A single catalog and categorisation method

Abstract

The present invention is a single catalog and categorisation method in which data can be entered through a seller screen (10) or received from an integrator (20) with the help of an API, and it enables the creation of single product catalogs for products to be easily found on online shopping platforms, the control of the accuracy of product data, and the assignment of products to the correct categories.

Description

[0001] A SINGLE CATALOG AND CATEGORISATION METHOD

[0002] Technical Field

[0003] The invention relates to a single catalog and categorisation method.

[0004] More specifically, it relates to a method for creating single product catalogs to ensure that products can be easily found on online shopping platforms, for checking the accuracy of product data, and for assigning products to the correct categories, wherein data can be entered through a seller screen or received from an integrator with the help of an API.

[0005] State of the Art

[0006] Nowadays, shopping habits are rapidly shifting to online platforms in line with technological developments. People have the opportunity to quickly access the products they want and compare prices without going to physical stores. Besides saving time, online shopping makes it possible for users to easily meet their needs by offering a wider range of products. On the other hand, e- commerce sites, which offer budget-friendly options with campaigns and discounts, make users' purchasing processes more economical. This transformation provides flexibility to both consumers and businesses, helping them adapt to the dynamics of modern life.

[0007] E-commerce is a digital trade model where products and services are bought and sold over the internet. E-commerce platforms act as a bridge between buyers and sellers, enabling trade to be conducted quickly and efficiently. On these platforms, sellers can display their products and provide details such as price information and images, while buyers can easily search for, examine, and purchase the products they want using secure payment methods. E-commerce systems generally facilitate the shopping experience by offering features such as user-friendly interfaces, order tracking systems, multiple payment options, and customer support. Warehousing, logistics, and delivery processes are also important parts of this ecosystem, ensuring that products reach the consumer quickly and reliably.

[0008] In e-commerce platforms, a single catalog refers to a structure where all products on the platform are uniquely listed, and each product is defined only once. The single catalog aims to prevent the same product from being repeatedly uploaded by different sellers, thereby preventing duplicate entries and improving the user experience. The single catalog allows buyers to see the prices and conditions of different sellers for the same product on a single page, thus facilitating the comparison and purchasing processes. In a single catalog, products are grouped according to their unique attributes, and sellers compete by adding their price, stock, and shipping information for the same product. This system ensures that users have a better shopping experience.

[0009] In single catalogs, it is very important that the buyer sees only a single product page, even though the same product is offered by different sellers. This approach prevents the same product from appearing on repetitive and redundant pages on the platform, ensuring that users have a more organized and effective shopping experience. The differentiation of sellers on a single product page by specifying their own price, stock status, and shipping options allows users to make the most suitable choice by comparing the prices and service conditions offered by different sellers on a single page. If a single catalog is not provided, the same product is listed repeatedly on the platform, and the user encounters unnecessary clutter in the search results. This situation makes it difficult for buyers to access the information they are looking for.

[0010] The accuracy and appropriateness of product data in the single catalog on e-commerce platforms are very important. Products must be classified in the correct categories, and as many attributes as possible should be added. This makes it easier for buyers to find and compare the right products. If the data is not accurate, users may encounter misleading information and purchase the wrong products. Furthermore, having products in the wrong categories causes confusion in search results and negatively affects the user experience. The lack of product attributes makes it difficult for buyers to make a decision. The efficiency of single catalogs depends on checking the accuracy and appropriateness of product data, classifying them in the correct categories, and recording the maximum number of product attributes. Existing single catalog solutions are insufficient in terms of process automation and labor.

[0011] Nowadays, artificial intelligence and machine learning touch every aspect of our lives by offering revolutionary solutions in many fields such as health, finance, education, and commerce. These technologies provide significant advantages such as speeding up processes, reducing costs, and improving the user experience. Artificial intelligence and machine learning techniques have recently started to be used to improve existing single catalog solutions. However, the techniques are not yet able to assist existing single catalog solutions at a sufficient level. Existing machine learning and artificial intelligence-based solutions, while playing an important role in automating these processes, are insufficient in many cases. Problems such as incorrect categorization, missing product attributes, and incorrect data matching both negatively affect the user experience and reduce the platform's reliability. Therefore, there is a need for more advanced algorithms, customized artificial intelligence models, and high-accuracy data validation systems. Such a solution both increases user satisfaction and enhances the platform's competitive strength.

[0012] US2004267614A1 discloses a customized e-commerce catalog management and system based on customer segmentation. It aims to offer personalized shopping experiences with product presentations based on customer segmentation and a pricing structure tied to commercial agreements. The personalized catalog structure increases customer satisfaction while offering a competitive advantage to e-commerce companies.

[0013] US2007150376A1 discloses a catalog viewing method and system, aiming to display catalog items in a static view and provide quick access to detailed information for a selected item. The system offers detailed views and a comparison view for users to examine and compare multiple products at the same time.

[0014] US2008255963A1 discloses a method that visualizes products in an e-catalog with different graphic styles based on the user's purchase history, providing the user with the ability to distinguish between products they have previously purchased and those they have not. In the related document, catalog information organized in a hierarchical tree structure is retrieved from a database via a device and presented to the user. However, the related document does not provide a solution for single product catalogs used in e-commerce platforms.

[0015] In conclusion, there is a need for a method that allows for the control of the accuracy and appropriateness of product data, enables the classification of products into correct categories, and facilitates the recording of the maximum number of product attributes.

[0016] Object and Brief Description of the Invention

[0017] An object of the invention is to enable the creation of single product catalogs so that products can be easily found on online shopping platforms.

[0018] Another object of the invention is to enable the control of the accuracy of product data.

[0019] Another object of the invention is to enable the assignment of products to the correct categories.

[0020] A single catalog and categorisation method, wherein data is enterable through a seller screen or received from an integrator via an API, comprising the process steps of: creating single product catalogs for products to be easily found on online shopping platforms, controlling the accuracy of product data, and assigning products to correct categories; through a product upload server creating product upload interface data and a product upload API, a back-office server creating comparison results interface data and performing product matching operations, a validation server creating error interface data and update error interface data and validating products according to a rule set, a catalog server recording incoming product data and containing product catalog data, an inventory server containing the seller’s inventory, and a data collection server creating update interface data,

[0021] - transmitting product upload interface data created in the product upload server to the seller screen, entering product data in the product upload interface, and / or transmitting product data from the integrator, the product data via an API created in the product upload server, and transmitting the product data to the back-office server for comparison,

[0022] - comparing, in the back-office server, the newly incoming product data with the previous product data according to a threshold value,

[0023] - transmitting, from the back-office server, the incoming product data to the validation server for validation before being added to the catalog as a new product, if there is no match above a certain value between the incoming product data and one of the previous product data according to the comparison result,

[0024] - transmitting, from the back-office server, a comparison results interface data created in the back-office server to the seller screen for seller approval, if there is a match above a certain value between the newly incoming product data and one of the previous product data according to the comparison result,

[0025] - checking, on the seller screen, the incoming product data, entering data into the comparison results interface indicating 'new product' or 'one of the previous products' for the incoming product data, and transmitting the entered data to the back-office server,

[0026] - transmitting, from the back-office server, the incoming product data to the validation server for validation before being added to the catalog as a new product, if, following data verification, data indicating 'new product' has been entered into the comparison results interface,

[0027] - transmitting, from the back-office server, the incoming product data to the inventory server for addition to the inventory and recording in the inventory server, if, following data verification, data indicating 'one of the previous products' has been entered into the comparison results interface,

[0028] - subjecting, in the validation server, the incoming product data to a validation rule set, and, if no error exists in the product data, transmitting it to the catalog server for addition to the catalog as a new product, or, if an error exists in the new product data, transmitting error interface data created in the validation server to the seller screen for checking, - performing, on the seller screen, checks on the product data in the error interface, and, following error rectification through data addition / removal in the error interface, transmitting the rectified product data to the catalog server for addition to the catalog as a new product, and recording the product data in the catalog server,

[0029] - transmitting the new product data from the catalog server to the data collection server for sale on the online shopping platform,

[0030] - transmitting, from the data collection server, update interface data created in the data collection server to the seller screen to enable updating of the new product data,

[0031] - updating, on the seller screen, the new product data through data addition / removal in the update interface, and transmitting the updated product data to the validation server for validation before being added to the catalog as an updated product,

[0032] - subjecting, in the validation server, the updated product data to the validation rule set, and, if no error exists in updated product data, transmitting it to the catalog server for addition to the catalog and recording it in the catalog server, or, if an error exists in the updated product data, transmitting update error interface data created in the validation server to the seller screen for checking,

[0033] - performing, on the seller screen, checks on the updated product data in the update error interface, and, following error rectification through data addition / removal in the update error interface, transmitting the rectified updated product data to the catalog server for addition to the catalog, and recording the product data in the catalog server.

[0034] The invention relates to a single catalog and categorisation method, in which data can be entered through a seller screen or received from an integrator with the help of an API, and it relates to a method for creating single product catalogs for products to be easily found on online shopping platforms, for controlling the accuracy of product data, and for ensuring the assignment of products to the correct categories.

[0035] Brief Description of the Figures

[0036] Figure - 1 A representative view of the elements of the single catalog and categorisation method and the interactions between the elements is provided.

[0037] Reference Numerals

[0038] 10 Seller screen

[0039] 20 Integrator

[0040] 30 Product upload server

[0041] 40 Back-office server 50 Validation server

[0042] 60 Catalog server

[0043] 70 Inventory server

[0044] 80 Data collection server

[0045] Detailed Description of the Invention

[0046] When a buyer performs any search on an e-commerce platform, the relevant products must be listed. The fundamental features that constitute a single catalog are that the sellers selling the product are listed under the product with their own sales prices, the listing is done for different products, and the same product is not listed repeatedly. To create a single catalog, a matching process must be performed while products are being entered into the system. The invention relates to a method for creating single product catalogs for products to be easily found on online shopping platforms, for controlling the accuracy of product data, and for ensuring the assignment of products to the correct categories.

[0047] The servers to be mentioned in the description refer to computing environments that can receive data from other environments (database, different servers, cloud environments, etc.) with different tools (API, etc.) or transmit data to the environments, and can record the data they receive or transmit. In other words, the servers are hardware or software that provide a specific service such as data storage, sharing, hosting web pages, running applications, or meeting client requests and providing them with the necessary data or service. invention basically comprises a seller screen (10), an integrator (20), a product upload server (30), a back-office server (40), a validation server (50), a catalog server (60), an inventory server (70), and a data collection server (80).

[0048] The seller screen (10) refers to a structure that enables the display of interfaces, data entry, and monitoring of shopping processes. Here, the seller screen (10) can be a data processing device (e.g., a computer), a mobile device (e.g., a phone), etc., or just a display (e.g., a monitor). The seller screen (10) is in interaction with the product upload server (30), the back-office server (40), the validation server (50), the catalog server (60), the inventory server (70), and the data collection server (80).

[0049] The integrator (20) refers to automation tools that bring different systems together to facilitate e- commerce operations. The tools are software that provides data transfer between e-commerce platforms and accounting programs, stock tracking systems, cargo companies, and marketplace management tools. Sellers aim to manage order management, stock updates, invoicing, and cargo processes from a single platform using this software. Sellers can sometimes work with integrators (20). In the invention, integrators (20) transmit product data (information) to the product upload server (30) via an API (Application Programming Interface) provided (opened) from the product upload server (30) to be added to the catalog. The integrator (20) is in interaction with the product upload server (30) and the inventory server (70).

[0050] The seller can request an update from the seller screen (10) and with the help of integrators (20).

[0051] The product upload server (30) refers to a server where the product upload interface data transmitted to the seller screen (10) and the product upload API provided to the integrator (20) are created. The product upload interface is transmitted to the seller screen (10) with the help of a tool (e.g., API). The seller records product data either individually by entering data into the product upload interface displayed on the seller screen (10) or in bulk using an Excel file. The data recorded in the product upload interface is transmitted to the relevant server that needs it via a data transmission tool. The created interface data is saved in the memory unit of the product upload server (30). The product upload server (30) is in interaction with the seller screen (10), integrators (20), and the back-office server (40).

[0052] The back-office server (40) refers to a server where comparison processes between the product data in the catalog and the newly entered product data are carried out according to a threshold value, and where the comparison results interface data is created. The threshold value comparison refers to the process of measuring / calculating the proximity of the vectors of the product whose data is entered by the seller or integrator (20) and the products previously recorded in the back-office server (40). As a result of the comparison, if the products previously recorded in the back-office server (40), in other words, the products in the catalog, match the products whose data is entered by the seller or integrator (20) above a certain threshold value (in proximity), the newly entered product is considered a potential match. The back-office server (40) is in interaction with the seller screen (10), the product upload server (30), the validation server (50), and the inventory server (70).

[0053] The back-office server (40) contains an artificial neural network. In the preferred embodiment of the invention, a CNN artificial neural network is used in the back-office server (40). CNN is a subfield of machine learning, using multi-layered artificial neural networks and automatically learning features within the data. The CNN model in the back-office server (40) is an open-source model trained with a large amount of text (e-commerce specific data), and vectors to identify the products are generated with the help of this model. Here, besides storing the vectors of all products in the catalog, vectors are also created for newly entered products. The CNN model is a neural network that supports 16 different languages. The data belonging to this model, the model's algorithm data, cleaning and preprocessing algorithm data, generated vector data, the model's current data, created interface data, data of all products in the catalog, data of each newly entered product, hierarchical data, attribute data that products must have for categories, prohibited word data, VAT data of categories, installment data, dimensional weight (desi) data, and metadata related to e-commerce single catalog processes are saved in the memory unit of the back-office server (40). In other words, product categories, the category tree, mandatory and optional product attributes for product types and their possible values, installment rules based on product type, and VAT values based on product type are kept hierarchically in the back-office server (40) and are served (transmitted) to other servers when needed. When a product is entered, its correct category, number of installments, and VAT value are determined by the algorithms and models in the back-office server (40).

[0054] Some examples of data held in the back-office server (40) are given below;

[0055] Example of Category Tree:

[0056] XYZ.com»Fashion Guru» Women» Clothing»Vest XYZ» Fashion Guru» Men»Giyim»Vest Example of Product Attributes (for Vest) a. Pattern (optional) b. Season (optional) c. Gender (mandatory) d. Color (mandatory, variant attribute) e. Size (mandatory, variant attribute)

[0057] Example of Possible Values for Vest Size Attribute: a. S b. M c. L d. XL e. XXL

[0058] Example of Installment Rules:

[0059] For Vest product type: 12 installments

[0060] Example of VAT Rules:

[0061] Vest product type: 10%

[0062] The validation server (50) refers to a server where product data is checked, an error interface data and an update error interface data are created, and product data is validated according to a rule set. The validation server (50) contains three different machine learning models to facilitate and automate validation processes. In the preferred embodiment of the invention, model 1 , model 2, and EfficientNet models are used in the validation server (50). The validation server (50) is in interaction with the seller screen (10), the back-office server (40), and the catalog server (60).

[0063] The catalog server (60) refers to a server where all data related to products is recorded and stored. The catalog server (60) contains the catalog where products are listed with all their attributes. In the catalog server (60), product attribute extraction is performed from images and texts using machine learning models and text-based methods to infer and add to the catalog the product attributes not provided by the seller. The catalog server (60) is in interaction with the seller screen (10), the validation server (50), and the data collection server (80).

[0064] The catalog server (60) uses a machine learning model to recommend categories. In the preferred embodiment of the invention, the CBOW model is used in the catalog server (60) to recommend categories. Category recommendation serves to place products opened in incorrect categories, either intentionally or unintentionally for fraud purposes by sellers, into their correct categories. In this model, a solution is produced for the multi-class text classification problem with an imbalanced dataset by using the CBOW algorithm.

[0065] The catalog server (60) uses a machine learning model to extract product attributes from images. In the preferred embodiment of the invention, the YOLOv8 model is used in the catalog server (60) to extract product attributes from images. While extracting product attributes from images, a YOLOv8- based object recognition model is trained on online fashion sales data and used, and the detected products are classified according to their attributes in hierarchical category breakdowns with a cascaded model structure.

[0066] The catalog server (60) uses a machine learning model to extract product attributes from features such as product name, product description, etc. In the preferred embodiment of the invention, an LLM (Large Language Model) model is used in the catalog server (60) to extract product attributes from features such as product name, product description, etc., and preferably ChatGPT is used. For example, it can be determined with the LLM solution that the hard disk with code HD13500 has a capacity of 500GB.

[0067] The catalog server (60) also includes text-based methods to extract product attributes. In the preferred embodiment of the invention, REGEX (Regular Expression) methods are used in the catalog server (60) to extract product attributes from features such as product name, product description, etc. For example, if "white" is in the product name but the color attribute is empty, this can be completed with the REGEX solution.

[0068] In the catalog server (60), by using the machine learning and text-based methods, it is ensured that the seller can easily find products according to the desired attributes. Some product and variant attributes must have standard values defined in the back-office server (40). For example, a "vest" product's size attribute must have one of the defined values S, M, L, XL, XXL. If it does not have one of the defined values, the text-based methods and machine learning models described above, which are used for attribute extraction, add the most appropriate value to the product attributes from the entered name, attributes, and images of the product. For example, if the size attribute is entered as 36 and the text "Small" is in the product name, or if the size information is S in the images, these are determined by text and image machine learning models, and the attribute value S is added to the product attributes. More specifically, in the catalog server (60), if there are missing and / or enrichable attributes in the product data received from the seller or integrator (20), they are found and defined with the REGEX solution, LLM solution, CBOW solution, and YOLOv8 solution.

[0069] Products are created in the catalog with all the information of the main product and the variant. The catalog data is served to the relevant servers with a data transmission messaging structure (e.g., Apache Kafka) to be used in product listing / detail / filtering / search applications on the web / mobile application. Having many product attributes and having them as defined standard values as much as possible serves to help the buyer easily find the product in product search and filtering, while the singularization of the brand contributes to the correct management of traffic for SEO (Search Engine Optimization). The product category is determined by a machine learning method from the product name and attributes. When a product is added to the catalog, a product quality score is also determined based on the existing product attributes. This score is used as an input when listing search results and encourages the seller to increase product attributes. The following is an example of a JSON message transmitted to the catalog server (60).

[0070] {

[0071] "Jd" : 0bjectld("57690dllc0140edcb8731944"),

[0072] "createdAt" : ISODate("2021-03-05T09:04:08.967+0000"),

[0073] "createdBy" : "user-0",

[0074] "fields" : {

[0075] "UrunAciklamasi" : {

[0076] "value" : "MADE FROM 1ST CLASS POLYESTER FABRIC. CORNICE MOUNTING APPARATUS ARE INCLUDED IN THE PRODUCT. There is a 4 cm margin in roller blinds. For example, when you buy a 60 cm wide product, the net width of the curtain is 56 cm.", "mandatory" : false SODMTP10398", false false ", false olblue.jpg", false D122", false : { * Polyester Roller Blind P-122 Petrol blue Scalloped Edge 80x200 ", false : { false 5589083680", false false d" : { false *", "mandatory" : false

[0077] },

[0078] "Theme" : {

[0079] "value" : "Plain",

[0080] "mandatory" : false

[0081] },

[0082] "FabricType" : {

[0083] "value" : "Polyester",

[0084] "mandatory" : false

[0085] },

[0086] "CurtainFeatures" : {

[0087] "value" : "Polyester",

[0088] "mandatory" : false

[0089] },

[0090] "Color" : {

[0091] "value" : "Petrol blue",

[0092] "mandatory" : false

[0093] },

[0094] "FabricWeave" : {

[0095] "value" : "Polyester",

[0096] "mandatory" : false

[0097] },

[0098] "Packagecontents" : {

[0099] "value" : "Curtain, cornice mounting apparatus, Invoice",

[0100] "mandatory" : false

[0101] },

[0102] "Dimensions" : {

[0103] "value" : "80X200",

[0104] "mandatory" : false

[0105] },

[0106] "merchantSku" : {

[0107] "value" : "EAKSODMTP10398",

[0108] "mandatory" : false

[0109] }

[0110] },

[0111] "retail" : false,

[0112] "listingStatusList" : [

[0113] ],

[0114] "isMatched" : true,

[0115] "merchant" : "user-3",

[0116] "merchantSku" : "EAKSODMTP10398",

[0117] "status" : "MATCHED",

[0118] "validationstatus" : "FAILED", "productType" : {

[0119] "name" : "Curtains",

[0120] "productTypeld" : Numberlnt(997),

[0121] "buyingCategoryld" : Numberlnt(80794182),

[0122] "buyingCategoryName" : "Curtain",

[0123] "parentBuyingCategoryld" : Numberlnt(80089022), "parentBuyingCategoryName" : "HomeTextileStore", "deleted" : false

[0124] },

[0125] "uploadDate" : ISODate("2016-06-21T09:46:00.000+0000"), }

[0126] The following is an example of a JSON message transmitted from the product catalog to the data collection server (80).

[0127] "InstallmentRate": 12,

[0128] "BuyingCategoryld": "80794416",

[0129] "Variants": [

[0130] {

[0131] "Id": "HBCV000060ZS1U",

[0132] "Name": "Printify Samsung Galaxy A50 Cover Nostalgia Pattern Designed Transparent Super Silicone

[0133] Case",

[0134] "TaxVatRate": "20",

[0135] "Freight": "1.00",

[0136] "WinnerFreight": "1.00",

[0137] "Barcodes": [

[0138] "prtsprslknsmA50043d"

[0139] ],

[0140] "Media": [

[0141] {

[0142] "ImageFileName": "110000643300800.jpg",

[0143] "LinkFormat":

[0144] "https: / / productimages.hepsiburada.net / s / 777 / {size} / 110000643300800.jpg",

[0145] }],

[0146] "Video": [],

[0147] "WarrantyPeriod": 6, "Variantclassifications": [

[0148] {

[0149] },

[0150] {

[0151] "Classld": "529",

[0152] "ClassName": "Phone Cases",

[0153] "Attributeld": "compatible_brand", "Displayvalue": "s***", "Name": "Compatible Brand", "Value": "4316"

[0154] },

[0155] {

[0156] "Classld": "529",

[0157] "ClassName": "Phone Cases",

[0158] "Attributeld": "case_type",

[0159] "DisplayValue": "Back Cover", "Name": "Case Type", "Value": "14065"

[0160] },

[0161] {

[0162] "Classld": "529",

[0163] "ClassName": "Phone Cases",

[0164] "Attributeld": "material_type", "DisplayValue": "Silicone", "Name": "Material Type", "Value": "9077"

[0165] },

[0166] ],

[0167] "CreateDate": "2024-03-02T21:ll:19.8600000" }

[0168] }

[0169] The inventory server (70) refers to a server where sellers can track their product names and stocks. Product name data and stock data are saved in the memory unit of the inventory server (70). The inventory server (70) updates itself for every product added to and removed from the catalog. The inventory server (70) contains the inventory that sellers can see. If a seller makes a change regarding a product in the inventory, the data is transmitted to the data collection server (80). The seller screen (10) and the integrator (20) are in interaction with the inventory server (70). Sellers can access the inventory directly or through integrators (20). Sellers can create an update request for the products in their inventory directly or through integrators (20). On the inventory server (70), thanks to the created interfaces, inventory information can be tracked on the seller screen (10). The inventory server (70) is in interaction with the seller screen (10), the integrator (20), and the back-office server (40).

[0170] The data collection server (80) refers to a server that collects and combines product data from the back-office server (40) and the catalog server (60) with different data not present in the servers (price, campaign, etc.) from different sources (websites, devices, APIs), presents the combined data, and where an update interface data is created. The data collection server (80) provides a centralized view by taking data from distributed sources and particularly facilitates data management processes. The data collection server (80) is in interaction with the seller screen (10) and the catalog server (60).

[0171] Before the working method of the invention is explained in detail, some definitions will be made. The interface creation process to be explained refers to a traditional series of interface creation processes. The interface is designed (drawn, visualized) using a design tool (Figma, Adobe, etc.) according to the need, and details such as user interactions and interface layout are determined. The designed interface goes through a series of frontend development processes (HTML, CSS, JavaScript, React, Swift, etc.) and is brought into a coded state. The coded interface goes through a backend development process (RESTful API, GraphQL, etc.), and the elements in the interface (button, etc.) are brought into a coded state to become functional (e.g., receiving / transmitting data from a server). In the backend development process, the data to be displayed on the interface is processed to be suitable for the frontend side. In the backend-frontend communication process, in other words, the data received from or transmitted to the server is processed in the backend process to comply with certain formats (JSON, XML). Finally, the code data of the interface is converted into visual data (rendering process) with the help of a browser (web-based interface, etc.) or a graphics engine (SurfaceFlinger, UlKit). The necessary calculations for visual-graphic processes are performed by means of a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). For example, while the drawing of images and animations can be done in the GPU, operations related to data processing and business logic processes can be done in the CPU. The interface data, converted into visual data, is transmitted to a display (monitor, television, computer, phone, etc.) with the help of a tool. Here, the tools refer to different hardware or software elements (API, HDMI, GPU, etc.). In the invention, the API tool is used for the interactions of the seller screen (10), which represents a display.

[0172] The operation of the invention involves a series of process steps. First, a product upload interface data created in the product upload server (30) is transmitted to the seller screen (10) with the help of a tool. Subsequently, sellers enter product data in the product upload interface displayed on the seller screen (10). Here, the seller can enter product data individually into the product upload interface or can enter data into the product upload interface in Excel format. Sellers working with integrators (20) transmit their product data to the integrators (20), and the integrators (20) perform the product data entry. In other words, the integrator (20) transmits the product data with the help of an API created in the product upload server (30). The product data transmitted from the product upload interface on the seller screen (10) and from the integrators (20) to the product upload server (30) is transmitted from the product upload server (30) to the back-office server (40) for comparison via a data transmission tool (Rsync, Apache Kafka, etc.). In the invention, the data communication tool used for data transmission and reception between servers is Apache Kafka, and the data type is JSON.

[0173] In the back-office server (40), the newly incoming product data and the previous product data are compared according to a threshold value. The matching process is performed with a CNN (Convolutional Neural Network) model in the back-office server (40) using product name, product attributes, product brand, and product image data. In the back-office server (40), the matching process is done by comparing the newly entered product data with other product data in the catalog with the help of the CNN.

[0174] In the back-office server (40), in the first stage, some preprocessing steps such as removing special characters from the product data registered in the catalog and converting all characters to lowercase are performed with the help of a traditional cleaning algorithm. The preprocessed product data is given as input to the CNN model to generate vector data of this data (texts). In the back-office server (40), the vector data generated by the CNN model is added (recorded) to a search engine (Milvus, Pinecone, Elasticsearch, etc.), and the search vector data is given to the CNN model to distinguish the products with the closest attributes to the searched product.

[0175] In the back-office server (40), additional checks such as barcode and product attributes are added after the search results to increase the precision and recall performance of the CNN model. In the back-office server (40), when a new product is entered into the product upload server (30), the entered product data is transmitted from the product upload server (30) to the back-office server (40), and a product vector is extracted from the product's attribute data with the help of CNN and subjected to a similarity search (vector search) with the vectors of products previously registered in the memory unit of the back-office server (40). According to the vector similarity search, the new product data is matched with the product having the closest vector that matches above a certain threshold value. As a result of the matching, if the product is found in the catalog data, the seller is expected to open a product listing with their own price and stock information. If not found, it is subjected to validation to be added as a new product. In traditional methods, when product matching is done by barcode, only the seller's data is trusted, which leads to duplicate products in the catalog. In our invention, however, thanks to these operations performed in the back-office server (40), a much more consistent matching is achieved by using not only the barcode but also many other attributes such as product name, attributes, and images with machine learning and vector search, and the single catalog is secured.

[0176] In the back-office server (40), if there is no match above a certain value between the incoming product data and one of the previous product data according to the comparison made based on a predetermined threshold value, the incoming product data is transmitted to the validation server (50) via a data transmission tool to be validated before being added to the catalog as a new product. In the back-office server (40), if there is a match above a certain value between the newly incoming product data and one of the previous product data according to the comparison result, the comparison results data is transmitted to the seller screen (10) by means of a comparison results interface data created in the back-office server (40) to be presented for the seller's approval.

[0177] On the seller screen (10), the incoming product data is checked by the seller in the comparison results interface, and the seller enters the data input corresponding to "new product" or "one of the previous products" for the incoming product data into the comparison results interface, and the data entered into the interface is transmitted to the back-office server (40) with the help of a tool for "new product" or "one of the previous products" control, to be compared with previous products.

[0178] In the back-office server (40), after checking the data, if a data input corresponding to "new product" has been made to the comparison results interface, the incoming product data is transmitted to the validation server (50) via a data transmission tool to be validated before being added to the catalog as a new product.

[0179] In the back-office server (40), after checking the data, if a data input corresponding to "one of the previous products" has been made to the comparison results interface, the incoming product data is transmitted to the inventory server (70) to be added to the inventory, and the product data is recorded in the inventory server (70).

[0180] In the validation server (50), the product data coming from the back-office server (40) for validation is subjected to a validation rule set, and if there is no error in the product data, it is transmitted to the catalog server (60) to be added to the catalog as a new product; if there is an error in the new product data, an error interface data created in the validation server (50) is transmitted to the seller screen (10) with the help of a tool for the incoming product data to be checked. The rule set algorithm here includes checks such as the fields entered for the product not being empty, the product being singular, and no prohibited words in the entered product data (Table 1).

[0181] Table 1 : Non-visual validation checks (rule set) In the validation server (50), visual data checks are also performed. Here, the control of images containing +18 content or child models, which can be done with manual checks in the prior art, is quickly performed by a machine learning method trained specifically for this control.

[0182] In the validation server (50), different models are trained for visual validation. Validation operations are performed using the models, which are; the EfficientNet model, and in addition to this model, models named model 1 and model 2.

[0183] In the validation server (50), model 1 , which we can call the base model, is trained to perform easy tasks. Model 1 is used as the basic feature extractor and is based on the EfficientNet architecture. For model 1 , a ready-made model was not used directly, but it has been customized for e-commerce and the single catalog. Model 1 is trained by removing some layers from the neural network according to the features needed for validation processes and connecting a classifier fully connected neural network instead.

[0184] In the validation server (50), model 2, which we can call the advanced model, is trained to perform sensitive tasks. For model 2, in addition to the feature extractor EfficientNet model for more complex or sensitive tasks, specially developed convolutional neural network features have been used. The two extracted features have been combined with an attention mechanism. The resulting feature has been connected to a fully connected neural network and trained with regression or classification targets according to the module's requirements.

[0185] In the validation server (50), during the inference phase where the trained models will be used, the models are designed by paying attention to the number of parameters in order to meet the requirements of high response speed and high classification performance. While designing the architecture, the feature outputs of the EfficientNet model trained with a dataset and the outputs of a sub-model 1 that extracts a separate feature are used by combining them with an attention approach. With this approach, it is aimed for the model to understand the regions in the images that need to be focused on. In this way, when investigating the visual resolution quality, errors in the details of the images are detected with higher success. In the preferred embodiment of the invention, the EfficientNet model is trained with the ImageNet dataset.

[0186] In the validation server (50), visual checks are performed for the headings given in Table 2. For this, the images of the products existing in the catalog (visual data) are initially labeled by manual marking, and then machine learning models are trained with these images. Afterwards, manual feedback is added to the labels made by the model, and the models continue to be trained at regular intervals. Then, visual checks are performed with the results of the models.

[0187] Table 2: Visual validation checks

[0188] When product data is entered (seller screen (10) or integration), it is added as a main product and variants with different attributes attached to it. For example, an XYZ brand vest product may have different variants according to S, M, L, XL, XXL size attributes and white, black, gray color attributes. This situation allows other variants of the same product to be easily found while looking at one product. With this approach, in the validation server (50), product attributes are subjected to validation once, and variant attributes are subjected to validation for each variant. The validations are performed using model 1 , model 2, and the EfficientNet model.

[0189] On the seller screen (10), checks related to the product data in the displayed error interface are performed by the seller. After the seller rectifies the errors by adding / removing data in the error interface, the product data with rectified errors is transmitted to the catalog server (60) to be added to the catalog as a new product, and the product data is recorded in the catalog server (60). The new product data is transmitted from the catalog server (60) to the data collection server (80) via a data transmission tool to be sold on the online shopping platform.

[0190] In the data collection server (80), an update interface data created in the data collection server (80) is transmitted to the seller screen (10) by means of a tool so that the new product data can be updated. On the seller screen (10), the new product data is updated by the seller adding / removing data in the displayed update interface, and the updated product data is transmitted to the validation server (50) with the help of a data transmission to be validated before being added to the catalog as an updated product. In the validation server (50), the updated product data is subjected to the validation rule set, and if there is no error in the updated product data, the updated product data is transmitted to the catalog server (60) with the help of a data transmission tool to be added to the catalog, and the product data is recorded in the catalog server (60). In the validation server (50), the updated product data is subjected to the validation rule set, and if there is an error in the updated product data, an update error interface data created in the validation server (50) is transmitted to the seller screen (10) by means of a tool for the updated product data to be checked. On the seller screen (10), in the displayed update error interface, the seller performs checks on the updated product, and after rectifying the errors by adding / removing data in the update error interface, the updated product data with rectified errors is transmitted to the catalog server (60) with the help of a tool to be added to the catalog, and the product data is recorded in the catalog server (60).

[0191] The present invention, by using machine learning methods, enables the creation of a single product catalog, ensures that products are assigned to the correct categories, checks the accuracy of product data and its suitability for sale, enriches the catalog by filling in product information not sent by the seller, and differs from the prior art in all these aspects.

Claims

CLAIMS1. A single catalog and categorisation method, wherein data is enterable through a seller screen (10) or received from an integrator (20) via an API, characterized in that the invention comprises the process steps of: creating single product catalogs for products to be easily found on online shopping platforms, controlling the accuracy of product data, and assigning products to correct categories; through a product upload server (30) creating product upload interface data and a product upload API, a back-office server (40) creating comparison results interface data and performing product matching operations, a validation server (50) creating error interface data and update error interface data and validating products according to a rule set, a catalog server (60) recording incoming product data and containing product catalog data, an inventory server (70) containing the seller’s inventory, and a data collection server (80) creating update interface data, transmitting product upload interface data created in the product upload server (30) to the seller screen (10), entering product data in the product upload interface, and / or transmitting product data from the integrator (20), the product data via an API created in the product upload server (30), and transmitting the product data to the back-office server (40) for comparison, comparing, in the back-office server (40), the newly incoming product data with the previous product data according to a threshold value, transmitting, from the back-office server (40), the incoming product data to the validation server (50) for validation before being added to the catalog as a new product, if there is no match above a certain value between the incoming product data and one of the previous product data according to the comparison result, transmitting, from the back-office server (40), a comparison results interface data created in the back-office server (40) to the seller screen (10) for seller approval, if there is a match above a certain value between the newly incoming product data and one of the previous product data according to the comparison result, checking, on the seller screen (10), the incoming product data, entering data into the comparison results interface indicating 'new product' or 'one of the previous products' for the incoming product data, and transmitting the entered data to the back-office server (40),transmitting, from the back-office server (40), the incoming product data to the validation server (50) for validation before being added to the catalog as a new product, if, following data verification, data indicating 'new product' has been entered into the comparison results interface, transmitting, from the back-office server (40), the incoming product data to the inventory server (70) for addition to the inventory and recording in the inventory server (70), if, following data verification, data indicating 'one of the previous products' has been entered into the comparison results interface, subjecting, in the validation server (50), the incoming product data to a validation rule set, and, if no error exists in the product data, transmitting it to the catalog server (60) for addition to the catalog as a new product, or, if an error exists in the new product data, transmitting error interface data created in the validation server (50) to the seller screen (10) for checking, performing, on the seller screen (10), checks on the product data in the error interface, and, following error rectification through data addition / removal in the error interface, transmitting the rectified product data to the catalog server (60) for addition to the catalog as a new product, and recording the product data in the catalog server (60), transmitting the new product data from the catalog server (60) to the data collection server (80) for sale on the online shopping platform, transmitting, from the data collection server (80), update interface data created in the data collection server (80) to the seller screen (10) to enable updating of the new product data, updating, on the seller screen (10), the new product data through data addition / removal in the update interface, and transmitting the updated product data to the validation server (50) for validation before being added to the catalog as an updated product, subjecting, in the validation server (50), the updated product data to the validation rule set, and, if no error exists in updated product data, transmitting it to the catalog server (60) for addition to the catalog and recording it in the catalog server (60), or, if an error exists in the updated product data, transmitting update error interface data created in the validation server (50) to the seller screen (10) for checking, performing, on the seller screen (10), checks on the updated product data in the update error interface, and, following error rectification through data addition / removal in the update error interface, transmitting the rectified updated product data to thecatalog server (60) for addition to the catalog, and recording the product data in the catalog server (60).

Images