Electronic device and control method therefor

Abstract

Disclosed is an electronic device. The electronic device comprises: an interface for receiving an input of data on prices and sales amounts of a plurality of products; a memory for storing an artificial intelligence model; and a processor, wherein the processor obtains data of a first unit period preset on the basis of a learning time point and data of a second unit period before the first unit period, among data classified on the basis of a release time of each of the plurality of products, retrains, through verification using the data of the first unit period, the artificial intelligence model that has been trained by using the data of the second unit period, and when data on a target product is input through the interface, provides a prediction value corresponding to the target product by using the input data and the re-trained artificial intelligence model.

Description

Electronic device and method of controlling the same

[0001] The present disclosure relates to an electronic device and a control method thereof, and more particularly, to an electronic device and a control method thereof that obtains changes in sales volume according to changes in product price by utilizing machine learning.

[0002] For a company to efficiently plan production, distribution, sales, and inventory, it is essential to predict product sales volumes, i.e., consumer demand. The most significant influence on consumer demand is the product's selling price, known as price elasticity of demand. Price elasticity is an indicator of the extent to which changes in the final market price affect changes in demand or sales volume, and can indicate how the market reacts to price fluctuations.

[0003] However, in the past, there was a problem in that price variables were not considered as factors for predicting product sales, and product sales predictions were mainly dependent on data values ​​entered immediately before.

[0004] According to the present disclosure, an electronic device includes an interface for receiving data on prices and sales volumes of a plurality of products, a memory for storing an artificial intelligence model, and a processor. The processor obtains data for a first unit period set based on a learning time point and a second unit period prior to the first unit period from among data classified based on the release dates of each of the plurality of products. The processor retrains the artificial intelligence model trained using data for the second unit period through verification using data for the first unit period. When data on a target product is input through the interface, the processor provides a predicted value corresponding to the target product using the input data and the retrained artificial intelligence model.

[0005] Meanwhile, a method for controlling an electronic device according to one or more embodiments of the present disclosure includes the steps of: receiving data on prices and sales volumes of a plurality of products; classifying the data based on the release time of each of the plurality of products; acquiring data of a first unit period preset based on a learning time point and a second unit period prior to the first unit period from among the classified data; training an artificial intelligence model using the data for the second unit period; retraining the artificial intelligence model through verification using the data for the first unit period; and, when data for a target product is input, providing a predicted value corresponding to the target product using the input data and the retrained artificial intelligence model.

[0006] Meanwhile, a computer-readable recording medium including a program for executing a control method of an electronic device according to one or more embodiments of the present disclosure includes a step of receiving data on prices and sales volumes of a plurality of products, a step of classifying the data based on the release time of each of the plurality of products, a step of acquiring data of a first unit period preset based on a learning time point and a second unit period prior to the first unit period from among the classified data, a step of training an artificial intelligence model using data for the second unit period, a step of retraining the artificial intelligence model through verification using data for the first unit period, and a step of providing a predicted value corresponding to the target product using the input data and the retrained artificial intelligence model when data for the target product is input.

[0007] FIG. 1 is a block diagram showing the configuration of an electronic device according to various embodiments of the present disclosure.

[0008] FIG. 2 is a diagram illustrating a linear regression distribution for determining a critical period according to various embodiments of the present disclosure.

[0009] FIG. 3 is a diagram illustrating determining a critical period based on data of multiple products according to various embodiments of the present disclosure.

[0010] FIG. 4 is a diagram for explaining the operation of an artificial intelligence model according to various embodiments of the present disclosure.

[0011] FIG. 5 is a diagram showing a sales status prediction result according to various embodiments of the present disclosure on a display.

[0012] FIGS. 6 to 9 are flowcharts for explaining a method of controlling an electronic device according to various embodiments of the present disclosure.

[0013] The terms used in this specification will be briefly explained, and the present disclosure will be described in detail.

[0014] The terms used in the embodiments of this disclosure have been selected from widely used, current terms, taking into account the functions of this disclosure. However, these terms may vary depending on the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in certain cases, terms may be arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the description of the relevant disclosure. Therefore, the terms used in this disclosure should not be defined simply as names of terms, but rather based on the meanings of the terms and the overall content of this disclosure.

[0015] In this specification, expressions such as “has,” “can have,” “includes,” or “may include” indicate the presence of a feature (e.g., a number, function, operation, or component such as a part), and do not exclude the presence of additional features.

[0016] In this disclosure, expressions such as “A or B,” “at least one of A and / or B,” or “one or more of A or / and B” can include all possible combinations of the listed items. For example, “A or B,” “at least one of A and B,” or “at least one of A or B” can all refer to (1) including at least one A, (2) including at least one B, or (3) including both at least one A and at least one B.

[0017] As used herein, the expressions “first,” “second,” “first,” or “second,” etc., may describe various components, regardless of order and / or importance, and are only used to distinguish one component from another, but do not limit the components.

[0018] When it is said that a component (e.g., a first component) is “operatively or communicatively coupled with / to” or “connected to” another component (e.g., a second component), it should be understood that the component may be directly coupled to the other component, or may be connected through another component (e.g., a third component).

[0019] Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "comprise" or "consist of" are intended to indicate the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but should be understood not to preclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0020] In the present disclosure, a "module" or "part" performs at least one function or operation and may be implemented as hardware or software, or as a combination of hardware and software. Furthermore, multiple "modules" or multiple "parts" may be integrated into at least one module and implemented as at least one processor (not shown), excluding any "modules" or "parts" that need to be implemented as specific hardware.

[0021] An embodiment of the present disclosure will be described in more detail with reference to the attached drawings below.

[0022] FIG. 1 is a block diagram illustrating the configuration of an electronic device according to various embodiments of the present disclosure. The electronic device (100) of FIG. 1 may be implemented as various types of devices, such as a server device, a computer, a laptop PC, a smartphone, a tablet PC, a kiosk, etc. In addition, it may be implemented as any device equipped with at least one processor capable of data processing.

[0023] According to FIG. 1, an electronic device (100) includes an interface (110), a memory (120), and a processor (130).

[0024] The interface (110) is configured to receive various data from a user, external memory, or an external device. For example, the interface (110) can receive data on the prices and sales volume of multiple products. The interface (110) may include a communication interface, an operation interface, and an input / output interface.

[0025] For example, a communication interface is a configuration for performing communication with at least one external device. The communication interface may include at least one wireless communication module, at least one wired communication module, etc. Each communication module may be implemented in the form of at least one hardware chip. The wireless communication module may include at least one module among a Wi-Fi module, a Bluetooth module, an infrared communication module, or other communication modules. In addition, the communication interface may include at least one communication chip that performs communication according to various wireless communication standards such as Zigbee, 3G (3rd Generation), 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), LTE-A (LTE Advanced), 4G (4th Generation), 5G (5th Generation), etc. The wired communication module may include, for example, at least one among a LAN (Local Area Network) module, an Ethernet module, a pair cable, a coaxial cable, a fiber optic cable, or a UWB (Ultra Wide-Band) module. The communication interface is implemented in various forms like this, and by performing communication with an external device, various data related to product price and sales can be received from the external device.

[0026] The operation interface is a configuration for receiving user input. The operation interface may include various buttons, a touch screen, etc. provided on the main body of the electronic device (100). Using the operation interface, the user can directly input various data related to product price and sales into the electronic device (100).

[0027] The input / output interface is a configuration for inputting and outputting various external signals. The input / output interface can be connected to various external memories or external sources (e.g., web servers, user terminal devices, etc.) and can input and receive various data. The input / output interface can be implemented as at least one interface among HDMI (High Definition Multimedia Interface), MHL (Mobile High-Definition Link), USB (Universal Serial Bus), USB C-type, DP (Display Port), Thunderbolt, VGA (Video Graphics Array) port, RGB port, D-SUB (Dsubminiature), and DVI (Digital Visual Interface). At least some of the input / output interfaces may be connected to a communication interface. For example, the input / output interface can transmit information received from an external device to the communication interface or transmit information received through the communication interface to the external device. The electronic device (100) can directly read or receive various data related to product prices and sales stored in an external memory or external source connected through the input / output interface.

[0028] The memory (120) can store at least one command, data, program, etc. required for the operation of the electronic device (100). For example, the memory (120) can store data on the prices and sales volume of multiple products and an artificial intelligence model. The memory (120) may be implemented in the form of a memory embedded in the electronic device (100) or in the form of a memory that can be detachably attached to the electronic device (100) depending on the purpose of data storage. For example, data for operating the electronic device (100) may be stored in a memory embedded in the electronic device (100), and data for expanding the functions of the electronic device (100) may be stored in a memory that can be detachably attached to the electronic device (100).

[0029] In the case of memory embedded in the electronic device (100), it may be implemented as at least one of volatile memory (e.g., dynamic RAM (DRAM), static RAM (SRAM), or synchronous dynamic RAM (SDRAM)), non-volatile memory (e.g., one time programmable ROM (OTPROM), programmable ROM (PROM), erasable and programmable ROM (EPROM), electrically erasable and programmable ROM (EEPROM), mask ROM, flash ROM, flash memory (e.g., NAND flash or NOR flash), hard drive, or solid state drive (SSD)).

[0030] The memory (120) may be implemented as a single memory that stores data generated in various operations according to the present disclosure, but is not limited thereto, and the memory (120) may be implemented to include multiple memories that each store different types of data or each store data generated in different stages.

[0031] The memory (120) can store various data received through the interface (110), for example, various data related to product price and sales.

[0032] The processor (130) is a component that is connected to each component of the electronic device (100) and controls the overall operation of the electronic device (100). The processor (130) may be implemented as a digital signal processor (DSP), a microprocessor, a GPU (Graphics Processing Unit), an AI (Artificial Intelligence) processor, an NPU (Neural Processing Unit), or a TCON (Time Controller). However, the processor (130) is not limited thereto, and may include one or more of a central processing unit (CPU), an MCU (Micro Controller Unit), an MPU (Micro Processing Unit), a controller, an application processor (AP), a communication processor (CP), or an ARM processor, or may be defined by the relevant terminology. In addition, the processor (130) may be implemented as a SoC (System on Chip), an LSI (Large Scale Integration) having a built-in processing algorithm, or may be implemented in the form of an ASIC (Application Specific Integrated Circuit), or an FPGA (Field Programmable Gate Array).

[0033] In addition, a processor (130) for executing an artificial intelligence model according to an embodiment may be implemented through a combination of a general-purpose processor such as a CPU, an AP, a DSP (Digital Signal Processor), a graphics-only processor such as a GPU, a VPU (Vision Processing Unit), or an artificial intelligence-only processor such as an NPU, and software.

[0034] The processor (130) can be controlled to process input data according to predefined operating rules or artificial intelligence models stored in memory. Alternatively, if the processor (130) is a dedicated processor (or an artificial intelligence-dedicated processor), it can be designed with a hardware structure specialized for processing a specific artificial intelligence model. For example, hardware specialized for processing a specific artificial intelligence model can be designed as a hardware chip such as an ASIC or an FPGA.

[0035] When the processor (130) is implemented as a dedicated processor, it may be implemented to include a memory for implementing the embodiments of the present disclosure, or it may be implemented to include a memory processing function for utilizing external memory. The processor (130) may be implemented as one or more processors.

[0036] Meanwhile, the function related to artificial intelligence according to the present disclosure may be operated through a processor (130) and a memory (120). One or more processors are controlled to process input data according to predefined operation rules or artificial intelligence models stored in the memory (120). Alternatively, if one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model. The predefined operation rules or artificial intelligence models are characterized by being created through learning.

[0037] Here, "created through learning" means that a basic artificial intelligence model is learned using a learning algorithm using a large amount of learning data, thereby creating a predefined set of operating rules or an artificial intelligence model configured to perform a desired characteristic (or purpose). This learning may be performed on the device itself on which the artificial intelligence according to the present disclosure is performed, or may be performed through a separate server and / or system. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0038] An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values ​​and performs neural network operations by calculating the results of previous layers and the multiple weights. The multiple weights of the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights may be updated during the learning process to reduce or minimize the loss or cost values ​​obtained by the artificial intelligence model.

[0039] The artificial neural network may include a Conditional Neural Process (CNP) or a Deep Neural Network (DNN), and examples thereof include, but are not limited to, a Convolutional Neural Network (CNN), a Deep Neural Network (DNN), a Recurrent Neural Network (RNN), a Restricted Boltzmann Machine (RBM), a Deep Belief Network (DBN), a Bidirectional Recurrent Deep Neural Network (BRDNN), a Generative Adversarial Network (GAN), a Latent Neural Process (LNP), Attentive Neural Processes (ANP), Residual Neural Processes (RNP), Doubly Stochastic Neural Processes (DSNP), Transformer Neural Processes (TNP), Memory-Augmented Neural Networks (MANN), Simple Neural Attentive Meta-Learner (SNAIL), or Deep Q-Networks.

[0040] The processor (130) can collect or receive data on the prices and sales volume of multiple products through the interface (110) and store the data in the memory (120). Specifically, the processor (130) can collect data on the prices and sales volume of products through data from a market research institute, a company's SCM (Supply Chain Management) system, etc.

[0041] In this case, the processor (130) may collect data on prices and sales volumes of various products by category and store the data in the memory (120). For example, if the target product is a washing machine, the processor (130) may collect data on prices and sales volumes of products belonging to the washing machine category. Categories may be classified by product type, or may classify various products by use, or may be classified by various criteria such as by manufacturer or price range. For example, since a washing machine is a type of home appliance, it may be classified into the home appliance category along with air conditioners, TVs, refrigerators, vacuum cleaners, etc.

[0042] The processor (130) may collect additional variable data via the interface (110) and store it in the memory (120). The additional variable data may include information on additional variables that may affect product sales, in addition to the product price. For example, the additional variable data may include variable data related to basic product characteristics that do not change over time, such as the product's size. Furthermore, the additional variable data may include variable data related to peak product sales periods, such as periods when product demand significantly increases or decreases (e.g., year-end, holidays, new semesters, Christmas, Black Friday). The additional variable data may also include variable data related to economic indicators, such as exchange rates, price indices, unemployment rates, money supply, and interest rates. Generally, product sales volume is influenced by external economic conditions, so variable data related to economic indicators can be an important factor in predicting product sales volume. The processor (130) may collect variable data related to economic indicators through national statistical systems, etc.

[0043] Additional variable data may include variable data related to product demand, product supply, product inventory, and product production costs. The processor (130) may collect data on product demand, product supply, product inventory, and product production costs through a SCM system or the like.

[0044] Additional variable data can include variable data based on the product's sales region. For example, refrigerators and air conditioners may sell more in warm temperate or tropical regions than in colder regions, given the same price. Conversely, heating appliances may sell more in colder regions than in tropical regions, given the same price.

[0045] Alternatively, preferred specifications may vary depending on the sales region, and factors such as exchange rates at the time of export to that region or transportation costs may also vary. For example, in the case of TVs, preferred sizes may differ between Korea and the US, and Korean and US sales prices may differ due to exchange rate fluctuations, oil price fluctuations, and other factors. Regional influences can also be incorporated into additional variable data and used for AI model training and prediction.

[0046] In the present disclosure, data on product prices and sales volumes and additional variable data may be provided as time series data, but are not limited thereto.

[0047] The processor (130) selects data stored in the memory (120) according to preset learning selection criteria. The learning selection criteria may be used to select learning data for training an artificial intelligence model used in the electronic device (100). For example, the learning selection criteria may be used to select data that has passed a critical period of time since its release.

[0048] The processor (130) classifies the selected data based on the release date of each of the plurality of products. The processor (130) obtains data of a first unit period preset based on the learning time point from among the classified data, and data of a second unit period prior to the first unit period. For example, if the learning time point is a certain point in 2023, the first unit period may be from January to December 2022, and the second unit period may be from January to December 2021. This is merely an example, and the unit period setting criteria and the length of the period may be variously changed. For example, if the learning time point is November 2023, the first unit period may be one year prior to that point, that is, from November 2022 to October 2023, and the second unit period may be prior to October 2022.

[0049] Specifically, the processor (130) may remove data up to a preset threshold period (e.g., 8 weeks) based on the release date of each of the multiple products from the price and sales data of the multiple products input through the interface (110), and classify data after the threshold period based on the release date of each of the multiple products. For example, if data on the price and sales of Product A released in January 2020 is input through the interface (110), the processor (130) may remove data for 8 weeks from January 2020 of Product A, and select data thereafter.

[0050] Typically, when a product is launched, sales may be low for a certain period immediately after launch, as the product may not fully penetrate the market. Conversely, sales may be high due to the effects of the new product launch and advertising.

[0051] In other words, data collected early in the product's life cycle contains significant noise, as factors other than price play a significant role. For this reason, the processor (130) can eliminate data prior to a critical period of time, based on the release date of each product, to improve the accuracy of the training data.

[0052] When multiple products exist, the processor (130) can remove data for a critical period based on the release date of each of the multiple products. That is, data for products whose release date falls within the first unit period among the entire unit period, as well as data for products subject to sales status, can be removed and used for the critical period based on the release date. The critical period may be the same period calculated by analyzing the sales volume of multiple products, but is not necessarily limited thereto, and may be adaptively changed for each product to suit its sales characteristics.

[0053] When data from a first unit period and data from a second unit period are extracted, the processor (130) can train an artificial intelligence model using these data. Specifically, the processor (130) retrains an artificial intelligence model trained using data from a second unit period through verification using data from a first unit period following the second unit period. The processor (130) stores the trained or retrained artificial intelligence model in the memory (120).

[0054] When data regarding a target product is input via the interface (110), the processor (130) can use the input data and a retrained artificial intelligence model to provide a predicted value corresponding to the target product. The target product may be a product for which a user wishes to predict sales volume. For example, if a user specifies a washing machine and inputs its expected price, the processor (130) can predict the sales volume if the washing machine is sold at that price and provide the predicted result.

[0055] FIG. 2 is a diagram illustrating a linear regression distribution for determining a critical period according to various embodiments of the present disclosure. In FIG. 2, the x-axis represents the log price, and the y-axis represents the log sales volume. Furthermore, for convenience of explanation, the following description related to FIG. 2 assumes that the product launch period is the 8th week (W8).

[0056] According to FIG. 2, the processor (130) can obtain data on the relationship between log prices and log sales volumes based on data on prices and sales volumes of a plurality of classified products. The processor (130) can perform linear regression analysis on the sales volumes of each of the plurality of products during a sales period based on the obtained data on the relationship between log prices and log sales volumes, thereby obtaining a linear regression line (210) representing the price elasticity of each product.

[0057] Price elasticity can be estimated by performing linear regression on the relationship between the logarithm of price and the logarithm of sales volume, and then selecting a statistically significant model. In economics, assuming price elasticity is constant, the logarithm of price and the logarithm of sales volume form a linear relationship, and the slope of the linear regression line can be expressed as price elasticity.

[0058] The processor (130) can produce a linear regression line (210) through linear regression analysis of the log price and log sales data. In this case, the processor (130) may set a main sales period based on the product sales volume to increase the accuracy of the linear regression line (210), and perform linear regression analysis of the log price and log sales data during the set main sales period. For example, the linear regression line (210) of FIG. 2 represents a case in which a linear regression analysis is performed by setting the period from the 34th week (W34) to the 52nd week (W52), when the product sales volume is high, as the main sales period based on the product launch date (W8).

[0059] The processor (130) may obtain the standard deviation (σ) of the calculated residual values ​​in the process of performing linear regression analysis, and based on the obtained standard deviation (σ), may set a confidence interval (220) indicating the error range between the linear regression line (210) and the actual input data. For example, in FIG. 2, the processor (130) sets a confidence interval (-2.58σ, 2.58σ) (220) in which the error (Error) value between the linear regression line (210) and the actual data indicates a 99% confidence level.

[0060] The processor (130) can identify the entry point at which the deviation of each sales amount enters within a preset confidence interval (220) based on the acquired linear regression line (210). In this case, the processor (130) can predict the sales amount of each of a plurality of products using an artificial intelligence model learned based on learning data for a second unit period, and can obtain the deviation between the predicted sales amount and the input sales amount by comparing the predicted sales amount and the input sales amount. In Fig. 2, it can be confirmed that the deviation of the data first enters within the preset confidence interval (220) at the 19th week (W19) (230) based on the product launch time (W8). The processor (130) can obtain the entry period taken to enter within the confidence interval (220) from the product launch time (W8) as 11 weeks.

[0061] The processor (130) may determine the entry period between the product launch date and the entry point as a threshold period for filtering and selecting data on prices and sales volumes of multiple products. In the example described above, since W19-W8=11 weeks, the threshold period may be set to 11 weeks. The processor (130) may remove data up to the threshold period based on the product launch date, and classify data after the threshold period based on the launch date of each of the multiple products. For example, if the product launch date is the 12th week (W12), the processor (130) may remove data for 11 weeks based on the launch date (W12), and classify and select data after the 23rd week (W23).

[0062] FIG. 3 is a diagram illustrating determining a threshold period based on data for multiple products according to various embodiments of the present disclosure. In FIG. 3 , the x-axis represents the entry period from the launch date of each product to the entry point, and the y-axis represents the number of products corresponding to each entry period. In FIG. 3 , the units for the period and number of products are omitted.

[0063] According to FIG. 3, the processor (130) obtains data on the relationship between log prices and log sales volumes based on data on prices and sales volumes of a plurality of products, and performs linear regression analysis on the sales volumes of each of the plurality of products during a sales period based on the obtained data to obtain a linear regression line representing the price elasticity of each product. In addition, the processor (130) can identify an entry point at which the deviation of each sales volume enters a preset confidence interval based on the linear regression line. Since specific linear regression lines, confidence intervals, and entry points have been described in the above-described section, a redundant description thereof will be omitted.

[0064] As shown in FIG. 3, the processor (130) can identify the number of products with the same entry period from the product launch date to the entry point for each entry period. For example, in FIG. 3, the number of products that entered within the confidence interval after 1 week from the product launch date is 18, and the number of products that entered within the confidence interval after 2 weeks from the product launch date is 25. In addition, the number of products that entered within the confidence interval after 15 weeks from the product launch date is 9, and the number of products that entered within the confidence interval after 25 weeks from the product launch date is 1.

[0065] The processor (130) may determine the entry period (310) of a product number corresponding to a preset ratio based on the total number of products calculated by adding up the number of products for each entry period as a critical period. For example, if the ratio for determining the entry period (310) in FIG. 3 is set to 85%, the processor (130) may determine 12 weeks, corresponding to 85% of the total number of products calculated by adding up the number of products for each entry period, as the critical period.

[0066] In this case, the processor (130) may preset a reference threshold value of the critical period and store it in the memory (120), and if the calculated critical period value is smaller than the reference threshold value, the critical period may be determined using the preset reference threshold value. For example, if 8 weeks are set as the reference threshold value, since the critical period calculated in FIG. 3 is 12 weeks, 12, which is larger than 8 weeks, is determined as the final critical period. The processor (130) may classify data after the critical period, from which data from the release date of the product to the preset critical period is removed, based on the release dates of each of the multiple products.

[0067] The processor (130) may select price and sales data of a plurality of products based on at least one of the initial sales volume, release date, and number of initial release data of each of the plurality of products, and may classify the selected data based on the release date of each product. Specifically, the processor (130) may filter and remove data of products whose initial sales volume for a preset period of time after a threshold period is below a preset standard sales volume, and classify the selected data based on the release date of each product.

[0068] The processor (130) may remove data on products whose release dates are after a preset reference time from the data entered through the interface (110) or the price and sales volume data of multiple products after a classified threshold period, and classify the data based on the release dates of each product. For example, the processor (130) may remove data on products whose release dates are later than a preset number of weeks from the data entered after the classified threshold period, and classify the selected data based on the release dates of each product.

[0069] The processor (130) may filter and remove data of products for which the number of initial launch data for each product during a preset period after a critical period is less than or equal to a preset reference data number, and may classify the selected data based on the launch time of each product.

[0070] The processor (130) trains an artificial intelligence model using data for a second unit period based on the product launch date among the classified data. The artificial intelligence model can be implemented in various types as described above, but for convenience of explanation, the following description assumes an artificial intelligence model using a Conditional Neural Process (CNP).

[0071] Unlike recurrent neural network (RNN)-based models (such as RNN, GRU, and LSTM models) that assume recurrent relationships between time-series data and perform value prediction, the CNP learning model performs conditional probability distribution prediction for predicted values ​​based on stochastic process theory. Therefore, while RNN-based models primarily rely on the previously input data value to predict the next data value, the CNP model can equally use all data values ​​input up to the present to predict the probability distribution of the next data value.

[0072] FIG. 4 is a diagram illustrating the operation of an artificial intelligence model according to various embodiments of the present disclosure. Specifically, FIG. 4 illustrates predicting price and sales volume changes and calculating price elasticity using a Conditional Neural Process (CNP). In FIG. 4, x represents product price data, y represents product sales volume data, and f represents additional variable data. In FIG. 4, r represents a vector encoding x, y, and f, and g (421) and h (411) represent a fully connected neural network (FCNN). In addition, in FIG. 4, μ represents the average sales volume data of the target product, and σ represents the standard deviation sales volume data of the target product.

[0073] According to FIG. 4, the CNP model may include an encoder (410) and a decoder (420). Furthermore, the encoder (410) and the decoder (420) may include a fully connected neural network (FCNN). The processor (130) may train the artificial intelligence model using at least one of the price and sales data of each product and additional variable data.

[0074] As shown in Fig. 4, when at least one of the price and sales volume data and additional variable data of each product stored in the memory (120) is input to the FCNN (411) of the encoder (410), the processor (130) can compress the data input to the encoder (410) regardless of the length of each data and encode it into a single vector (412). The encoded vector (412) is transmitted to the decoder (420) of the CNP model. The vector (422) input to the decoder (420) can include learning information about the price and sales volume data of each product and the additional variable data.

[0075] The FCNN (421) of the decoder (420) includes the vector (422) encoded and transmitted from the encoder (410) and the price data (x) of the target product. t ), additional variable data of the target product (f t ) can be input. The processor (130) can obtain parameters of a specific distribution for the sales volume of the target product or a probabilistic distribution for the sales volume of the target product by using the FCNN (421) of the decoder (420). For example, if the specific distribution for the sales volume of the target product represents a normal distribution, the processor (130) can obtain the mean (μ) and variance data (σ) for the sales volume of the target product. 2 ) can be used to obtain parameters related to the object.

[0076] In this case, among the data on prices and sales volumes of multiple products, the data for the second unit period may include information on the time series data of the products. Accordingly, the processor (130) may calculate the likelihood of the actual sales volume for the product for the second unit period based on the distribution of sales volume predicted based on the data for the second unit period. The processor (130) may adjust the parameter values ​​of the artificial intelligence model so that the average of the calculated likelihood values ​​is maximized. The processor (130) may repeatedly perform learning until the average of the calculated likelihood values ​​converges to a preset constant value or more, and may store the performed learning model in the memory (120).

[0077] When performing learning using an artificial intelligence model, learning results can vary with each learning cycle. To address this learning variation, the processor (130) can repeatedly perform learning and determine the final parameter values ​​for sales prediction based on the average value of the learning model.

[0078] Meanwhile, if additional variable data related to the price or sales volume of a product is input through the interface (110), the processor (130) can adjust each data by reflecting each additional variable data to the data for the first unit period and the data for the second unit period, respectively. As described above, the additional variable data may include at least one of an exchange rate, a price index, an unemployment rate, a currency supply, an interest rate, a product supply amount, a product inventory amount, a product production cost, and whether a product is in a peak season. The processor (130) can train an artificial intelligence model using the data adjusted by reflecting the additional variable data.

[0079] The processor (130) verifies the learned artificial intelligence model using data from the first unit period among the classified data, and retrains the artificial intelligence model based on the verification result data. In this case, the processor (130) may remove data from the data from the first unit period up to a predetermined threshold period based on the release dates of each of the multiple products. Since the specific method for determining the threshold period has been described in the above section, a redundant description will be omitted.

[0080] Taking the CNP model illustrated in FIG. 4 as an example, as described above, the processor (130) inputs data of the second unit period into the encoder (410) to train the artificial intelligence model, and transmits the trained encoding vector to the decoder (420). The processor (130) inputs data of the first unit period into the decoder (420) of the trained artificial intelligence model to obtain sales status prediction data for products of the first unit period, and compares the obtained sales status prediction data with actual sales amount data of the first unit period stored in the memory (120) to verify the prediction performance of the artificial intelligence model.

[0081] The processor (130) can repeatedly perform a verification process of an artificial intelligence model using data for a first unit period for a plurality of hyperparameters, and can calculate hyperparameters of the artificial intelligence model based on the verification result data. Here, hyperparameters represent variables that are directly set to control the learning process when learning a model. While parameters of a general learning model are variables that are adjusted by the model during the learning process, such as weights or biases, hyperparameters represent external variables that are set to control the learning model. For example, hyperparameters may include at least one of the number of layers of an artificial neural network model, the number of hidden units, the size of a learning rate, the number of epochs, the batch size, and the number of neurons.

[0082] Layers process the input data of an artificial intelligence model and extract output data. In the case of the CNP model in Figure 4, layers are included in the fully connected neural network (FCNN) used in the encoder and decoder. The learning rate determines the degree to which the learning model adjusts parameters at each learning step, the epoch determines how many times the learning data is repeated during learning, and the batch size determines the amount of learning data used in a single learning step.

[0083] The processor (130) can verify an artificial intelligence model based on the parameters of each of the multiple products, and can adjust the learning model by performing retraining. For example, the processor (130) can verify an artificial intelligence model based on at least one of the initial sales volume, launch date, and number of initial launch data for each of the multiple products.

[0084] The processor (130) may repeatedly perform retraining while adjusting the hyperparameters of the artificial intelligence model and the use of variables such as the number of initial release data without first applying a filter for the product's initial sales volume and release date, and may determine the final learning model based on the results of the learning. In the case where there are multiple models with similar performance during the iterative learning process, the processor (130) may determine the model with the smaller initial release data value as the final learning model.

[0085] The processor (130) may repeatedly perform retraining while changing the initial sales volume value for the hyperparameters of the learning model determined based on the initial data variables of the product launch, and determine the final learning model based on the learning performance results. The processor (130) may set the case where there is no filter for the initial sales volume as the initial sales volume baseline, and may sequentially increase the initial sales volume value from 1 to check the change in performance for the verification result data. In this case, the processor (130) may determine the initial sales volume value that exhibits the maximum performance based on the results of the repeatedly performed retraining, and if there are multiple learning models with similar performance, may determine the learning model with the smaller initial sales volume value as the final learning model. If the parameters determined as the learning results do not have higher performance than the initial sales volume baseline, the processor (130) may perform retraining without applying the used parameters. For example, if the initial sales amount value is determined to be not higher than the reference value based on the relearning result, the processor (130) may perform relearning without setting a filter for the determined initial sales amount value.

[0086] The processor (130) may repeatedly perform retraining while changing the product release time for hyperparameters determined based on the product's initial release data and initial sales volume variables, and determine a final learning model based on the learning performance results. In this case, the processor (130) may set a case in which the learning model does not have a filter for the product's release time as a release time baseline, and may check the change in model performance for the learning data while decreasing the release time from the second release time after the first release time to the first release time. The processor (130) may determine the release time that exhibits the maximum performance based on the performance change of the verified model. If there are multiple models with similar performance, the processor (130) may determine the learning model with a larger release time value as the final learning model. In addition, if the parameters determined as learning results do not have higher performance than the release time baseline, the processor (130) may perform retraining without applying the used parameters.

[0087] The processor (130) may input additional variable data related to the price or sales volume of a product through the interface (110), and when data for a second unit period adjusted to reflect each additional variable data is learned in the artificial intelligence model, the artificial intelligence model may be retrained using data for the first unit period that has been adjusted.

[0088] When data on a target product is input via the interface (110), the processor (130) uses the input data and a retrained artificial intelligence model to provide a sales forecast result for the target product. In this case, the data on the target product may include price and sales data from the early stage of the product's launch.

[0089] The processor (130) can predict the sales results of a target product using sales data of the target product, additional variable data, and an artificial intelligence model. For example, in the case of the CNP model, the processor (130) inputs the sales data of the target product and each additional variable data into a decoder to obtain the output value of the artificial intelligence model, and can predict the sales distribution of the target product based on the output value. The processor (130) can obtain the sales distribution of the target product using [Mathematical Formula 1] below.

[0090]

[0091] In [Equation 1], is a task represents the dataset, and x is the task For input data, Y represents the output data to be predicted in response to the input data. For example, x can be the price data of a product, and Y can be the sales volume data of the product. In this case, It takes two input data and internal parameters In combination with, a prediction can be made on the Y value. The processor (130) can predict a single value of the Y value, and can also predict the probability distribution of the Y value.

[0092] For example, in the case of the CNP learning model, the processor (130) inputs a dataset to the encoder. By inputting, average, sum, element-wise min / max operations, etc. can be performed, and then a single vector can be generated. The processor (130) can predict output data (e.g., sales data of the target product) corresponding to input data (e.g., price data of the target product) through a decoder. In [Mathematical Formula 1] may correspond to a decoder.

[0093] The encoder can compress the dataset secured so far and stored in the memory (120) into a single latent representation vector. The processor (130) can directly compress the dataset stored in the memory (120) into a single vector through the encoder's parameters without an external device. Alternatively, if the electronic device (100) is connected to an external device, the processor (130) can update the vector by reflecting a new dataset received from the external device. When new input data (e.g., price condition data of a target product) comes in based on the latent representation vector compressed by the encoder, the processor (130) can use the decoder to predict a single value or expected distribution of the corresponding output data (e.g., sales volume data of the target product).

[0094] In this case, the sales distribution of the target product can be expressed as in [Mathematical Formula 2] below.

[0095]

[0096] In [Equation 2], are the parameters of the artificial intelligence model learned through the learning and relearning process, represents a compressed encoding vector by inputting price and sales data up to point k into the encoder. After learning is complete, it becomes a fixed value when predicting the sales volume of the target product. is the price data of the target product input to the decoder, represents additional variable data of the target product input to the decoder. It represents the sales volume of the target product produced through the learning results of the artificial intelligence model.

[0097] In this way, the electronic device according to various embodiments of the present disclosure can calculate the distribution of expected sales corresponding to the price conditions when the price conditions of the target product are input, as shown in [Mathematical Formula 2]. In addition, based on additional variable data, changes in sales due to peak seasons such as Black Friday and holidays, external economic indicators, and changes in SCM data can be reflected in the distribution of expected sales. In addition, the electronic device according to various embodiments of the present disclosure can calculate the distribution of expected sales according to the parameter, as shown in [Mathematical Formula 2]. Since it is fixed when predicting the sales volume of the target product, even if new learning data is input, new data can be added and reflected directly to the learned model without having to perform separate retraining.

[0098] When additional variable data and price and sales volume data of the latest product belonging to the first unit period or the second unit period are input through the interface (110), the processor (130) can compress the input data in the encoder of the artificial intelligence model to obtain an updated encoding vector.

[0099] The processor (130) may use an artificial intelligence model to calculate a predicted sales volume distribution from the price data of the target product, and may also use the sales volume distribution to calculate an expected value of the predicted sales volume. By repeatedly performing the process of calculating the expected value of the predicted sales volume while varying the price data, the processor (130) may calculate the price elasticity at each price point using the following [Mathematical Formula 3], which calculates the price elasticity at each price point.

[0100]

[0101] In [Mathematical Equation 3], E represents price elasticity, Q represents sales volume data, and P represents price data. represents the difference in sales volume (or fluctuation in sales volume) between two preset points in time, It represents the price difference (or price fluctuation) between two predetermined points in time.

[0102] [Mathematical expression 3] can be transformed into [Mathematical expression 4] below.

[0103]

[0104] In [Equation 4], is the conditional price data of the target product, represents the expected value of the predicted sales volume of the target product.

[0105] The processor (130) calculates the price difference between two points in time ( ) is reduced very small and [Equation 4] is calculated, so that the given price ( ) can be used to calculate the price elasticity value. In addition, the processor (130) By greatly increasing [Equation 4], the average price elasticity value in the corresponding section can be calculated.

[0106] As such, electronic devices according to various embodiments of the present disclosure can flexibly predict price elasticity based on the predicted sales volume distribution once the conditional price for the target product is determined. Furthermore, since the learned model does not significantly change even when new training data is input, sales volume of the target product can be reliably predicted. Electronic devices according to various embodiments of the present disclosure can accurately predict sales volume of the target product even when initial data on the target product is insufficient. Furthermore, because training is performed by collecting data on prices and sales volumes of various products within the same category, sales volume of the target product can be quickly predicted even when the target product changes.

[0107] [Table 1] below shows the results of predicting product sales volume compared to linear regression and Gaussian process when initial data for the target product is small.

[0108] The number of initial data used for the linear regression Gaussian Process prediction of this invention (j value) 66 10 14 6 10 14 weighted MAPE (%) 28.7 12 23.5 6 6 0.5 14 7.44 10 1.1 15 2.0 9 4 6.11

[0109] When examining the results of calculating MAPE (Mean Absolute Percentage Error) using the sales of products as a weight for the sales volume prediction results of the target product, it can be confirmed that the electronic device according to the present disclosure produces more accurate and stable prediction results than the results predicted through linear regression and Gaussian process.

[0110] The processor (130) may provide the prediction results in various ways. For example, the processor (130) may provide the sales status prediction results by transmitting them to an external device, such as a user's mobile phone or other terminal device, through a communication interface among the interfaces (110).

[0111] Alternatively, if the electronic device (100) includes a display, the processor (130) may display a UI screen indicating the sales status prediction result.

[0112] The processor (130) can visualize the sales distribution of the target product in the form of a table, chart, picture, etc. and display it on the display. The display can be implemented as a display including a self-luminous element or a display including a non-luminous element and a backlight. For example, it can be implemented as various types of displays such as an LCD (Liquid Crystal Display), an OLED (Organic Light Emitting Diodes) display, an LED (Light Emitting Diodes), a micro LED, a Mini LED, a PDP (Plasma Display Panel), a QD (Quantum dot) display, a QLED (Quantum dot light-emitting diodes), etc.

[0113] According to various embodiments, the electronic device (100) may not include a display and may be used in connection with an external display device. In this case, the electronic device (100) may configure a UI screen including predicted sales data for a target product, and transmit the data and a control signal for displaying the data to the external display device.

[0114] The electronic device (100) may also provide prediction results through a small display or speaker that can display only simple information such as text information.

[0115] FIG. 5 is a diagram illustrating a sales status prediction result according to various embodiments of the present disclosure on a display. In FIG. 5, the x-axis represents the log price of the target product, and the y-axis represents the log sales volume of the target product.

[0116] According to FIG. 5, the processor (130) can display the sales distribution of the target product predicted using the artificial intelligence model on the display. For example, as shown in FIG. 5, the processor (130) can display the average value (510) of the sales distribution and the 95% confidence interval (520) excluding the upper and lower 2.5% ranges on the display. In addition, the processor (130) can adjust additional variable data and display the predicted result separately on the display. FIG. 5 shows the predicted result for the peak season displayed separately with a dotted line. For example, as shown in FIG. 5, the processor (130) can display the average value (530) of the sales distribution reflecting the additional variable data and the 95% confidence interval (540) reflecting the additional variable data on the display.

[0117] Although Fig. 5 illustrates a case where the distribution of sales volume by price of a target product is displayed in a graph format, the present invention is not necessarily limited to this, and the prediction result data may be displayed in various forms. For example, if a user inputs a product name and price, the processor (130) may directly display the sales volume as a number. Alternatively, if the user inputs a target sales volume, the processor (130) may directly display the product price that must be determined to sell that target volume. Based on data such as that shown in Fig. 5, the processor (130) may detect a numerical value corresponding to the product price or sales volume.

[0118] FIGS. 6 to 9 are flowcharts for explaining a method of controlling an electronic device according to various embodiments of the present disclosure.

[0119] According to FIG. 6, an electronic device receives data on prices and sales volumes of multiple products (S610). In this case, the electronic device may receive data on prices and sales volumes in the form of time-series data. The electronic device may also receive additional variable data about the products. The additional variable data may include variable data related to peak sales seasons, such as year-end, holidays, new semesters, Christmas, and Black Friday. In addition, the additional variable data may include variable data related to economic indicators, such as exchange rates, price indices, unemployment rates, money supply, and interest rates, variable data related to product demand, product supply, product inventory, product production costs, and SCM (supply chain management) data.

[0120] The electronic device classifies the input data based on the release dates of each of the multiple products (S620). In this case, the electronic device can remove data from the input data up to a predetermined threshold period based on the release dates of each of the multiple products, and classify data after the threshold period based on the release dates of each of the multiple products.

[0121] An electronic device may filter price and sales data for multiple products based on at least one of the initial sales volume, release date, and number of initial release data for each of the multiple products, and may also classify the selected data based on the release date of each product. For example, the electronic device may remove products whose initial sales volume is below a preset threshold based on the release date of each product. The electronic device may also remove data for products whose release date is after a preset threshold or data for products whose release date is too old. The electronic device may also filter and remove data for products whose number of initial release data is below a preset threshold, and classify the selected data based on the release date of each product.

[0122] In Fig. 6, data is illustrated and described as being directly classified by an electronic device, but this is not necessarily limited to the electronic device, and the electronic device may also receive data classified by another device.

[0123] The electronic device acquires data from a first unit period set based on the learning time point among the classified data and a second unit period prior to the first unit period (S630). For example, if the target product is scheduled to be released in 2023, the data for the first unit period can be set to data for products released in 2022, and the data for the second unit period can be set to data for products released before 2021, thereby classifying the input data.

[0124] The electronic device trains an artificial intelligence model using data from a second unit period (S640). The electronic device may train the artificial intelligence model using at least one of the price and sales data for each product and additional variable data. In this case, the electronic device may also train the artificial intelligence model by incorporating additional variable data that does not change over time into the price and sales data.

[0125] The electronic device retrains the learned artificial intelligence model through verification using data for a first unit period (S650). The electronic device can input data for the first unit period into the learned artificial intelligence model using data for a second unit period to obtain a sales distribution for products for the first unit period. The electronic device can verify the performance of the artificial intelligence model by comparing the sales data for the first unit period obtained through the artificial intelligence model with actual sales data for products for the first unit period. In this case, if the sales data for the first unit period obtained through the artificial intelligence model deviates from a preset confidence interval based on the verification result data, the electronic device can adjust the data for the first unit period to perform retraining.

[0126] Although FIG. 6 illustrates and describes the training and retraining of an artificial intelligence model as being performed entirely on an electronic device, this is not necessarily limited to the present invention, and the training of an artificial intelligence model may be performed on other electronic devices. That is, the electronic device may perform a retraining operation on an artificial intelligence model trained using data from a second unit period through verification using data from a first unit period.

[0127] When data regarding a target product is input, the electronic device uses the input data and the retrained artificial intelligence model to provide a predicted value corresponding to the target product (S660). Based on the retraining results of the artificial intelligence model, if the sales data for the first unit period acquired through the artificial intelligence model satisfies a preset confidence interval, the electronic device may input at least one of the price conditions and additional variable data regarding the target product into the retrained artificial intelligence model to provide a predicted sales status result for the target product. In this case, the electronic device may visualize the predicted value in the form of a table, chart, figure, or the like and display it on a display.

[0128] Figure 7 illustrates an example of a method for determining learning selection criteria for selecting data.

[0129] According to FIG. 7, an electronic device can obtain data on the relationship between log prices and log sales volumes based on data on prices and sales volumes of multiple products (S710). Furthermore, the electronic device can perform linear regression analysis on the sales volumes of each of the multiple products during their respective sales periods based on the obtained data to obtain a linear regression line representing the price elasticity of each product (S720). In this case, the electronic device can also set a key sales period based on the sales volume of the products to increase the accuracy of price elasticity, and perform linear regression analysis on the log prices and log sales volume data during the set key sales period. Since the specific linear regression line has been described in the above-mentioned section, a redundant description will be omitted.

[0130] The electronic device can obtain the entry point at which the deviation of each sales volume from the linear regression line falls within a preset confidence interval (S730). In this case, the electronic device can use an artificial intelligence model trained on learning data for a second unit period to predict the sales volume of each of multiple products, and compare the predicted sales volume with the product sales volume for the input second unit period to obtain the deviation.

[0131] The electronic device can determine the entry period between the launch date and the entry point as the critical period (S740). The electronic device can set the critical period based on the learning selection criteria described above. Specifically, the electronic device can remove data up to the critical period from the input data of multiple products based on the launch date of each product, and then classify the selected data after the critical period based on the launch date of each of the multiple products.

[0132] Figure 8 is a flowchart illustrating another method for determining learning selection criteria.

[0133] According to FIG. 8, an electronic device can obtain data on the relationship between log prices and log sales volumes based on data on prices and sales volumes of a plurality of products (S810). Furthermore, the electronic device can perform linear regression analysis on the sales volumes of each of the plurality of products during a sales period based on the obtained data to obtain a linear regression line representing the price elasticity of each product (S820). Furthermore, the electronic device can obtain an entry point at which the deviation of each sales volume based on the linear regression line enters within a preset confidence interval (S830). In FIG. 8, the step of obtaining data on the relationship between log prices and log sales volumes (S810) to the step of obtaining an entry point at which the deviation enters within the confidence interval (S830) may be performed in the same manner as the step of obtaining data on the relationship between log prices and log sales volumes (S710) of FIG. 7 to the step of obtaining an entry point at which the deviation enters within the confidence interval (S730).

[0134] An electronic device can repeatedly perform steps (S810) to (S830) using data on prices and sales volumes of multiple products. Based on the result data obtained through the repeated performance, the electronic device can obtain the number of products with the same entry period between the launch date and the entry point of each product for each entry period (S840).

[0135] The electronic device can determine the entry period for a predetermined percentage of the total number of products, calculated by adding up the number of products for each entry period, as the threshold period (S850). Once the threshold period is determined, the electronic device can remove data from the price and sales data of multiple products entered, based on the release dates of each of the multiple products, up to the threshold period determined based on the release dates of each of the multiple products. The electronic device can then classify data after the threshold period based on the release dates of each of the multiple products.

[0136] Figure 9 is a flowchart illustrating a method for learning an artificial intelligence model.

[0137] According to FIG. 9, the electronic device can receive additional variable data related to the price or sales volume of the product (S910).

[0138] When additional variable data is input, the electronic device can adjust each data by reflecting the additional variable data in the data for the first unit period and the data for the second unit period, respectively (S920). For example, the electronic device can receive data on the price and sales volume of a product to which each additional variable data is reflected, classify the data by each additional variable data, and store it in memory. In this case, the data on the price and sales volume of a product to which the additional variable data is reflected may exhibit a different sales volume distribution than the data for a product to which the additional variable data is not reflected. As shown in Fig. 5, the average value of the sales volume distribution to which the additional variable data is reflected may be higher than the average value of the sales volume distribution to which the additional variable data is not reflected, or conversely, may exhibit a lower average value.

[0139] The electronic device can train an artificial intelligence model using data from the adjusted second unit period (S930). For example, the electronic device can train an artificial intelligence model by reflecting additional variable data for each period corresponding to each additional variable data in the data from the second unit period.

[0140] The electronic device may retrain the artificial intelligence model using data for the adjusted first unit period (S940). The electronic device may train the artificial intelligence model by inputting price conditions or additional variable conditions for the data for the adjusted first unit period into the trained artificial intelligence model by reflecting additional variable data. In this case, the electronic device may perform training using a model corresponding to the additional variable conditions for the data for the first unit period. The electronic device may compare the training result using the data for the first unit period with the actual data for the first unit period stored in the memory, and verify the performance of the artificial intelligence model. If the performance is not satisfactory based on the verification result of the artificial intelligence model (for example, if the sales volume distribution obtained using the artificial intelligence model falls outside a preset confidence interval), the electronic device may retrain the artificial intelligence model using data for the adjusted second unit period. Here, the additional variable data may include at least one of an exchange rate, a price index, an unemployment rate, a currency supply, an interest rate, a product supply amount, a product inventory amount, a product production cost, and whether the product is in a peak season.

[0141] The various methods described in FIGS. 6 to 9 can be performed by an electronic device having the configuration of FIG. 1, but are not necessarily limited thereto, and may be performed by an electronic device having at least some configurations different from or added to the configuration.

[0142] Additionally, in the various embodiments described above, the unit period is divided into two, such as a first unit period and a second unit period, and data for each unit period is used for learning or relearning. However, this is not necessarily limited to this. For example, the unit period may be set to a single unit period, or it may be subdivided into three or more units.

[0143] While various embodiments have been described above, these embodiments need not necessarily be implemented individually and may be combined with other embodiments. That is, at least some of the embodiments described above may be combined, in whole or in part, with other embodiments and implemented together in a single device.

[0144] Furthermore, the above-described embodiments described training an AI model using data on product price and sales volume, as well as additional variable data, to predict sales results for a target product. However, in addition to product price, product specifications can also influence sales volume. Therefore, sales volume data based on various specifications, such as product size and function, can also be used for training or prediction.

[0145] In addition, although the above-described embodiments describe an electronic device independently training an artificial intelligence model and using the same to predict sales results, at least some steps of these operations may be linked with other devices. For example, if the electronic device (100) is implemented as a TV and connected to an external server device, the electronic device (100) may receive data on the price or sales volume of a target product from a user and transmit the data to the external server device. The external server device may train an artificial intelligence model and store the data in advance in the same manner as described in the various embodiments described above. When the external server device receives information on the target product from the electronic device (100), the external server device may use the information and the artificial intelligence model to predict the sales results of the target product. The external server device may transmit the prediction result to the electronic device (100), and the electronic device (100) may display the prediction result.

[0146] Alternatively, the electronic device (100) may directly collect data required for learning an artificial intelligence model from an external server device, transmit the data to the external server device, and use the data to perform learning. Additionally, the electronic device (100) and the external server device may be interconnected in various ways to implement the above-described embodiments.

[0147] Meanwhile, according to an embodiment of the present disclosure, the various embodiments described above may be implemented as software including instructions stored in a machine-readable storage medium that can be read by a machine (e.g., a computer). The device may include an electronic device according to the disclosed embodiments, as a device that can call instructions stored in the storage medium and operate according to the called instructions. When an instruction is executed by a processor, the processor may directly or under the control of the processor perform a function corresponding to the instruction using other components. The instruction may include code generated or executed by a compiler or interpreter. The machine-readable storage medium may be provided in the form of a non-transitory storage medium. Here, 'non-transitory' means that the storage medium does not contain a signal and is tangible, but does not distinguish between data being stored semi-permanently or temporarily in the storage medium.

[0148] Furthermore, according to one embodiment of the present disclosure, the method according to the various embodiments described above may be provided as included in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., compact disc read-only memory (CD-ROM)) or online through an application store (e.g., Play Store™). In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily generated in a storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.

[0149] In addition, each of the components (e.g., modules or programs) according to the various embodiments described above may be composed of a single or multiple entities, and some of the corresponding sub-components described above may be omitted, or other sub-components may be further included in various embodiments. Alternatively or additionally, some components (e.g., modules or programs) may be integrated into a single entity, which may perform the same or similar functions as those performed by each of the corresponding components prior to integration. Operations performed by modules, programs or other components according to various embodiments may be executed sequentially, in parallel, iteratively or heuristically, or at least some operations may be executed in a different order, omitted, or other operations may be added.

[0150] In this way, the electronic device and the control method thereof according to various embodiments of the present disclosure can train an artificial intelligence model using data on prices and sales volumes of multiple products belonging to a single category, and predict the price elasticity of the target product and the relationship between price and sales volume based on the training results. In addition, the electronic device and the control method thereof according to various embodiments of the present disclosure can enhance the accuracy of the training model by reflecting additional variable data, such as price, whether it is a peak season, changes in economic indicators, and changes in supply chain variables, which affect product sales volume.

[0151] Although the preferred embodiments of the present disclosure have been illustrated and described above, the present disclosure is not limited to the specific embodiments described above, and various modifications may be made by a person having ordinary skill in the art to which the present disclosure pertains without departing from the gist of the present disclosure as claimed in the claims, and such modifications should not be understood individually from the technical idea or prospect of the present disclosure.

Claims

1. In electronic devices, Interface for entering data on prices and sales volumes of multiple products; Memory for storing artificial intelligence models; and Processor; including; The above processor, Based on the release date of each of the above multiple products, the data of the first unit period set based on the learning time and the second unit period prior to the first unit period are acquired, The artificial intelligence model trained using data during the second unit period is retrained through verification using data during the first unit period. An electronic device that provides a prediction value corresponding to the target product by using the input data and the re-learned artificial intelligence model when data for the target product is input through the interface.

2. In paragraph 1, The above processor, An electronic device, wherein when data on the prices and sales volumes of the above-mentioned plurality of products are input through the interface, data up to a critical period based on the release date of each of the above-mentioned plurality of products are removed, and data after the time is classified based on the release date of each of the above-mentioned plurality of products.

3. In paragraph 2, The above processor, An electronic device, which obtains data on the relationship between log price and log sales volume based on data on the prices and sales volume of the plurality of products, performs linear regression analysis on the sales volume during the sales period of each of the plurality of products based on the obtained data, obtains a linear regression line representing the price elasticity of each product, identifies an entry point at which the deviation of each sales volume based on the linear regression line enters a preset confidence interval, and determines the entry period from the launch time to the entry point as the critical period.

4. In paragraph 3, The above processor, An electronic device that uses the artificial intelligence model learned based on learning data during the second unit period to predict the sales amount of each of the plurality of products, and obtains the deviation by comparing the predicted sales amount and the input sales amount.

5. In paragraph 2, The above processor, Based on the data on the prices and sales volumes of the above multiple products, data on the relationship between the log price and the log sales volume is obtained, and based on the obtained data, a linear regression analysis is performed on the sales volume during the sales period of each of the above multiple products to obtain a linear regression line representing the price elasticity of each product, and an entry point at which the deviation of each sales volume enters within a preset confidence interval based on the linear regression line is identified. An electronic device that identifies the number of products with the same entry period from the launch time to the entry point for each entry period, and determines the entry period of a number of products corresponding to a preset ratio based on the total number of products calculated by adding up the number of products for each entry period as the threshold period.

6. In paragraph 1, The above processor, When additional variable data related to the price or sales volume of the product is input through the interface, the additional variable data is reflected in the data for the first unit period and the data for the second unit period, respectively, to adjust each data, train the artificial intelligence model using the adjusted data for the second unit period, and retrain the artificial intelligence model using the adjusted data for the first unit period. An electronic device wherein the above additional variable data includes at least one of an exchange rate, a price index, an unemployment rate, a money supply, an interest rate, a product supply quantity, a product inventory quantity, a product production cost, and whether the product is in a peak season for sales.

7. In paragraph 6, The above processor, An electronic device that selects the price and sales data of the plurality of products based on at least one of the initial sales volume, release date, and number of initial release data of each of the plurality of products, and classifies the selected data based on the release date of each product.

8. In paragraph 1, The above processor, The hyperparameters of the artificial intelligence model relearned using the data during the first unit period are calculated, The above hyperparameters are, An electronic device including at least one of the number of layers of an artificial neural network model, the number of hidden units, the size of a learning rate, and the number of epochs.

9. In a method for controlling an electronic device, A step of entering data on prices and sales volumes of multiple products; A step of classifying the above data based on the launch date of each of the plurality of products; A step of acquiring data of a first unit period set based on a learning time point among the classified data and a second unit period prior to the first unit period; A step of training an artificial intelligence model using data during the second unit period; A step of retraining the artificial intelligence model through verification using data during the first unit period; and A control method, comprising: a step of providing a predicted value corresponding to the target product by using the input data and the re-learned artificial intelligence model when data for the target product is input.

10. In paragraph 9, A step of removing data up to a critical period based on the release date of each of the above multiple products; and A control method further comprising a step of classifying data after the above time based on the release time of each of the plurality of products.

11. In paragraph 10, The step of removing data up to the above critical period is: A step of obtaining data on the relationship between log price and log sales volume based on data on the prices and sales volume of the plurality of products; A step of performing a linear regression analysis on the sales volume during the sales period of each of the plurality of products based on the acquired data to obtain a linear regression line representing the price elasticity of each product; A step of obtaining an entry point at which the deviation of each sales amount based on the linear regression line enters within a preset confidence interval; and A control method, comprising: a step of determining an entry period between the launch time and the entry time as the critical period; 12. In paragraph 10, The step of removing data up to the above critical period is: A step of obtaining data on the relationship between log price and log sales volume based on data on the prices and sales volume of the plurality of products; A step of performing a linear regression analysis on the sales volume of each of the plurality of products during the sales period based on the acquired data to obtain a linear regression line representing the price elasticity of each product; A step of obtaining an entry point at which the deviation of each sales amount based on the linear regression line enters within a preset confidence interval; A step of obtaining the number of products with the same entry period between the launch time and the entry time for each entry period; and A control method, comprising: a step of determining an entry period for a number of products corresponding to a preset ratio based on the total number of products calculated by adding up the number of products for each entry period as the threshold period; 13. In paragraph 9, A step of entering additional variable data related to the price or sales volume of the above product; A step of adjusting each data by reflecting the additional variable data to the data for the first unit period and the data for the second unit period, respectively; A step of training the artificial intelligence model using data during the adjusted second unit period; and Further comprising a step of retraining the artificial intelligence model using data during the first unit period that has been adjusted; A control method wherein the above additional variable data includes at least one of an exchange rate, a price index, an unemployment rate, a money supply, an interest rate, a product supply quantity, a product inventory quantity, a product production cost, and whether the product is in a peak season for sales.

14. In paragraph 9, The step of classifying the above data based on the release date of each of the above multiple products is: A step of selecting the price and sales data of the plurality of products based on at least one of the initial sales volume, launch date, and number of initial launch data of each of the plurality of products; and A control method, comprising: a step of classifying selected data based on the launch date of each product; 15. A computer-readable recording medium including a program for executing a method for controlling an electronic device, The above control method is, A step of entering data on prices and sales volumes of multiple products; A step of classifying the above data based on the launch date of each of the plurality of products; A step of acquiring data of a first unit period set based on a learning time point among the classified data and a second unit period prior to the first unit period; A step of training an artificial intelligence model using data during the second unit period; A step of retraining the artificial intelligence model through verification using data during the first unit period; and A computer-readable recording medium, comprising: a step of providing a predicted value corresponding to the target product by using the input data and the re-learned artificial intelligence model when data for the target product is input.

Images