System for calculating fair prices and predicting price fluctuations through AI-based watch trading data analysis

Abstract

The present invention relates to a system for calculating a fair price and predicting price fluctuations through AI-based analysis of watch trading data. The system may include a transaction data collection module for collecting transaction data, a data preprocessing module for separating abnormal transaction data, a price analysis module for generating feature vectors and performing AI learning and AI inference, and a service provision module for providing watch valuation information and notifications. The price analysis module may generate feature vectors from preprocessed transaction data, train an AI model using actual transaction results, and calculate a fair price correction value, a probability of price increase, a probability of price decrease, and a probability of price maintenance using the trained model. Furthermore, the system may derive a final calculated price by combining a relationship-based primary calculated price with AI inference results, and control whether to trigger a notification by considering volatility indicators, price fluctuation probabilities, and the battery status of the user terminal unit. Accordingly, the present invention can perform fair price calculation and price fluctuation prediction reflecting complex patterns in watch trading data.

Description

Technology Field

[0001] The present invention relates to a system for calculating the fair price of a watch and predicting price fluctuations through AI-based transaction data analysis. More specifically, the invention relates to a system for calculating the fair price and predicting price fluctuations through AI-based watch transaction data analysis that dynamically adjusts the data collection cycle according to server resources and communication status, filters abnormal transactions based on the physical condition of the watch and seller reliability, generates feature vectors from preprocessed transaction data, calculates fair price correction values ​​and price fluctuation probability values ​​through AI learning and AI inference using actual transaction history, and provides valuation information and notifications by reflecting market scarcity, user response, and the hardware status of the terminal. Background Technology

[0002] Recently, the second-hand market for high-end watches, which are recognized and traded as a type of asset, has been growing rapidly. Even for the same model, the transaction price of a second-hand watch varies significantly depending on the year of manufacture, condition, and the presence or absence of accessories, and scarcity value based on the quantity of items available in the market acts as a major factor in determining the price.

[0003] Conventional pricing methods are limited to merely presenting average transaction prices within a specific period, failing to immediately reflect the supply and demand imbalances in the rapidly changing market. Furthermore, problems frequently arise where the reliability of calculated fair prices is low because these methods fail to effectively filter out abusive practices—such as artificially inflating view counts to manipulate the popularity of specific models—or bait listings where items are falsely registered without actual possession.

[0004] Meanwhile, even for watches of the same brand or model, the price formation structure can vary non-linearly depending on the manufacturing date, condition of preservation, availability of accessories, transaction history, recent market interest, and response to external events. Such complex transaction patterns are difficult to adequately reflect using fixed average calculation methods or single rule-based computations alone, and price judgment criteria need to be updated as new valid transaction data accumulates. Therefore, the introduction of an AI-based analysis structure capable of extracting feature values ​​from transaction data and learning from actual transaction results to calculate fair price correction values ​​and price fluctuation probabilities is required.

[0005] Furthermore, regarding the transmission of price fluctuation alerts to user terminals, issues concerning hardware control are continuously being raised, as the indiscriminate sending of alerts whenever minute market fluctuations occur excessively consumes the terminal's battery resources and increases user fatigue. Therefore, there is a need to introduce an information delivery mechanism that ensures the stability of system resources while providing highly accurate clock value information and taking into account the hardware status of the terminal. The problem to be solved

[0006] The main problem that the present invention aims to solve is as follows.

[0007] The first objective is to secure the latest market data without omission while maintaining system stability by detecting server computational load and network latency in real time and dynamically adjusting the data collection cycle.

[0008] The second task is to enhance the consistency of analysis data by linking physical depreciation factors, such as scratches on watches or the presence of accessories, to price deviation analysis and comprehensively evaluating the seller's past history, thereby clearly distinguishing between legitimate low-priced listings and fraudulent bait listings.

[0009] The third task is to derive an objective, real-time fair transaction price by calculating the scarcity value based on the remaining supply quantity in the market and the ratio of user views to the year of manufacture, while applying log relationships to defend against price distortions caused by specific manipulation activities.

[0010] The fourth task is to establish a hardware control system that efficiently manages the energy consumption of the terminal while not missing important price fluctuation information by organically combining price volatility indicators and the remaining battery level of the user terminal to adjust the notification transmission threshold.

[0011] The fifth task is to construct input data suitable for AI training and AI inference by generating a feature vector from preprocessed transaction data that includes brand, model name, manufacturing date, condition grade, physical depreciation value, remaining listing quantity, user responsiveness, bid price information, and external event information.

[0012] The 6th task is to enable complex transaction patterns that are difficult to reflect solely by relational-based primary calculated prices to be reflected in the form of fair price correction values ​​by training an AI model using actual normal transaction history and price trend information.

[0013] Task 7 is to calculate the probability of a price increase, a price decrease, and a price maintenance within a preset prediction period as AI inference results, and to combine said results with volatility indicators to control alert levels more precisely.

[0014] The 8th task is to maintain the stability of system operation by calculating the reliability of the AI ​​inference results together, applying the result based on the relationship first when the inference reliability is below a preset threshold value, and reflecting the AI ​​correction result only when the inference reliability is above the threshold value.

[0015] Task 9 is to enable the AI ​​model to adapt to changes in transaction data even during long-term operation by retraining and updating it in response to the accumulation of new normal transaction data or an increase in prediction errors.

[0016] The problems of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0018] According to an embodiment of the present invention for solving the above problem

[0019] A user terminal unit that transmits and receives data through a communication network; and

[0020] Includes a server unit that computes and processes data related to watch trading;

[0021] The above server unit is,

[0022] A transaction data collection module that collects transaction data including watch transaction price, transaction time, watch brand, model name, condition grade, and transaction frequency information from multiple communication channels;

[0023] A data preprocessing module that normalizes the collected transaction data and identifies and separates abnormal transaction data;

[0024] A price analysis module that receives valid transaction data separated from the above abnormal transaction data, calculates the real-time fair price of the target watch, and predicts future price volatility; and

[0025] A service providing module that controls watch value evaluation information to be transmitted to the user terminal unit and the frequency of notification occurrence based on the analysis results of the price analysis module above; is provided and executed.

[0026] The above price analysis module may further include: a feature generation module that generates a feature vector from preprocessed transaction data; an AI learning module that learns model parameters using the feature vector and actual execution results; an AI inference module that calculates a fair price correction value and a price fluctuation probability value using the learned model parameters; and a model update module that updates the model parameters according to the accumulation of new normal execution data or an increase in prediction error.

[0027] The feature generation module generates a feature vector including at least one of a brand, model name, date of manufacture, condition grade, physical depreciation value, presence of warranty, presence of box, average transaction price, remaining quantity of listings, cumulative views, seller cancellation history, number of buy orders, number of sell orders, number of article distributions, and inflation rate; the AI ​​learning module performs learning using the feature vector and at least one of an actual transaction price, actual transaction completion status, and a result of price increase, decrease, or maintenance after a preset prediction period as label data; the AI ​​inference module calculates a fair price correction value reflected in the relationship-based first calculated price and the probability of price increase, probability of price decrease, and probability of price maintenance within a preset prediction period; and the model update module can control relearning to be performed when new normal transaction data is secured above a preset cumulative standard or when the prediction error exceeds a preset increase standard.

[0028] The above transaction data collection module is,

[0029] To calculate a dynamic collection cycle, which is a time interval for collecting data in response to the volatility of watch transaction frequency in the online market, a control command is automatically transmitted to suppress server resource exhaustion by performing a composite operation of an increase resulting from applying an exponential relationship to the current memory buffer occupancy of the server unit and a decrease resulting from applying a logarithmic relationship to the new transaction post registration frequency of the target watch model, thereby extending the dynamic collection cycle to protect hardware resources when the buffer occupancy increases, and controlling the shortening range to converge gradually and smoothly when the new transaction post registration frequency explodes.

[0030] The above transaction data collection module is,

[0031] Measure the communication delay time with the data collection target server and extend the sum in a linear proportional relationship with the dynamic collection cycle, determine whether the buffer occupancy rate exceeds a preset risk threshold, and if it exceeds, omit the calculation of other variables and immediately transmit an emergency blocking command to fix the dynamic collection cycle to a preset maximum idle period,

[0032] The above risk threshold is set as a dynamic threshold that the system periodically updates by inversely calculating the average buffer occupancy rate at points during a past preset period when the communication delay time of the server unit surged by more than a specific rate compared to normal times, thereby structurally preventing system failure, and

[0033] The above data preprocessing module is,

[0034] Automatically transmits a control command to prevent the deletion of normal price drop data due to physical damage by misidentifying it as abnormal transaction data, by calculating the difference between the collected target transaction price and the average transaction price within a preset period, and extending the normal allowable range of the difference value in an inverse relationship with the physical depreciation figure derived by counting the presence of scratches on the target watch and the number of missing parts through text and image analysis.

[0035] The above data preprocessing module is,

[0036] By counting the number of past abnormal transaction cancellations by the seller who registered the target watch, adding a penalty value proportional to the said cancellation count in a logarithmic manner, and applying an exponential decay relationship to the time interval between the past point in time when the said transaction data occurred and the current data processing point to calculate a final abnormal transaction score that is reduced, a control signal is transmitted to limit the influence of short-term malicious bait listings and block the influence of old noise data.

[0037] The above data preprocessing module is,

[0038] If an exceptional situation is detected where the seller intentionally leaves the text body of the post blank so that the physical condition cannot be determined, the physical depreciation value of the target watch is forcibly assigned to 0, signifying the highest condition, thereby controlling the application of the normal allowable range most strictly.

[0039] If the above-mentioned final abnormal transaction score exceeds the filtering threshold, the above-mentioned target transaction price is isolated, and the consistency of the database can be maintained through a structure in which the filtering threshold is dynamically set by normalizing the average distribution of calculated scores of fraudulent listings detected during a past preset period and tracking the mode interval excluding preset lower noise. Effects of the invention

[0041] The system for calculating fair prices and predicting price fluctuations through AI-based watch trading data analysis according to the present invention can maintain a balance between the updateability of market data and server resource management by adjusting the data collection cycle according to the buffer occupancy rate of the server unit, the frequency of new post registration, and the communication delay time.

[0042] In addition, the present invention can be utilized to distinguish between normal depreciation listings and abnormal transaction data by calculating an abnormal transaction score that reflects both the physical depreciation value of the watch and the seller's cancellation history.

[0043] In addition, the present invention derives a real-time calculated price by reflecting the remaining quantity of listings, cumulative views, and inflation rate together, thereby allowing for a more detailed reflection of the current market situation compared to a simple average transaction price.

[0044] In addition, the present invention allows for the simultaneous delivery of notifications and management of terminal power consumption by adjusting whether to generate a notification by considering the variability indicator and the remaining battery level of the user terminal unit together.

[0045] In addition, the present invention can reflect a transaction pattern that comprehensively reflects brand, year, condition, and market response in the form of a fair price correction value by generating a feature vector from preprocessed transaction data and performing AI learning using actual transaction history.

[0046] In addition, the present invention can be operated in a way that reduces the deviation from the actual transaction price compared to a single average price calculation method by reflecting both the relationship-based primary calculated price and the AI ​​inference-based correction value together.

[0047] In addition, the present invention can provide price fluctuation prediction results in a more detailed form of information by calculating the probability of a price increase, the probability of a price decrease, and the probability of maintaining the price, and determining the alert level by combining the probability values ​​with a volatility indicator.

[0048] In addition, the present invention can update model parameters according to the accumulation of new normal transaction data or an increase in prediction error, thereby enabling the implementation of a continuous analysis structure that responds to changes in the trading environment.

[0049] The effects according to the present invention are not limited to the matters described above and may vary depending on the specific implementation form. Brief explanation of the drawing

[0051] Figure 1 illustrates an integrated flowchart between all components according to the present invention. Figure 2 illustrates a dynamic data collection cycle control flowchart according to the present invention. FIG. 3 illustrates a flowchart of physical state-based abnormal transaction filtering according to the present invention. Figure 4 illustrates a multidimensional fair price calculation evaluation flowchart according to the present invention. FIG. 5 illustrates a price prediction and terminal notification control flowchart according to the present invention. Specific details for implementing the invention

[0052] Hereinafter, various embodiments are described in more detail with reference to the attached drawings. The embodiments described in this specification may be modified in various ways. Specific embodiments may be depicted in the drawings and described in detail in the detailed description. However, specific embodiments disclosed in the attached drawings are intended only to facilitate understanding of various embodiments. Accordingly, the technical concept is not limited by specific embodiments disclosed in the attached drawings, and it should be understood that it includes all equivalents or substitutions that fall within the spirit and scope of the invention.

[0053] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but these components are not limited by the aforementioned terms. The aforementioned terms are used solely for the purpose of distinguishing one component from another.

[0054] In this specification, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. When a component is described as being “connected” or “connected” to another component, it should be understood that it may be directly connected to or connected to that other component, or that there may be other components in between. On the other hand, when a component is described as being “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

[0055] Meanwhile, a "module" or "part" for a component as used in this specification performs at least one function or operation. Furthermore, a "module" or "part" may perform a function or operation by hardware, software, or a combination of hardware and software. Additionally, a plurality of "modules" or a plurality of "parts," excluding a "module" or "part" that must be performed on specific hardware or on at least one processor, may be integrated into at least one module. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0056] In addition, power, power transmission, and control therefor for the following assembly configurations and embodiments, including "by control," follow conventional technology including terminals, applications, hardware control modules, etc., so they are omitted to avoid redundancy.

[0057] In addition, the operation embodiments and configurations described in a general manner without being explained in detail below follow the prior art and are omitted in order to focus on describing the purpose of the present invention and the resulting effects.

[0058] Furthermore, in describing the present invention, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description is abbreviated or omitted.

[0059] Meanwhile, the artificial intelligence (AI)-based analysis described herein may include a series of processing structures that generate feature vectors from preprocessed transaction data, learn model parameters using actual execution history or price trend results, and calculate fair price correction values ​​and price fluctuation probability values ​​using the learned model parameters. Additionally, the AI-based analysis may be performed in parallel with relational equation-based calculation results and may be selectively reflected depending on the reliability of the inference.

[0060] A system for calculating an appropriate price and predicting price fluctuations through AI-based watch trading data analysis according to one embodiment of the present invention may include a server unit (100) that generates data collection, analysis, and control commands, and a user terminal unit (200) that is connected to the server unit (100) via a communication network to visually display information and transmit hardware status data.

[0061] At this time, the server unit (100) may be configured to include a transaction data collection module (110) that obtains watch transaction information from a plurality of external communication channels, a data preprocessing module (120) that reviews whether the collected data is false positive and refines valid data, a price analysis module (130) that generates a value indicator for each watch based on normalized data, and a service provision module (140) that transmits the analysis results to the user terminal unit (200) and adjusts the notification frequency in response to the physical state of the user terminal unit (200).

[0062] Furthermore, the price analysis module (130) may be equipped with a fair price calculation model (131) that derives real-time value by combining remaining market listings and user response, and a price prediction model (132) that derives future price changes by analyzing market bid-ask imbalances and the frequency of external events as sub-components.

[0063] Additionally, the price analysis module (130) may further include a feature generation module (133) that generates a feature vector from preprocessed transaction data, an AI learning module (134) that learns model parameters using the feature vector and actual execution results, an AI inference module (135) that calculates an appropriate price correction value and a price fluctuation probability value using the learned model parameters, and a model update module (136) that updates model parameters based on new normal execution data or changes in prediction errors. At this time, the appropriate price calculation model (131) and the price prediction model (132) may operate in conjunction with the feature generation module (133), the AI ​​learning module (134), the AI ​​inference module (135), and the model update module (136), respectively.

[0064] The server unit (100) acts as the entity for operating the system of the present invention, performs large-scale computation and data processing, and transmits control commands to communicate with multiple terminals. The server unit (100) automatically transmits commands to drive a transaction data collection module, a data preprocessing module, a price analysis module, and a service provision module equipped in an internal storage space through a processor. The server unit (100) preprocesses, analyzes, and predicts the collected watch transaction data, and transmits and sends control signals to control user terminal units based on the derived results. The server unit (100) automatically transmits commands to acquire watch transaction-related data from multiple external communication channels via a wired or wireless communication network and to receive physical hardware status data of the corresponding terminal from the connected terminal. In the event that the amount of computation increases rapidly due to an increase in the trading volume of a specific watch model, the server unit (100) may select a computation node among a plurality of computation nodes provided in advance whose current usage rate is below a preset standard, and distribute at least one computation task among the transaction data collection module (110), the data preprocessing module (120), the price analysis module (130), and the service provision module (140). At this time, the distribution of the computation tasks may be performed based on the length of the task queue, the current usage rate per node, and the memory occupancy rate. The user terminal unit (200) transmits a control signal in which the user communicates with the server unit (100) to request watch value evaluation information and receives the analysis results and notifications. The user terminal unit (200) automatically transmits a command to visually display the appropriate price and volatility graph of the watch on the display in accordance with the control command of the server unit (100). The user terminal unit (200) transmits a control command to output vibration or sound of a push notification or to block the notification in the background to conserve power, depending on the received notification control signal.The above user terminal unit (200) calls the application program interface of the battery management system and operating system provided inside the terminal and automatically transmits a command to obtain current battery remaining percentage data. The above user terminal unit (200), implemented as a portable electronic device such as a smartphone or tablet computer carried by the user, transmits a control signal to delay the discharge of the device by restricting the reception of general warning notifications from the above server unit (100) when the battery remaining amount drops below 15%.

[0065] The transaction data collection module (110) collects transaction data related to watch transactions from multiple communication channels and transmits a control command to dynamically adjust the collection cycle according to the amount of data inflow and hardware load. Based on a preset basic collection cycle, the transaction data collection module (110) calculates a dynamic collection cycle by performing a complex calculation of the exponential relationship regarding the buffer occupancy rate of the server unit (100), the logarithmic relationship regarding the frequency of new transaction posts registered for the target watch model, and the proportional relationship regarding the communication delay time, and automatically transmits a command to control the scheduler. The transaction data collection module (110) obtains the buffer occupancy rate through an operating system resource monitoring tool, obtains the communication delay time by measuring the round-trip time of packets transmitted to an external server, and transmits a control signal to obtain the frequency of post registration through an external channel application program interface. When the transaction registration frequency increases to 500 per 10 minutes due to news of the discontinuation of a popular watch model, the transaction data collection module (110) automatically transmits a control command to reduce the omission of transaction history without causing server load by gradually shortening the collection cycle by reflecting the increase in registration frequency on a logarithmic scale. The data preprocessing module (120) transmits a control signal to identify and filter out false listings and abnormal transaction data that may distort the subsequent price analysis algorithm among the collected transaction data. The data preprocessing module (120) automatically transmits a command to derive the deviation between the target transaction price and the average transaction price and to adjust the allowable deviation range in an inverse relationship with the physical depreciation value. The data preprocessing module (120) calculates an abnormal transaction score after applying a penalty in a logarithmic relationship to the seller's past number of abnormal transaction cancellations and undergoing time depreciation, and if the calculated score exceeds a preset dynamic threshold, transmits a control command to isolate the corresponding data.

[0066] At this time, the data preprocessing module (120) can calculate a primary raw abnormal transaction value based on the deviation ratio between the target transaction price and the average transaction price, the physical depreciation value, the seller cancellation history, and the elapsed time since the transaction occurred. Additionally, the data preprocessing module (120) can convert the raw abnormal transaction value into a final abnormal transaction score in the range of 0 to 100 points according to a preset score conversion function. For example, as the price deviation ratio increases and the seller cancellation history increases, the raw abnormal transaction value may increase; as the physical depreciation value increases, the normal allowable range expands and the raw abnormal transaction value may be mitigated; and as the elapsed time since the transaction occurs becomes longer, the influence may be reduced by time attenuation. Accordingly, the filtering threshold may have an initial setting value of 80 points based on the final abnormal transaction score rather than the raw abnormal transaction value, and may be periodically corrected according to the recent misclassification rate.

[0067] The data preprocessing module (120) can perform text analysis at the morpheme level on the title and body of a post, and perform object recognition or surface defect detection on attached images to extract depreciation items corresponding to scratches, glass damage, bezel damage, missing warranty, or missing box. The data preprocessing module (120) can calculate a physical depreciation value by summing the scores for each extracted depreciation item. When a listing for a watch without a warranty is registered at a price lower than the average price, the data preprocessing module (120) recognizes this as a reasonable depreciation based on the physical condition and transmits a control signal to preserve it as valid data. The data preprocessing module (120) automatically transmits a control command to improve the consistency of the database by separating listings from sellers with a high cancellation history as bait listings.

[0068] The price analysis module (130) calculates the current fair price of the target clock based on preprocessed transaction data and transmits a higher-level control command that comprehensively predicts future price fluctuations. The price analysis module (130) manages the internally equipped fair price calculation model (131) and price prediction model (132) and automatically transmits a command to allocate memory so that the two models can perform calculations in parallel. The price analysis module (130) transmits a data routing control signal that aggregates the calculated price indicators and volatility indicators and transmits them to the service provision module (140). The price analysis module (130) automatically transmits a command to receive and obtain normalized valid transaction history data and abnormal transaction score analysis results from the data preprocessing module (120). To ensure that the price calculation operation and the bid price volatility analysis operation do not conflict with each other, the price analysis module (130) transmits a control command that coordinates the distribution of system resources by dynamically adjusting the operation priority. The above-mentioned fair price calculation model (131) automatically transmits a control signal to derive a real-time fair price for a watch by reflecting the scarcity and vintage demand of the current market based on the average price of past transactions. The above-mentioned fair price calculation model (131) transmits a command to determine the final price by adding a scarcity weight inversely proportional to the quantity of remaining listings within a specific period, calculating vintage demand by applying a logarithmic relationship to the ratio of user views relative to the year of manufacture, and synthesizing the macroeconomic inflation rate. The above-mentioned fair price calculation model (131) can transmit a control command to obtain the number of remaining listing posts maintaining a sales status and the cumulative number of views of detail pages through a crawling engine, and to obtain consumer price inflation rate data compared to the same month of the previous year from a public data provision system.In addition, even when the cumulative number of views increases rapidly due to a specific operation, the above-mentioned fair price calculation model (131) can control the price reflection rate relative to the standard calculated price so that it does not exceed a preset upper limit by applying a demand correction term based on a log relationship and an upward upper limit threshold together. The above-mentioned price prediction model (132) automatically transmits a control command to derive a short-term volatility indicator of future watch prices by analyzing market bid-ask imbalances and the frequency of external news events. The above-mentioned price prediction model (132) transmits a command to calculate a bias imbalance index by deriving the log absolute value of the ratio of the number of waiting buy orders to the number of sell orders. The above-mentioned price prediction model (132) automatically transmits a control signal to derive a final volatility indicator by applying a square root relationship to the number of news article distributions for the target watch brand to reduce media noise and then summing them up. The above price prediction model (132) obtains the number of buy and sell orders waiting through a request for order book information from a trading platform and transmits a control command to obtain the number of related articles distributed within a specified time by parsing the search query response from a pre-configured news portal. Even if related articles are distributed explosively within a day due to a rumor of discontinuation of a specific brand, the above price prediction model (132) automatically transmits a command to calculate a stable indicator so as not to overreact to temporary noise by reducing the impact through square root calculation.

[0069] The service providing module (140) transmits the derived value assessment information and volatility indicator to the user terminal unit and transmits a control command to dynamically adjust the frequency of notification occurrences in conjunction with the battery status of the terminal. The service providing module (140) automatically transmits a control signal that processes the volatility indicator into visual chart data and transmits it. The service providing module (140) transmits a control command to transmit a push notification trigger signal to the terminal only when the notification control index derived by calculating the magnitude of the calculated volatility indicator and the remaining battery level of the user terminal unit (200) in an inverse relationship exceeds a dynamically set threshold. The service providing module (140) receives periodic status packets from the connected user terminal unit (200) to obtain remaining battery data and automatically transmits a command to obtain the volatility indicator calculated from the price prediction model (132) through the internal memory bus. Even in situations of high market volatility, if the remaining battery level of the user terminal unit (200) is 10% or less, the service providing module (140) may extend the transmission cycle of general warning notifications or limit the maximum number of transmissions per unit time. Additionally, the service providing module (140) may manage the power consumption of the user terminal unit (200) by selectively transmitting only emergency notifications corresponding to the upper volatility range.

[0070] The AI ​​learning module (134) can construct a learning dataset using normal transaction data refined through the transaction data collection module (110) and the data preprocessing module (120). The learning dataset may consist of input data including brand, model name, manufacturing date, condition grade, physical depreciation value, remaining listing quantity, cumulative views, seller cancellation history, number of buy orders, number of sell orders, number of article distributions, and inflation rate, and label data including actual transaction price, actual transaction completion status, and price increase, decrease, or maintenance results after a preset prediction period. Additionally, the AI ​​learning module (134) can maintain the consistency of the learning dataset by excluding from the learning target data determined to be false listings, data where transaction cancellation has been confirmed, data where posting date or transaction date information is missing, and data isolated by the data preprocessing module (120).

[0071] The feature generation module (133) can generate a feature vector from the training dataset or the transaction data to be inferred. The feature vector may include at least one of a brand code, model code, manufacturing date information, condition grade, physical depreciation value, presence or absence of a warranty, presence or absence of a box, average transaction price, remaining quantity of listings, cumulative views, ratio of views to years of manufacture, seller cancellation history, number of buy orders, number of sell orders, number of article distributions, and inflation rate. Additionally, the feature generation module (133) can be configured to perform normalization processing on numeric variables and perform pre-set code value conversion or embedding conversion on categorical variables so that they can be input to the AI ​​learning module (134) and the AI ​​inference module (135).

[0072] The AI ​​learning module (134) can perform learning using the feature vector and label data generated by the feature generation module (133). At this time, the AI ​​learning module (134) can calculate the error between the predicted value and the actual value and iteratively update the model parameters in a direction that reduces the error. Additionally, the AI ​​learning module (134) can separate a portion of the entire learning dataset into verification data to check the verification error or classification accuracy at the end of the learning process, and can control the learning of the model parameters to be reflected in the operating environment only when they exceed a preset performance standard.

[0073] In a process for dynamically controlling the data collection cycle in response to server load and market volatility, the transaction data collection module (110) variably adjusts the data collection cycle based on the memory buffer occupancy of the server unit (100) and the registration frequency of the target data. The process performs a key logic that maintains system stability and secures the latest market transaction information without omission by controlling the collection speed in an exponential or logarithmic relationship according to the hardware load status.

[0075] In the abnormal transaction data filtering process based on physical depreciation and seller reliability, the data preprocessing module (120) calculates the physical defect status of the watch and the seller's past transaction cancellation history to identify and separate abnormal transaction data among the collected data. The process ensures the consistency of the analysis data by precisely distinguishing between normal low-priced listings and false bait listings by applying a correction to the allowable range based on physical depreciation factors in the simple price deviation analysis.

[0076] In a multidimensional fair price calculation process that reflects market scarcity and user response indices, the fair price calculation model (131) derives a real-time fair transaction price based on scarcity value according to the quantity of remaining listings in the market and the ratio of user views relative to the year of manufacture. The fair price calculation model (131) can set the real-time fair transaction price calculated according to relational equation-based calculations as the first price, and the AI ​​inference module (135) can receive the feature vector used when calculating the first price and calculate a fair price correction value. At this time, the fair price correction value can be applied in the form of a multiplier factor multiplied by the first price or a correction value added to or subtracted from the first price. In addition, the fair price calculation model (131) can suppress the final price from fluctuating excessively due to short-term external noise by limiting the fair price correction value so that it does not exceed a preset allowable range. The above process quantifies the imbalance between supply and demand and the level of interest in the actual market, while generating objective valuation indicators by applying convergence logic of logarithmic relationships to block price distortions caused by specific abusive behaviors.

[0077] In the process of controlling notifications linked to price volatility indicators and terminal hardware status, the service providing module (140) determines whether to send a push notification by combining the volatility indicator derived from the price prediction model (132) with the remaining battery level of the user terminal unit (200). The price prediction model (132) can calculate the probability of a price increase, the probability of a price decrease, and the probability of maintaining the price within a preset prediction period using the calculated price received from the fair price calculation model (131) and the feature vector generated by the feature generation module (133). For example, the prediction period can be set to at least one of 7 days, 30 days, or 90 days, and the price prediction model (132) can calculate the probability value for each prediction period separately. Additionally, the price prediction model (132) can determine the grade of a general caution notification, a warning notification, or an emergency notification by using the probability value together with the relationship-based volatility indicator. The AI ​​inference module (135) can calculate an inference reliability value along with a fair price correction value or a price fluctuation probability value. If the inference reliability value is greater than or equal to a preset threshold value, the fair price calculation model (131) or the price prediction model (132) can reflect the AI ​​inference result in the final result calculation. For example, the inference reliability threshold value may be set to 0.65, and the fair price correction value or the price fluctuation probability value may be reflected in the final result only when the inference reliability value calculated by the AI ​​inference module (135) is 0.65 or higher. On the other hand, if the inference reliability value is less than 0.65, the relationship-based primary calculated price or the relationship-based volatility indicator may be applied first, and the reflection ratio of the AI ​​inference result may be reduced or omitted. Accordingly, the system can maintain a balance between the utilization of the AI ​​inference result and stable operation. Accordingly, the system can maintain a balance between the utilization of the AI ​​inference result and stable operation.The above model update module (136) can control retraining to be performed when new normal transaction data is secured at a level exceeding a preset cumulative threshold, or when the average absolute error or classification error of the price prediction model (132) exceeds a preset increase threshold. For example, the cumulative threshold for the new normal transaction data can be set to 500 cases or more over the past 7 days, and the increase threshold for the average absolute error can be set to 10% or more compared to the previous threshold. Accordingly, the above model update module (136) can initiate retraining when new normal transaction data accumulates at 500 cases or more or when the average absolute error increases by 10% or more. At this time, the above model update module (136) can compare the verification performance of the model before retraining and the model after retraining, and can reflect the updated model parameters in the operating environment only when the verification performance has improved. In addition, the above model update module (136) can record the update time and perform retraining by incrementally reflecting only the normal transaction data accumulated since the previous update. The above process predicts volatility by analyzing market bid-ask imbalances and external events, and simultaneously performs a control function that efficiently manages hardware resources by dynamically adjusting the notification threshold according to the power status of the receiving terminal.

[0079] The transaction data collection module (110) transmits a control command to dynamically adjust the time interval for collecting data in response to the variability of the online market's clock transaction frequency. The transaction data collection module (110) automatically transmits a control signal to minimize data omissions that may occur when the market's transaction frequency surges, while simultaneously preventing physical overload and communication bandwidth exhaustion of the server unit (100) caused by excessive data requests. To improve the conventional fixed collection cycle method, the transaction data collection module (110) calculates the final collection cycle by combining three system variables—buffer occupancy rate, frequency of new post registration, and communication delay time—based on a preset basic collection cycle, and transmits an update command. When the current memory buffer occupancy rate of the server unit (100) increases, the transaction data collection module (110) transmits a control command to extend the collection cycle in an exponential relationship. Since linear extension alone may make it difficult to prevent sudden memory overflow, the transaction data collection module (110) automatically transmits a command to activate a defense mechanism for hardware protection. When the frequency of new transaction posts for the target watch model increases, the transaction data collection module (110) transmits a command to adjust the collection cycle so that it is shortened. Since there is a risk of server resources being exhausted if the collection cycle is reduced simply inversely proportional to the registration frequency, the transaction data collection module (110) automatically transmits a signal to control the reduction of the collection cycle so that it converges gradually and smoothly by applying a log relationship. As the communication delay time with the target server increases, the transaction data collection module (110) transmits a control command to extend the collection cycle in a linear proportional relationship to resolve network bottlenecks.

[0081] The transaction data collection module (110) transmits a command to obtain percentage data of memory usage measured in 1-second intervals through the operating system resource monitoring tool of the server unit (100). For example, when 12.8GB of the total 16GB memory is used, the transaction data collection module (110) obtains 80% of the data and automatically transmits it internally. The transaction data collection module (110) transmits a command to physically count and obtain the number of posts of the target watch model newly registered in the last 10 minutes through an application program interface provided to an external communication channel. For example, when 50 posts are registered in 10 minutes, the transaction data collection module (110) obtains the value 50 and automatically transmits it internally. The transaction data collection module (110) transmits a command to measure and obtain the round-trip time in milliseconds from sending a ping test packet to the target server and receiving a response. For example, when the round-trip time takes 200ms, the transaction data collection module (110) obtains and automatically transmits the value 200. The transaction data collection module (110) transmits a command to immediately transmit the final dynamic collection cycle value calculated by combining the above conditions to an internal scheduler. The transaction data collection module (110) automatically transmits a control command to update the operation schedule so that the scheduler initializes the previously scheduled collection timer and executes the next data crawling or interface request when a newly calculated cycle time, such as 45 seconds, has elapsed.

[0083] The transaction data collection module (110) transmits a command to primarily collect and normalize data on the buffer occupancy rate, new post registration frequency, and communication delay time of the server unit (100) at a preset update cycle, such as every minute. The transaction data collection module (110) automatically transmits data determining whether the collected buffer occupancy rate exceeds a preset risk threshold, such as 95%. If it exceeds the threshold, the transaction data collection module (110) omits calculations on other variables and immediately transmits an emergency blocking command to fix the collection cycle to a preset maximum idle period, such as 3600 seconds. If it does not exceed the threshold, the transaction data collection module (110) automatically transmits a control command to calculate a preliminary collection cycle based on a preset basic collection cycle, such as 300 seconds, by multiplying by an exponential increase according to the buffer occupancy rate, multiplying by a logarithmic decrease of the new post registration frequency, and multiplying by an increase proportional to the communication delay time.

[0084] At this time, the transaction data collection module (110) may not determine the preliminary collection cycle as a result of a single continuous formula, but may apply different correction rules depending on the buffer occupancy rate. For example, if the buffer occupancy rate is less than 50%, the shortening correction based on the frequency of new post registration may be applied first, and if the buffer occupancy rate is 50% or more but less than 80%, the shortening correction based on the registration frequency and the extension correction based on the communication delay time may be applied together. In addition, if the buffer occupancy rate is 80% or more but less than 95%, the extension correction based on the buffer occupancy rate and the communication delay time may be applied first rather than the shortening correction based on the registration frequency, thereby switching to protection mode. Accordingly, the transaction data collection module (110) may shorten the collection cycle stepwise during normal times, forcibly extend the collection cycle during high-load periods, and apply a maximum rest period of 3600 seconds when the buffer occupancy rate is 95% or more.

[0085] The transaction data collection module (110) verifies that the calculated preliminary collection cycle does not exceed the range of the preset minimum collection cycle, such as the maximum rest period and 10 seconds, and then confirms it as the final dynamic collection cycle and transmits a command to register it to the scheduler. The transaction data collection module (110) transmits an exception handling command if an exception occurs in which the frequency of new post registrations cannot be counted normally due to a data communication failure. In this case, the transaction data collection module (110) does not treat the variable value as 0, but instead automatically transmits a control command to maintain the stability of the logic and prevent the collection cycle from diverging abnormally by performing calculations using the average value of the past three most recently collected normally.

[0087] Assuming a situation where a specific limited edition watch is discontinued and transaction posts surge to 500 within 10 minutes, the transaction data collection module (110) detects the increase in registration frequency and transmits a command to immediately shorten the collection cycle to a minimum limit, such as 10 seconds. While conventional fixed-cycle methods may cause omissions in rare transaction data, the transaction data collection module (110) according to the present invention automatically transmits a control signal to collect transaction history while reducing omissions. Assuming a situation where the buffer occupancy rate of the server unit (100) reaches 90% as the cycle shortening continues, the transaction data collection module (110) transmits a command to significantly extend the collection cycle according to a structure designed so that the exponential relationship of the buffer occupancy rate takes precedence over the logarithmic relationship of the registration frequency. Through this, the transaction data collection module (110) automatically transmits a control command to defend against system load. When the transaction data collection module (110) sets the basic collection cycle to 300 seconds, it transmits the result of calculating the dynamic collection cycle based on changes in environmental variables to internal memory. The transaction data collection module (110) determines the collection cycle to be 295 seconds when the buffer occupancy rate is 20%, the registration frequency is 5 cases per 10 minutes, and the communication delay is 50ms, and transmits a command to maintain a normal stable state. The transaction data collection module (110) automatically transmits a control signal to shorten the collection cycle to 180 seconds when the buffer occupancy rate is 22%, the registration frequency is 15 cases, and the communication delay is 60ms. The transaction data collection module (110) transmits a command to shorten the collection cycle to 110 seconds when the buffer occupancy rate is 25%, the registration frequency is 50 cases, and the communication delay is 80ms. The transaction data collection module (110) automatically transmits a command to continuously shorten the collection cycle to 65 seconds when the buffer occupancy rate is 30%, the registration frequency is 150 cases, and the communication delay is 150ms.The transaction data collection module (110) transmits a command to determine the collection cycle to 45 seconds by mitigating the reduction range by log scale even if the frequency increases when the buffer occupancy is 45%, the registration frequency is 300 cases, and the communication delay is 300ms. The transaction data collection module (110) automatically transmits a signal to extend the collection cycle to 80 seconds by reflecting the occurrence of delay when the buffer occupancy is 60%, the registration frequency is 320 cases, and the communication delay is 500ms. The transaction data collection module (110) transmits a command to forcibly switch the collection cycle to 250 seconds as the buffer occupancy increases when the buffer occupancy is 80%, the registration frequency is 350 cases, and the communication delay is 800ms. The transaction data collection module (110) automatically transmits a control command to significantly extend the collection cycle to 850 seconds for hardware protection when the buffer occupancy is 90%, the registration frequency is 400 cases, and the communication delay is 1200ms. The above transaction data collection module (110) transmits an emergency command to enter a maximum idle period of 3600 seconds due to exceeding a 95% risk threshold when the buffer occupancy rate is 96%, the registration frequency is 100 cases, and the communication delay is 500ms. The above transaction data collection module (110) automatically transmits a command to normalize the collection cycle to 210 seconds when the system stabilizes when the buffer occupancy rate is 50%, the registration frequency is 20 cases, and the communication delay is 100ms.

[0089] The transaction data collection module (110) transmits a control command according to a dynamic control judgment rule for the data collection cycle. As a single priority condition, if the buffer occupancy rate exceeds 95%, the transaction data collection module (110) automatically transmits a command to set the collection cycle to a preset maximum idle period and perform emergency blocking, regardless of the registration frequency and communication delay. As a multiple complex operation condition, if the buffer occupancy rate is 95% or less, the registration frequency increases, and the communication delay is normal, the transaction data collection module (110) transmits a control signal to gradually shorten the collection cycle in a logarithmic relationship. As a multiple complex operation condition, if the buffer occupancy rate is 95% or less, the registration frequency increases, and a delay occurs, the transaction data collection module (110) automatically transmits a command to adjust the cycle by offsetting the shortened registration frequency and the extended delay time. As a multiple complex operation condition, if the buffer occupancy rate is on an upward trend, the transaction data collection module (110) transmits a control command to extend the collection cycle in an exponential relationship, prioritizing server protection, regardless of whether the registration frequency is maintained or increased. The transaction data collection module (110) transmits control commands based on a structure suitable for effectively collecting external data that fluctuates within limited physical computing resources when determining the collection cycle. The transaction data collection module (110) automatically transmits a signal that contributes to system stability by controlling the speed at which hardware limits are reached by applying an exponential relationship to the buffer occupancy variable. The transaction data collection module (110) transmits a command that supports more stable system operation by preventing the server from attempting excessive requests when the frequency increases by applying a logarithmic relationship to the registration frequency. The transaction data collection module (110) transmits control commands that coordinate two conflicting elements, data updating and hardware stability, through the combination of multiple conditions of independent physical and environmental variables.The transaction data collection module (110) automatically transmits control commands that help maintain the availability of the entire system even in situations of increased traffic by offsetting bias toward specific conditions through exponential and logarithmic relationships. The transaction data collection module (110) transmits a command to set the default collection cycle, which is the normal stable collection interval calculated through a network load test during the initial operation of the system, to 300 seconds. The transaction data collection module (110) automatically transmits a command to update the above value based on the average value of server communication log data over the past month. The transaction data collection module (110) transmits a signal to specify a maximum idle period of 3600 seconds during which no collection requests are performed in order to prevent overload of the server unit (100). The transaction data collection module (110) automatically transmits a control command to set the risk threshold, which is the hardware defense standard value, to 95%, which is the value set just before the operating system swaps memory and a physical disk bottleneck occurs.

[0090] Among the input values ​​used in Equations 1 through 4 described below, percentage values ​​can be used by dividing by 100 to convert them into normalized values ​​ranging from 0 to 1. Additionally, time values, such as communication delay time and the elapsed time after a transaction, can be converted into dimensionless values ​​by dividing by a pre-set reference time. For example, the communication delay time can be converted into a delay coefficient converted to a reference unit of 100ms, and the elapsed time after a transaction can be converted into an elapsed coefficient converted to a reference unit of 30 days. Furthermore, the inflation rate can be reflected not as the percentage value itself, but in the form of a multiplier factor calculated by adding the inflation rate to 1.

[0091] The dynamic collection cycle calculation method driven by the transaction data collection module (110) can be implemented by multiplying the basic collection cycle by an exponential increase term of the normalized buffer occupancy rate and a linear increase term of the normalized communication delay time, and dividing by a logarithmic decay term of the frequency of new post registration. At this time, the buffer occupancy rate can be normalized by dividing the percentage value by 100, and the communication delay time can be used by converting it to a standard unit of 100ms.

[0092]

[0093] As shown in [Equation 1] above, the risk threshold can be calculated based on the average buffer occupancy value at points in time when the communication delay time over the past 30 days was 3 times or more than the normal average, and if the calculated value is less than 90%, it can be corrected to 90%, and if it exceeds 95%, it can be corrected to 95%. Accordingly, the risk threshold can be set as a dynamic threshold that is periodically updated within the range of 90% or more and 95% or less.

[0094] [Table 1-1] below shows the results of a simulation performed 10 times with increased buffer occupancy and communication delay time in a system load test environment, through which an objective risk threshold is derived.

[0095] [Table 1-1: Server Load Simulation and Threshold Derivation (Unit: %, ms)]

[0096]

[0097] Table 1-1 above is an example of simulation results performed while varying server load conditions, and Table 1-2 above is an example of simulation results comparing the fixed cycle method and the dynamic collection cycle method under the same transaction surge conditions. According to the tables above, it can be observed that communication delay and data loss tend to increase as the buffer occupancy rate enters the range of 90% or higher, and it can be confirmed that the dynamic collection cycle control method contributes to mitigating both data loss and server overload simultaneously.

[0098] [Table 1-2] below is the result of comparing and verifying the data omission prevention effect between the present invention, which applies the relationship and threshold value derived above, and the prior art using a simple fixed period.

[0099] [Table 1-2: Verification of Collection Effectiveness in Situations of Transaction Surge (Condition: 500 Transactions in 10 Minutes)]

[0100]

[0101] The simulation results confirm that the present invention can work to mitigate data loss and server overload by applying a logarithmic relationship to the registration frequency to smoothly adjust the reduction range of the collection cycle, while prioritizing the application of extension correction in the rising sections of buffer occupancy and communication delay time.

[0103] The data preprocessing module (120) transmits a control command to identify and remove false listings and abnormal transactions among the collected transaction data that may distort the subsequent price calculation model. The data preprocessing module (120) automatically transmits a control signal to ensure the integrity of the database by preventing normal used depreciation data from being misidentified as an outlier based solely on price differences, and by suppressing bait listings by malicious sellers from damaging the system's standard price. To improve the conventional simple statistical outlier removal method, the data preprocessing module (120) transmits a command to calculate a final outlier transaction score by complexly calculating the physical depreciation factors of the watch, the seller's history, and the flow of time. The data preprocessing module (120) automatically transmits a control command to derive the difference value between the target transaction price and the average transaction price. The data preprocessing module (120) transmits a command to expand the normal allowable range of the difference value in an inverse relationship with the physical depreciation value, so that the greater the physical depreciation, such as scratches on the watch or missing parts, the more it can be considered a normal price drop. The data preprocessing module (120) automatically transmits a control signal to impose a penalty by counting the number of past abnormal transaction cancellations of the same seller. Since the score may diverge excessively if a penalty is imposed linearly based on the number of cancellations, the data preprocessing module (120) transmits a command to control the penalty increase by applying a logarithmic relationship so that the penalty increase becomes gentler as the number of cancellations increases. To reduce the impact of past abnormal transaction data on the current model, the data preprocessing module (120) automatically transmits a control command to reduce the abnormal transaction score in an exponential decay relationship as the time interval between the time of the transaction and the current processing time becomes longer.

[0104] The data preprocessing module (120) transmits a command to calculate and obtain an absolute difference of 1.5 million won between the collected target transaction price of 8.5 million won and the average transaction price of the last 3 months stored in the database of 10 million won. The data preprocessing module (120) analyzes the text and image data of the sales post to obtain a physical depreciation value 2, which is the sum of one glass scratch and one no warranty, and automatically transmits it internally. In this case, the data preprocessing module (120) first calculates a raw abnormal transaction value reflecting the price deviation ratio, cancellation history penalty, and time reduction, and then converts the raw abnormal transaction value into a final abnormal transaction score in the range of 0 to 100 points. For example, in the above case, if the normal depreciation range is also confirmed, the final abnormal transaction score may be converted to less than 80 points and maintained as valid transaction data; conversely, if seller cancellation history is repeatedly accumulated under the same price deviation conditions, the final abnormal transaction score may exceed 80 points and be classified as an isolation target. The data preprocessing module (120) transmits a control command to obtain a value of 5 by counting the transaction cancellation history based on buyer reports over the past year that matches the corresponding seller identifier within the platform. The data preprocessing module (120) calculates the difference between the original posting date of the target data and the current server calculation date to obtain and automatically transmit data with a time interval of 60 days. If the final abnormal transaction score derived by comprehensively calculating the above conditions exceeds a preset threshold, the data preprocessing module (120) transmits a control command to tag the data as an outlier. The data preprocessing module (120) automatically transmits a command to immediately exclude the tagged data from the input data for calculating the appropriate price of the subsequent price analysis module and transfer it to an isolation database.The data preprocessing module (120) transmits a control command to calculate a first absolute deviation score indicating how far the price of the collected transaction data deviates from the average price. The data preprocessing module (120) automatically transmits a control signal to discount the first absolute deviation score based on the parsed physical depreciation figures. The data preprocessing module (120) transmits a command to add a penalty to the score derived stepwise by applying a log weight to the number of past transaction cancellations by the seller. The data preprocessing module (120) automatically transmits a command to determine the final abnormal transaction score by applying exponential depreciation according to the time interval of the data. The data preprocessing module (120) transmits a control command to separate the data if the final abnormal transaction score exceeds a threshold of 80 points, and to merge it into the main database as normalized valid data if it is less than or equal to 80 points.

[0105] The data preprocessing module (120) automatically transmits a command to perform exception handling when a situation occurs where the seller intentionally leaves the body of the post blank so that the physical depreciation value cannot be determined. In this case, the data preprocessing module (120) transmits a control command to forcibly assign the physical depreciation value of the watch to 0. As the depreciation value becomes 0 and the allowable range for price deviation becomes strict, the data preprocessing module (120) automatically transmits a command to maintain the reliability of the system by filtering out listings that are excessively cheaper than the average price without a description of the condition. Assuming a situation where a watch with an average price of 10 million won is listed for 7 million won because the glass is broken and there is no warranty, the data preprocessing module (120) transmits a control signal to expand the allowable deviation range by reflecting the physical depreciation value. Through this, the data preprocessing module (120) automatically transmits a control command to recognize the listing as a reasonably low-priced listing due to poor condition and preserve it as valid data. Assuming a situation where a seller lists a watch in top condition for 9 million won, which is cheaper than the average price, while hiding a history of 10 past cancellations, the data preprocessing module (120) transmits a command to apply a strong seller history penalty. The data preprocessing module (120) automatically transmits a control command to filter the listing as a bait listing by raising the abnormal transaction score above a threshold.

[0106] The data preprocessing module (120) transmits a control command based on an abnormal transaction filtering judgment rule according to physical condition and seller history. The data preprocessing module (120) automatically transmits a control signal classifying the data as an outlier if the price deviation is greater than 30%, the physical depreciation value is 0, and the seller cancellation history is normal. The data preprocessing module (120) transmits a command to recognize the data as valid data if the physical depreciation value is 3 or higher, even if the price deviation is greater than 30%. The data preprocessing module (120) automatically transmits a control command to classify the data as an outlier by applying a log weighting penalty if the price deviation is within an appropriate range of 10% or less and the physical depreciation value is 0, but the seller cancellation history is 5 or higher. For old data where the price deviation is greater than 30%, the physical depreciation value is 0, and the cancellation history is 5 or higher, the data preprocessing module (120) transmits a control signal that maintains the outlier state even when exponential depreciation is applied. The data preprocessing module (120) automatically transmits a command to link collected text and image variables to price deviation correction variables, as watch transaction data is accompanied by physical substance. The data preprocessing module (120) transmits a control command to implement a data preprocessing architecture suitable for refining used transaction data by suppressing the repetitive behavior of malicious users on a log-by-log basis. The data preprocessing module (120) automatically transmits a control signal to increase the suitability of filtering through the combination of physical and historical variables. The data preprocessing module (120) transmits a control command to reduce noise data flowing into the subsequent fair price calculation model, thereby improving the stability of input data for the subsequent fair price calculation model and aiding in price prediction. The data preprocessing module (120) automatically transmits a command to calculate an abnormal transaction score, which is an index that quantifies the degree to which the transaction data violates the logic of normal market price formation.The data preprocessing module (120) transmits a control command that sets a threshold of 80 points, derived from the average score distribution of false listings detected over the past year, as the threshold for determining whether to merge or separate the data. The data preprocessing module (120) automatically transmits a control command that reduces the impact of past noise on current price analysis by applying an exponential decay relationship in which the influence of a variable continuously decreases at a constant rate over time.

[0107] The data preprocessing module (120) can calculate a price deviation ratio by dividing the absolute deviation between the target transaction price and the average transaction price by the average transaction price. Additionally, the physical depreciation value can be calculated by summing the depreciation item scores corresponding to at least one of scratches, glass damage, bezel damage, missing warranty, and missing box. For example, one glass scratch can be calculated as 1 point, a missing warranty as 1 point, and exterior damage as 2 points, and then summed. Additionally, the time interval between the time of the transaction and the current processing time can be normalized by an elapsed coefficient converted to 30 days as one unit, and the abnormal transaction score can be calculated by reflecting the log penalty of the seller's cancellation history in the price deviation ratio, followed by sequentially applying an allowable range correction based on the physical depreciation value and an exponential depreciation of the elapsed coefficient.

[0108]

[0109] The above relationship 2 is an example for calculating raw abnormal transaction values, and the data preprocessing module (120) can convert the raw abnormal transaction values ​​into a final abnormal transaction score in the range of 0 to 100 points according to a preset score conversion function.

[0110] The above filtering threshold can be set as the center value of the mode interval excluding the bottom 10% noise interval after normalizing the distribution of abnormal transaction scores of transaction data classified as fraudulent listings over the past year. The initial setting can be 80 points and can be periodically adjusted based on the recent misclassification rate.

[0111] [Table 2-1] below shows the distribution of outlier transaction scores calculated by applying the above relationship to 10 virtual transaction data sets with various variable combinations. (Higher scores indicate a higher probability of outliers.)

[0112] [Table 2-1: Simulation of Abnormal Transaction Score Calculation (Condition: Average Price 10 Million Won)]

[0113]

[0114] Table 2-1 above is a simulation example comparing the abnormal transaction score calculation structure assuming transaction data with various combinations of variables.

[0115] In addition, the data preprocessing module (120) can transmit a control command that reduces noise data flowing into the subsequent fair price calculation model to improve the stability of the input data.

[0116] [Table 2-2: Comparative Verification of Filtering Logic (Condition: Total data 1,000 records, including 100 actual fake records)]

[0117]

[0118] As shown in [Table 2-2] above, the comparison results confirm that the logic of the present invention, which reflects physical depreciation figures, seller cancellation history, and time depreciation together, can reduce false detections and improve the accuracy of identifying false listings compared to a method that uses only simple price deviation criteria.

[0120] The above-mentioned fair price calculation model (131) transmits a control command to dynamically calculate the most reasonable standard price for watch trading at the current time based on collected and preprocessed transaction data. The above-mentioned fair price calculation model (131) moves away from the conventional method of relying solely on past static average prices and automatically transmits a control signal to reflect the imbalance between supply and demand in the price by synthesizing the scarcity of items remaining in the current market, real-time user interest, and macroeconomic indicators. To prevent errors in simple average calculation, the above-mentioned fair price calculation model (131) sets the past average transaction price as the basic standard and then transmits a command to derive a real-time fair transaction price by complexly calculating the remaining quantity of items, detail page views, year of manufacture, and inflation rate. The above-mentioned fair price calculation model (131) automatically transmits a control command to calculate the remaining quantity of the corresponding model released in the online and offline markets during a specific period. The above fair price calculation model (131) transmits a command to add a scarcity premium to the above basic standard by applying a relationship inversely proportional to the quantity of the above remaining items, reflecting the characteristic that purchasing competition intensifies as the number of remaining items decreases. The above fair price calculation model (131) automatically transmits a control signal that derives the ratio of user click views on the detail page of the corresponding watch model relative to the time elapsed from the year of manufacture of the watch to the present. The above fair price calculation model (131) transmits a command to reflect in the price if the number of views is high despite the watch being old, as this signifies an increase in demand as a vintage model. To suppress manipulation of view counts by an automatic input program, the above fair price calculation model (131) automatically transmits a command to control the value fluctuation range to converge smoothly by applying a logarithmic relationship to the above ratio and reflecting the calculated value in a proportional relationship.The above-mentioned fair price calculation model (131) transmits a control command that applies a real asset defense effect against a decline in currency value by collecting the fluctuation range of the macroeconomic price inflation rate at the time of calculation and additionally reflecting it in a proportional relationship with the above-mentioned calculated price.

[0121] The above fair price calculation model (131) automatically transmits a command to obtain basic reference data, such as 15 million won, by calculating the average of valid transaction data over the past month that has passed through the data preprocessing module. The above fair price calculation model (131) transmits a control command to obtain a remaining quantity of listings, such as 3, by physically counting unique listing posts that are currently in a sales state among the data collection channels. The above fair price calculation model (131) automatically transmits a command to derive the ratio of views relative to the year of manufacture by using 98 months, which is the number of months elapsed from the manufacturing year / year month (such as January 2018) to the present, as the denominator, and 5,000 times, which is the cumulative number of clicks on the detail page of the model over the past week, as the numerator. The above fair price calculation model (131) transmits a control signal to obtain a fluctuation range of the inflation rate, such as 2.5%, by checking the consumer price inflation rate compared to the same month of the previous year through an external public data application interface. The above-mentioned fair price calculation model (131) automatically transmits a command to control the real-time clock fair transaction calculation price derived by synthesizing the above conditions to be transmitted to the service provision module and visually displayed on the screen of the user terminal unit. The above-mentioned fair price calculation model (131) transmits a control command to input the above-mentioned calculation price as baseline data for predicting future price volatility in a subsequent price prediction model. The above-mentioned fair price calculation model (131) may transmit a control command to derive a first calculation price by sequentially summing the scarcity correction rate based on the remaining quantity of listings, the demand correction rate based on the cumulative number of views and the number of elapsed months, and the price correction rate based on the inflation rate, by setting the preprocessed past average transaction price of the target clock model as the reference calculation price.At this time, the scarcity correction rate may be limited so as not to exceed a preset maximum correction rate, while increasing as the remaining quantity of listings decreases; the demand correction rate may be calculated by applying a logarithmic relationship to the value obtained by dividing the cumulative number of views by the number of elapsed months; and the price correction rate may reflect the consumer price inflation rate compared to the same month of the previous year. For example, if the consumer price inflation rate is 2.5%, it may be reflected as a price correction rate of 0.025. In addition, the first calculated price may be calculated by adding multiple correction rates to the basic average price. Accordingly, the fair price calculation model (131) can prevent excessive price amplification while gradually reflecting multiple correction rates to the standard calculated price. Furthermore, the fair price calculation model (131) may provide the first calculated price as an input value to a subsequent AI inference module (135), and determine the final calculated price by reflecting the fair price correction value and inference reliability calculated by the AI ​​inference module (135).

[0122] The above fair price calculation model (131) automatically transmits an exception handling command to forcibly add a default value of 1 to the remaining quantity of listings to prevent mathematical calculation errors that may occur when the remaining quantity of listings of a discontinued rare model becomes 0. The above fair price calculation model (131) transmits a command to control the application of the scarcity premium safely at a preset maximum value while preventing errors through this exception handling. The above fair price calculation model (131) automatically transmits a command to control the upper limit to further guarantee system stability, even though the rate of increase is slowed down by the logarithmic relationship in the event that the number of views explodes infinitely due to abnormal access. The above fair price calculation model (131) transmits a control signal to maintain the stability of the logic by presetting the maximum limit range at which the user reaction index raises the price to a maximum of 15% relative to the base price. Assuming a situation where the number of views of a 20-year-old watch has recently exploded, the above fair price calculation model (131) detects that the ratio of the number of views to the year of manufacture has surged and automatically transmits a command to add a vintage premium. The above fair price calculation model (131) transmits a control command to calculate an increased fair price that meets current market demand, unlike the conventional method that suggests a low price in the past due to a lack of recent transaction history. Assuming a situation where a seller generates 100,000 views in one day using a macro program to raise the price, the above fair price calculation model (131) automatically transmits a command to offset by applying a logarithmic relationship to the view count ratio. The above fair price calculation model (131) transmits a control signal that effectively suppresses price distortion caused by abnormal manipulation by activating upper limit exception processing.

[0123] The above fair price calculation model (131) transmits control commands based on price calculation judgment rules according to remaining listings and user reactions. When the number of remaining listings is large, the ratio of views to years is low, and macroeconomic prices are stable, the above fair price calculation model (131) automatically transmits a command to slightly lower the basic average price by reflecting oversupply. When the number of remaining listings is small, the ratio of views to years is moderate, and macroeconomic prices are stable, the above fair price calculation model (131) transmits a control signal to add a scarcity premium by reflecting a seller's market. When the number of remaining listings is very low at the level of 0 to 1, the ratio of views is very high, and macroeconomic prices are rising, the above fair price calculation model (131) automatically transmits a command to derive the maximum calculated price by maximizing scarcity, reflecting vintage value in a logarithmic manner, and applying inflation. The above fair price calculation model (131) transmits a control command to suppress the price increase by applying log offsets and upper limits, regardless of macroeconomic prices, when the remaining quantity of listings is moderate and the view count ratio explodes abnormally. The above fair price calculation model (131) automatically transmits a control signal with a structure suitable for reflecting the special market logic of vintage demand, where value actually increases over time. The above fair price calculation model (131) transmits a command to software-implement market economic characteristics by deriving a ratio by combining the view count variable with the year variable. The above fair price calculation model (131) automatically transmits a control command that helps overcome the lagging nature of the average of past data through the multiple combination of scarcity and dynamic response variables. The above fair price calculation model (131) transmits a signal to provide three-dimensional watch valuation information that best fits the current market situation by tracking the user's immediate click response and the increase or decrease in market listings in real time and reflecting them in the price.The above-mentioned fair price calculation model (131) automatically transmits a command to determine the real-time watch fair transaction calculation price, which is the final reference price corrected by reflecting the current market status and user response, while based on the past average price. The above-mentioned fair price calculation model (131) transmits a control command to multiply the scarcity reflection weight, which is differentially assigned according to the market dominance of the brand or the grade of the model, by the reciprocal of the remaining quantity of listings. The above-mentioned fair price calculation model (131) automatically transmits a control signal to defend against distortion of the price calculation logic caused by manipulation of view counts by applying a logarithmic relationship in which the growth rate of the output value gradually decreases even if the input value increases.

[0124] The above-described fair price calculation model (131) can calculate a first calculated price by adding a scarcity correction rate, a demand correction rate, and a price correction rate to the basic average price. At this time, the scarcity correction rate may be increased as the quantity of remaining listings decreases, but may be limited so as not to exceed a preset maximum correction rate, the demand correction rate may be calculated by applying a logarithmic relationship to the value obtained by dividing the cumulative number of views by the number of elapsed months, and the price correction rate may reflect the consumer price inflation rate compared to the same month of the previous year as is. For example, if the consumer price inflation rate is 2.5%, it may be reflected as a price correction rate of 0.025.

[0125] In addition, the above first calculated value can be calculated as shown in the following relationship.

[0126] 1st Calculated Price ∝ Base Average Price × (1 + Scarcity Adjustment Rate + Demand Adjustment Rate + Price Adjustment Rate)

[0127] Here, the scarcity correction rate can be calculated by multiplying the reciprocal of the remaining quantity of listings by a preset scarcity weight, and the demand correction rate can be calculated by multiplying the value obtained by applying a logarithmic relationship to the ratio of cumulative views to the number of elapsed months by a preset demand weight. Accordingly, the appropriate price calculation model (131) can prevent excessive price amplification while gradually reflecting multiple correction rates to the basic average price.

[0128] At this time, the threshold for the upper limit of the rise, intended to prevent the calculated price from being distorted due to an infinite surge in views caused by specific abuse (macro) attacks, is calculated based on the distribution of short-term price increase rates of the same brand group or same model group for which sufficient normal transactions have been secured over the past six months, and can be set at a maximum of 15% relative to the reference calculated price. Accordingly, even if the demand correction rate or the fair price correction value increases simultaneously, the first calculated price or the final calculated price may be limited so as not to rise by more than 15% relative to the reference calculated price. Accordingly, even if the cumulative views increase rapidly due to view count manipulation or external noise, the price reflection rate based on the demand correction term may be limited so as not to exceed the threshold for the upper limit of the rise.

[0129] [Table 3-1] below shows the results of simulating the impact of simple proportional calculation and the logarithmic relationship calculation of the present invention on price when a macro attack is introduced into a watch under the same conditions and the cumulative number of views increases exponentially.

[0130] [Table 3-1: View Count Manipulation Abuse Simulation (Conditions: Base Average Price 10 Million Won, Same Model Year)]

[0131]

[0132] In the above [Table 3-1], the conventional technology causes the price to skyrocket unrealistically after the 5th round, paralyzing the system. On the other hand, the present invention can control the price reflection rate so that it does not exceed 15% of the standard calculated price even if an abnormal increase in views occurs, by applying a log relationship and an upward upper limit threshold. Accordingly, it is reasonable to set the upward upper limit threshold for abuse prevention to 15% of the standard calculated price.

[0133] [Table 3-2: Comparison Verification of Calculated Prices Based on Reapplication of Past Transaction History]

[0134]

[0135] Table 3-2 above is an example of comparing the primary calculated price based on the relationship formula and the final calculated price reflecting AI correction with the actual execution price by reapplying transaction data with a history of past normal execution. Accordingly, it can be confirmed that the final calculated price reflecting AI correction can act to reduce the deviation from the actual execution price compared to the primary calculated price based on the relationship formula.

[0137] The price prediction model (132) and the service provision module (140) analyze market bid-ask imbalances and the frequency of external news events to derive an indicator of future watch price volatility and transmit a control command to dynamically adjust the frequency of notification occurrences according to the hardware status of the user terminal unit (200) receiving the indicator. The price prediction model (132) and the service provision module (140) proactively identify signs of short-term price fluctuations and automatically transmit a control signal to reduce battery depletion and user fatigue of the user terminal unit (200) caused by frequent notification transmission. The price prediction model (132) transmits a command to perform a complex calculation of two market variables for price volatility prediction and one physical variable for terminal control. The price prediction model (132) transmits a control command to calculate the ratio of the number of currently waiting buy orders to the number of sell orders. The above price prediction model (132) automatically transmits a signal that converts the ratio into a fluctuation index by applying the absolute value after taking the natural logarithm of the ratio, in order to measure the degree of imbalance regardless of scale when a bias toward either buying or selling occurs. The above price prediction model (132) transmits a command to control external event noise by reflecting the number of news article distributions related to the announcement of new products or discontinuation of the target brand. The above price prediction model (132) automatically transmits a control signal that adds by applying a square root relationship to the number of news article distributions to suppress the excessive expansion of the volatility index caused by a short-term increase in media exposure. The above service provision module (140) transmits a command to check the current remaining battery level of the user terminal unit (200) when providing the calculated volatility index to the user as a notification. The above service provision module (140) automatically transmits a command to control the notification so that the notification is activated only in designated extreme fluctuation situations when the battery is low, by adjusting the notification threshold upward in an inverse relationship with the remaining battery level.

[0138] The above price prediction model (132) transmits a control command to obtain 150 first-round buy orders and 30 sell orders at a specific point in time by physically counting them through the order book application interface of a used goods trading platform. The above price prediction model (132) automatically transmits a signal to obtain 16 data points by web crawling the number of articles newly distributed within the last 24 hours using the target watch brand and model name as keywords on a pre-configured news portal site. The above service provision module (140) transmits a command to obtain a current battery level percentage of 15% by calling the hardware status application interface of the operating system of the dedicated application installed on the above user terminal unit (200). The above service provision module (140) automatically transmits a control signal to utilize the derived future price volatility indicator to generate a visual graph evaluating the value stability of the watch. The service providing module (140) transmits a control command to transmit or block a push notification trigger signal to the operating system background of the user terminal unit (200) according to the calculated indicator and the calculation result of the remaining battery amount. The price prediction model (132) automatically transmits a command to set a basic volatility coefficient by deriving the standard deviation of transaction prices within a past preset period. The price prediction model (132) transmits a control signal to calculate the log absolute value of the ratio of 150 buy orders and 30 sell orders and multiply it by the basic volatility coefficient. The price prediction model (132) automatically transmits a command to derive the final future price volatility indicator by taking the square root of 4 of the number of 16 news articles that occurred within 24 hours and adding it to the preceding result. The service providing module (140) confirms that the battery of the user terminal unit (200) is in a low-power state of less than 15% and transmits a control command to block the transmission of notifications at the volatility indicator of a general warning stage.The above service provision module (140) automatically transmits a command to generate an exceptional notification only when the indicator exceeds the highest threshold, which is preset to the severe stage.

[0139] The above price prediction model (132) transmits a control command to perform exception handling to prevent calculation errors or failure of the logarithmic function from occurring when the number of sell or buy orders becomes zero due to infrequent trading. The above price prediction model (132) automatically transmits a signal to maintain the stability of the logic by forcibly adding a default value of 1 to the numerator and denominator of the order count, respectively. Assuming a situation where the number of articles surges to 10,000 in one day due to a rumor of a specific model being discontinued, the above price prediction model (132) transmits a command to reduce noise to a level of 100 by applying a square root function instead of deriving an indicator linearly proportional to the number of articles. The above price prediction model (132) automatically transmits a control command to prevent the system from causing errors due to external issues and to derive a stable indicator. Assuming a situation where the volatility indicator remains consistently high but the user's smartphone battery has 10% remaining, the service providing module (140) transmits a control signal that increases the notification threshold inversely proportional to the remaining battery level. The service providing module (140) automatically transmits a dynamic control command that limits the transmission frequency of general caution notifications and manages the power consumption of the user terminal. The service providing module (140) transmits a control command based on a notification control judgment rule according to market volatility and the terminal battery status. The service providing module (140) automatically transmits a command to transmit a regular report notification when the bid-ask imbalance is close to 1:1, the number of articles with square roots applied is small, the volatility indicator is in a stable stage, and the remaining battery level is sufficient at 80%. The service providing module (140) transmits a control signal to transmit a volatility occurrence notification once when bid-ask imbalance occurs, the number of articles is at a normal level, the volatility indicator is in a caution stage, and the remaining battery level is 50%.The above service provision module (140) automatically sends a command to block the transmission of notifications by prioritizing power preservation when the indicator is at the warning stage and the battery level is insufficient at 15%, even if the price skew occurs and the number of articles increases explosively. The above service provision module (140) recognizes the situation as an emergency when the indicator is at the severe stage due to extreme price skew and an explosive increase in the number of articles and the battery level is insufficient at 10%, and sends a control command to send a maximum warning notification once as an exception.

[0140] The service provision module (140) transmits a command that considers the remaining battery level so that software operating in a mobile environment can be run in accordance with the physical resource limits of the terminal. The service provision module (140) automatically transmits a signal that drives a hardware-software fusion control rule suitable for a mobile-based service provision system by determining whether to output a notification by combining the calculated value of the volatility indicator and the remaining battery level of the terminal in an inverse relationship. The price prediction model (132) and the service provision module (140) transmit a control command that helps reduce computing resources and battery consumption of the mobile terminal while predicting bid-ask imbalances and external noise through the combination of data analysis and hardware control logic. The system automatically transmits a command that drives a structure that includes a hardware control methodology beyond a simple business model. The price prediction model (132) transmits a control signal that derives a future price volatility indicator, which is a numerical value that quantifies the possibility of a sudden price change in the near future by calculating the current market bid-ask imbalance and external event noise. The service provision module (140) automatically transmits a command to determine a notification by defining the state in which the calculated volatility indicator reaches a value within the top 5% as the highest warning stage, which is a severe stage requiring user judgment. The price prediction model (132) transmits a control command to defend against logic malfunction caused by media noise by applying a square root relationship, which is a mathematical relationship that adjusts the calculation result value to increase gradually even if a specific variable value increases rapidly.

[0141] The above service provision module (140) can calculate an order imbalance coefficient by applying the logarithmic absolute value to the ratio of the number of buy orders and sell orders aggregated in the same time window, multiplying it by the square root of the number of news article distributions, and then dividing it by the percentage value of the remaining battery of the user terminal unit (200). Accordingly, as the percentage value of the remaining battery decreases, the notification control index can increase even under the same price volatility conditions.

[0142] Additionally, the notification occurrence threshold may be set according to a power conservation standard that sums the screen power coefficient corresponding to the display size of the user terminal unit (200) and the average power consumption coefficient when receiving a push notification once. At this time, the notification occurrence threshold may be adjusted stepwise according to the battery level range. For example, if the battery level is 20% or more, a basic threshold is applied; if the battery level is 10% or more but less than 20%, 1.5 times the basic threshold is applied; and if the battery level is less than 10%, the number of transmissions of general warning notifications may be limited and only emergency notifications may be selectively transmitted. For example, if the battery level is less than 15%, the notification occurrence threshold may be increased by 1.5 times the basic value, and if the battery level is less than 10%, the maximum number of transmissions of general warning notifications may be limited to once per hour.

[0143]

[0144] The 'notification threshold' that determines whether to send or block notifications is not arbitrarily specified by the user, but is dynamically adjusted upward according to a preset battery conservation requirement based on the display size of the user terminal unit and data on the average power consumption consumed by the communication module upon receiving push notifications.

[0145] [Table 4-1] below shows the simulation results of the notification control index calculation based on changes in battery level and the corresponding notification blocking / transmission actions.

[0146] [Table 4-1: Simulation of Battery Level-Linked Terminal Control Index (Conditions: Bid-ask imbalance and article noise are identical)]

[0147]

[0148] As a result of the simulation, it can be confirmed that as the battery level decreases, the notification threshold is gradually raised and the frequency of sending general warning notifications tends to decrease. Accordingly, the service providing module (140) can control the reduction of unnecessary general notifications in situations where price volatility is not significant.

[0149] [Table 4-2: Verification of Terminal Battery Consumption Based on Notification Control Logic (Observation After 12 Hours)]

[0150]

[0151] In periods of high bid-ask volatility, the present invention utilizes an alert control structure that reflects both the remaining battery level and market volatility to limit the number of transmissions of general alerts while maintaining the possibility of delivering emergency alerts. Accordingly, the comparison results confirm that this can function to maintain the delivery of key price fluctuation information while mitigating the battery consumption of the terminal.

[0153] The server unit (100) detects market trading signs for a specific discontinued vintage watch model and transmits a control command to organically initiate the entire valuation process. The transaction data collection module (110) detects a situation where the frequency of new transaction posts registered on an external communication channel increases to 500 per 10 minutes due to rumors of the discontinuation of the watch model and automatically transmits a control signal to dynamically adjust the collection cycle. The transaction data collection module (110) applies a logarithmic relationship to the registration frequency to gradually shorten the collection cycle, and at the same time, if the buffer occupancy of the server unit (100) reaches 90%, it prioritizes applying an exponential relationship to extend the cycle, thereby protecting hardware resources and transmitting a command to transmit the collected data to the internal bus. The data preprocessing module (120) automatically transmits a filtering control command to separate normal depreciation data and false listings intended to manipulate market prices from the received transaction data. The data preprocessing module (120) transmits a control signal to transfer data to a valid database, determining that data with physical depreciation values, such as one broken glass and one missing warranty, identified through text and image analysis among listings registered at 30% lower than the average price is a reasonable listing according to its condition. The data preprocessing module (120) automatically transmits a control command to isolate the data by applying a log weighting penalty to listings of sellers that are registered at a price lower than the average price despite being in the highest condition with a physical depreciation value of 0, and have a past transaction cancellation count of 10 times, thereby deriving an abnormal transaction score of 80 points or higher.

[0154] The price analysis module (130) receives only valid data normalized through the isolation measure and transmits control commands to input the internal fair price calculation model (131) and the price prediction model (132). The fair price calculation model (131) automatically transmits a control signal to add a scarcity premium by physically counting that the remaining number of listings for the corresponding vintage model is low at 3 based on the basic average price of the past month. Even if 500,000 macro attacks are introduced to manipulate the number of views, the fair price calculation model (131) applies a logarithmic relationship to the view count ratio relative to the year of manufacture to converge the price calculation fluctuation rate to within 15%, thereby defending against distortion and transmitting a command to derive a real-time fair transaction price for the watch that reflects the inflation rate. The above price prediction model (132) automatically transmits a control command to derive the future price volatility indicator to the highest warning level by applying the absolute logarithm to the ratio of 150 buy orders and 30 sell orders waiting in the market in conjunction with the above calculation price and taking the square root of the number of 10,000 articles distributed within 24 hours.

[0155] The service providing module (140) transmits a control signal to check the remaining physical battery level of the terminal before providing the derived value assessment result and volatility indicator to the user terminal unit (200). The service providing module (140) recognizes that the remaining battery level of the user terminal unit (200) is very low at 10% and automatically transmits a control command to systematically block frequent warning notifications. The service providing module (140) confirms that the volatility indicator is at the highest warning level and transmits a control command to send an emergency notification once exceptionally to help the user respond and preserve the terminal's power. The user terminal unit (200) automatically transmits a command to activate the display according to the received exception notification signal to visually display comprehensive watch value assessment information and emergency fluctuation situations. The server unit (100) transmits a series of control signals that collect the results generated by the sequential interaction of the transaction data collection module (110), the data preprocessing module (120), the price analysis module (130), and the service provision module (140), and stably provide dynamic changes in the market to the user.

[0156] In one embodiment, the AI ​​learning module (134) may use 5,000 watch transaction datas confirmed to have been successfully concluded over the past 12 months as a learning dataset. At this time, the feature generation module (133) may generate a feature vector for each transaction data that includes brand, model name, manufacturing date, condition grade, physical depreciation value, remaining quantity of listings, cumulative views, seller cancellation history, number of buy orders, number of sell orders, and number of article distributions. When the fair price calculation model (131) derives a first calculated price of a specific watch as 16.2 million won through relational expression-based calculation, the AI ​​inference module (135) may calculate a fair price correction value of 1.04 and an inference reliability of 0.78, and accordingly, the final calculated price may be calculated as approximately 16.84 million won. For example, if the basic average price of the specific watch is 15 million won, and the scarcity correction rate is 0.02, the demand correction rate is 0.035, and the price correction rate is 0.025, the fair price calculation model (131) can derive a first calculated price of approximately 16.2 million won by adding the multiple correction rates to 15 million won. Subsequently, if the AI ​​inference module (135) calculates a fair price correction value of 1.04 and an inference reliability of 0.78, the fair price calculation model (131) can derive a final calculated price of approximately 16.84 million won by reflecting the fair price correction value to the first calculated price. Additionally, the price prediction model (132) can calculate a probability of a price increase of 0.71, a probability of a price decrease of 0.12, and a probability of maintaining the price of 0.17 within 30 days at the same point in time, and if the volatility indicator exceeds a warning threshold, the service provision module (140) can send a control command to generate an emergency notification.

[0157] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. It is understood that various modifications can be made by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols

[0159] Server unit (100) Transaction data collection module (110) Data preprocessing module (120) Price analysis module (130) Fair price calculation model (131) Price prediction model (132) Service provision module (140) User terminal unit (200)

Claims

Claim 1 A user terminal unit that transmits and receives data through a communication network; and a server unit that computes and processes data related to watch transactions; wherein the server unit comprises: a transaction data collection module that collects transaction data including information on watch transaction prices, transaction times, watch brands, model names, condition grades, and transaction frequency from multiple communication channels; a data preprocessing module that normalizes the collected transaction data and determines and separates abnormal transaction data; a price analysis module that receives valid transaction data from which abnormal transaction data has been separated, calculates the real-time fair price of a target watch, and predicts future price volatility; and a service provision module that controls the frequency of notifications and watch valuation information to be transmitted to the user terminal unit based on the analysis results of the price analysis module; wherein the price analysis module comprises: a feature generation module that generates a feature vector from the preprocessed transaction data; an AI learning module that learns model parameters using the feature vector and actual transaction results; an AI inference module that calculates a fair price correction value and a price fluctuation probability value using the learned model parameters; and a model update module that updates the model parameters according to the accumulation of new normal transaction data or an increase in prediction error.It may further include, and the transaction data collection module, in order to calculate a dynamic collection cycle, which is a time interval for collecting data in response to the volatility of clock transaction frequency in the online market, performs a composite operation of an increase obtained by applying an exponential relationship to the current memory buffer occupancy of the server unit and a decrease obtained by applying a logarithmic relationship to the frequency of new transaction post registrations of the target clock model, and automatically transmits a control command to suppress server resource exhaustion by extending the dynamic collection cycle to protect hardware resources when the buffer occupancy increases, and controlling the shortening range to converge gradually and gently when the frequency of new transaction post registrations explodes, and the transaction data collection module measures the communication delay time with the data collection target server and extends and sums it in a linear proportional relationship with the dynamic collection cycle, and determines whether the buffer occupancy exceeds a preset risk threshold; if it exceeds, it transmits an emergency blocking command to immediately fix the dynamic collection cycle to a preset maximum idle period by omitting the calculation of other variables, wherein the risk threshold is periodically updated by the system by inversely calculating the average buffer occupancy at points in time when the communication delay time of the server unit exploded by more than a specific ratio compared to normal times during a preset past period. A system for calculating fair prices and predicting price fluctuations through AI-based clock trading data analysis, which structurally prevents system failure by setting dynamic thresholds. Claim 2 delete Claim 3 delete Claim 4 A system for calculating fair prices and predicting price fluctuations through AI-based watch transaction data analysis, wherein the data preprocessing module calculates the difference between the collected target transaction price and the average transaction price within a preset period, and automatically transmits a control command to prevent the deletion of normal price drop data due to physical damage by misjudging it as abnormal transaction data by extending the normal allowable range of the difference value in an inverse relationship with the physical depreciation figure derived by counting the presence or absence of scratches on the target watch and the number of missing parts through text and image analysis. Claim 5 In claim 4, the data preprocessing module counts the number of past abnormal transaction cancellations of a seller who registered a target watch, adds a penalty value proportional to the number of cancellations in a logarithmic relationship, and calculates a final abnormal transaction score by applying an exponential decay relationship to the time interval between the past point in time when the transaction data occurred and the current point in time when the data is processed, thereby transmitting a control signal to limit the influence of short-term malicious bait listings and block the influence of old noise data, thereby a system for calculating a fair price and predicting price fluctuations through AI-based watch transaction data analysis. Claim 6 In claim 5, the data preprocessing module controls the application of the normal allowable range most strictly by forcibly assigning the physical depreciation value of the target watch to 0, signifying the highest grade state, when an exceptional situation is detected where the seller intentionally leaves the text body of the post blank so that the physical state cannot be identified, and isolates the target transaction price when the final abnormal transaction score exceeds the filtering threshold, wherein the filtering threshold maintains the consistency of the database through a structure in which the average distribution of calculated scores of false listings detected during a past preset period is normalized and the mode interval excluding preset lower noise is tracked.

Images