prediction operation on the context input vector for each candidate to calculate an expected reward value based on the revisit probability, sales growth rate, margin retention rate, and participation rate, selects the discount candidate combination with the largest expected reward value among the plurality of discount candidate combinations, and determines the discount candidate combination with the largest expected reward value as the discount period, the discount rate, and the maximum allowable discount rate.
Citation Information
Patent Citations
Smart ordering system based on user terminal usage patterns identified through artificial intelligence
KR1020250113292A
Method and device for providing promotional information to a user terminal using a neural network
KR102740580B1
Method for providing integrated online and offline inventory management and customer acquisition solutions through ai model-based customer information and payment data analysis
KR102851024B1
