Marketing terminology translation apparatus and method using a category map

The marketing term translation device addresses e-commerce cross-border challenges by translating product information into category-specific marketing terms, enhancing sales and reducing costs through automated processes.

KR103046873B1Active Publication Date: 2026-11-11GOPHERSOFT INC
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
KR1020250186653
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Priority Date
2024-12-04
Filing Date
2025-12-01
Publication Date
2026-11-11
Estimated Expiration
2045-12-01

AI Technical Summary

Technical Problem

E-commerce cross-border business faces challenges in automating processes such as product sourcing, creating product introduction pages, setting prices considering international shipping costs, and responding to buyer inquiries, with conventional translation technologies failing to translate terms suitable for product sales activities and not distinguishing between categories like fashion and electronics.

Method used

A marketing term translation device using a category map that classifies categories and translates product information into marketing terms, utilizing a category-specific word dictionary generated through artificial intelligence learning, with a similarity threshold of 85% for word replacement.

Benefits of technology

Enhances sales by providing category-specific translations, reducing human resource consumption, and lowering costs by automating processes like product sourcing, registration, and customer service.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112025134856671-PAT00021_ABST
    Figure 112025134856671-PAT00021_ABST
Patent Text Reader

Abstract

An apparatus and method for translating marketing terms using a category map are disclosed. An apparatus for translating marketing terms according to one embodiment of the present invention includes a database for storing data and a processor for controlling the translation apparatus. When product information is input, the processor uses a category map stored in the database to determine a category that matches the input product information, translates the input product information into a target language to generate basic translation data including at least one basic translation word, searches for a marketing word of the determined category, calculates a semantic similarity value between at least one basic translation word and the searched marketing word, and generates marketing translation data according to the calculated semantic similarity value.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a device and method for translating marketing terms using a category map, and more specifically, to a device and method for translating marketing terms that classifies categories using a machine-learned open market category map and translates language into marketing terms according to the classified categories. Background Technology

[0002] E-commerce cross-border business, which involves overseas purchasing agency services in the e-commerce market, generally follows the processes of product sourcing, product processing, product listing, order collection, product delivery, and customer service. In other words, sellers locate products from overseas, translate their descriptions or modify their images, and register them on open marketplaces. The registered products are displayed to buyers (consumers), and when a buyer makes a purchase, the order is received by the seller. The seller purchases the goods on behalf of the buyer and ships them to the buyer's address. Furthermore, the seller must resolve buyer inquiries throughout this entire process.

[0003] However, the lack of solutions to automate each of these processes presents a problem in that sellers must resolve all difficulties themselves. For example, there are challenges such as the difficulty of finding marketable products during the sourcing process, the difficulty of creating new product introduction pages for domestic customers to recognize even after finding products, the difficulty of considering many factors like international shipping costs when setting prices, the difficulty of not being able to ship domestically immediately upon receiving an order, and the difficulty of responding to buyer inquiries.

[0004] In particular, while the accuracy of translations utilizing translation systems has improved, there is a problem in that the terms are not translated into marketing terms suitable for product sales activities in open markets. Furthermore, despite the existence of optimal marketing terms for each category, conventional translation technology uses uniform translation that does not consider category characteristics. Consequently, there is a problem in that it fails to distinguish between the fashion category and the electronics category, even though the same word is used with different marketing terms. The problem to be solved

[0005] The present invention aims to solve the aforementioned problems by providing a marketing term translation device and method capable of translating product information and details into marketing terms corresponding to classified categories. means of solving the problem

[0006] A marketing term translation device using a category map according to an embodiment of the present invention for achieving the above objective includes a database for storing data and a processor for controlling the translation device, wherein the processor loads a category map stored in the database, translates information introducing products according to each country, matches the translated product information with the loaded category map, generates processed data using the matched data and marketing word data, and completes the marketing translation using the generated processed data.

[0007] And the processor compares the category of the generated processed data with the category of the marketing word data, compares the similarity between the meaning of the translated word of the data that matches the category and the meaning of the marketing word, and if the similarity of the comparison result is 85% or more, the word can be replaced with the marketing word to complete the marketing translation.

[0008] In addition, the category map may be a map generated by collecting category data and word dictionaries used in multiple open markets, matching the collected category data with the word dictionaries to generate category-specific word dictionaries, performing artificial intelligence learning using the category-specific word dictionaries as training data, and using the trained artificial intelligence model.

[0009] And the information introducing the above product may include at least one of the product name, detailed information, and options.

[0010] Meanwhile, a method for translating marketing terms using a category map according to an embodiment of the present invention for achieving the above objective may include the steps of: loading a category map stored in a database; translating product introduction information to suit each country; matching the translated product information with the loaded category map; generating processed data using the matched data and marketing word data; and completing the marketing translation using the generated processed data.

[0011] And the step of completing the marketing translation may include a step of comparing the category of the generated processed data with the category of the marketing word data, a step of comparing the similarity between the meaning of the translated word of the data that matches the category and the meaning of the marketing word, and a step of completing the marketing translation by replacing the word with a marketing word if the similarity of the comparison result is 85% or more.

[0012] In addition, the category map may be a map generated by collecting category data and word dictionaries used in multiple open markets, matching the collected category data with the word dictionaries to generate category-specific word dictionaries, performing artificial intelligence learning using the category-specific word dictionaries as training data, and using the trained artificial intelligence model.

[0013] And the information introducing the above product may include at least one of the product name, detailed information, and options. Effects of the invention

[0014] According to various embodiments of the present invention as described above, by providing translations specialized for sales, it is possible to promote sales to consumers and achieve the effect of increasing revenue. In addition, since there is no need to individually verify translations, it also results in the effect of reducing human resource consumption and costs. Brief explanation of the drawing

[0015] FIG. 1 is a conceptual diagram illustrating the operation of an automated product sourcing device for e-commerce cross-border according to an embodiment of the present invention. FIG. 2 is a schematic block diagram for explaining the configuration of a product sourcing automation device according to one embodiment of the present invention, FIG. 3 is a conceptual diagram for explaining the operation of a sourcing module according to an embodiment of the present invention, FIG. 4 is a conceptual diagram for explaining the operation of a processing module according to an embodiment of the present invention, FIG. 5 is a conceptual diagram for explaining the operation of a registration module according to an embodiment of the present invention, FIG. 6 is a conceptual diagram for explaining the operation of an order processing module according to an embodiment of the present invention, FIG. 7 is a conceptual diagram for explaining the operation of a purchasing module according to an embodiment of the present invention, FIG. 8 is a conceptual diagram for explaining the operation of a delivery module according to an embodiment of the present invention, FIG. 9 is a conceptual diagram for explaining the operation of a CS module according to an embodiment of the present invention, FIGS. 10 to 13 are flowcharts for explaining a control method of an automation device according to an embodiment of the present invention, FIG. 14 is a conceptual diagram illustrating a category mapping method according to an embodiment of the present invention, FIG. 15 is a conceptual diagram for explaining a method of matching product categories using a category map according to an embodiment of the present invention, FIG. 16 is a flowchart illustrating a machine learning-based category linkage method for mapping categories of different open markets according to an embodiment of the present invention, FIG. 17 is a flowchart illustrating a method for generating a set of similar words according to an embodiment of the present invention, Fig. 18 is a reference diagram for explaining the category vectorization process, Figure 19 is a diagram of experimental data regarding the relationship between similarity and the angle between vectors, FIG. 20 is a conceptual diagram illustrating a method for translating product information and detailed information into marketing terms using a category map according to an embodiment of the present invention, FIG. 21 is a flowchart illustrating a method for translating marketing terms according to an embodiment of the present invention, FIG. 22 is an example diagram of the category map creation and management process, FIG. 23 is an example diagram of the category inference and matching process, FIG. 24 is an example diagram of a process substituted into marketing terms, and, Figure 25 is an example diagram of a process for dynamically determining a threshold value and a process for comparing a similarity value with a threshold value after determining the threshold value. Specific details for implementing the invention

[0016] Various embodiments of this document are described below with reference to the accompanying drawings. However, this is not intended to limit the technology described in this document to specific embodiments and should be understood to include various modifications, equivalents, and / or alternatives to the embodiments of this document. Similar reference numerals may be used for similar components in connection with the description of the drawings.

[0017] In this document, expressions such as 'have,' 'can have,' 'include,' or 'can include' refer to the existence of the relevant feature (e.g., components such as numerical values, functions, actions, or parts) and do not exclude the existence of additional features.

[0018] In this document, expressions such as 'A or B', 'at least one of A and / or B', or 'one or more of A and / or B' may include all possible combinations of items listed together. For example, 'A or B', 'at least one of A and B', or 'at least one of A or B' may refer to cases including (1) at least one A, (2) at least one B, or (3) both at least one A and at least one B. Expressions such as 'first', 'second', 'first', or 'second' used in this document may modify various components regardless of order and / or importance, and are used only to distinguish one component from another and do not limit said components.

[0019] As used in this document, the expression 'configured to' may be replaced, depending on the context, with, for example, 'suitable for,' 'having the capacity to,' 'designed to,' 'adapted to,' 'made to,' or 'capable of.' The term 'configured to' does not necessarily mean 'specifically designed to' in hardware. Instead, in some situations, the expression 'device configured to' may mean that the device is 'capable of' doing something together with other devices or components.

[0020] The terms used in this specification are for the purpose of describing embodiments and are not intended to limit or / or restrict the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as "comprising" or "having" are intended to indicate the existence of the features, numbers, 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, actions, components, parts, or combinations thereof.

[0021] In the embodiments, a 'module' or 'part' performs at least one function or operation and may be implemented in hardware or software, or a combination of hardware or software. Additionally, a plurality of 'modules' or a plurality of 'parts' may be integrated into at least one module and implemented by at least one processor, except for a 'module' or 'part' that needs to be implemented in specific hardware.

[0022] The present invention will be described in detail below using the attached drawings.

[0023] FIG. 1 is a conceptual diagram illustrating the operation of a product sourcing automation device (1000) for e-commerce cross-border according to an embodiment of the present invention. The product sourcing automation device (1000) may operate in a SaaS form in a cloud environment or in an on-premises form. In the example of FIG. 1, for convenience in understanding each function and operation of the product sourcing automation device (1000), it will be described by dividing it into a total of seven modules. Furthermore, the product sourcing automation device (1000) will be described primarily in terms of operating in a SaaS-type cloud.

[0024] Referring to FIG. 1, the product sourcing automation device (1000) can transmit and receive data with domestic and foreign open market servers (2000), delivery agency servers (3000), and shopping mall servers (4000) and perform overseas purchasing agency operations. Additionally, the product sourcing automation device (1000) visualizes data related to products, purchases, payments, delivery, and customer inquiries so that it can be checked on a seller terminal (5000) and / or an administrator terminal (6000). Referring to FIG. 1, the terms used in this specification will be defined. A seller refers to a person who acts as an agent for a buyer's purchase of goods through overseas purchasing agency. A seller also refers to a user of the automation device (1000) who has obtained usage rights in the form of a subscription, etc. A buyer refers to a general consumer. An open market refers to a platform where a seller sells goods to a buyer. A shopping mall refers to a platform for a seller to source goods requested by a buyer. While open markets and shopping malls may be the same platform, shopping malls are used to refer to places where sellers source products, whereas open markets are used to refer to places where sellers sell the products they have sourced.

[0025] The functions and operations of the product sourcing automation device (1000) can be broadly divided into seven categories, and these are named as the sourcing module (1100), processing module (1200), registration module (1300), order processing module (1400), purchasing module (1500), delivery module (1600), and CS module (1700).

[0026] FIG. 2 is a schematic block diagram for explaining the configuration of a product sourcing automation device (1000) according to an embodiment of the present invention. Referring to FIG. 2, the product sourcing automation device (1000) may include a database (10), a communication unit (20), and a processor (30). In addition, it is obvious that it may include other components such as an input unit (not shown), a display (not shown), and a memory (not shown).

[0027] The database (10) can store data necessary to operate the product sourcing automation device (1000). Additionally, the database (10) can perform operations such as moving, copying, or deleting previously stored data under the control of the processor (30). For example, the database (10) may include various data such as keyword data, shopping mall-related data, product data, infringement determination data, data on similar products, option data, order data, inventory data, and delivery data.

[0028] Memory (not shown) can be used as a concept including a database (10). Memory (not shown) can store various modules, software, functions, AI-trained models, data, etc. for operating a product sourcing automation device (1000). For example, a function that performs an operation to extract product data can be stored in memory (not shown), and at least one AI-trained model for appropriately translating for each product category can be stored.

[0029] The communication unit (20) communicates with external servers and devices such as an open market server (2000), a delivery agency server (3000), a shopping mall server (4000), a seller terminal (5000), and an administrator terminal (6000). For example, the product sourcing automation device (1000) can access a webpage on an external server through the communication unit (20) to scrape product data.

[0030] To this end, the communication unit (20) can use wired communication methods such as LAN, HDMI, etc., and wireless communication methods such as wireless LAN, NFC, IR communication, Zigbee communication, Bluetooth, etc.

[0031] The processor (30) can control the remaining configuration of the product sourcing automation device (1000). For example, the processor (30) can control the database (10) to store the final product data by including data for similar products in the first temporary data.

[0032] The processor (30) can be implemented as a single CPU to perform product sourcing, product information processing, product registration, order processing, etc., and may also be implemented as multiple processors and IPs that perform specific functions. The specific operation of the processor (30) will be described later.

[0033] FIG. 3 is a conceptual diagram for explaining the operation of a sourcing module (1100) according to an embodiment of the present invention. Referring to FIG. 3, the sourcing module (1100) may include a popular search term extraction function (1110), a product data extraction function (1120), a product filter function (1130), and a similar product extraction function (1140). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description.

[0034] The processor (30) can call the popular search term extraction function (1110) to extract and store keywords that have been searched more than a preset frequency. This performs the function of finding keywords that match trends in order to source products likely to be sold in large quantities. The processor (30) can scrape web pages. And when the processor (30) detects a preset word indicating a popular search term during scraping, it can extract the keyword and store it in the database (10). For example, the processor (30) can extract and store popular keywords by using the detection of a word such as 'Popular Search Terms TOP 10' on the web page being scraped as a trigger.

[0035] More specifically, when a pre-set word representing popular search terms, such as 'Top 10 Popular Search Terms', is identified, the processor (30) can analyze the scraped webpage to identify the linked search engine. Then, the processor (30) can extract keywords that have been searched more than a pre-set frequency through the identified search engine and store them in the database (10). For example, the processor (30) can extract popular keywords through the trend analysis function of the search engine.

[0036] Next, the processor (30) can call the product data extraction function (1120) to collect information about shopping malls that are selling products corresponding to the extracted keywords. The processor (30) can search on the shopping platform using the keywords extracted and stored in the database (10). The shopping platform may be various overseas open markets or a company's own mall. The processor (30) can scrape the URL address, shopping mall name, etc. of the shopping mall searched on the shopping platform and store them in the database (10). In the following, actions such as receiving data from the shopping mall may be interpreted as receiving data from the shopping mall server (4000). For convenience of explanation, the shopping mall and the shopping mall server (4000) are used interchangeably.

[0037] And the processor (30) can call the product filter function (1130) to collect product data, and at the same time, delete product data for products that cannot be sold due to the risk of infringing intellectual property rights. The processor (30) can store the product data of the shopping mall in the database (10) using the collected shopping mall URL address.

[0038] Specifically, the processor (30) can scrape product data from a shopping mall using a stored shopping mall URL address. At this time, the processor (30) can store the product data from the shopping mall as first temporary data. Then, the processor (30) can compare the stored first temporary data with infringement judgment data.

[0039] The infringement determination data may include prohibited word data and deletion data. The prohibited word data may be a pre-set data set based on words registered as trademarks. The deletion data may be a data set in which the names, trademarks, names, etc. of products that are deleted as products that cannot be sold through the operation of the product filter function (1130) are collected.

[0040] If, as a result of comparison with the infringement determination data, it is determined that the first temporary data corresponds to the infringement determination data, the processor (30) can determine that the first temporary data is data regarding a product that cannot be sold and delete it. Then, the processor (30) can manage the first temporary data determined as data that cannot be sold by adding it as deleted data. Through this, the data set for determining the risk of intellectual property infringement can be continuously updated.

[0041] The processor (30) may call the similar product extraction module (1140) to search for and manage similar products together using the stored product data. The processor (30) may extract product images constituting the first temporary data. Then, the processor (30) may search for similar products using the extracted product images. Subsequently, the processor (30) may control the database (10) to include data regarding the similar products searched in the first temporary data and store them as final product data. Additionally, the processor (30) may prevent duplicate data collection when searching for similar product data by using the shopping mall name stored together when storing the shopping mall URL address.

[0042] FIG. 4 is a conceptual diagram for explaining the operation of a processing module (1200) according to an embodiment of the present invention. Referring to FIG. 4, the processing module (1200) may include an option page creation function (1210), an option translation function (1220), and a detail page creation function (1230). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description.

[0043] The processor (30) calls the option page creation function (1210) to perform the task of creating multiple options for a single product from data of a product to be sold, specifically multiple similar products sourced from multiple shopping malls. Specifically, the processor (30) can classify the data for similar products stored through the sourcing module (1100) into options. The processor (30) can control the database (10) to store the product data of the first shopping mall from which the product is to be sourced as the first temporary data. The processor (30) can also search for similar products from at least one shopping mall different from the first shopping mall using the product images constituting the first temporary data. The data for the searched similar products can be stored as final product data along with the first temporary data.

[0044] The processor (30) can use the stored final product data to generate option data to be attached to each of the multiple products searched in the first shopping mall and at least one shopping mall different from the first shopping mall. The processor (30) can also use the generated option data to generate an option page that configures the searched 'multiple products' into 'other options of one product'.

[0045] More specifically, when generating option data, the processor (30) can compare the first temporary data of the stored final product data with data for similar products. The processor (30) can extract differences between these multiple products. The processor (30) can match the extracted differences with pre-set option classification criteria. By matching with option classification criteria having a certain standard, the creation of option pages can be done consistently. For example, the option classification criteria can be composed of standardized specifications such as color, size, dimensions, and length. The processor (30) can generate option data based on the matched option classification criteria.

[0046] Next, the processor (30) can generate an option page with a function added to select a product corresponding to each option based on the generated option data. In some cases, the number of option data may exceed one; in such cases, the option page is generated based on the first option, and options after the second option can be combined and displayed. When multiple option data exist, which option is set as the first option can be determined according to the option selection priority set according to the product category. For example, depending on the product category, the criteria for customer selection may be color or size.

[0047] The processor (30) can set different prices for each product corresponding to each option when creating an option page. Since products with different options are likely to be sourced from different shopping malls, prices may not be consistent. Setting option prices is intended to adjust for this price inconsistency.

[0048] Specifically, the processor (30) can set the option price for each product corresponding to each option based on the price at which the product corresponding to each option is sold at the shopping mall sourcing the product, shipping costs by product category, exchange rates, margins, payment fees, and shipping agency information.

[0049] And the processor (30) can determine the regular price and the selling price to be entered on the option page based on the set option price. For example, the regular price can be set to a price corresponding to 100% of the option price. The selling price is set to a price equal to or lower than the option price, and based on the data of each item used to set the option price, the processor (30) can set the selling price to a value between 100% and 50% of the option price by deriving an item that can lower the price.

[0050] The processor (30) can perform appropriate translations for each product category by calling the option translation function (1220). Specifically, the processor (30) can select one of a plurality of trained translation models based on the category of the target product for which the option page was created. The plurality of trained translation models may be artificial intelligence models that have learned the language, terminology, nuances, etc. commonly used in the field based on the category. The processor (30) can translate the option page using the selected trained translation model.

[0051] Next, the processor (30) can create a product detail page to be registered on the open market server (2000) by calling a detail page creation function (1230). Specifically, the processor (30) can create a product detail page by combining a translated option page, a header image file, a footer image file, and a product page containing the saved final product data. Then, the processor (30) can control the database (10) to save the created product detail page.

[0052] When generating a product detail page, the processor (30) can identify images or text related to prices present in the product page included in the stored final product data. For example, it identifies an image that has a portion of the image containing the overseas local selling price. Then, the processor (30) can remove the identified images or text. Of course, appropriate image processing can be performed so that the removed portion does not look awkward. Through this, it is possible to prevent price information different from the selling price set through the option price from being included in the product detail page.

[0053] FIG. 5 is a conceptual diagram for explaining the operation of a registration module (1300) according to an embodiment of the present invention. Referring to FIG. 5, the registration module (1300) may include a product registration function (1310) and a product initialization function (1320). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description. In the following description, operations such as receiving data from an open market may be interpreted as receiving data from an open market server (2000). For convenience of explanation, the terms open market and open market server (2000) will be used interchangeably.

[0054] The processor (30) can register products on multiple open markets by calling the product registration function (1310). To register on each open market, data for product registration must be generated to meet the requirements of each open market. To this end, the processor (30) can generate a data format that meets the respective requirements for registering on at least one open market. The processor (30) can convert product data, including a previously stored product detail page, into a JSON format and apply the converted product data to the generated data format to generate data that can be used for product registration.

[0055] The processor (30) can register a product on behalf of the seller in a corresponding open market using the generated product registration data. At this time, authentication of the seller is required for product registration. To this end, when product registration data is generated, the processor (30) can control the communication unit (20) to transmit authentication for product registration to the seller's terminal (5000). When a key value is received from the seller terminal (5000) through the communication unit (20), the processor (30) can complete authentication by comparing the received key value with the key value stored in the database (10). Once authentication is completed, the processor (30) can proceed with product registration in the open market using the product registration data.

[0056] And when the processor (30) receives from the open market server (2000) via the communication unit (20) that product registration is complete, it can match the registered product with the seller and store it in the database (10). By matching the seller with the product, it is possible to request payment from the seller when an order for the product is placed, notify the seller when a CS occurs, and perform actions such as settling the sale to the seller when sales revenue is generated.

[0057] The processor (30) can call the product initialization function (1320) to initialize the product registration and proceed with new product registration. Since open markets are platforms where products reflecting trends sell well, it is not appropriate to register and sell the same product for a long period without change. This is especially true for sellers who generate revenue through overseas purchasing agency. For this reason, the processor (30) can execute product registration initialization after a preset period has elapsed following the completion of product registration. For example, initialization can be executed after one month has elapsed since product registration, or initialization can be executed on the 1st of every month.

[0058] Specifically, the processor (30) can delete all products for which product registration has been completed from the open market. It can also classify products stored in the database (10) to select products for new product registration. Even after product registration, products are continuously added to the database through the sourcing module (1100) and the processing module (1200). The processor (30) can classify products stored in the database (10) into products that have been sold, products that have not been sold, and new sourced products. Products that have been sold can be classified based on whether there is a history of sales. Products that have not been sold and new sourced products can be classified based on the date they were registered in the database (10).

[0059] Next, the processor (30) can adjust the ratio of products of each category through a distribution model. For example, the distribution model may be a fully trained model that takes at least one of the sales frequency, the set sales price, and the subscription period of the seller as input values ​​and outputs a product distribution ratio that maximizes the total sales amount. Once the optimal product distribution ratio is determined, the processor (30) can proceed with product registration again according to the adjusted ratio of products. The process of proceeding with product registration again is the same as the process of registering products on an open market by calling the product registration function (1310) described above.

[0060] FIG. 6 is a conceptual diagram for explaining the operation of an order processing module (1400) according to an embodiment of the present invention. Referring to FIG. 6, the order processing module (1400) may include an inventory check function (1410), a margin calculation function (1420), a personal customs clearance code check function (1430), and an order approval function (1440). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description.

[0061] The processor (30) can collectively collect and process orders from various open markets operated through the order processing module (1400). In particular, it collects information on the inventory quantity and fluctuating prices of shopping malls from which products are sourced, and if the pre-set conditions are not met, it cancels the order to prevent the seller from suffering a loss.

[0062] When a processor (30) receives an order signal through a communication unit (20) from at least one open market server (2000) where a product is registered, it can call an inventory check function (1410) to determine whether the inventory quantity of the product corresponding to the received order signal is greater than or equal to the order quantity. Specifically, the processor (30) can transmit a signal to a shopping mall server (4000) that sources the product to check the inventory quantity of the product, and determine whether the order can be processed by comparing the product inventory quantity information received from the shopping mall server (4000) with the order quantity.

[0063] If it is determined that the inventory quantity of a product is greater than or equal to the order quantity, the processor (30) calls the margin calculation function (1420) to calculate the margin. At this time, the margin may be a margin expressed as a monetary amount or a margin rate expressed as a percentage. The processor (30) can control the communication unit (20) to receive product price information from the shopping mall that has verified the product inventory quantity. Then, the processor (30) can calculate the margin by comparing the received product price information with the sales price registered on the open market server (2000). Since there is a possibility that the shopping mall from which the product is sourced has changed the price, a new margin is calculated and compared with the margin set when the sales price was previously determined. The processor (30) can determine that the condition is satisfied if the current margin is equal to or higher than the margin set when the sales price was determined. Conversely, if the margin set when the sales price was determined is lower than the current margin, the processor can determine whether the condition is satisfied by determining whether it corresponds to an allowable margin.

[0064] If the calculated margin does not satisfy these preset conditions, the processor (30) can control the communication unit (20) to send a signal to the open market server (2000) that received the order signal to cancel the order corresponding to the received order signal. Conversely, if the calculated margin satisfies the preset conditions, the processor (30) can control the database (10) to change the state of the received order signal to an order-available state.

[0065] Next, the processor (30) can call the personal customs clearance code verification function (1430) to check whether information required for overseas direct purchase agency has been entered. When it is confirmed that the state stored in the database (10) has been changed to an orderable state, the processor (30) can determine whether a value corresponding to the personal customs clearance code exists. If it is determined that a value corresponding to the personal customs clearance code has been entered in advance, the processor (30) can control the database (10) to change the orderable state to a delivery ready state. Conversely, if there is no value corresponding to the personal customs clearance code, the processor (30) can wait until a value is entered.

[0066] When it is confirmed that the status has changed to a shipping preparation state, the processor (30) can call the order approval function (1440) to approve the order. Then, when the order is approved, the processor (30) can control the communication unit (20) to transmit information regarding the approved product to the seller terminal (5000). Although the seller does not need to directly approve the order through the automation system, this is to notify the seller of the details regarding the approved order.

[0067] FIG. 7 is a conceptual diagram for explaining the operation of a purchasing module (1500) according to an embodiment of the present invention. Referring to FIG. 7, the purchasing module (1500) may include a payment request function (1510), a payment approval function (1520), an order ordering function (1530), and an order details transmission function (1540). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description.

[0068] The processor (30) can automate the batch processing of purchasing agency from various shopping malls according to the order history of multiple open market servers (2000) through the purchasing module (1500). The processor (30) can control the database (10) to change the status from the shipping preparation state to the order approval state when an order is approved. When it is confirmed that the status has been changed to the order approval state, the processor (30) can control the communication unit (20) to call the payment request function (1510) and send a payment request message regarding the details of the sale price to the seller terminal (5000). The processor (30) can also control the database (10) to change the status to the payment request state.

[0069] Next, the processor (30) calls the payment approval function (1520) to proceed with the payment. When a payment request signal is received from the seller terminal (5000), the processor (30) can control the communication unit (20) to retrieve the details to be paid from the database (10), configure the screen, and transmit the configured screen to the seller terminal (5000). When a signal indicating that the payment has been approved is received from the seller terminal (5000), the processor (30) can control the database (10) to change the state to a payment approval state.

[0070] When it is confirmed that the payment approval status has changed, the processor (30) can order products by calling the order order function (1530). The processor (30) can retrieve a list of products in the payment approval status from the database (10), create an order form suitable for each shopping mall server (4000), and transmit it. Additionally, the processor (30) can proceed with payment by utilizing the payment method provided by each shopping mall server (4000). The processor (30) can control the database (10) to change the status to order order completed for products for which payment has been completed at the shopping mall.

[0071] Subsequently, the processor (30) can control the communication unit (20) to call the order details transmission function (1540) to transmit details of the completed shopping mall payment to the shipping agency server (3000). The content transmitted to the shipping agency server (3000) may include the origin country code, destination country code, shipping method, customs clearance classification, personal customs clearance unique code, recipient information, product information, inspection options, whether freight shipping fees have been paid, and whether customs duties and taxes have been paid. In response to this, the shipping agency server (3000) transmits the destination country tracking number of the transmitted product. The processor (30) can control the database (10) to store the destination country tracking number received through the communication unit (20).

[0072] FIG. 8 is a conceptual diagram for explaining the operation of a delivery module (1600) according to an embodiment of the present invention. Referring to FIG. 8, the delivery module (1600) may include a shipping label verification function (1610), a shipping label transmission function (1620), and a delivery tracking function (1630). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description.

[0073] The processor (30) can check the shipping label in transit by calling the shipping label verification function (1610) at preset time intervals. The shipping label information can be received from the delivery agency server (3000). The processor (30) can also control the database (10) to update the verified information. The processor (30) can configure a screen displaying the delivery status and control the communication unit (20) to transmit the configured screen to the seller terminal (5000).

[0074] Next, the processor (30) can call the shipping label transmission function (1620) to transmit the destination country shipping label number to the open market server (2000) where the purchase occurred, thereby enabling domestic delivery tracking. While delivery is in progress overseas, delivery tracking is not possible through the open market server (2000), but after the cargo is transported to the domestic market, delivery tracking becomes possible through the open market server (2000).

[0075] The processor (30) calls the delivery tracking function (1630) to provide the delivery status to the seller terminal (5000). The processor (30) can visualize the delivery status of the database (10) in real time. The processor (30) can also control the communication unit (20) to transmit the visualized information to the seller terminal (5000). Since delivery tracking is not possible through the open market server (2000) when the product is being shipped overseas, when the processor (30) receives a signal from the seller terminal (5000) requesting that the buyer be notified about the product being shipped overseas, the processor (30) can control the communication unit (20) to transmit a notification message about the current delivery status overseas to the buyer.

[0076] FIG. 9 is a conceptual diagram for explaining the operation of a CS module (1700) according to an embodiment of the present invention. Referring to FIG. 9, the CS module (1700) may include a return processing function (1710), an exchange processing function (1720), a cancellation processing function (1730), an inquiry confirmation function (1740), and an inquiry processing function (1750). Since these modules and functions are called and operated through a processor (30), for convenience of explanation, the names of the processor (30) and each module and function will be used interchangeably in the following description.

[0077] The processor (30) can check the history of inquiries made to the seller through the open market server (2000) via the CS module (1700) and automatically take action.

[0078] The processor (30) may call the return processing function (1710) at preset intervals to request data of the product for which a return request has occurred from the open market server (2000). Based on the received data, the processor (30) updates the database (10). The product for which a return request is in progress may be visualized by the processor (30) and transmitted to the administrator terminal (6000). When a return approval signal is received from the administrator terminal (6000), the processor (30) may control the database (10) so that information indicating that the return has been approved is updated.

[0079] The processor (30) may call the exchange processing function (1720) at preset intervals to request data of the product for which an exchange request has occurred from the open market server (2000). Based on the received data, the processor (30) updates the database (10). The product for which an exchange request is in progress may be visualized by the processor (30) and transmitted to the administrator terminal (6000). When an exchange approval signal is received from the administrator terminal (6000), the processor (30) may control the database (10) so that information indicating that the exchange has been approved is updated.

[0080] The processor (30) may call the cancellation processing function (1730) at preset intervals to request data of the product for which a cancellation request has occurred from the open market server (2000). Based on the received data, the processor (30) processes the order cancellation through the following process. First, the processor (30) determines whether the order processing has already begun or has already begun. If the order processing has already begun, the processor (30) may control the database (10) to immediately cancel the order and update the product data. Conversely, if the order processing has already begun, the processor (30) rejects the cancellation request. The processor (30) may also control the communication unit (20) to send a notification message to the buyer stating that cancellation is difficult because the product purchase or delivery process is in progress, but that a return is possible in the future.

[0081] The processor (30) can collect inquiry history from the open market server (2000) by calling an inquiry confirmation function (1740) at preset intervals. The collected inquiry history is stored in the database (10), and the processor (30) can visualize the history and control the communication unit (20) to transmit the visualized history to the seller terminal (5000) and the administrator terminal (6000).

[0082] The inquiry processing function (1750) is called by the processor (30) when an operation signal is received from the administrator terminal (6000) through the communication unit (20). The processor (30) can use a chatbot to explore the context of the inquiry made by the buyer and classify it into simple inquiries and complex inquiries. If classified as a simple inquiry, the processor (30) can send a rule-based answer to the open market server (2000). If classified as a complex inquiry, the processor (30) can send the answer received from the administrator terminal (6000) to the open market server (2000).

[0083] According to various embodiments of the present invention as described above, a seller who sells products on an open market through overseas purchasing agency can reduce the effort required to find products, process products, register products, purchase products when an order is received, process delivery, and perform customer service including customer management, and can also reduce the costs incurred.

[0084] Furthermore, this cost-saving sales process allows buyers to quickly receive products at lower prices and enables prompt action regarding defective goods.

[0085] In the following, additional explanations will be provided through flowcharts to aid in understanding the various embodiments of the present invention described above. The specific control method of the automation device (1000) described through the flowchart corresponds to the contents of the automation device (1000) described above.

[0086] FIG. 10 is a flowchart for explaining a control method of an automation device (1000) according to an embodiment of the present invention. In particular, FIG. 10 describes a method for automating product sourcing. According to FIG. 10, the automation device (1000) can scrape web pages (S1010). If a preset word representing a popular search term is identified during scraping, the automation device (1000) can extract and store keywords that have been searched more than a preset frequency (S1020). Specifically, if a preset word representing a popular search term is identified, the automation device (1000) can analyze the scraped web page to determine what the linked search engine is. Then, through the identified search engine, keywords that have been searched more than a preset frequency can be extracted and stored. For example, keywords that have been searched more than a preset frequency can be extracted through the trend analysis function of the search engine.

[0087] Next, the automation device (1000) can search on a shopping platform using the extracted and stored keywords (S1030). Then, the automation device (1000) can store the shopping mall URL address found on the shopping platform (S1040). At this time, in addition to the shopping mall URL address, the shopping mall name, etc., can be additionally stored. For example, the shopping mall name can be used as a keyword to prevent duplicate searches when searching for similar products.

[0088] The automation device (1000) can store product data of a shopping mall using a stored shopping mall URL address (S1050). Specifically, the automation device (1000) can store the product data of the shopping mall as first temporary data. Then, it can determine whether the first temporary data is data regarding a product that can be sold by comparing it with infringement determination data. If it corresponds to the infringement determination data, the first temporary data can be determined as data that cannot be sold and deleted. The infringement determination data is divided into prohibited word data and deleted data, and the first temporary data that was deleted because it corresponded to the infringement determination data can be added to the deleted data. Through this, the effect of continuously updating the infringement determination data can be achieved.

[0089] Additionally, the automation device (1000) can search for similar products using product images constituting the first temporary data. Data regarding the searched similar products can be included in the first temporary data and stored as final product data.

[0090] FIG. 11 is a flowchart for explaining a control method of an automation device (1000) according to an embodiment of the present invention. In particular, FIG. 11 describes a method for automating product page processing. Referring to FIG. 11, the automation device (1000) can store product data from a first shopping mall to source products as first temporary data (S1110). Then, the automation device (1000) can search for similar products from at least one shopping mall different from the first shopping mall using product images constituting the first temporary data (S1120), and can store the data for the similar products searched in the first temporary data as final product data (S1130).

[0091] Next, the automation device (1000) can generate option data to be attached to each of a plurality of products searched in the first shopping mall and at least one shopping mall different from the first shopping mall using the stored final product data (S1140). Specifically, the automation device (1000) can compare the first temporary data of the stored final product data with the data for similar products searched, and extract the differences between the first shopping mall and the plurality of products searched in the at least one shopping mall different from the first shopping mall. Then, the extracted differences can be matched with pre-set option classification criteria to generate option data.

[0092] And the automation device (1000) can generate an option page that configures multiple searched products as different options of a single product using the generated option data (S1150). That is, it can generate an option page that includes a function to select a product corresponding to each option based on the generated option data.

[0093] And the automation device (1000) can generate an option page and determine and add a sales price. The automation device (1000) can set an option price for each product corresponding to each option based on the price at which the product corresponding to each option is sold in a shopping mall that sources the product, shipping costs by product category, exchange rates, margins, payment fees, and shipping agent information. Then, based on the set option price, the regular price and sales price to be entered on the option page can be determined, and the sales price can be set to be less than or equal to the option price.

[0094] Additionally, the automation device (1000) may also translate an option page. Specifically, the automation device (1000) may select one of a plurality of fully trained translation models based on the category of the target product for which the option page was created. Then, the option page may be translated using the selected fully trained translation model. The automation device (1000) may create and save a product detail page by combining the translated option page with a header image file, a footer image file, and a product page containing the saved final product data. When creating the product detail page, the automation device (1000) may identify and remove images or text related to prices present in the product page included in the saved final product data.

[0095] FIG. 12 is a flowchart for explaining a control method of an automation device (1000) according to an embodiment of the present invention. In particular, FIG. 12 describes a method for automating the process of registering a product on an open market. According to FIG. 12, the automation device (1000) can generate a data format in accordance with each requirement for registration on at least one open market (S1210). Then, the automation device (1000) converts product data including a previously stored product detail page into a JSON format (S1220), and applies the converted product data to the generated data format to generate data that can proceed with product registration (S1230).

[0096] Next, the automation device (1000) can register a product on a corresponding open market on behalf of the seller using the generated product registration data (S1240). Specifically, when the product registration data is generated, the automation device (1000) can transmit authentication for product registration to the seller's terminal. Then, it performs authentication by comparing the key value received from the seller's terminal with the key value stored in the database, and if authentication is completed, it can register the product on the open market. When it receives from the open market server that the product registration is complete, the automation device (1000) can match the registered product with the seller and store them (S1250).

[0097] Additionally, the automation device (1000) can perform product registration initialization after a preset period has elapsed following the completion of product registration. Specifically, the automation device (1000) can start the initialization by deleting all products for which product registration has been completed from the open market. Subsequently, the automation device (1000) can classify products stored in the database into products that have been sold, products that have not been sold, and newly sourced products. Then, the automation device (1000) can proceed with product registration again by adjusting the ratio of products in each category through a distribution model. For example, the distribution model may be a completed learning model that takes at least one of the sales frequency, the set sales price, and the subscription period of the seller as input values ​​and outputs a product distribution ratio that maximizes the total sales amount.

[0098] FIG. 13 is a flowchart for explaining a control method of an automation device (1000) according to an embodiment of the present invention. In particular, FIG. 13 describes a method for automating the operation of processing multiple orders received from multiple open markets. Referring to FIG. 13, when an order signal is received from at least one open market server, the automation device (1000) can determine whether the inventory quantity of a product corresponding to the received order signal is greater than or equal to the order quantity (S1310). If the inventory quantity is less than the order quantity, the automation device (1000) can cancel the order. Conversely, if it is determined that the inventory quantity of a product is greater than or equal to the order quantity, the automation device (1000) can calculate a margin (S1320). Specifically, the automation device (1000) can receive product price information sold at a shopping mall that has confirmed the product inventory quantity, and calculate a margin by comparing the received product price information with the sales price registered for the product. Then, the automation device (1000) can determine whether the calculated margin satisfies a preset condition.

[0099] If the calculated margin satisfies the preset conditions, the automation device (1000) can change the state of the received order signal to an order-available state (S1330). Conversely, if the calculated margin does not satisfy the preset conditions, the automation device (1000) can cancel the order corresponding to the received order signal.

[0100] When it is confirmed that the ordering status has been changed to an orderable state, the automation device (1000) can check whether a value corresponding to the personal customs clearance code exists. If it is determined that a value corresponding to the personal customs clearance code exists, the automation device (1000) can change the orderable state to a shipping preparation state. Conversely, if it is determined that a value corresponding to the personal customs clearance code does not exist, the automation device (1000) can wait until a value corresponding to the personal customs clearance code is entered. When it is confirmed that the status has been changed to a shipping preparation state, the automation device (1000) can approve the order and transmit information regarding the approved product to the seller terminal.

[0101] In addition, the method of automating the collection of orders from various open markets and purchasing products from each shopping mall, the method of automating the confirmation of product delivery, and the method of automating the processing of CS such as buyer's return requests and inquiries are replaced by the description of the automation device (1000) described above.

[0102] FIG. 14 is a conceptual diagram illustrating a category mapping method according to an embodiment of the present invention. The above-described automation device (1000) may be named differently according to its function, and in the embodiment of FIG. 14, it may be named a category linkage device (1000).

[0103] The category linkage device (1000) can link categories of different open markets by mapping them based on machine learning. The category linkage device (1000) can collect category data and word dictionaries used in multiple open markets. Then, it can generate a category-specific word dictionary by matching the collected category data with the word dictionary. Subsequently, it can perform artificial intelligence learning using the category-specific word dictionary as training data and generate a category map using the trained artificial intelligence model.

[0104] Specifically, when generating a category map, the category linkage device (1000) inputs category data of a new open market into an artificial intelligence model that has undergone training, and can determine a set of similar words that may include the input category data of the new open market. Then, the category map can be constructed according to the determined set of similar words.

[0105] Additionally, when determining a set of similar words, the category linkage device (1000) can determine whether the input category data of a new open market has a similarity of 85% or more to a specific set of similar words. If it is determined that it has a similarity of 85% or more, the category linkage device (1000) can place the category data of the new open market in a specific set of similar words. Conversely, if it is determined that it has a similarity of less than 85%, the category linkage device (1000) can place it in a set containing the similar words most similar to the category data of the new open market.

[0106] Referring again to FIG. 14, the category linkage device (1000) can perform data learning using a machine learner on category data and a word dictionary used in various open markets. The category data is matched with the word dictionary, and similar words can be formed into sets. This is done by training the category data of various open markets using a machine learner. Next, when a new category is input based on the learned category data, the set of similar words that may contain the new category is determined based on a similarity of 85%, and if it is lower than that, it can be assigned to the set with the highest similar word. By repeating this process, a category map is constructed. Once all category data is processed, the category map is stored in a database and can be utilized as a category matching solution for open markets.

[0107] By utilizing the category linkage device (1000) of Fig. 14, when there is a need to link a new open market, categories can be matched quickly without spending human resources and time. Additionally, it is possible to obtain various information through data analysis using a category map. Furthermore, it can contribute to safe sales and purchasing activities by minimizing the need for trademark disputes by category in the e-commerce industry, and it also results in a reduction in the development requirements for category classification work when developing e-commerce solutions.

[0108] FIG. 15 is a conceptual diagram illustrating a method for matching product categories using a category map according to an embodiment of the present invention. The above-described automation device (1000) may be named differently depending on its function, and in the embodiment of FIG. 15, it may be named a category linkage device (1000).

[0109] The category linkage device (1000) can input a product to be sold into an artificial intelligence model to extract a similar category. Then, using the generated category map, it can match the extracted category with the category of the open market. Subsequently, the category linkage device (1000) can use the matched category number to proceed with matching with the product to be sold. In addition, the category linkage device (1000) can use the matched category number to register the product in the open market where the product is to be newly registered.

[0110] Referring again to FIG. 15, the category linkage device (1000) can obtain a set of category synonyms using a machine learning system and method that collects and maps categories of a marketplace. Then, a product to be sold in an open market can be specified. Next, similar categories can be extracted by inputting the product into a machine learning model. Then, the extracted category can be matched with the category of the open market. In addition, the product can be matched using the matched open market category number, and the product can be registered using the matched category number.

[0111] By utilizing the category linkage device (1000) of Fig. 15, the effect of shortening category search time is achieved by utilizing a machine learning system and method that collects and maps categories of a marketplace. Additionally, it has the effect of making it suitable for developing and integrating a solution for registering products. In cases where different or similar categories exist in an open market, product registration can be carried out effectively through this. Furthermore, trademark disputes can be minimized based on data utilizing product and category maps.

[0112] FIG. 16 is a flowchart illustrating a machine learning-based category linkage method for mapping categories of different open markets according to an embodiment of the present invention. Referring to FIG. 16, a category linkage device (1000) can collect and store category data used in multiple open markets (S1610). For example, the category linkage device (1000) can extract a category tree structure by exploring category pages of multiple open markets. Original category data is collected through the extraction of the tree structure. Alternatively, the category linkage device (1000) may collect original category data through APIs provided by each of the multiple open markets. The category linkage device (1000) can store the collected original category data in an internal database (10) in JSON or XML format. At this time, the collected original category data can be converted into category data and stored along with metadata such as a category ID, category name, hierarchical structure (parent-child relationship), and collection date and time. The category linkage device (1000) can track the change history through version management of the category data using the metadata stored together.

[0113] Next, the category linkage device (1000) can preprocess the collected and stored category data to construct a word dictionary and generate a set of similar words (S1620). Specifically, the category linkage device (1000) performs the following process.

[0114] Referring to FIG. 17, the category linkage device (1000) can normalize the collected and stored category data into text (S1710). This converts the category names into a standardized format. For example, categories that have the same meaning even if they are written differently, such as “men’s athletic shoes,” “men’s athletic shoes,” and “men-athletic shoes,” should be organized in the same format. To this end, the category linkage device (1000) can perform preprocessing to remove unnecessary symbols or spaces. Subsequently, the category linkage device (1000) can break down the category names into words through morphological analysis and tokenization.

[0115] And the category linkage device (1000) can build a word dictionary by extracting words from text-normalized category data (S1720). This can be described as a process of creating a dictionary by extracting words used in all collected categories. For example, the category linkage device (1000) can extract all words such as “athletic shoes,” “running shoes,” “men,” “sports shoes,” and “children” and input them into the word dictionary. In the word dictionary, each word can be stored by matching it with a unique word ID. While building the word dictionary, the category linkage device (1000) can identify words with similar meanings and quantify and analyze the influence each word has on category classification. Through this, the category linkage device (1000) can build a word database specialized for the product category area.

[0116] Next, the category linkage device (1000) can perform category vectorization by converting category names into N-dimensional semantic vectors (S1730). The category vectorization process will be explained again below. Subsequently, the category linkage device (1000) can generate a set of synonyms using the converted N-dimensional semantic vectors (S1740). The category linkage device (1000) can form a set of synonyms by finding categories among several open markets that have substantially the same meaning. Each set has a representative value, and this representative value later serves as a criterion for determining which set a new category name belongs to.

[0117] Below, the category vectorization process will be described in detail with reference to the example in FIG. 18. The category linkage device (1000) can tokenize a category name by separating it into word units. Tokenization means separating a category name into word units. In the example in FIG. 18, when the category name “men’s sneakers” is entered, the category linkage device (1000) can tokenize this category name by separating it into word units of “men” and “sneakers.” Subsequently, the category linkage device (1000) can convert each tokenized word into a word ID by looking it up in a word dictionary. In the example in FIG. 18, through word dictionary lookup, “men” can be converted into ID_1523 and “sneakers” into ID_4892. This process is intended to break down the compound name into semantic units to independently identify the meaning of each word, and to maintain consistency by converting it into a standardized ID through a word dictionary.

[0118] Next, the category linkage device (1000) can convert each converted word ID into an N-dimensional numeric vector. The category linkage device (1000) can convert the word ID into a vector of a predetermined fixed dimension (N-dimensional) using a pre-trained word embedding model (Word2Vec, GloVe, etc.) or sentence embedding model (BERT, etc.). In the example of FIG. 18, ID_1523 can be converted into an N-dimensional numeric vector such as [0.23, -0.45, 0.78, 0.31, … , 0.12]. Similarly, ID_4892 can be converted into an N-dimensional numeric vector such as [-0.12, 0.67, 0.34, -0.21, … , -0.56]. Each number constituting each N-dimensional vector (numerical vector and semantic vector common) represents the strength of a specific semantic attribute. And each dimension of the N-dimensional vector can represent 'major product category', 'target customer', 'use and purpose', and 'other semantic attributes'. For example, dimensions 1 to 50 may correspond to major product category attributes such as clothing, electronics, and food; dimensions 51 to 100 may correspond to target customer attributes such as men, women, and children; and dimensions 101 to 150 may correspond to use and purpose attributes such as exercise, work, daily life, and travel. Additionally, dimensions 151 to N may correspond to other semantic attributes. For example, if the dimension 70 value of the "Male" vector is 0.9, it may mean that the male attribute is very strong, and if the dimension 120 value of the "Sneaker" vector is 0.8, it may mean that the exercise use attribute is strong.

[0119] Next, the category linkage device (1000) can integrate the N-dimensional numerical vectors of each converted word ID to convert them into an N-dimensional semantic vector corresponding to the category name. Since multiple words are included in a single category name, a process of generating a category integration vector is required. For example, the category linkage device (1000) can generate a category integration vector by performing a simple average calculation or by performing a weighted average calculation. If there is a word that has a greater influence on product selection based on a value evaluated in the dictionary, the category linkage device (1000) can assign a weight to the N-dimensional numerical vector corresponding to that word ID. For example, in the case of a simple average calculation, the category name “Men’s Sneakers” consists of “Men” and “Sneakers,” and each N-dimensional numerical vector representing “Men” and “Sneakers” will be determined. By adding these two N-dimensional numerical vectors and dividing by 2, an N-dimensional numerical vector regarding “Men’s Sneakers” is generated. Next, the category linkage device (1000) can perform vector normalization to make the length of the vector 1, thereby finally converting it into an N-dimensional semantic vector. Through vector normalization, fair comparison becomes possible regardless of the length (number of words) of the category name.

[0120] Returning to the example of Fig. 16, the category linkage device (1000) can perform artificial intelligence learning to model the relationships between categories using preprocessed category data as training data (S1630). The learning process is as follows. First, in the initial learning, category data collected from various open markets is analyzed to identify which categories are similar to each other. In this process, the similarity between categories is calculated, and similar items can be grouped into a single set. Next, in the continuous learning, whenever a new category is added or an existing category is changed, the category linkage device (1000) learns this and can automatically update the category map. Then, in the verification and improvement process, the performance of the learned artificial intelligence model is continuously evaluated, and user feedback is reflected to improve accuracy.

[0121] Next, the category linkage device (1000) can perform category matching using a trained artificial intelligence model (S1640). When new category data is input, the category linkage device (1000) can determine a set of similar words that may include the input new category data using the trained artificial intelligence model. Then, according to the determined set of similar words, the category linkage device (1000) can construct a category map. The category linkage device (1000) can match the input new category data to multiple open market categories using the constructed category map.

[0122] Let us examine in more detail the process of determining a set of similar words when new category data is input. The category linkage device (1000) can calculate the similarity between the input new category data and a specific set of similar words. This process begins with the category linkage device (1000) converting the category name of the input new category data into an N-dimensional semantic vector. Subsequently, the similarity can be calculated by performing a cosine similarity calculation between the converted N-dimensional semantic vector and the N-dimensional semantic vector of a specific set of similar words that has already been derived. Similarity can be calculated by calculating the inner product of two N-dimensional semantic vectors and expressing it as a percentage. The reason for calculating the inner product is that if the directions of two vectors are similar, their meanings are similar. In this application, experiments were conducted to determine the degree of semantic similarity depending on the angle between two vectors. As shown in FIG. 19, it was confirmed that when the angle between the vectors is a threshold of 32 degrees, the categories are sufficiently semantically similar to be classified into the same set. When expressed as similarity, this corresponds to a similarity of 85%. And cases that are slightly related and require administrator or user verification were analyzed to correspond to a similarity of 70%. These two similarity values ​​can serve as a reference point for determining a set of similar words.

[0123] If the calculated similarity is 85% or higher, the category linkage device (1000) can place new category data in a specific set of similar words for which similarity calculation was performed. If the calculated similarity is less than 85% and 70% or higher, the category linkage device (1000) can request verification by outputting the set containing the similar words with the highest similarity to the administrator terminal (6000) or user terminal (5000). When the verification result is received through the communication unit (20), the category linkage device (1000) can place new category data in the verified set of similar words. Finally, if the calculated similarity is less than 70%, the category linkage device (1000) can create a new similarity set and place new category data.

[0124] FIG. 20 is a conceptual diagram illustrating a method for translating product information and detailed information into marketing terms using a category map according to an embodiment of the present invention. The above-described automation device (1000) may be named differently according to its function, and in the embodiment of FIG. 20, it may be named a marketing term translation device (1000) using a category map.

[0125] The marketing term translation device (1000) can load a category map stored in a database. The marketing term translation device (1000) can translate product introduction information according to each country. The translated product information can be matched with the loaded category map. Subsequently, the marketing term translation device (1000) can generate processed data using the matched data and marketing word data, and complete the marketing translation using the generated processed data.

[0126] Referring again to FIG. 20, the marketing term translation device (1000) can obtain a category map by utilizing a machine learning system and method that collects and maps categories of a marketplace. Then, information introducing the product, such as the product name, details, and options, can be translated according to each country. Subsequently, the marketing term translation device (1000) can match the translated product information with the category map. Then, the marketing term translation device (1000) can perform secondary processing using the matched data and marketing word data. Secondary processing is performed by comparing the category of the matched data with the category of the marketing word data. The similarity between the meaning of the translated word of the data with the matching category and the meaning of the marketing word is distinguished based on 85%, and words with a similarity exceeding 85% are replaced with marketing words, and secondary processing can be completed. The secondary processed data is replaced according to the product information, and marketing translation can be completed.

[0127] By utilizing the marketing term translation device (1000) of Fig. 20, the need for human resources is minimized, and a cost reduction effect is achieved. Additionally, there is no need to check translations one by one, and it can be used as a very significant support factor in conducting global e-commerce activities. Furthermore, by using translations specialized for sales, sales can be promoted to more consumers, and an increase in revenue can be expected.

[0128] FIG. 21 is a flowchart illustrating a method for translating marketing terms according to an embodiment of the present invention. Referring to FIG. 21, when product information to be translated into marketing terms is input, the marketing term translation device (1000) can determine a category matching the product information using a category map stored in a database (10) (S2110). This is intended to resolve the problem caused by existing technologies using a uniform translation method that does not consider this point, even though optimal marketing terms should be applied for each category. The category map is the same concept as the category map described above. To explain once again, the category map can be created and updated as follows. For example, the marketing term translation device (1000) collects and stores category data used in multiple open markets, preprocesses the collected and stored category data to construct a word dictionary and generate a set of synonyms, performs artificial intelligence learning to model the relationships between categories using the preprocessed category data as training data, and whenever new category data is input, uses the artificial intelligence model that has undergone the learning A set of synonyms that may include the new category data entered above can be determined, and a category map can be constructed according to the determined set of synonyms. Please refer to FIG. 22 for an example of the category map creation and management process.

[0129] Specifically, the marketing term translation device (1000) can generate multiple field information by analyzing input product information. The multiple field information is information corresponding to product name, detailed information, option name, option value, brand name, and model name. Subsequently, the marketing term translation device (1000) can extract keywords from these multiple field information. The keyword extraction method or translation method may vary depending on which field information the term is included in. By comparing the words included in each category constituting the category map with the extracted keywords, the marketing term translation device (1000) can determine a predetermined number of candidate categories. By selecting multiple top categories with high similarity rather than a single category, the marketing term translation device (1000) can select the optimal marketing term. Furthermore, the marketing term translation device (1000) can calculate a reliability score for each of the determined candidate categories. The category with the highest reliability score is determined as the category that matches the input product information. When calculating the reliability score, it is calculated based on keyword match, category hierarchy, and past matching success rate. Please refer to Fig. 23 for the category inference and matching process.

[0130] Next, the marketing term translation device (1000) can generate basic translation data by translating input product information into a target translation language (e.g., English, Chinese, Japanese, Spanish, etc.) (S2120). The basic translation data includes at least one basic translation word. The marketing term translation device (1000) performs translation that considers context rather than simple word-unit translation. The context referred to here means that it can reflect the product category.

[0131] Next, the marketing term translation device (1000) can retrieve marketing words of a determined category (S2130). In the example of FIG. 21, step S2120 is described as proceeding prior to step S2130, but these two steps may proceed in parallel, or step S2130 may proceed prior to it. A marketing word database may be constructed in the database (10) for each category. The marketing term translation device (1000) retrieves marketing words for a determined category using the marketing word database constructed in this way. The marketing word database has a structure of Category ID, Original Term, Marketing Term, Semantic Vector, Usage Frequency, Effectiveness Indicator, and Metadata. Specifically, the Category ID is a category identifier to which the corresponding marketing term applies, the Original Term is a general translation term, and the Marketing Term refers to a marketing term actually used in an open market. The Semantic Vector refers to a semantic vector extracted through a natural language processing model as described above. Usage Frequency refers to the usage frequency and popularity of the corresponding marketing term, Effectiveness Indicator refers to marketing effectiveness indicators such as click-through rate and conversion rate, and Metadata refers to country, time period, trend information, etc.

[0132] Next, the marketing term translation device (1000) can calculate a semantic similarity value between a basic translation word and a marketing word (S2140). The process of calculating the semantic similarity value is similar to the process of generating N-dimensional semantic vectors as described above and calculating similarity based on the angle between each N-dimensional semantic vector. The marketing term translation device (1000) can convert at least one basic translation word and a retrieved marketing word into N-dimensional semantic vectors, respectively. Then, the marketing term translation device (1000) can calculate a semantic similarity value based on the angle between each converted N-dimensional semantic vector.

[0133] Next, the marketing term translation device (1000) can generate marketing translation data based on the calculated semantic similarity value (S2150). By comparing the semantic similarity value calculated in the previous step S2140 with a threshold value, the method of generating marketing translation data changes. One of the features of the present invention is that it uses a dynamically adjusted similarity threshold value rather than a fixed threshold value. First, how marketing translation data is generated based on the result of comparison with the threshold value will be explained, and the process of determining the dynamic final threshold value will be explained later.

[0134] The marketing term translation device (1000) can compare the calculated semantic similarity value with a determined final threshold. If the calculated semantic similarity value is greater than or equal to the final threshold, the marketing term translation device (1000) can replace at least one basic translation word with the retrieved marketing word. The marketing term translation device (1000) can select several marketing terms with high similarity as candidates. The marketing term translation device (1000) can select the optimal marketing term from the candidates by comprehensively considering usage frequency, effectiveness indicators, recency, etc. After adjusting the form of the selected marketing term to fit the context of the product information, the marketing term translation device (1000) can generate marketing translation data by replacing the basic translation word with the adjusted marketing word. Please refer to FIG. 24 for the process of replacement with marketing terms. Conversely, if the calculated semantic similarity value is less than the determined final threshold, the marketing term translation device (1000) can retain at least one basic translation word. The retained basic translation word is ultimately generated as marketing translation data.

[0135] The marketing term translation device (1000) can determine the final threshold value in two steps. First, a similarity threshold value for each category is determined, and then the similarity threshold value for each category is adjusted based on the results of analyzing product attributes to determine the final threshold value.

[0136] The marketing term translation device (1000) can determine an optimal threshold by analyzing past translation data for each category and comprehensively considering the translation success rate, sales growth rate, and user feedback. For example, a relatively low threshold may be applied to categories with a high success rate and a high sales growth rate, but conversely, a higher threshold must be applied to sensitive categories.

[0137] The marketing term translation device (1000) can fine-tune the threshold according to product attributes in addition to the basic threshold for each category. The similarity threshold for each category can be adjusted by reflecting the price range of the product, whether it is a branded product, and the popularity of the category to which the product belongs. For example, in the case of high-priced products, it is necessary to reduce the risk of mistranslation by applying a higher threshold. And in the case of branded products, stricter standards need to be applied to protect the brand name. Conversely, a relatively lower threshold can be allowed for popular categories because more marketing term data has been accumulated. Please refer to FIG. 25 for the process of dynamically determining the threshold and the process of comparing the similarity value and the threshold after determining the threshold.

[0138] The methods described above may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

[0139] As described above, although the present disclosure has been explained by limited embodiments and drawings, the present disclosure is not limited to the above embodiments, and various modifications and variations are possible from this description by those skilled in the art to which the present disclosure belongs. Therefore, the scope of the present disclosure should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof. Explanation of the symbols

[0140] 1000: Product Sourcing Automation Device 10: Database 20: Communications Department 30: Processor 1100: Sourcing Module 1200: Machining Module 1300: Registration Module 1400: Order Processing Module 1500: Purchase Module 1600: Delivery Module 1700: CS Module 2000: Open Market Server 3000: Forwarding Service Server 4000: Shopping mall server 5000: Seller's terminal 6000: Administrator Terminal

Claims

Claim 1 A marketing term translation device using a category map comprises: a database for storing data; and a processor for controlling the translation device. The processor, when product information is input, uses a category map stored in the database to determine a category that matches the input product information, translates the input product information into a target language to generate basic translation data including at least one basic translation word, searches for a marketing word of the determined category, calculates a semantic similarity value between the at least one basic translation word and the searched marketing word, and generates marketing translation data according to the calculated semantic similarity value. The processor adjusts a category-specific similarity threshold based on product attributes to determine a final threshold. If the calculated semantic similarity value is greater than or equal to the determined final threshold, the at least one basic translation word is replaced with the searched marketing word to generate the marketing translation data. If the calculated semantic similarity value is less than the determined final threshold, the at least one basic translation word is retained to generate the marketing translation data. Claim 2 A marketing term translation device according to claim 1, wherein the category map is generated and updated by a method of collecting and storing category data used in multiple open markets, preprocessing the collected and stored category data to construct a word dictionary and generate a set of synonyms, performing artificial intelligence learning to model the relationships between categories using the preprocessed category data as training data, determining a set of synonyms that may include the newly input category data using the artificial intelligence model that has undergone the learning whenever new category data is input, and configuring a category map according to the determined set of synonyms. Claim 3 A marketing term translation device according to claim 1, wherein the processor analyzes the input product information to generate a plurality of field information corresponding to product name, detailed information, option name, option value, brand name, and model name, extracts keywords from the generated plurality of field information, determines a predetermined number of candidate categories by comparing the words included in each category constituting the category map with the extracted keywords, calculates a reliability score for each of the determined candidate categories, and determines the category with the highest reliability score as the category matching the input product information. Claim 4 In paragraph 3, the processor is a marketing term translation device that calculates the reliability score based on keyword matching, category hierarchy, and past matching success rate. Claim 5 A marketing term translation device according to claim 1, wherein the processor converts the at least one basic translation word and the searched marketing word into N-dimensional semantic vectors, respectively, and calculates the semantic similarity value based on the angle between the converted N-dimensional semantic vectors. Claim 6 delete Claim 7 A marketing term translation device according to claim 1, wherein the processor analyzes past translation data of each category to determine a similarity threshold for each category reflecting translation success rate, sales growth rate, and user feedback, and adjusts the similarity threshold for each category reflecting the price range of the product, whether it is a branded product, and the popularity of the category to which the product belongs to determine the final threshold. Claim 8 A method for translating marketing terms using a category map, comprising: a step in which, when product information is input, a processor determines a category that matches the input product information by using a category map stored in a database; a step in which the processor translates the input product information into a target language and generates basic translation data including at least one basic translation word; a step in which the processor looks up a marketing word of the determined category; a step in which the processor calculates a semantic similarity value between the at least one basic translation word and the looked-up marketing word; and a step in which the processor generates marketing translation data according to the calculated semantic similarity value. The step of generating marketing translation data according to the calculated semantic similarity value comprises: a step of determining a final threshold by adjusting a category-specific similarity threshold based on product attributes; and a step in which, if the calculated semantic similarity value is greater than or equal to the determined final threshold, the at least one basic translation word is replaced with the looked-up marketing word to generate the marketing translation data, and if the calculated semantic similarity value is less than the determined final threshold, the at least one basic translation word is retained to generate the marketing translation data. Claim 9 In claim 8, the category map collects and stores category data used in multiple open markets, preprocesses the collected and stored category data to construct a dictionary and generate a set of synonyms, performs artificial intelligence learning to model the relationships between categories using the preprocessed category data as training data, and whenever new category data is input, uses the artificial intelligence model that has undergone the learning A marketing term translation method that is generated and updated by determining a set of similar words that may include the new category data entered above, and constructing a category map according to the determined set of similar words. Claim 10 In claim 8, the step of determining a category matching the input product information comprises: a step of generating a plurality of field information corresponding to a product name, detailed information, option name, option value, brand name, and model name by analyzing the input product information; a step of extracting keywords from the generated plurality of field information; a step of determining a predetermined number of candidate categories by comparing the words included in each category constituting the category map with the extracted keywords; a step of calculating a reliability score for each of the determined candidate categories; and a step of determining the category with the highest reliability score as the category matching the input product information. Claim 11 In claim 10, the step of calculating the reliability score is a marketing term translation method that calculates the reliability score based on keyword matching, category hierarchy, and past matching success rate. Claim 12 A method for translating marketing terms according to claim 8, wherein the step of calculating a semantic similarity value between at least one basic translation word and the searched marketing word comprises: a step of converting each of the at least one basic translation word and the searched marketing word into an N-dimensional semantic vector; and a step of calculating the semantic similarity value based on the angle between the converted N-dimensional semantic vectors. Claim 13 delete Claim 14 A method for translating marketing terms according to claim 8, wherein the step of determining the final threshold includes: a step of determining a similarity threshold for each category by analyzing past translation data of each category and reflecting translation success rate, sales growth rate, and user feedback; and a step of determining the final threshold by adjusting the similarity threshold for each category by reflecting the price range of the product, whether it is a branded product, and the popularity of the category to which the product belongs.

Citation Information

Patent Citations

  • The method and apparatus for automatic translating online shopping mall website and product database

    KR101666611B1

  • Text translation for ecommerce

    KR1020180049827A

  • Search method for goods based on online shopping mall, apparatus and system using said method

    KR1020200046184A

  • Method, Apparatus and System for Translating Web Page

    KR1020230104435A