WO2024005242A1

WO2024005242A1 - Method for processing art work transaction-related data

Info

Publication number: WO2024005242A1
Application number: PCT/KR2022/009501
Authority: WO
Inventors: 이은우; 이승행; 김항주; 신규식; 최윤정; 이유정; 최정아
Original assignee: 주식회사 투게더아트
Priority date: 2022-07-01
Filing date: 2022-07-01
Publication date: 2024-01-04
Also published as: KR20240003772A

Abstract

A method for processing art work transaction-related data of the present invention, may comprise the steps of: generating work data including classification or similarity information for a plurality of works through data preprocessing of calculating text similarity and image similarity of raw data including information about the plurality of works; obtaining a search condition including at least one of a work name, an author name, an image, a category, a size, a price, and a production year; providing selection data including information about at least one work, on the basis of the work data and the search condition; and calculating a first predicted purchase price for a first work included in the selection data, by using machine learning on the basis of first price data including price information according to time.

Description

How we process data related to art transactions

The present invention relates to a method of processing data related to art trade, and more specifically, to a method of calculating the predicted purchase price of an artwork using machine learning.

Recently, as interest in investing in art has increased, interest in art trading or art auctions has also increased. In addition, transactions in which artworks are divided into pieces and owned by multiple people are becoming more frequent. Recently, the 'online auction' platform has been applied to art trading or auctions, and online art trading services are also being provided.

However, because art transaction data is mixed online and offline, it is not easy to personally know how much it is worth or how much is an appropriate purchase price. Additionally, it is difficult to personally know how much profit can be made in the future at the current purchase price and when is the right time to sell. Therefore, a platform is needed that collects mixed transaction data, organizes and analyzes it, and provides predicted purchase prices and returns.

An object of the present invention relates to a method of processing data related to art trade to calculate the predicted purchase price of an artwork using machine learning.

According to an embodiment of the present invention, a method of processing data related to art trade that calculates a predicted purchase price of an artwork using machine learning may be provided.

1 is an environmental diagram of an art trade service providing system according to an embodiment.

Figure 2 is a block diagram of a server providing an art trading service according to an embodiment.

Figure 3 is a flowchart of a method for processing data related to art trade according to an embodiment.

Figure 4 is a flowchart of a raw data preprocessing method according to an embodiment.

Figure 5 is an example to explain raw data preprocessing according to an embodiment.

Figure 6 is a flowchart of a raw data preprocessing method according to another embodiment.

Figure 7 is an example to explain raw data preprocessing according to another embodiment.

Figure 8 is an example to explain raw data preprocessing according to another embodiment.

Figure 9 is a flowchart of a method for selecting and providing works according to an embodiment.

Figure 10 is a flowchart of a method for calculating a predicted purchase price and rate of return according to an embodiment.

Figure 11 is a flowchart of a method for calculating a predicted purchase price and rate of return according to another embodiment.

Figure 12 is a flowchart of a method for recruiting joint buyers according to an embodiment.

Figure 13 is a diagram showing an example screen of an art trading service according to an embodiment.

A method of processing art trade-related data according to an embodiment is a method of processing art trade-related data performed by at least one processor, which includes calculating text similarity and image similarity for raw data containing information about a plurality of works. Generating work data including classification or similar information for the plurality of works through data preprocessing; Obtaining search conditions including at least one of the work name, artist name, image, category, size, price, and production year; providing selection data including information about at least one work based on the work data and the search conditions; And it may include calculating a first predicted purchase price for the first work included in the selection data using machine learning based on first price data including price information over time.

Here, generating the work data includes vectorizing text included in the raw data; Calculating text similarity for each item based on the vectorized text; and grouping works for each item based on the text similarity.

Here, generating the work data includes extracting feature points of the image included in the raw data; calculating image similarity based on the feature points; and grouping works based on the image similarity.

Here, the step of generating the work data includes generating a histogram based on pixel values of the image included in the raw data; calculating image similarity based on the histogram; and grouping works based on the image similarity.

Here, the step of generating the work data includes extracting a density distribution based on the pixel value of the image included in the raw data; calculating image similarity based on the density distribution; and grouping works based on the image similarity.

Here, the step of providing the selection data may include, if there is no work matching the search condition, selecting a similar work that matches at least one item among the items included in the search condition.

Here, calculating the first predicted purchase price includes extracting a first autocorrelation function and a first partial autocorrelation function based on the first price data; setting an autoregressive order, a difference order, and a moving average order based on the first price data, the first autocorrelation function, and the first partial autocorrelation function; Calculating second price data by log-transforming or differentiating the first price data using the autoregressive order, the difference order, and the moving average order; Confirming the normality of the second price data and adjusting the autoregressive order, the difference order, and the moving average order; and calculating the first predicted purchase price for the first work based on the adjusted autoregressive order, difference order, and moving average order.

Here, when there is no price data for the first work, the first price data is price data for a second work that is a similar work that matches at least one item among the items included in the information on the first work. You can.

Here, inputting the first price data into a forget gate and an update gate; calculating a candidate group based on the output result of the forgetting gate; calculating a hidden layer based on the output result of the update gate and the calculated candidate group; and calculating the first predicted purchase price for the first work based on the calculated hidden layer.

Here, extracting a user who has entered the search condition in the past or a user who has a history of purchasing a similar work that matches at least one item among the items included in the search condition; And for group purchase of the first work, transmitting group purchase proposal data including information about the first work, the first predicted purchase price, and information on the joint buyer recruitment status to the user's terminal. It can be included.

Here, the method may further include calculating a future price trend and rate of return for the first work based on the adjusted autoregressive order, difference order, moving average order, and the first predicted purchase price.

Here, the step of calculating a future price trend and rate of return for the first work may be further included based on the calculated hidden layer and the first predicted purchase price.

Here, a computer program stored in a computer-readable recording medium may be provided to execute the art trade-related data processing method.

The embodiments described in this specification are intended to clearly explain the idea of the present invention to those skilled in the art to which the present invention pertains, and the present invention is not limited to the embodiments described in this specification, and the present invention is not limited to the embodiments described in this specification. The scope should be construed to include modifications or variations that do not depart from the spirit of the present invention.

The terms used in this specification are general terms that are currently widely used as much as possible in consideration of their function in the present invention, but this may vary depending on the intention of those skilled in the art, precedents, or the emergence of new technology in the technical field to which the present invention belongs. You can. However, if a specific term is defined and used with an arbitrary meaning, the meaning of the term will be described separately. Therefore, the terms used in this specification should be interpreted based on the actual meaning of the term and the overall content of this specification, not just the name of the term.

The drawings attached to this specification are intended to easily explain the present invention, and the shapes shown in the drawings may be exaggerated as necessary to aid understanding of the present invention, so the present invention is not limited by the drawings.

In this specification, if it is determined that a detailed description of a known configuration or function related to the present invention may obscure the gist of the present invention, the detailed description thereof will be omitted as necessary.

Referring to FIG. 1, an art trade service providing system according to an embodiment may include a server 1000 and at least one user terminal 2000.

The server 1000 is a central component of the art trade service provision system and can function as an overall control unit of the art trade service provision system.

The server 1000 is connected to at least one user terminal 2000 and can transmit and receive data. According to one embodiment, the server 1000 is a server of an art trading platform and can provide information about art works, information about art auctions, information about joint purchases of art works, etc. to the user terminal 2000.

The user terminal 2000 may be a device of a consumer using an art trading platform. Specifically, the user terminal 2000 may be the consumer's mobile device or a server of the consumer's computer.

A consumer who is a user of the user terminal 2000 can use the service provided by the server 1000 through the user terminal 2000. Specifically, the user can search for art works, obtain information about art works, purchase art works, receive recommendations for art works, and receive predicted purchase prices for art works through the user terminal 2000.

Although the art trade service providing system according to an embodiment of the present invention has been briefly described above, the connection relationship between each component may be modified and implemented.

For example, the first terminal and the second terminal may communicate directly, and the first terminal or the second terminal may each receive information from the server 1000. Also, for example, the server 1000 is implemented in the form of a program (e.g., an application) installed on the first terminal or the second terminal, thereby implementing an art trading platform in which only bilateral communication between the first terminal and the second terminal is performed. It can be.

Additionally, in FIG. 1, the number of servers 1000 is shown as one, but the number is not limited to this, and a plurality of servers may exist according to each role.

Referring to FIG. 2, the server 1000 may include a control unit 1100, a communication unit 1200, an analysis unit 1300, a storage unit 1400, and a display unit 1500.

Figure 2 shows five components included in the server 1000, but the illustrated components are not essential, and the server 1000 may have more or fewer components. Additionally, each component of the server 1000 may be physically included in one server, or may be a distributed server distributed for each function.

The control unit 1100 may oversee the operation of the server 1000. Specifically, the control unit 1100 includes at least one processor and can execute the operations of each department by sending control commands to the communication unit 1200, analysis unit 1300, storage unit 1400, and display unit 1500. there is. The control unit 1100 may be a program instruction executed by a computer device.

Unless otherwise specified below, the operation of the server 1000 may be interpreted as being performed under the control of the control unit 1100. Additionally, in the specification of the present invention, the performance of the control unit 1100 may be described with the processor as the main agent.

The communication unit 1200 can connect the server 1000 and an external device to communicate. That is, the communication unit 1200 can transmit and receive data with an external device. For example, the communication unit 1200 may exchange data with the user terminal 2000. The communication unit 1200 may be a program instruction executed by a computer device.

According to one embodiment, the communication unit 1200 may receive information about a work of art from the user terminal 2000. Additionally, the communication unit 1200 may receive information about the artwork selected or selected by the user from the user terminal 2000.

According to one embodiment, the communication unit 1200 may receive data including search conditions entered by the user in a search box to search for a work of art from the user terminal 2000. Additionally, the communication unit 1200 may receive user information about the user's gender, age, artist of interest, favorite painting style, taste, etc. from the user terminal 2000.

The communication unit 1200 may be a communication module that supports at least one of a wired communication method and a wireless communication method. For example, the communication unit 1200 may acquire data from an external device using communication methods such as Bluetooth, Zigbee, BLE (Bluetooth Low Energy), and RFID.

The analysis unit 1300 may perform calculations such as searching for a work of art or calculating an expected purchase price of a work of art based on data received from the user terminal 2000. Additionally, the analysis unit 1300 can predict the sale price and sale time according to the expected rate of return of the artwork, based on price data including information on the price of the artwork. The analysis unit 1100 may be a program instruction executed by a computer device.

The analysis unit 1300 may perform each step included in the art transaction-related data processing method shown in FIG. 3. Details about this will be described later.

The storage unit 1400 can store various data and programs necessary for the server 1000 to operate. The storage unit 1400 may store information acquired by the server 1000. The storage unit 1400 may be a program instruction executed by a computer device.

For example, the storage unit 1400 may store data received from the user terminal 2000 acquired by the communication unit 1200. Specifically, the storage unit 1400 may store user data including user information and preferences. Additionally, the storage unit 1400 may store information about the user's search history and purchase history based on the user's search conditions and the user's purchase history.

Also, for example, the storage unit 1400 may store data including past art transaction details necessary for the server 1000 to provide services. At this time, the past art transaction history data stored in the storage unit 1400 may be updated by information received by the communication unit 1200 or continuously added/modified by the server 1000 itself.

The storage unit 1400 can store data temporarily or semi-permanently. Examples of the storage unit 1400 include a hard disk drive (HDD), solid state drive (SSD), flash memory, read-only memory (ROM), and random access memory (RAM). Or there may be cloud storage, etc. However, the storage unit 1400 is not limited to this and may be implemented with various modules for storing data.

The display unit 1500 can output visual information. For example, the display unit 1500 may be connected to a mobile application display or a computer display to output visual information related to an art trading service. The display unit 1500 may be a program instruction executed by a computer device.

For example, the display unit 1500 may be connected to the display of the user terminal 2000 and output information about the work of art and analysis of the work of art. Additionally, the display unit 1500 may output a screen containing information about the joint purchase of artwork.

Referring to FIG. 3, a method of processing data related to art trade according to an embodiment includes preprocessing raw data (S100), obtaining search conditions entered by the user to search for art (S200), and applying the search conditions to the search conditions. A step of selecting works based on the work and providing information about the work (S300), a step of calculating the predicted purchase price and rate of return for the selected works (S400), and a step of recruiting joint buyers for joint purchase of art works (S500) ) may include.

The raw data preprocessing step (S100) may be a step of organizing transaction data (raw data) of past artworks according to certain standards. At this time, raw data may be offline/online transaction details or auction details for multiple works of art.

Transaction details or auction details for multiple works of art may be sorted or organized differently for each business partner. For example, a specific author's name or work name may be written in English or in Korean. Therefore, a preprocessing step to organize raw data according to certain standards may be necessary.

Preprocessing of raw data may be based on text or images of the work. At this time, the text may include at least one of the artist's name for the work, the artist's year of birth, work name, production year, category, size (horizontal, vertical), estimated lower price, estimated upper price, transaction/successful bid, transaction date, and customer.

Preprocessing of raw data will be described later with reference to FIGS. 4 to 8.

The step of acquiring the search conditions entered by the user to search for a work of art (S200) may be a step of obtaining a search keyword entered by the user into the user terminal 2000. Search conditions may be text-based, image-based, or include both text and images.

At this time, if the search condition is text, the search condition must be at least one of the following: artist name, artist's year of birth, work name, production year, category, size (width, height), estimated lower price, estimated upper price, transaction price/successful bid, transaction date, and customer. It can be included. When the search condition is an image, the search condition may be a work image or may include at least part of the work image.

The step of selecting works based on search conditions and providing information about the works (S300) is a step of selecting works that match the search conditions obtained in step S200 and providing information about the works to the user about the selected works. It can be.

In step S300, if there is a work matching the search condition, the processor may provide information about the matching work to the user. However, if there are no works matching the search conditions, the processor may select works similar to the search conditions and provide information about the selected works to the user.

For example, if the search conditions obtained in step S200 are author - Alena Shymchonak, production year - 2020, the processor selects at least one work that matches the search conditions from the work data generated by preprocessing the raw data and provides it to the user. Information can be provided to. However, if there is no work matching the search conditions in the work data, the processor may select at least one work similar to the search conditions and provide information to the user.

The process of selecting works similar to the search conditions will be described later with reference to FIG. 9.

The step of calculating the predicted purchase price and rate of return for the selected works (S400) is a step of calculating the predicted purchase price and rate of return for the selected works using machine learning based on price data including past price information. You can. At this time, if there is past price data for the selected work, the price data is the price data for the selected work. However, if there is no past price data for the selected work, the price data may be price data for works similar to the selected work.

In step S400, the processor can calculate the predicted purchase price of the work through time-series analysis using machine learning. In addition, based on the predicted purchase price, the rate of return and sale period over time can be predicted and information provided to the user.

The step of calculating the predicted purchase price and rate of return for a work using machine learning will be described later with reference to FIGS. 10 and 11.

The step of recruiting joint buyers for joint purchase of an artwork (S500) may be a step performed in the process of a user purchasing a work after step S400. Specifically, if the price of the work is greater than the user's budget or if the user wishes to make a joint purchase, the processor may perform step S500.

The step of recruiting joint buyers (S500) may be a step of recruiting other users who want to jointly purchase the work selected by the user. According to one embodiment, the processor may advertise information about the work using a banner or the like on an art trading platform. According to another embodiment, other users similar to the user's work tastes may be selected and information about the work may be provided to the other users.

The details of selecting other users will be described later with reference to FIG. 12.

Figure 4 is a flowchart of a raw data preprocessing method according to an embodiment. Specifically, Figure 4 is a flowchart of a method for preprocessing raw data based on text.

Referring to FIG. 4, the raw data preprocessing method according to one embodiment includes a step of vectorizing text (S110), a step of calculating text similarity (S120), and a step of grouping works based on text similarity (S130). can do.

The step of vectorizing the text (S110) is the text included in the raw data: author's name, author's year of birth, work name, production year, category, size (horizontal, vertical), estimated lower price, estimated upper price, transaction price/successful bid, transaction date, and customer. This may be a step of vectorizing at least one of the steps. Before the step of vectorizing the text (S110) is performed, morphological analysis of the text may be performed.

Specifically, the processor can generate a text vector corresponding to the text by adding up all one-hot vectors corresponding to each of the characters included in the raw data. Alternatively, the processor can use Word2Vec to generate a text vector corresponding to the text. The processor can generate a text vector included in each item of raw data (author's name, work title, production year, etc.).

The step of calculating text similarity (S120) may be a step of calculating text similarity based on the text vector generated in step S110. Specifically, the processor can calculate the similarity between texts based on the cosine similarity or singular value decomposition of the vector.

Specifically, the processor may calculate the similarity between texts included in each item. At this time, by calculating the text similarity, the processor can confirm that the author's name, work title, etc. written differently due to spacing, capitalization, etc. are the same text.

The step of grouping works based on text similarity (S130) may be a step of classifying works corresponding to texts whose text similarity is above a certain value into one group based on the text similarity calculated in step S120.

For example, the text similarity between the first author's name and the second author's name may be 90% or more. At this time, the processor may determine that the first author name and the second author name are author names for the same author, and classify the first work corresponding to the first author name and the second work corresponding to the second author name into the same group.

The reason why the processor groups works may be to select similar works when no works match the search conditions, or to recruit joint buyers.

Specifically, if there is no work that matches the first search condition, the work of the first author included in the first search condition may be selected. At this time, the processor may provide the user with information about the works of the group classified as works of the first author. Additionally, in order to recruit joint buyers for the first work, the processor may select users who have a history of purchasing works by the first author of the first work.

As above, the processor can group works with text similarity above a certain value based on the text. At this time, the processor can group the works by item (author's name, work title, production year, etc.).

Referring to Figure 5, you can check the work data created by the processor preprocessing the raw data according to the items (transaction date, author, author's year of birth, title, production year, etc.). Figure 5 may show data for a group classified as 'writer'. However, in Figure 5 as well, the works can be reclassified/regrouped again by production year, size, or business partner.

As shown in Figure 5, the processor can collect and organize transaction data scattered offline/online through data preprocessing to generate work data including classification or similar information for a plurality of works. At this time, similar information may include information on the group to which the work belongs. Specifically, similar information may include information about a group in which works with similar items (author, work name, production year, etc.) are classified.

Figure 6 is a flowchart of a raw data preprocessing method according to another embodiment. Figure 6 is a flowchart of a method for preprocessing raw data based on the image of the work.

Referring to FIG. 6, the raw data preprocessing method according to another embodiment includes extracting feature points, histograms, or density distributions of images (S140), calculating image similarity (S150), and selecting works based on the image similarity. A grouping step (S160) may be included.

The step of extracting feature points, histograms, or density distributions of an image (S140) may be a step of analyzing pixel values of the image. The processor can extract edges that can be characteristic points of the image based on differences in pixel values of the image. Additionally, the processor can generate a histogram using pixel values of the image. Additionally, the processor may extract a density distribution based on the pixel value of the image and the density of the pixel value.

Specifically, the processor may extract the edge of the image based on the difference in pixel values between neighboring pixels of the image. Additionally, the processor can generate a histogram according to the brightness of the image or a histogram according to RGB. Additionally, the processor can generate a density distribution based on the feature points, corners, and colors of the image using the mean-shift algorithm.

The step of calculating image similarity (S150) may be a step of calculating the similarity between images based on the feature points, histogram, or density distribution extracted in step S140.

The processor may calculate the image similarity between the first image and the second image based on the feature points extracted in step S140. Specifically, the processor may compare a feature point of a first image at a specific location with a pixel of the specific location at a second location, and calculate image similarity by contrasting the feature points. Additionally, the processor may calculate image similarity based on the distance between the first and second feature points included in the first image and the distance between the third and fourth feature points included in the second image.

The processor may calculate the image similarity between the first image and the second image based on the histogram generated in step S140. Specifically, the processor may calculate image similarity by comparing the distribution and value of a histogram generated based on brightness and RGB in the first image and the second image. At this time, the processor can compensate for the fact that the histogram may be unrelated to the pixel location by considering all histograms generated based on brightness, R, G, and B.

The processor may calculate the image similarity between the first image and the second image based on the density distribution generated in step S140. Specifically, the processor may compare the density distribution of the feature points of the first image with the density distribution of the feature points of the second image. Additionally, the processor may compare the color-based density distribution of the first image with the color-based density distribution of the second image.

The processor may calculate image similarity based on each of the feature points, histogram, and density distribution, or may calculate image similarity based on a combination of two or more of the feature point, histogram, and density distribution.

The step of grouping works based on image similarity (S160) may be a step of classifying images with image similarity above a certain value into one group based on the image similarity calculated in step S150.

For example, the image similarity between the first image and the second image may be 90% or more. At this time, the processor may determine that the first image and the second image are images of the same work, and classify the first work corresponding to the first image and the second work corresponding to the second image into the same group.

As above, the processor can group works with image similarity above a certain value based on the image. At this time, the processor can group the works by item (author's name, work title, production year, etc.).

Figure 7 is an example to explain raw data preprocessing according to another embodiment. Figure 7 is an example of a method for calculating image similarity based on feature points of the image.

Referring to FIG. 7, the processor can extract feature points 110 of the two images. The processor can determine the similarity of the images by comparing the extracted feature points. In FIG. 7, a work 140 that is similar to the sample work 120 and a work 130 that is dissimilar may be selected based on feature point extraction.

Figure 8 is an example to explain raw data preprocessing according to another embodiment. Figure 8 is an example of a method for calculating image similarity based on a histogram.

Referring to FIG. 8 , the processor may generate a brightness histogram 220, an R histogram 230, a G histogram 240, and a B histogram 250, respectively, of the sample image 210. The processor may select an image similar to the sample image 210 based on the plurality of

histograms

220, 230, 240, and 250.

Figure 9 is a flowchart of a method for selecting and providing works according to an embodiment. Figure 9 is a flowchart of a method for recommending similar works when there are no works that match the search conditions.

Referring to FIG. 9, a method of selecting and providing works according to an embodiment includes the steps of distinguishing items included in a search condition (S310) and selecting a group that matches the first search condition for the first item. (S320), a step of selecting a work that matches the second search condition for the second item among the selected group (S330), and a step of providing information about the selected work (S340).

The step S310 of classifying items included in the search condition may be a step of classifying the items of the search condition in order to search for groups classified in step S130 or step S160. For example, if the search conditions are Author - Alena Shymchonak, Production Year - 2020, Size (Width) - 40 or more, Size (Height) - 60 or more, Category - Oil, the processor has the items necessary to select the works. You can check the author, production year, size (horizontal), size (vertical), and category.

The step of selecting a group that matches the first search condition for the first item (S320) may be a step of confirming a group that matches the search condition for the item classified in step S310. In the above example, if the first item is an author, the processor may select a group classified by the author 'Alena Shymchonak'.

The step (S330) of selecting works that match the second search conditions for the second item among the selected groups selects works that match the search conditions for the second item from among the works included in the group selected in step S320. This may be a selection stage.

In the example above, if the second item is a category in the group classified by the artist 'Alena Shymchonak', the processor can select works of the category 'Oil'. Alternatively, if the second item is a production year, the processor may select works with a production year of 2020 from a group classified by author 'Alena Shymchonak'.

In the step (S340) of providing information about the selected work, the processor provides the artist's name, year of birth, work name, production year, category, size (width, height), estimated lower limit price, and estimated upper limit price for the work selected in step S330. , This may be a step of transmitting data including at least one of the predicted purchase price, rate of return, transaction price/successful bid, transaction date, and customer to the user terminal 2000.

As above, if there is a work that matches the search conditions, the processor provides the user with information about the matched work, or if there is no matching work, the processor provides the user with information about similar works that satisfy part of the search conditions. You can (recommend works).

Figure 10 is a flowchart of a method for calculating a predicted purchase price and rate of return according to an embodiment. Figure 10 is a flowchart of a method for calculating the predicted purchase price and rate of return by machine learning using the ARIMA model.

Referring to FIG. 10, the method of calculating the predicted purchase price and rate of return according to an embodiment includes obtaining and analyzing price data (S411), extracting an autocorrelation function and a partial autocorrelation function (S412), and autoregressive Setting the order, difference order, and moving average order (S413), performing log transformation or difference (S414), adjusting the autoregressive order, difference order, and moving average order (S415), calculating the predicted purchase price It may include a step (S416) and a step of calculating future price trends and rates of return (S417).

The step of acquiring and analyzing price data (S411) may be a step of acquiring price data including a time-series price change trend based on past transaction history for the work and pre-processing and/or analyzing it. When calculating the predicted purchase price and rate of return for the first work, past price data of the first work can be analyzed.

However, if there is no past price data for the first work, the predicted purchase price and rate of return for the first work can be calculated using past price data for the second work similar to the first work. For example, if there is no historical price data for the first work by artist A, the processor can use historical price data for the second work produced in the same year as the first work by artist A.

The step of extracting the autocorrelation function and the partial autocorrelation function (S412) may be a step of extracting a function for using the ARIMA (Autoregressive Integrated Moving Average) model based on the price data obtained in step S411.

AutoCorrelation Function (ACF) refers to a series of autocorrelation according to time lag, and can be calculated using [Equation 1] below.

The processor can determine the normality of price data using the autocorrelation function. At this time, stationary may mean that the average and variance are constant over time. Because non-stationary data is difficult to analyze, the processor may perform log transformation and/or differencing to ensure that the price data is stationary.

Partial AutoCorrelation Function (PACF) refers to a series of partial autocorrelations according to time lag, and can be calculated using [Equation 2] and [Equation 3] below.

By calculating the partial autocorrelation function, the processor can determine only the correlation between the first price and the second price by removing the influence of prices at other times between the first price at the first time and the second price at the second time. You can.

The step of setting the autoregressive order, the difference order, and the moving average order (S413) may be a step of setting parameters to perform log transformation or difference. Specifically, the autoregressive order (p) refers to the number of time lags, the difference order (d) refers to the number of times data has been subtracted from the past value, and the moving average order (q) refers to the order of the moving average model. You can. At this time, the difference order can be determined by the degree of normality of the price data.

In the step of performing log transformation or difference (S414), for stationarity of the price data, the processor log-transforms the price data obtained in step S414 based on the set autoregressive order, difference order, and moving average order, or difference ( differential), or it may be a step that performs both log transformation and difference. Price data that has been log-transformed or differenced may exhibit stationarity. Specifically, the processor can confirm normality by recalculating the autocorrelation function and partial autocorrelation function of the price data on which log transformation or difference has been performed.

The step (S415) of adjusting the autoregressive order, difference order, and moving average order is performed when the stationarity of the price data on which the log transformation or difference was performed in step S414 is not confirmed, the parameters autoregressive order, difference order, and moving average. This may be a step of setting the order to be different from the value in step S413. For example, if the normality of the price data is not confirmed in step S414, the processor may increase the difference order and increase the number of times log transformation or difference is performed.

The step of calculating the predicted purchase price (S416) may be a step of calculating the predicted purchase price of the work using converted or differentiated price data when the normality of the price data is confirmed in step S414 and/or step S415. Specifically, the processor may input price data into a machine learning model modeled by a regression order, difference order, and moving average order with adjusted values, and output a predicted purchase price.

The step of calculating the future price trend and rate of return (S417) may be a step of calculating the predicted purchase price for a future point in time and the rate of return at that point based on the machine learning model used in step S416. At this time, the processor may reflect the predicted purchase price output in step S416 back into the model, remodel it, and then predict the purchase price and rate of return at a future point in time. At this time, for remodeling, the processor may set a new autoregressive order, difference order, and moving average order for the normality of the new price data including the predicted purchase price output in step S416.

Figure 11 is a flowchart of a method for calculating a predicted purchase price and rate of return according to another embodiment. Figure 11 is a flowchart of a method for calculating the predicted purchase price and rate of return by machine learning using the GRU model.

Referring to FIG. 11, a method of calculating the predicted purchase price and rate of return according to another embodiment includes obtaining and analyzing price data (S421), inputting the price data into a forget gate and an update gate (S422), and forgetting. A step of calculating a candidate group based on the output result of the gate (S423), a step of calculating a hidden layer based on the output result of the update gate and the calculated candidate group (S424), and a step of calculating a predicted purchase price based on the calculated hidden layer ( It may include steps S425) and calculating future price trends and rates of return (S426).

The step of acquiring and analyzing price data (S421), similar to step S411 of FIG. 10, is a step of acquiring price data including a time-series price change trend based on past transaction history for the work and pre-processing and/or analyzing it. You can. However, if there is no past price data for a specific work, past price data for works similar to the specific work may be used.

The step of inputting price data into the forget gate and the update gate (S422) is the step of inputting the price data into the forget gate to initialize a certain portion of the price data and inputting it into the update gate to determine the reflection ratio of past price information and current price information. It can be.

Specifically, the processor may initialize a certain portion of the price data by multiplying the previous hidden layer by a value of 0 or 1 using the output of the sigmoid function in the forget gate. Additionally, the processor may determine the ratio of past and present information reflection by multiplying the output of the sigmoid by the current price information and multiplying the past price information by 1 minus the output of the sigmoid in the update gate.

The step of calculating a candidate group based on the output result of the forgetting gate (S423) may be a step of calculating a candidate group of the predicted purchase price using price data partially initialized by the forgetting gate. The processor calculates the candidate group by using the output result of the forgetting gate rather than using the past hidden layer information as is.

The step of calculating a hidden layer based on the output result of the update gate and the calculated candidate group (S424) may be a step of calculating a hidden layer based on the price data considering the past and present reflection ratios and the candidate group calculated in step S423. .

The step of calculating the predicted purchase price based on the calculated hidden layer (S425) may be a step of inputting price data into a machine learning model modeled with the hidden layer calculated in step S424 and outputting the predicted purchase price.

The step of calculating the future price trend and rate of return (S426) may be a step of calculating the predicted purchase price for a future point in time and the rate of return at that point in time, based on the machine learning model used in step S425. At this time, the processor may reflect the predicted purchase price output in step S425 back into the model, remodel it, and then predict the purchase price and rate of return at a future point in time. At this time, the processor may recalculate the candidate group and hidden layer calculated in steps S424 and S425 in order to use new price data including the predicted purchase price output in step S425 for remodeling.

The processor can calculate the predicted purchase price using both the ARIMA model of FIG. 10 and the GRU model of FIG. 11 and then measure the accuracy of the two models. At this time, accuracy can be measured by confirmation of past data, expert opinion, actual transaction or auction price, etc. The processor can provide the user with a predicted purchase price output from the model with higher accuracy among the two models.

Referring to FIG. 12, a method of recruiting joint buyers according to an embodiment may include extracting users (S510) and providing information related to joint purchasing to the extracted users (S520).

The user extraction step (S510) may be a step of extracting users who are likely to have a purchase intention for a specific work. Specifically, when the first user enters the first search condition and selects to purchase the first work, the processor may extract a second user who has entered the first search condition in the past. Alternatively, the processor may extract a second user who has previously entered a search condition in which the detailed search condition for at least one item among the items included in the first search condition is the same. Alternatively, in addition to searches, the processor can also extract users with purchase/auction history.

For example, if the search conditions are writer-Alena Shymchonak, production year-2020, size (width)-40 or more, size (height)-60 or more, and category-Oil, the processor will use the same search conditions as the above search conditions. You can extract users who have searched in the past.

Also, in the above example, the processor can extract users who have searched in the past for the following search terms: Author - Alena Shymchonak, Production Year - 2020. That is, the processor may extract a second user who has previously entered a second search condition that matches only part of the search condition searched by the first user.

Alternatively, the processor may extract other users matching part of the user information based on the user information (including gender, age, artist of interest, favorite painting style, taste, etc.) acquired by the communication unit 1200. It may be possible.

The step of providing information related to a group purchase to the extracted users (S520) may be a step of providing information suggesting a group purchase to the users extracted in step S510. Specifically, the processor may transmit data containing information about the predicted purchase price, rate of return, sale period, and group purchase solicitation status of the work to the extracted user's terminal.

As above, if a user wants to jointly purchase a specific work, the processor can extract users who have similar tastes or are likely to purchase the specific work, and provide data for joint purchase to the extracted users. there is.

Referring to FIG. 13, the art trading platform of the present invention can provide not only information about the work (production year, category, size, etc.), but also future price trends, predicted purchase price, and rate of return when searching for a work or artist. Information provided by the platform can be provided to users in the form of tables or graphs.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with limited examples and drawings, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

Claims

In a method of processing data related to art trades performed by at least one processor,

Generating work data including classification or similar information for a plurality of works through data preprocessing to calculate text similarity and image similarity for raw data containing information about a plurality of works;

Obtaining search conditions including at least one of the work name, artist name, image, category, size, price, and production year;

providing selection data including information about at least one work based on the work data and the search conditions; and

Comprising the step of calculating a first predicted purchase price for the first work included in the selection data using machine learning based on first price data including price information over time.

How we process data related to art transactions.
According to paragraph 1,

The step of generating the work data is,

vectorizing text included in the raw data;

Calculating text similarity for each item based on the vectorized text; and

Including the step of grouping works for each item based on the text similarity.

How we process data related to art transactions.
According to paragraph 1,

The step of generating the work data is,

extracting feature points of images included in the raw data;

calculating image similarity based on the feature points; and

Including grouping works based on the image similarity.

How we process data related to art transactions.
According to paragraph 1,

The step of generating the work data is,

generating a histogram based on pixel values of the image included in the raw data;

calculating image similarity based on the histogram; and

Including grouping works based on the image similarity.

How we process data related to art transactions.
According to paragraph 1,

The step of generating the work data is,

extracting a density distribution based on pixel values of images included in the raw data;

calculating image similarity based on the density distribution; and

Including grouping works based on the image similarity.

How we process data related to art transactions.
According to paragraph 1,

The step of providing the selection data is,

If there is no work matching the search conditions, selecting similar works that match at least one item among the items included in the search conditions.

How we process data related to art transactions.
According to paragraph 1,

The step of calculating the first predicted purchase price is,

extracting a first autocorrelation function and a first partial autocorrelation function based on the first price data;

setting an autoregressive order, a difference order, and a moving average order based on the first price data, the first autocorrelation function, and the first partial autocorrelation function;

Calculating second price data by log-transforming or differentiating the first price data using the autoregressive order, the difference order, and the moving average order;

Confirming the normality of the second price data and adjusting the autoregressive order, the difference order, and the moving average order; and

Comprising the step of calculating the first predicted purchase price for the first work based on the adjusted autoregressive order, difference order, and moving average order.

How we process data related to art transactions.
In clause 7,

If there is no price data for the first work, the first price data is price data for a second work that is a similar work that matches at least one item among the items included in the information on the first work.

How we process data related to art transactions.
According to paragraph 1,

Inputting the first price data into a forget gate and an update gate;

calculating a candidate group based on the output result of the forgetting gate;

calculating a hidden layer based on the output result of the update gate and the calculated candidate group; and

Comprising the step of calculating the first predicted purchase price for the first work based on the calculated hidden layer.

How we process data related to art transactions.
According to clause 9,

If there is no price data for the first work, the first price data is price data for a second work that is a similar work that matches at least one item among the items included in the information on the first work.

How we process data related to art transactions.
According to paragraph 1,

extracting users who have entered the search conditions in the past or users who have a history of purchasing similar works that match at least one item among the items included in the search conditions; and

For group purchase of the first work, transmitting group purchase proposal data including information on the first work, the first predicted purchase price, and information on the joint buyer recruitment status to the user's terminal. doing

How we process data related to art transactions.
In clause 7,

Based on the adjusted autoregressive order, difference order, moving average order, and the first predicted purchase price, further comprising calculating a future price trend and rate of return for the first work.

How we process data related to art transactions.
According to clause 9,

Based on the calculated hidden layer and the first predicted purchase price, further comprising calculating a future price trend and rate of return for the first work.

How we process data related to art transactions.
A computer program stored in a computer-readable recording medium to execute the art trade-related data processing method according to claims 1 to 13.