CO-Purchase System for Health Foods and Related Products Based on Slow Aging Recipe Content
The intelligent automation system addresses inefficiencies in content commerce by automating buyer identification, predicting optimal sales times, and sending personalized notifications, enhancing marketing efficiency and user experience.
Patent Information
- Application Number
- KR1020250184492
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2025-11-27
- Filing Date
- 2025-11-27
- Publication Date
- 2026-11-06
- Estimated Expiration
- 2045-11-27
AI Technical Summary
Existing content commerce models for health foods and related products face inefficiencies due to manual comment verification, inaccurate sales timing, and lack of data connectivity, leading to missed sales opportunities and poor customer experience.
An intelligent automation system that automatically identifies potential buyers through keyword analysis, predicts repurchase times based on consumption cycles, and sends personalized notifications using a group buying server with integrated data analysis and graph neural networks.
Enhances marketing efficiency, secures stable sales, and improves user experience by accurately targeting potential customers and optimizing sales timing, thereby increasing repurchase rates and content effectiveness.
Smart Images

Figure 112025133966964-PAT00002_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a system that produces and disseminates recipe content utilizing health foods (e.g., olive oil, kale, berries, mineral salt, Manuka honey, etc.) and related products tailored to the 'slow aging' trend, automatically identifies and collects the purchasing intent of potential buyers through comment keywords in conjunction with this, analyzes the expected time of product consumption to determine the time to resume group purchasing, and conducts personalized notifications and group purchasing based on this. Background Technology
[0002] Recently, in the health food and related product market, the Content Commerce model—which gains customer trust by providing product recipes or consumption routines through social media channels such as YouTube and Instagram—has become mainstream. Existing technology relied on a method where content was created and uploaded, and then sales links were delivered to customers who expressed interest in participating in group purchases via comments or direct messages (DMs). However, this approach has clear limitations, such as the manual process of sifting through potential customers from numerous comments expressing purchase intent, and the loss of opportunity costs resulting from missed comments or delayed responses that cause customers' purchase intentions to fade.
[0003] Existing technology required sellers to manually verify comments, identify user IDs, and send direct messages (DMs), even when customers used explicit keywords such as "I want to participate in the group purchase." Since this is performed manually, it is costly and time-consuming, and customer experience (CX) can deteriorate due to delays in responding to the surge in comments, especially during group purchase periods. Furthermore, there was a risk that unnecessary notifications would be sent to comments with negative intent, such as "I wish the group purchase wouldn't happen," leading to problems with low accuracy and inefficiency in marketing targeting.
[0004] In existing group buying methods, it was common for sellers to arbitrarily set a date ('the 1st of every month') or rely on intuitive judgment ('three weeks after the last stock was depleted') to determine the timing of the next group purchase. Despite the fact that it is crucial to launch sales at the precise timing when customer repurchase demand arises due to the nature of consumable health foods and related products, existing technology lacked a system capable of scientifically analyzing and predicting past sales data and the average consumption cycle per customer. Consequently, this resulted in inefficiencies that failed to optimize sales timing, such as sending notifications when customers had not yet used up the product (causing customer dissatisfaction) or receiving late notifications when the product was already depleted (loss of repurchase opportunities).
[0005] Existing technologies were limited primarily to tallying only the final sales volume regarding "how much was sold." In other words, they failed to analyze and connect by content type which content (e.g., morning routines vs. cooking recipes) led potential customers to actually convert into buyers. Furthermore, they lacked the capability to quantitatively evaluate the effectiveness of notifications by measuring the actual purchase conversion rate relative to the notification results. This absence of data connectivity and analysis capabilities prevented the establishment of a scientific feedback loop for marketing strategies aimed at boosting sales performance, causing problems that made the continuous improvement of content creation and sales strategies difficult.
[0006] Existing automated notification technologies generally operated based on active customer behaviors, such as "saving" or "requesting restock notifications." However, for the consumable health foods and related products covered in this invention, there was no technology capable of providing notifications by predicting when a product will run out without the customer explicitly checking inventory. Thus, an intelligent automated notification system that automatically calculates the expected repurchase time by combining customer purchasing behavior data and product consumption cycle data, and sends personalized notifications at that time, was a core area that could not be realized in existing technologies.
[0007] In summary, existing content commerce models suffer from fundamental problems, including (1) manual response to comments and inaccurate targeting, (2) determination of sales timing without scientific basis, and (3) a lack of data connectivity between marketing activities and sales performance. This invention aims to overcome these limitations of existing technology and present a new technical solution that enhances operational efficiency and accurately responds to customer demand by integrating the entire process from content creation to sales and repurchase prediction into an intelligent automation system based on natural language processing. Prior art literature
[0008] Korean Patent Publication No. 10-2025-001563 The problem to be solved
[0009] The technical problem that the present invention aims to solve is as follows.
[0010] 1. Automatic identification and acquisition of potential buyers: The purpose is to increase marketing efficiency by automatically detecting comments expressing intent to participate in group purchases left by an unspecified number of people on social media or video content platforms using specific keywords (e.g., "I want to participate in the group purchase," "Please share the link"), and organizing those users into a list of potential buyers.
[0011] 2. Repurchase Cycle Optimization: The purpose is to maximize sales opportunities by comprehensively analyzing previous group purchase sales volume and average consumption data by product to automatically calculate the time when users are expected to want to repurchase after using up the product, and utilizing this to determine the next group purchase date.
[0012] 3. Personalized Purchase Notification: This aims to prevent the omission of information delivery and increase the purchase conversion rate by automatically sending notification messages and purchase links to identified potential buyers via methods such as personal DMs and emails at the start of a group purchase.
[0013] 4. Strengthening Content-Commerce Integration: The purpose is to increase the lifetime value of customers by continuously providing content related to 'slow aging' recipes and consumption methods to impress upon users the efficacy and necessity of the product, and by facilitating the transition from content viewing to actual purchase. means of solving the problem
[0014] To solve the above-mentioned problem, the present invention provides the following means.
[0015] First, the system includes a group buying server (100) for uploading certain content and sending notification messages to potential buyers based on comments on the uploaded content, and the group buying server (100) is for producing and disseminating content regarding 'slow aging' recipes and consumption methods, and includes a content provision unit (110) for uploading content regarding 'slow aging' recipes and consumption methods to a content platform and managing views, reactions, and group buying connection links (URLs) included within the content, a potential buyer comment collection unit (120) for automatically detecting and collecting specific keywords among video content comments and identifying author IDs, a group buying resumption time determination unit (130) for loading information on the final sales date (D_last), sales quantity (Q_sold), and average user consumption cycle (Cycle_avg) of the previous group buying from the DB of the group buying server (100) and providing a group buying start date (D_GO), and is implemented within the group buying server (100). A group purchase notification unit (140) that receives the group purchase start date (D_GO) confirmed by the group purchase resumption time determination unit (130) and automatically sends personalized notification messages and purchase links to targets based on the list of potential buyers generated by the potential buyer comment collection unit (120); a sales performance analysis unit (150) that collects all data generated throughout the entire process of the system, analyzes the purchase conversion rate by content type, analyzes the repurchase rate and purchase frequency of potential customers, generates the analysis results in the form of a visualized report, and transmits the report results to the content provision unit (110) and the group purchase resumption time determination unit (130) to induce strategy modification; and constructs a heterogeneous graph between users, products, and interaction events including the customer ID obtained from the potential buyer comment collection unit (120), and applies a graph neural network (GNN) model to the heterogeneous graph to calculate the group unit diffusion probability (P_spread) for each user node.A system for joint purchasing of health foods and related products based on slow aging recipe content is provided, characterized by including a social graph-based group purchase amplification unit (160) that identifies a group invitation initiator to induce chain purchases based on the calculated group unit diffusion probability (P_spread) and transmits it to a group purchase notification unit (140).
[0016] Secondly, to proceed with the above system, the present invention provides a method for group purchasing health foods and related products based on slow-aging recipe content, comprising: a first step of creating 'slow-aging recipe' content and distributing it to an external content platform with a traceable link (URL_tracking) inserted to generate content response data; a second step of periodically collecting comment data from the content platform; a third step of determining whether the current time has reached the previously predicted optimal group purchase resumption date (D_GO); a fourth step of automatically sending personalized notification messages and purchase tracking links to the target of the potential buyer list in accordance with the confirmed optimal group purchase resumption date (D_GO); a fifth step of collecting data including sales records, buyer IDs, and repurchase status during the group purchase period from the sales / payment system and storing it in a DB; and a sixth step of analyzing the purchase conversion rate (CR), repurchase rate, and notification effect by content type using the collected data, and applying an NLP engine to the comments collected in the second step to analyze purchase intent keywords and positive sentiment scores, and registering user IDs whose scores are above a threshold in the potential buyer list DB. Effects of the invention
[0017] The low-speed aging recipe content-based health food and related product group purchasing system according to the present invention provides the following effects.
[0018] 1. Increased marketing efficiency: By using the comment keyword detection function, you can accurately target only potential customers with actual purchasing intent, significantly improving time and cost efficiency compared to general marketing targeting an unspecified audience.
[0019] 2. Securing sales stability and improving repurchase rates: Data-driven decision-making on when to resume group buying enables the resale of products at the 'most needed moment' when consumers have finished using the product, thereby dramatically increasing repurchase rates and establishing a stable sales cycle.
[0020] 3. Improvement of User Experience: By automatically providing group purchase notifications in a personalized manner, users receive necessary information in a timely manner without the hassle of having to search for the notification time themselves, thereby increasing convenience.
[0021] 4. Optimization of Content Creation and Marketing Strategies: Objectively identify which content (e.g., breakfast on an empty stomach, cooking recipes, etc.) has a high purchase conversion rate through sales performance analysis, and reflect this in future content planning and marketing strategies to create a virtuous cycle within the system. Brief explanation of the drawing
[0022] FIG. 1 is an overall configuration diagram of a low-speed aging recipe content-based health food and related product group purchasing system according to the present invention. Figure 2 is a detailed internal configuration block diagram of the joint purchasing server of the present invention. Figure 3 is a flowchart of the content and group purchase data of the present invention. Figure 4 is a detailed configuration and operation flowchart of the potential buyer comment collection unit. Figure 5 is a detailed block diagram of the joint purchasing resumption timing determination unit. Figure 6 is a detailed configuration of the sales performance analysis department and a flowchart of the analysis item processing. Figure 7 is a flowchart of the detailed operation of the group purchase notification unit and the integration of the message sending gateway. Figure 8 is a diagram showing the link and data integration method between the content provider and the group purchase server. FIG. 9 is a flowchart of the overall operation method of a joint purchasing system for health foods and related products according to the present invention. Figure 10 is a detailed flowchart of the method for identifying and listing potential buyers. Figure 11 is a detailed flowchart of the method for predicting and determining the timing of the resumption of group purchasing. Figure 12 is a detailed flowchart of a method for sending automated group purchase notifications to potential buyers. Figure 13 is a flowchart of the marketing strategy feedback method through the collection and analysis of sales performance data. Figure 14 is an explanatory diagram of the logic for calculating the average user consumption cycle by product. Figure 15 is an example of a screen for viewing slow aging recipe content and entering comments on a user terminal. Figure 16 is an example of a screen for managing the results of collecting potential buyer comments and the status of automatic listing. Figure 17 is an example of a screen showing the reception of a personalized group purchase notification message on a user terminal. Figure 18 is an example of a dashboard for predicting the timing of the resumption of group purchasing and managing inventory linkage. Figure 19 is an example of a visualization screen showing the results of a sales performance analysis (comparison of conversion rates by content). Figure 20 is an example diagram of the core table structure of the database (DB) of a group buying server. Specific details for implementing the invention
[0023] Hereinafter, various embodiments are described in more detail with reference to the attached drawings. The embodiments described in this specification may be modified in various ways. Specific embodiments may be depicted in the drawings and described in detail in the detailed description. However, specific embodiments disclosed in the attached drawings are intended only to facilitate understanding of various embodiments. Accordingly, the technical concept is not limited by specific embodiments disclosed in the attached drawings, and it should be understood that it includes all equivalents or substitutions that fall within the spirit and scope of the invention.
[0024] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but these components are not limited by the aforementioned terms. The aforementioned terms are used solely for the purpose of distinguishing one component from another.
[0025] Functions related to artificial intelligence according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.
[0026] The predefined rules of operation or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a basic artificial intelligence model is trained using a number of training data by a learning algorithm, thereby creating predefined rules of operation or artificial intelligence models configured to perform desired characteristics (or objectives). Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.
[0027] An artificial intelligence model can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple nodes and weight values, and performs neural network operations through calculations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, multiple weights can be updated so that the loss value or cost value obtained by the artificial intelligence model during the learning process is reduced or minimized. Additionally, to minimize the loss value or cost value, multiple weights can be updated in a direction that minimizes the gradient associated with the loss value or cost value. Artificial neural networks may include deep neural networks (DNNs), such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Restricted Boltzmann Machines (RBMs), Deep Belief Networks (DBNs), Bidirectional Recurrent Deep Neural Networks (BRDNNs), or Deep Q-Networks, but are not limited to the examples mentioned above.
[0028] A network is a network that serves as a transmission path for web pages; it may be a closed network such as a LAN (Local Area Network) or WAN (Wide Area Network), but it is desirable for it to be an open network such as the Internet. The Internet refers to a global open computer network structure that provides the TCP / IP protocol and various services existing at its upper layers, namely HTTP (HyperText Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), and NIS (Network Information Service).
[0029] Terminals can be implemented in various forms. For example, the terminals described in this specification may include mobile terminals such as smartphones, tablet PCs, PDAs, portable multimedia players, and MP3 players, as well as fixed terminals such as smart TVs and desktop computers.
[0030] In this specification, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. When a component is described as being “connected” or “connected” to another component, it should be understood that it may be directly connected to or connected to that other component, or that there may be other components in between. On the other hand, when a component is described as being “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.
[0031] Meanwhile, a "module" or "part" for a component as used in this specification performs at least one function or operation. Furthermore, a "module" or "part" may perform a function or operation by hardware, software, or a combination of hardware and software. Additionally, a plurality of "modules" or a plurality of "parts," excluding a "module" or "part" that must be performed on specific hardware or on at least one processor, may be integrated into at least one module. A singular expression includes a plural expression unless the context clearly indicates otherwise.
[0032] In addition, power, power transmission, and control therefor for the following assembly configurations and embodiments, including "by control," follow conventional technology including terminals, applications, hardware control modules, etc., so they are omitted to avoid redundancy.
[0033] In addition, the operation embodiments and configurations described in a general manner without being explained in detail below follow the prior art and are omitted in order to focus on describing the purpose of the present invention and the resulting effects.
[0034] Furthermore, in describing the present invention, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description is abbreviated or omitted.
[0036] The present invention relates to a system that produces and disseminates recipe content utilizing health foods (e.g., olive oil, kale, berries, mineral salt, Manuka honey, etc.) and related products tailored to the 'slow aging' trend, automatically identifies and collects the purchasing intent of potential buyers through comment keywords in conjunction with this, analyzes the expected time of product consumption to determine the time to resume group purchasing, and conducts personalized notifications and group purchasing based on this.
[0038] FIG. 1 is an overall configuration diagram of a joint purchasing system for health foods and related products based on slow-aging recipe content according to the present invention. Referring to FIG. 1, the joint purchasing server is a core element of the present invention that serves as a central control. The content provision platform is a platform for uploading content, which includes icons for YouTube, Instagram, TikTok, etc., and receives user comment data by transmitting it to the joint purchasing server. It also receives content viewing, comment input, and notifications through user terminals. Additionally, the sales / payment system is an external shopping mall or payment system where actual product orders and payments take place. It can receive sales data by transmitting it to the joint purchasing server and can receive data by linking the actual inventory status to the joint purchasing server via the product supply / inventory system.
[0039] The overall configuration diagram of the joint purchasing system for health foods and related products based on slow-aging recipe content according to the present invention illustrates the virtuous cycle structure of the invention, which integrates the content creation and dissemination area with the commerce and supply area, centered around the joint purchasing server (100). This system distributes recipe content through the content provision unit (110), and the potential buyer comment collection unit (120) automatically identifies purchase intent comments generated from user terminals to secure a list of potential customers. Subsequently, the joint purchasing resumption time determination unit (130) analyzes sales / inventory data to predict the optimal sales time, and the joint purchasing notification unit (140) automatically sends personalized notifications to potential customers at the predicted time. Finally, the sales performance analysis unit (150) comprehensively analyzes data from the entire process and provides feedback to the system regarding content creation and sales strategies, thereby completing an intelligent automation loop that maximizes operational efficiency and repurchase rates.
[0041] Here, health foods such as olive oil, kale, berries, mineral salt, and Manuka honey may be mainly natural superfoods or wellness ingredients, and related products may be secondary processed foods (concentrates and juices, powders, nutritional supplements, etc.), tools for consumption / storage (glass bottles, dispensers, wooden sticks, grinders, blenders, etc.), beauty / care products (natural soaps, scrubs, bath additives, etc.).
[0043] The joint purchasing server (100) of the present invention is a central system that integrates and controls all core components of the present invention and manages a database, and includes components such as a content provision unit (110), a potential buyer comment collection unit (120), a joint purchasing resumption time determination unit (130), a joint purchasing notification unit (140), and a sales performance analysis unit (150) in a modular form, and is responsible for stable operation and data processing 24 hours a day, thereby serving as a core infrastructure that achieves the technical purpose of the invention.
[0044] The above-mentioned group buying server (100) is intended to upload certain content and send notification messages to potential buyers based on comments on the uploaded content. It provides a system environment in which each module (110~150) functions independently while maintaining an interdependent relationship, and comprehensively controls function calls and data transmission. For example, when the potential buyer comment collection unit (120) generates a list, the server prepares to transmit this list to the group buying notification unit (140).
[0045] It provides infrastructure to handle all tasks requiring stable 24-hour operation without interruption, such as real-time / periodic communication with social platforms (comment collection), complex time-series analysis (determination of resumption time), and sending large-scale notifications (notification department).
[0047] The core role of the joint purchase server (100) is to collect, store, and manage all data generated throughout the entire invention process.
[0048] All key data, such as potential buyer IDs, comment content, views and reactions per content, sales volume from past group purchases, and average consumption cycles per product, are integrated and stored in the server's database.
[0049] The accumulated data becomes the sole source of 'knowledge' necessary for the prediction model of the group purchase resumption timing decision unit (130) and the generation of marketing feedback by the sales performance analysis unit (150). Without a server, all automation and intelligent functions are impossible.
[0051] The group buying server (100) automates the business cycle from content distribution to actual purchase and prediction of the next purchase time.
[0052] Customers whose purchase intent has been confirmed through the potential buyer comment collection unit (120) are automatically filtered and listed without manual work, thereby focusing marketing efforts on the users most likely to purchase.
[0053] The group purchase resumption timing determination unit (130) automatically sets up notifications to open a group purchase at the time of the user's repurchase demand predicted. This maximizes the repurchase rate and ensures that sales opportunities are not missed.
[0054] By automatically sending personalized notifications based on customer IDs at a time determined by the group purchase notification unit (140), necessary information is provided to the user in a timely manner, and the system secures a high purchase conversion rate.
[0056] The group purchase server (100) performs functions beyond simple execution, including measuring system operation performance and suggesting directions for improvement.
[0057] Through the sales performance analysis department (150), the content response (views, comments) and actual sales results (conversion rate, repurchase rate) are linked and analyzed. The analysis results (e.g., the conversion rate of the 'morning fasting routine' content is the highest) are fed back to the next content production strategy (content provision department) to complete a virtuous cycle structure of the system so that content with high sales efficiency is continuously produced.
[0059] FIG. 2 is a detailed internal configuration block diagram of the joint purchasing server of the present invention. The joint purchasing server (100) of the present invention is configured to include a potential buyer comment collection unit (120), a joint purchasing resumption time determination unit (130), a joint purchasing notification unit (140), and a sales performance analysis unit (150).
[0061] FIG. 3 is a flowchart illustrating the cyclic data flow in which data is generated, analyzed, and transmitted during the overall operation process of the ‘low-aging recipe content-based health food and related product group purchasing system’ of the present invention, clearly diagramming the automated virtuous cycle structure of content-commerce-feedback, which is a technical feature of the present invention. The flow of this diagram consists of the following five key steps, and this cyclic process is continuously repeated under the intelligent control of the group purchasing server (100). Content creation and uploading involves the content provision unit (110) creating ‘low-aging recipe’ content, distributing it to an external platform by inserting a traceable link, and generating content response data. Potential customer identification and listing involves the potential buyer comment collection unit (120) collecting comments on the content, identifying potential customers with purchasing intent through NLP analysis, and listing them in the potential buyer database. For the accumulation of sales / consumption data and the prediction of repurchase timing, the group purchase resumption timing determination unit (130) intelligently predicts and determines the optimal time when customer repurchase demand is expected, based on previous sales records and product-specific consumption cycle data. For personalized group purchase notifications, the group purchase notification unit (140) automatically sends personalized notification messages and purchase links to a list of potential customers in accordance with the predicted D_GO to induce actual purchase conversion. For purchase conversion and performance analysis, the sales performance analysis unit (150) analyzes integrated data such as conversion rates and repurchase rates by content, and sends the results back to the system as feedback for content creation strategies and prediction model improvements for the next cycle, thereby completing a cyclical operation that continuously improves the efficiency of the invention system.
[0063] Each component of the server of the present invention is described in detail as follows.
[0065] The content provision unit (110) is intended to produce and distribute content (videos, recipe books, etc.) for ‘slow aging’ recipes and consumption methods. Beyond simple uploading to content platforms (YouTube, Instagram, etc.), it centrally manages the number of views, reactions, and group purchase link (URL) included within each piece of content on the server, and transmits basic data for analyzing purchase conversion rates to the potential buyer comment collection unit and the sales performance analysis unit.
[0067] The content provider (110) is a component responsible for producing ‘slow aging recipe content’ which is the technical basis of the present invention, spreading it to the public through external content platforms such as YouTube, Instagram, and TikTok, and collecting all basic data (views, reactions, link clicks) generated from the spread content and transmitting it to the group purchase server (100).
[0068] In other words, beyond simple content creation functions, the primary purpose is to scientifically analyze the path of potential customer acquisition by inserting trackable links (Tracking URLs) directly connected to group purchases within the content and managing click-through rates and content response data.
[0070] The core functions of the content provider (110) are as follows.
[0071] The content provision department (110) plans and produces various forms of content (videos, recipe books, etc.) on the theme of 'slow aging', and manages distribution by selecting an external content platform suitable for each piece of content. The themes of the content are directly related to the invention, such as how to consume, the benefits, and recipes for using group purchase products (olive oil, Manuka honey, etc.).
[0072] In addition, a unique, traceable link (URL_{content_ID}) specialized for each content is generated by linking with the group purchase server (100). This link is inserted into the content's description box, pinned comment, or caption within the video, etc., so that when a user clicks it, the click data is recorded in the DB of the group purchase server (100).
[0073] The platform API where the content is uploaded is periodically called to collect basic content response data, such as the number of views, likes, shares, and user comments per content, and transmits it to the potential buyer comment collection unit (120) and the sales performance analysis unit (150).
[0074] The produced content is classified by topic (e.g., Type A: morning empty stomach intake method, Type B: cooking recipe, Type C: user challenge review), and metadata such as content ID, product relevance, and distribution platform is stored in the group purchase server (100) and used as a standard for subsequent analysis (150).
[0076] The core operation algorithm of the content provider (110) focuses on data linkage through a traceable link as follows. FIG. 8 is a diagram showing the link and data linkage method between the content provider and the group purchase server.
[0077] First, once the creation of new content is complete, a unique ID (Content_ID) is assigned to the content, and the associated group purchase product (Product_ID) information is finalized.
[0078] Second, based on Content_ID and Product_ID, a link generation module within the group purchase server (100) is called to obtain a unique, traceable link (URL_content_ID). (e.g., go.co / link?cid=abc&pid=oil)
[0079] Third, when uploading content to an external platform (e.g., YouTube), be sure to insert the issued URL_{content_ID} in the video description.
[0080] Fourth, when a user clicks URL_{content_ID} and moves to the actual group buying site, the group buying server (100) performs a redirection and records the following data in real time: click time, content ID (Content_ID), entry path (platform), user terminal information (optional) [Reaction data collection (113)]. Periodically (e.g., every hour), the latest reaction data, such as the number of views, likes, and comments associated with the Content_ID, is collected through the platform API and updated in the DB of the group buying server (100).
[0082] The most important technical feature of the content provision unit (110) is that, beyond simple 'channel operation,' it makes all content activities dependent on the system's data flow. Additionally, the sales performance analysis unit (150) can directly compare the number of clicks on URL_{content_ID} generated for each content with the actual purchase conversion rate, allowing for quantitative determination of which content type (A, B, C) contributes most to sales and optimization (feedback) of the next content production strategy.
[0083] The above content providing unit (110) provides initial data for automatically identifying and listing users with a high intention to purchase by immediately transmitting comment response data for the content to the potential buyer comment collection unit (120).
[0085] The potential buyer comment collection unit (120) automatically detects and collects specific keywords (e.g., "participate in group purchase", "please share the link") among video content comments and identifies the author ID.
[0086] The potential buyer comment collection unit (120) collects, analyzes, and filters comment data in real-time or periodically through a natural language processing (NLP) engine implemented inside the group buying server, using APIs of external content platforms or web crawling / parsing technology. This list is stored in a database (DB) and utilized by the group buying notification unit. Automatic detection and listing are impossible without a server.
[0088] The potential buyer comment collection unit (120) is implemented inside the group buying server (100) and is an intelligent module that monitors the comment area of an external content platform (YouTube, Instagram, etc.) distributed by the content provision unit (110) in real time or periodically, and automatically identifies, collects, and analyzes text containing the user's explicit / implicit purchase intention to generate a potential customer list.
[0089] The primary purpose of this module is to maximize marketing efficiency and purchase conversion rates by fully automating the manual process of checking comments and responding to customers, and by targeting group purchase notifications only to users with confirmed purchase intent.
[0091] The above potential buyer comment collection unit (120) collects all comment data associated with a specific content ID (Content_ID) along with metadata such as raw text, author ID, and time of creation by using an API (Application Programming Interface) of an external content platform or web crawling / parsing technology.
[0092] In addition, the potential buyer comment collection unit (120) primarily filters comments containing a predefined list of purchase-inducing keywords ("group purchase", "link", "participation", "I want to buy", "next sale") and product names covered by the content (e.g., "olive oil", "Manuka honey").
[0093] The potential buyer comment collection unit (120) uses an NLP engine implemented inside the group buying server (100) to analyze the context of comments containing keywords. It determines whether the comments contain positive purchase intent (e.g., 'Please share the link') or negative intent (e.g., 'It seems better not to do the group purchase'), and calculates a purchase intent score (Purchase Intent Score) from 0 to 100. It identifies the author ID (@UserID) of comments where the purchase intent is determined to be positive, and verifies duplication and validity by querying the DB to see if the ID is already listed or if it is a valid account capable of sending notifications. Finally, it stores the user IDs and content information whose purchase intent score is above a preset threshold as a list of potential buyers in the DB of the group buying server (100), and sets the status to 'Notification needed' so that it can be utilized by the group buying notification unit (140). Fig. 4 is a detailed configuration and operation flowchart of the potential buyer comment collection unit, which is referenced here.
[0095] The potential buyer comment collection unit (120) lists potential buyers through the following detailed steps, and this process is illustrated in FIG. 10.
[0096] First, a list of comments for new content is periodically received from the content provider (110), or comment data is loaded by calling the platform API. For each loaded comment (N), it is checked whether it is included in the core keyword list. If no keyword is detected, the comment is discarded and the process moves on to the next comment.
[0097] An NLP engine is applied to comments where keywords are detected to analyze the semantics of the text. To detect negative words, it is checked whether negative expressions such as 'can't it be?', 'I don't like it', or 'stop' are used in conjunction with the keywords. To calculate the intent score, if the purchase intent is positive, a purchase intent score (e.g., 80 points) is assigned based on the intensity of the context. It is determined whether the calculated purchase intent score exceeds a threshold (e.g., 70 points). The author ID of comments that pass the threshold is extracted, and it is checked in the DB to see if the corresponding ID is already registered in a valid list of potential buyers (to prevent duplicate registration). For valid, non-duplicate IDs, the Content_ID, Product_ID, purchase intent score, and listing time are stored in the group buying server DB, and the notification sending status for that ID is set to 'Active'.
[0099] The technical feature of the potential buyer comment collection unit (120) is that it converts comments, which are unstructured data, into structured potential customer data. In addition, by accurately identifying only a small number of users who actually intend to purchase among numerous comments, it enables accurate targeting instead of sending notifications indiscriminately to a large number of users, thereby reducing marketing costs.
[0100] The potential buyer comment collection unit (120) improves the customer experience (CX) that meets user expectations without manual response delays by completing preparations to provide notifications at the time explicitly requested by the user. In addition, beyond simple comment collection, it classifies the priorities of potential customers using a quantitative indicator called a 'purchase intention score' through NLP and provides this data as analysis input data to the sales performance analysis unit (150).
[0102] The Natural Language Processing Engine (NLP) applied to the potential buyer comment collection unit (120) is a software module that converts unstructured data (comments) in text form into structured information (purchase intention score, positive / negative) that a computer can understand and analyze. It is implemented inside the group buying server (100).
[0104] The above NLP engine includes specialized text classification and sentiment analysis models focused on the following functions.
[0105] First, Korean comments are separated into word (token) units, and the part of speech (noun, verb, etc.) and semantic unit (morpheme) of each word are identified. This lays the foundation for interpreting various expressions, such as "gonggu" and "buying and wanting," with the same intent.
[0106] Second, the overall purpose of the comments (questions, praise, complaints, purchase intent) is classified into categories. This acts as a filter to select only comments with a clear 'purchase intent' and pass them to the next stage.
[0107] Third, it determines whether the tone of the comments is positive, negative, or neutral. Through this, it serves to exclude the keyword ("group purchase") from the potential customer list even if it is included in a negative context, such as "I wish you wouldn't do a group purchase."
[0109] The potential buyer comment collection unit (120) can use an NLP engine to calculate the final purchase intent score (Purchase Intent Score) through the following three steps.
[0111] Step 1: Keyword-based context extraction
[0112] The NLP engine primarily receives only comments filtered by keyword filters (e.g., "link", "group buy", "purchase").
[0113] Identify the sentence structure based on key keywords in the comments, and extract the context of 3 to 5 words before and after the keywords to determine whether the comment is a request related to group purchasing.
[0115] Step 2: Sentiment Analysis and Negative Filtering
[0116] Determine whether the comments represent a positive purchase opportunity, or complaints or negative feedback.
[0117] Reflecting the characteristics of the Korean language, it detects negative adverbs and auxiliary verbs such as 'an', 'mot', 'ma-seyo', and 'eop-eyo'. Even in negative sentences of the request form, such as "Don't give me the link?", it distinguishes clear negative intentions like "The tool is not very good."
[0118] Purchase intention scores are increased when positive expressions such as "I will definitely buy it," "It's so good," and "Thank you" are used together.
[0120] Step 3: Calculate Purchase Intent Score and Targeting
[0121] The final purchase intention of the comments is quantified by synthesizing the analyzed results.
[0122] The NLP engine calculates a purchase intention score between 0 and 100 by combining the following elements.
[0124] Purchase Intention Score = (Base Score x W_1) + (Positive Sentiment Score x W_2) - (Negative Sentiment Score x W_3)
[0126] Here, W_1, W_2, and W_3 are weights representing the importance of each element.
[0128] W_1 is a weight that quantifies how much the comment content contains core keywords that explicitly indicate purchase intent. Since this is the most basic and powerful indicator for identifying potential customers, an initial weight of 0.5 (50%) is assigned. The technical definition of this weighting factor is to measure the presence and frequency of keywords directly related to purchase requests, such as "link," "group buy," and "I want to buy." Additionally, if specific product names covered in the content (e.g., "olive oil," "mineral salt") are mentioned together, it is considered as specific interest in a particular product rather than a simple inquiry, and an additional bonus point of 0.05 is assigned to reinforce the score.
[0130] W_2 is a weight used to determine whether the likelihood of an actual purchase is high, based on whether the comments contain positive and proactive sentiment. An initial weight of 0.3 (30%) is assigned to this factor. The technical definition of W_2 is to measure the presence of expressions that emphasize purchase intent or include positive evaluations, such as "definitely," "right away," "quickly," and "wait." Since this positive sentiment goes beyond simple keyword mentions and reflects a proactive willingness to execute that supports purchase intent, it serves as an important factor in constructing the score alongside W_1.
[0132] W_3 is a penalty factor used to filter out comments that should not be classified as potential customers because, despite containing keywords related to purchase intent, they carry negative contextual connotations. This factor is applied by deducting from the score, with a high initial penalty value, such as 0.8. The technical definition of W_3 is to measure the presence of expressions indicating a negative tone or an explicit refusal to purchase, such as "not," "cannot," "not really," or "wouldn't have been." Through this penalty factor, the system accurately excludes users without purchase intent, even if their comments contain keywords, thereby preventing the sending of unnecessary automated notifications.
[0134] The group buying server (100) can register only users who exceed a preset threshold (e.g., 70 points) using the purchase intention score as described above in the potential buyer list, and can provide the customer with the highest score with the opportunity to send priority notifications.
[0136] As such, unlike existing systems that merely check whether keywords are included, the present invention accurately identifies the complex context and sentiment of Korean comments (e.g., 'You still haven't opened the tool. Please give me the link quickly.') to select only genuine potential customers. Additionally, by filtering out comments with negative intent, it prevents the sending of unnecessary notifications and improves the customer experience by increasing the accuracy of notification sending. Furthermore, by comparing and analyzing the 'purchase intention score' and the actual 'purchase conversion' results in the sales performance analysis unit (150), the weights and thresholds of the NLP model for score calculation are continuously improved (retrained), thereby increasing the prediction accuracy of the system over time.
[0138] The group purchase resumption timing determination unit (130) is implemented within the group purchase server (100) and is an intelligent module that comprehensively analyzes sales data from previous group purchases, average user consumption cycle data by product, and current inventory status data. This module is responsible for the function of automatically predicting the optimal time when a user is expected to consume all purchased products and want to repurchase, and confirming the predicted time as the start date of the next group purchase. FIG. 5 is a flowchart for explaining the group purchase resumption timing determination unit.
[0139] The primary purpose of this module is to support a seamless health management routine by providing sellers with optimal sales timing to achieve the highest repurchase rates and revenue while minimizing the risk of inventory depletion, and by offering users necessary notifications when products are running low.
[0141] The group purchase resumption time determination unit (130) loads information on the last sales date (D_last), sales quantity (Q_sold), and average user consumption cycle (Cycle_avg) per product of the previous group purchase from the DB of the group purchase server (100). The consumption cycle data is stored internally (e.g., olive oil - 30 days, Manuka honey - 60 days) or uses values updated through in-depth analysis by the sales performance analysis unit (150).
[0142] The above-mentioned joint purchase resumption timing determination unit (130) calculates, through mathematical logic, the expected time (D_pred) when potential customers have finished consuming the product and repurchase demand will occur, based on D_last and Cycle_avg. This step performs the core intelligent prediction function of the invention. It checks the current inventory level and the expected delivery date (LT) in real time by linking with a product supply / inventory system (external). It determines whether the expected demand can be met if the joint purchase is initiated at the predicted time (D_pred).
[0143] In addition, the expected repurchase time (D_pred) and the inventory securing schedule (D_stock = current time + LT) are compared, and the time closest to D_pred and when sufficient inventory necessary for sales is secured is determined as the final group purchase resumption date (D_GO). Information regarding the determined group purchase resumption date (D_GO) is transmitted to the group purchase notification unit (140) to set a time trigger for sending notifications, and this prediction result is stored in the DB of the group purchase server (100).
[0145] The joint purchase resumption timing determination unit (130) determines the optimal D_GO through the following detailed steps, and this process is illustrated in FIG. 11.
[0146] The final sales date (D_last) and average consumption cycle (Cycle_avg) for product P are loaded from the DB of the group purchase server (100).
[0147] The point in time when the buyer is expected to have used up the product (D_pred) is calculated by adding the average consumption cycle to the date the final sale was completed, using the formula (D_pred = D_last + Cycle_avg).
[0148] Compare the current inventory (S_current) with the expected demand (Q_pred), and determine whether the total inventory (S_total) exceeds Q_pred by the time point D_pred due to the arrival of new items from the inventory system.
[0149] If there is sufficient inventory or it is scheduled to be secured at D_pred, set D_GO = D_pred.
[0150] If there is a shortage of inventory at time D_pred and it can only be secured after the lead date (LT), set [D_GO = D_last + Cycle_avg + T]. (Here, T reflects the lead delay period.)
[0151] Marketing data such as the ‘optimal day / period (e.g., last Friday of the month)’ provided by the sales performance analysis department (150) is reflected in the D_GO decision and slightly adjusted.
[0152] The basis for this prediction (sales volume, consumption cycle, inventory status) is recorded in the DB along with the final confirmed D_GO, and the D_GO is transmitted to the group purchase notification unit (140) to schedule the sending of the notification.
[0153] Figure 14 is an explanatory diagram of the logic for calculating the average user consumption cycle by product, briefly illustrating the above content, and relates to receiving the calculated data after data input and calculation of the expected repurchase time.
[0155] The joint purchase resumption timing determination unit (130) has the following technical features.
[0156] Unlike the existing method of simply opening group purchases at fixed intervals (e.g., the 1st of every month), the timing of sales is determined based on the user's actual consumption pattern (Cycle_avg). This maximizes the repurchase rate by providing notifications at the time when customers most need to repurchase.
[0157] By reflecting inventory status and delivery dates in real time, it aligns the timing of predicted supply and demand. This enables scientific operations that prevent lost sales opportunities due to inventory shortages or increased logistics costs caused by excess inventory.
[0158] The process of the seller manually analyzing data and deciding on the next sales date is eliminated, and the system automatically determines the optimal date and immediately links with the notification sending system (140), thereby maximizing operational efficiency.
[0160] The group purchase notification unit (140) is implemented inside the group purchase server (100) and is a module that receives the optimal group purchase start date (D_GO) determined by the group purchase resumption time determination unit (130) and automatically sends personalized notification messages and purchase links to targets based on the list of potential buyers generated by the potential buyer comment collection unit (120).
[0161] The primary purpose of this module is to maximize marketing efficiency and convert the high purchase intent of potential customers into actual purchase conversions by providing timely information so that potential customers who have expressed purchase intent do not miss the opportunity for group buying.
[0163] The main functions of the group purchase notification unit (140) are as follows.
[0164] First, in conjunction with the scheduling system of the group purchase server (100), the D_GO received from the group purchase resumption time determination unit (130) is set as the notification sending time (T_notify) and is continuously monitored. (e.g., 10:00 AM on the day of D_GO)
[0165] Second, when the notification time arrives, a list of valid potential buyer IDs is retrieved from the DB of the group purchase server (100) in which the purchase intention score is above the threshold and the notification sending status is set to 'active'.
[0166] Third, a personalized notification message is generated based on the potential buyer's ID and the product information (Product_ID) for which the purchase was requested. The message includes a group purchase tracking link (URL_tracking) and the notification recipient's ID (e.g., "Mr. / Ms. OOO").
[0167] Fourth, send personalized messages by integrating with external channel APIs (gateways) such as DM, email, and messenger. It may include retry logic in case of sending failure.
[0168] Fifth, the success or failure of sending notifications to each user is tracked, and the sending records (time of sending, message content, channel) are stored in the DB. This data is used as basic data for analyzing the effectiveness of notifications in the sales performance analysis department (150).
[0169] Figure 7 is a flowchart of the detailed operation of the group purchase notification unit and the integration of the message sending gateway. The notification creation trigger step operates on the date and time determined by the 'group purchase resumption time determination unit', the potential buyer list call step filters the list of DM reception consents and valid IDs from the DB, the message template load step loads a personalized message (ID, product name, link) template, the message gateway step indicates integration with an external channel sending system including DM (Direct Message) API, E-mail SMTP, messenger API, etc., and the notification sending and recording step sends the notification and records whether the sending was successful or failed in the DB.
[0171] The group purchase notification unit (140) sends personalized notifications through the following detailed steps, and this process is illustrated in FIG. 12.
[0173] [Notification Trigger Activation]: The notification sending process begins when the group purchase resumption date (D_GO) and time (T_notify) are reached.
[0174] [Get Target List]: Retrieve the list of targets for sending notifications (List_target) from the potential buyer DB.
[0175] [Start iteration]: Repeat the following steps for each potential buyer (User_N) in List_target.
[0176] [Generate Message]: Generate a personalized message by merging User_N's ID and associated product information into a template, and insert the corresponding group purchase tracking link (URL_tracking).
[0177] [Select Channel and Send]: Select User_N's preferred channel (DM, email, etc.) and attempt to send a notification via the corresponding channel's gateway API.
[0178] [Record Sending Result]: Records whether the sending was successful or failed in the DB, and if it fails, changes User_N's notification status to 'Inactive' or retryes via an alternative channel.
[0179] [End of loop]: Repeats until notifications are sent to all users in List_target.
[0181] The group purchase notification unit (140) has technical features of personalization, timeliness, and automation.
[0182] That is, through the prediction of the joint purchase resumption timing determination unit (130), a notification is sent at the very moment when the customer is expected to want to repurchase. This perfectly matches the customer's demand with the sales timing, dramatically improving the purchase conversion rate.
[0183] Group purchase notifications for specific products (e.g., olive oil) are sent only to users who have previously explicitly expressed an intention to purchase those products in the comments. This prevents indiscriminate promotional notifications and provides customers with a positive experience of being remembered and informed of their needs.
[0184] All marketing campaign processes, from identifying potential customers (120) to determining the timing (130) and sending the final notification (140), are performed automatically on the server without manual intervention, minimizing operating costs and human error.
[0185] In addition, the group purchase notification unit (140) can increase the notification delivery rate by simultaneously linking gateways of various social channels (DM, messenger, etc.) and delivering notifications to the platform primarily used by customers.
[0187] The sales performance analysis unit (150) is implemented inside the group buying server (100) and is an intelligent module that collects, connects, and analyzes all data (content response, potential customer intent, notification sending results, actual sales and repurchase records) generated throughout the entire process of the system, and calculates quantitative performance indicators regarding which marketing activity (content type, notification timing) achieved the highest purchase conversion rate.
[0189] The main purpose of this module is to go beyond simply aggregating sales results and to automatically feed the analyzed results to the content provision unit (110) and the joint purchase resumption time determination unit (130), thereby completing a virtuous cycle structure that continuously improves the operational efficiency and profitability of the invention system.
[0191] The sales performance analysis department (150) integrates and joins the following heterogeneous data from the DB of the group buying server (100).
[0192] 1) Content data (views, number of comments), 2) Potential customer data (purchase intent score, notification response rate), 3) Sales data (buyer ID, sales quantity, repurchase status). Data cleaning and synchronization tasks are performed for analysis.
[0193] We calculate the final purchase conversion rate relative to potential customer acquisition (link clicks or comments) by content type and platform. Through this, we quantitatively identify the most efficient marketing channels and content types (e.g., fasting routines vs. cooking recipes).
[0194] The repurchase timing, purchase frequency, and average consumption of individual customers are analyzed, and this data is fed back to the group purchase resumption timing determination unit (130). Through this process, the average user consumption cycle (Cycle_avg) for each product is regularly updated to increase the accuracy of the prediction.
[0195] The analyzed key indicators (conversion rate, repurchase rate, content efficiency) are automatically generated into a visualized report in the form of a dashboard (Fig. 19) so that the manager can intuitively understand them.
[0196] Based on the analysis results, specific feedback for the next marketing strategy (e.g., "increase the proportion of Type A content by 20%", "raise the threshold for NLP intention scores") is automatically sent to the content provision unit (110) and the potential buyer comment collection unit (120).
[0198] Referring to FIG. 6, the sales performance analysis unit (150) integrates (comment collection data + sales data + notification response data) into a single DB, analyzes the purchase conversion rate by content type (A: fasting routine, B: cooking recipe), analyzes the repurchase rate and purchase frequency of potential customers, generates the analysis results in the form of a visualized report, and transmits the report results to the content provision unit and the group purchase resumption timing decision unit to induce strategy modification.
[0200] The sales performance analysis department (150) performs strategic analysis through the following detailed steps to drive the virtuous cycle structure of the invention. This process is illustrated in FIG. 13.
[0202] [Data Integration]: Establish an integrated analysis environment by retrieving content, potential customer, and sales / repurchase data from a central database at the end of a group purchase or periodically (e.g., weekly).
[0203] [Calculation of Performance Indicators by Content]
[0204] - Conversion Rate (CR) Calculation: The conversion rate is calculated per content ID as (Number of final purchases) / (Number of tracking link clicks or lead list registrations).
[0205] - Comparison by Type: Compares conversion rates (CR) by group based on content metadata (Types A, B, C) to identify the highest-performing content types.
[0206] [Consumption Cycle Analysis]: Analyze purchase quantity and repurchase timing per customer to recalculate the current value of Cycle_avg for product P.
[0207] [Strategic Decision-Making and Feedback Generation]
[0208] Based on the CR comparison results by type, an instruction is sent to the content provision unit (110) to increase the production proportion of content types with superior performance during the next period. The recalculated Cycle_avg value is sent to the group purchase resumption time determination unit (130) to improve the accuracy of the next group purchase time prediction model. If the actual purchase rate differs from the expected purchase intention score, an instruction is given to adjust the NLP engine weights (W_1, W_2, W_3) of the potential buyer comment collection unit (120).
[0209] [Generate Report]: Visualizes analysis results and provides them to the administrator dashboard.
[0211] The technical features of the sales performance analysis department (150) are data connectivity and automated feedback loops.
[0212] The entire business process from content creation (110) to sales (external system), analysis (150), and strategic feedback (110, 130) is automatically circulated within the server. This is the only module that technically implements the core virtuous cycle structure of the invention.
[0213] Rather than simply looking at the result of 'high sales,' it accurately identifies 'who viewed which content bought it and when they will buy it again,' automatically guiding the system's resources (time, content creation) to be concentrated on the activities with the highest ROI (Return on Investment).
[0214] The sales performance analysis unit (150) provides a self-improvement function in which the accuracy of sales timing prediction improves as the system operates by continuously updating the prediction model of the joint purchase resumption timing determination unit (130) by reflecting the actual behavioral patterns of customers.
[0216] FIG. 9 is a flowchart of the overall operation method of a health food and related product group purchasing system according to the present invention. This figure is a detailed flowchart showing, step by step, the sequence of the overall operation method in which the system of the present invention operates intelligently from the start of content creation to the next group purchasing strategy feedback. This method shows a closed-loop structure in which each component is organically linked under the control of a group purchasing server (100).
[0217] Referring to FIG. 9, as an initial stage, the content provider (110) produces 'slow aging recipe' content, inserts a traceable link (URL_tracking), distributes it to an external content platform, and generates content response data. The potential buyer comment collection unit (120) periodically collects comment data from the content platform. An NLP engine is applied to the collected comments to analyze purchase intent keywords and positive sentiment scores, and user IDs whose scores are above a threshold are registered in the potential buyer list DB.
[0218] As the next step, the group buying server (100) determines whether the current time has reached the previously predicted optimal group buying resumption date (D_GO).
[0219] If (NO), the system continues with potential customer identification and data accumulation.
[0220] If (YES), proceed to the next prediction and decision step.
[0221] The group purchase notification unit (140) automatically sends personalized notification messages and purchase tracking links to potential buyers on the confirmed resumption date. It collects data such as sales records, buyer IDs, and repurchase status during the group purchase period from the sales / payment system and stores them in the DB. The sales performance analysis unit (150) analyzes the purchase conversion rate (CR), repurchase rate, and notification effectiveness by content type using the integrated data.
[0222] Based on the analyzed performance results, strategic feedback, such as adjusting the proportion of next-generation content production, updating the prediction model (Cycle_avg), and adjusting NLP weights, is transmitted to the relevant components of the system.
[0223] After incorporating the feedback, the system returns to the first step to prepare for the next group buying cycle and continues the operation cyclically.
[0225] Figure 15 is an example of a screen for viewing slow aging recipe content and entering comments on a user terminal.
[0226] Figure 16 is an example of a screen for managing the results of collecting potential buyer comments and the status of automatic listing.
[0227] Figure 17 is an example of a screen showing the reception of a personalized group purchase notification message on a user terminal.
[0228] Figure 18 is an example of a dashboard for predicting the timing of the resumption of group purchasing and managing inventory linkage.
[0229] Figure 19 is an example of a visualization screen showing the results of a sales performance analysis (comparison of conversion rates by content).
[0230] Figure 20 is an example diagram of the core table structure of the database (DB) of a group buying server.
[0232] The group buying server (100) may further include a social graph-based group buying amplification unit (160), and the social graph-based group buying amplification unit (160) is an intelligent diffusion module implemented inside the group buying server (100) that constructs a heterogeneous graph between user-user-product using customer IDs obtained from potential buyer comment collection unit (120) and interaction data on a social platform. By applying a deep learning model based on a graph convolutional neural network (GCN) or a graph neural network (GNN) to this graph, the main purpose is to predict not only the purchase conversion probability of individual potential customers but also the probability of group buying success and diffusion at the cluster level, and to automatically send a chain invitation bundle notification based on the predicted results to simultaneously maximize the diffusion and success rate of group buying.
[0234] The function and technical mechanism of the social graph-based group purchase amplification unit (160) are as follows.
[0235] The social graph-based group purchase amplification unit (160) constructs a composite graph in the DB that defines User, Product, and Participation Event as Nodes, and defines Share, Comment Friend Tag, and Co-purchase as Edges. This is a basic data structure for quantifying the influence of potential customers' relationship networks.
[0236] A Graph Neural Network (GNN) model is applied to the constructed graph structure to calculate the group-unit diffusion probability (P_spread) for spreading the purchase to neighboring networks (neighbor nodes) when a specific potential customer (node) receives a group purchase notification.
[0237] By analyzing the network centrality of potential customer nodes, high-leverage users with the highest potential as 'influencers' or 'invitation initiators' are identified and prioritized.
[0238] A strategy is established to form groups (clusters) centered on users with a high probability of diffusion, and to send chain invitation bundle notifications to members within each group through the group purchase notification unit (140).
[0240] The linkage structure and operation within the social graph-based group purchase amplification unit (160) and the group purchase server (100) are as follows.
[0241] The potential buyer comment collection unit (120) transmits data to the social graph-based group purchase amplification unit (160) as soon as it extracts the potential customer ID, thereby synchronizing the ID with the graph node.
[0242] When D_GO is determined in the group purchase resumption time determination unit (130), the group purchase amplification unit (160) identifies the group with the highest diffusion probability based on the latest graph of the DB.
[0243] The group purchase notification unit (140) sends notifications based on the ‘chain invitation bundle notification recipient list and group information’ received from the group purchase amplification unit (160), and uses a group invitation message template that is different from a general 1:1 notification.
[0244] The sales performance analysis unit (150) separately measures the 'group conversion rate' in addition to the individual purchase conversion rate and provides feedback data for improving the GNN model of the amplification unit (160).
[0246] The social graph-based group purchase amplification unit (160) adopts a Graph Convolutional Network (GCN) family GNN model to model the flow of information and the propagation of influence within a complex relationship network.
[0247] The GNN model is designed to effectively handle a heterogeneous graph structure in which Users, Products, and Events are defined as nodes, and interactions as edges.
[0248] Unlike existing deep learning models that analyze only independent data points (individual users), GNNs learn by integrating the features (H_u) of a node (user) with the features of connected neighbor nodes. Through this, network diffusion potential, such as 'the probability that friend $B$ will also purchase if user A makes a purchase', can be quantitatively predicted.
[0249] Feature Extraction and Aggregation Mechanism via GNN Models GNN models learn the influence of the relationship network by repeating the core processes of 'information aggregation' and 'transformation'.
[0250] The first layer of the GNN generates an initial embedding vector by receiving user node features, such as ① purchase intention score, ② number of interactions, and ③ total number of purchases, which are normalized in the potential buyer comment collection unit (120). Each user node (u) collects embedding information of all neighbor nodes (v) connected to it. Neighbor nodes include not only friend users, but also users who have purchased the same product or are tagged together in comments, and even interest product nodes.
[0251] Aggregated neighbor information is not simply summed; instead, weights are adjusted by applying a normalization factor based on the connectivity (degree) of each node. This dilutes information from relationships with weak connection strengths and integrates information from highly connected influencer nodes with appropriate weight, reflecting their reliability. Through this, the influence of the relationship network is effectively reflected while preventing information bias caused by specific hub nodes.
[0252] Aggregated neighbor information is transformed through a learnable weight matrix and combined with node u's own embedding information. Through this process, node u's embedding vector is updated into a new feature vector that encompasses node u's individual characteristics and the network's influence, which carries potential diffusion potential.
[0253] Specifically, the model receives past group purchase data (node characteristics and actual purchase conversion status) as input and learns a weight matrix through a backpropagation algorithm to minimize the prediction error of the 'group unit conversion rate (P_spread)'. This learning process is performed intensively during the early stages of system operation, and the model's weights are continuously fine-tuned using actual group purchase conversion rate data provided by the sales performance analysis unit (150) as the label.
[0255] The GNN model outputs the group spread probability (P_spread) at the final layer after undergoing an information aggregation process across multiple layers.
[0256] This probability value (0–1) is a quantitative prediction of the likelihood that the corresponding user node (u) will successfully complete a cluster purchase by inducing sequential purchases through the surrounding relationship network when it receives a group purchase notification.
[0257] For example, a user group with a P_spread of 0.90 (90%) is interpreted as having very high potential as an 'invitation initiator' beyond a simple 1:1 buyer. The social graph-based group purchase amplification unit (160) selects groups whose P_spread value exceeds a preset threshold (e.g., 0.85) as priority targets for 'chain invitation bundle notifications' to maximize the success rate of marketing at the group level. Through such a sophisticated learning mechanism, the GNN model provides a core technical foundation for proving the inventiveness of the invention by utilizing the latent interactions and influence of social networks beyond simple data analysis.
[0259] The social graph-based group purchase amplification unit (160) maximizes the spread and success rate of group purchases by utilizing the influence of the relationship network, rather than the existing individual user targeting method. This module constructs a heterogeneous graph of user-product-event and applies a graph neural network (GNN) model to calculate the 'group-unit spread probability (P_spread)', and this process follows the following specific technical procedure.
[0261] First, input data including potential customer IDs obtained from the potential buyer comment collection unit (120) is normalized. This process constructs a feature vector of the user node (u).
[0262] The three key feature values assigned to each user (u)—① potential purchase intention score (0~100 points), ② number of content interactions in the last 3 months (comments, shares, etc.), and ③ total number of purchases in previous group purchases (0 or more)—are normalized to values between 0 and 1.
[0263] Based on normalized feature values, heterogeneous graphs are constructed between users (friend relationships, tags), users and products (interested items), and users and events (past purchase events), and all relationship network information is modeled as nodes and edges.
[0265] Predictions are performed by applying a GNN model to the constructed graph. This process is a key step that combines the potential purchase intentions of individual users with the purchasing behaviors of neighboring users and the influence of the relationship network.
[0267] An initial embedding vector is generated based on the feature vector (three normalized values) of each user node (u). To proceed to the next step, node u aggregates the embedding information of its directly connected neighbor nodes (v). At this stage, instead of a simple summation method, the importance of the information is adjusted by applying a normalization factor proportional to the degree (number of relationships) between node u and neighbor v. The normalization factor applied is one commonly used in GCN models. It follows the concept of form, providing a symmetric normalization effect that distributes the influence of nodes with many connections and ensures that information from nodes with few connections is not lost.
[0268] This prevents the excessive reflection of information from highly connected 'hub' nodes (influencers) within the network and enhances the credibility of their influence on actual purchasing behavior.
[0269] The normalized neighbor information aggregation result is combined with the embedding information of node u itself, and an activation function (e.g., ReLU) is applied to update the feature vector of node u. By repeating this process through multiple layers, the indirect diffusion effect up to the second and third order relationship networks is reflected in the model.
[0270] The final diffusion probability is calculated based on the updated feature vector, and specific conditions for group-level optimization are applied during this process.
[0271] The final output layer of the GNN model calculates a group-unit diffusion probability (P_spread) between 0 and 1, which spreads the purchase to at least 2 people through the neighbor network when user node u receives a group purchase notification.
[0272] User nodes for which diffusion probability has been calculated are classified as 'High-Leverage Initiators' and form a cluster only if all of the following multiple conditions are met.
[0273] Condition A (spread potential): The calculated P_spread value must exceed a preset threshold value (e.g., 0.85). This threshold value (0.85) is set based on the average spread rate of groups where actual chain purchases occurred in past group purchases during the initial operation of the system, and is then automatically adjusted in 5% increments through feedback from the sales performance analysis unit (150).
[0274] Condition B (Network Centrality): The network centrality index (e.g., Betweenness Centrality) of the user node must be within the top 15% of the entire pool of potential customers. This guarantees the potential to play the role of an 'influencer' with the strongest information delivery power within the network, going beyond simply being a person with high purchase intent.
[0275] Condition C (Recent Activity): The user must have a history of at least 3 interactions within the last 7 days, such as viewing content or using the 'Slow Aging Routine Checklist'. Since potential customers with low current activity are unlikely to take immediate spreading action even if they receive a notification, this condition is used to select highly active customers to maximize the effectiveness of the notification.
[0277] A group (cluster) is formed around users (u) classified as 'group invitation initiators' who satisfy all of the above multiple conditions, and a 'list of targets for sending chain invitation bundle notifications' is sent to the group purchase notification unit (140).
[0279] Unlike conventional methods that select notification targets based on a single condition (e.g., whether only comments were posted), the present invention selects a final pool of candidates by quantitatively evaluating multiple independent conditions, such as potential intent, network influence, and recent activity. This configuration prevents false positives and yields results that improve group-based conversion rates to a statistically significant level compared to conventional methods.
[0281] The group buying server (100) may further include an individual and household unit consumption estimation unit (170), and the individual and household unit consumption estimation unit (170) is implemented within the group buying server (100) and is an intelligent prediction module for overcoming the limitations of the average consumption cycle (Cycle_avg) that was uniformly applied to all users for a single product.
[0282] The main purpose of this module is to significantly improve the prediction accuracy of the joint purchase resumption time determination unit (130) by integrating and analyzing various variables such as individual customers' ① purchase patterns (quantity, frequency), ② seasonality (seasonal factors), and ③ cohort information (single person / household / routine intensity), calibrating the consumption cycle for each customer group (cohort), and recalculating the repurchase time (D_GO) for each cohort.
[0284] The function and technical mechanism of the individual and household unit consumption estimation unit (170) are as follows.
[0285] The individual and household unit consumption estimation unit (170) classifies consumption pattern cohorts based on customer data provided by the sales performance analysis unit (150). The classification criteria include purchase quantity (e.g., 1 vs. 3), purchase unit (e.g., individual purchase vs. bulk bundle), and related content consumption patterns (e.g., whether or not to watch 'daily fasting routine' content intensively).
[0286] Past sales data is analyzed for each cohort to calculate the average consumption cycle (Cycle_cohort) per cohort. This cycle incorporates a seasonal variable (S_factor) to adjust for seasonal changes in consumption (e.g., increased olive oil consumption during the summer salad season).
[0287] Data is provided to the joint purchase resumption timing determination unit (130) to recalculate the optimal repurchase timing (D_GO_cohort) for each cohort using Cycle_cohort instead of the uniform overall average consumption cycle (Cycle_avg).
[0288] Based on the consumption cycle estimation results, information on the most suitable product capacity or bundle configuration for each cohort (e.g., single-person household) is generated and transmitted to the group purchase notification unit (140).
[0290] The individual and household unit consumption estimation unit (170) performs an intelligent procedure to classify customers into cohorts and apply multiple variables to calibrate the consumption cycle in order to solve the problem that even for the same product, the consumption cycle varies depending on the individual's living environment and consumption habits.
[0292] This module follows the following specific sequence to calculate the optimal consumption cycle (Cycle_cohort) per cohort to be transmitted to the joint purchase resumption timing determination unit (130).
[0294] The individual and household unit consumption estimation unit (170) first normalizes the customer-specific purchase history and content consumption data received from the sales performance analysis unit (150), and then sequentially applies the following multiple conditions (conditions D, E, F) to classify customers into cohort groups (C).
[0295] For each classified cohort C, ① average purchase frequency and ② average purchase quantity are extracted based on past purchase data.
[0296] Changes in consumption volume according to season (e.g., summer salad season) from the past sales records of product P are analyzed to determine the seasonality factor weight (S_factor), and this is applied to the current consumption pattern.
[0297] The calibrated consumption cycle (Cycle_cohort) of Cohort C is calculated by comprehensively combining the results with purchase quantity and the purchase frequency and seasonal weights.
[0298] To this end, we first calculate the basic rate of consumption index (R_consumption) for product P of Cohort C. This index comprehensively reflects consumption trends over time, including purchase frequency and seasonality factors.
[0299] It is assumed that the higher the average purchase frequency (F_purchase) for the corresponding product P in Cohort C (the more frequently it is purchased), the faster the consumption rate. The ratio obtained by dividing this F_purchase value by the overall average purchase frequency (F_avg) is reflected in the base consumption rate index to determine the relative consumption rate of the cohort.
[0300] A seasonal factor weight (S_factor) is applied to this ratio in a multiplicative manner. For example, if S_factor is 1.10 (season with a 10% increase in consumption), the base consumption rate index (R_consumption) is adjusted to increase by 1.10 times.
[0301] If R_consumption is greater than 1.0, it means that the product consumption rate of the cohort is faster than the overall average, and if R_consumption is less than 1.0, it means that the consumption rate is slower. As the R_consumption value increases, the Cycle_cohort value to be calculated in the next step becomes shorter.
[0303] Next, the calculated consumption rate index (R_consumption) is integrated with the purchase quantity (capacity) variable to calculate the final consumption cycle (Cycle_cohort).
[0305] To calculate the base consumption cycle, the system's overall average consumption cycle (Cycle_avg) is divided by the R_consumption value. By calculating this, if the consumption rate is fast (R_consumption > 1), it means that Cycle_base is shortened compared to Cycle_avg.
[0306] To adjust the purchase quantity, the ratio (Q_ratio = Q_cohort / Q_avg) obtained by dividing the average purchase quantity of cohort C (Q_cohort) by the overall average purchase quantity (Q_avg) is multiplied by Cycle_base to determine the final consumption cycle (Cycle_cohort). This process applies a capacity-proportional logic where the consumption period becomes longer as the purchase quantity increases (Q_ratio > 1). For example, in the case of a cohort that purchases 3 units at a time despite a fast consumption rate (frequency / seasonality), the notification timing becomes excessively early, so the purchase quantity is used to correct for this to prevent over-early notifications.
[0308] The calculated Cycle_cohort is transmitted to the joint purchase resumption time determination unit (130).
[0310] Here, the aforementioned seasonal factor weight (S_factor) is defined as a dimensionless coefficient indicating how much the demand for a specific period (month / week / quarter) exceeds or falls below the overall average. The S_factor is calculated as the ratio of the actual demand for a period to the overall average demand; to eliminate trends, it can be derived by dividing by a moving average or a regression-based trend value and then averaging the periodic indices.
[0311] As part of the calculation procedure, when determining the monthly index, the raw index is calculated by dividing the actual demand for each month by the annual monthly average demand, and the final S_factor(month) is obtained by averaging the indices for the same month year by year. If necessary, the trend is removed using a central moving average (or regression) before calculating the index to prevent growth or decline trends from being mistaken for seasonality.
[0312] For example, if the annual average demand is 1,000 and the demand in July is 1,300, the July S_factor is 1.30, which means it exceeds the average by 30%. In this invention, multiplicative seasonal correction is adopted by applying 1 / S_factor to the cohort consumption cycle formula; this ensures that the Cycle_cohort is shortened during peak seasons when S_factor > 1 and extended during off-seasons when S_factor < 1. The S_factor can be calculated on a monthly, quarterly, or weekly basis, and the choice depends on the business cycle and data availability. If a daily basis is required, it can be further subdivided by multiplying by a day of the week index (e.g., Monday effect). The S_factor is dynamically updated based on accumulated data, and a concurrent update method reflecting recent observations or a batch update method at the beginning of the year can be selected. Shortening the cycle for reflecting new data improves the accuracy of tracking changes in seasonal patterns. The calculated S_factor is input into the cohort-specific consumption cycle calibration function of the individual and household unit consumption estimation unit (170), and is also applied as a seasonal correction term when calculating D_GO in the group purchase resumption time determination unit (130), thereby adjusting the resumption time to advance during the peak season and retreat during the off-season. For example, if Cycle_avg for olive oil is 60 days and 54 days are derived through frequency and quantity weighting of Cohort C_A, and the S_factor for the corresponding month is 1.10, the final Cycle_cohort is shortened to 54 / 1.10=49.1 days, and the notification time is advanced by about 5 days. Conversely, if S_factor is 0.90, the Cycle_cohort is extended to 54 / 0.90=60 days, preventing an over-early notification.
[0315] The multiple conditions applied to classify the cohort (C) are as follows. These conditions determine the final cohort group through an AND combination method.
[0317] - Condition D (Purchase Quantity Unit): Cases where the average purchase quantity of the product over the past 6 months is 2 or more are classified as a 'Bulk Purchase Cohort (Household Unit).' This separates consumers who make large purchases at once, even if their consumption rate is slow, ensuring that the notification cycle should be set to a long duration.
[0318] - Condition E (Routine Concentration): A case in which the viewing and liking history of a specific routine content (e.g., 'morning fasting routine') among the records of the content provider (110) is 60% or more of the total content viewing history is classified as a 'routine concentration cohort'. This implies that since there is a high likelihood of using the product regularly every day, the consumption cycle should be predicted to be shorter than average.
[0319] - Condition F (Purchase Frequency): Cases where the repurchase interval of the product falls outside the ±15% range of the overall average consumption cycle (Cycle_avg) are classified as 'abnormal consumption cohorts'. Technical Significance: Reduces prediction errors by isolating cohorts whose consumption patterns deviate significantly from the average (e.g., very slow or fast).
[0321] When the calibrated consumption cycle (Cycle_cohort) is calculated, changes in each variable have the following meanings.
[0322] As purchase frequency increases (the more often one buys), the consumption rate of the corresponding cohort is interpreted to be faster, so the Cycle_cohort value decreases (the timing of repurchase notifications becomes earlier).
[0323] As the purchase quantity increases (the more you buy at once), it is interpreted that consumption takes longer with a single purchase, so the Cycle_cohort value increases (the timing of repurchase notifications is delayed).
[0324] As the seasonality factor weight (S_factor) increases beyond 1.0 (e.g., increased olive oil consumption in summer), the consumption rate is interpreted as faster, and the Cycle_cohort value decreases (the time to repurchase becomes earlier).
[0326] For example, the average consumption cycle of olive oil is Cycle_avg 60 days.
[0327] If Cohort C_A (Routine Concentration) has a purchase frequency 1.2 times higher than the average over 3 months, falls within the summer season (S_factor=1.1), and has an average purchase quantity of 1, the consumption rate index (R_consumption) is 1.2 × 1.1 = 1.32, meaning that consumption is 32% faster than the average, and the final consumption cycle is calculated as 60 days / 1.32 × 1.0, which is approximately 45.45 days.
[0328] This cohort sends notifications about 14.5 days earlier (on day 45) to ensure the routine is not interrupted.
[0330] For Cohort C_B (bulk purchase), if the purchase frequency over 3 months is the same as the average (1.0), there is no seasonal influence (S_factor=1.0), and the average purchase quantity is 3, the consumption rate index (R_consumption) is 1.0 × 1.0 = 1.0, which means that the consumption rate is the same as the average, and the final consumption cycle is calculated as 60 days / 1.32 × 3.0, which is approximately 180 days.
[0331] For this cohort, notifications are sent about 120 days late (day 180) to prevent unnecessary notifications.
[0333] Cycle_cohort, recalculated for each cohort, is transmitted to the joint purchase resumption time determination unit (130) and added to D_last of each cohort. As a result, unlike the conventional method that used a single average, the notification is sent 14.5 days earlier to cohort C_A and 60 days later to cohort C_B, thereby dramatically improving the timeliness of the notification and customer satisfaction.
[0335] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. Various modifications are possible by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols
[0336] Group purchase server (100) Content provider (110) Potential Buyer Comment Collection Department (120) Joint purchase resumption time determination unit (130) Group purchase notification department (140) Sales Performance Analysis Department (150)
Claims
Claim 1 The system includes a group buying server (100) for uploading certain content and sending notification messages to potential buyers based on comments of the uploaded content, and the group buying server (100) is for producing and disseminating content regarding 'slow aging' recipes and consumption methods, and includes a content provision unit (110) for uploading content regarding 'slow aging' recipes and consumption methods to a content platform and managing views, reactions, and group buying connection links (URLs) included within the content, a potential buyer comment collection unit (120) for automatically detecting and collecting specific keywords among content comments and identifying author IDs to generate a list of potential buyers, a group buying resumption time determination unit (130) for loading information on the final sales date (D_last), sales quantity (Q_sold), and average user consumption cycle (Cycle_avg) of the previous group buying from the DB of the group buying server (100) and providing a group buying start date (D_GO), and is implemented within the group buying server (100). A group purchase notification unit (140) that receives the group purchase start date (D_GO) confirmed by the group purchase resumption time determination unit (130) and automatically sends personalized notification messages and purchase links to targets based on the list of potential buyers generated by the potential buyer comment collection unit (120); a sales performance analysis unit (150) that collects all data generated throughout the entire process of the system, analyzes the purchase conversion rate by content type, analyzes the repurchase rate and purchase frequency of potential customers, generates the analysis results in the form of a visualized report, and transmits the report results to the content provision unit (110) and the group purchase resumption time determination unit (130) to induce strategy modification; and constructs a heterogeneous graph between users, products, and interaction events including the customer ID obtained from the potential buyer comment collection unit (120), and applies a graph neural network (GNN) model to the heterogeneous graph to calculate the group unit diffusion probability (P_spread) for each user node.A social graph-based group purchase amplification unit (160) that identifies a group invitation initiator to induce serial purchases based on the calculated group unit diffusion probability (P_spread) and transmits it to a group purchase notification unit (140); wherein the social graph-based group purchase amplification unit (160) classifies a group invitation initiator only when all of the following conditions are met: the condition that the group unit diffusion probability (P_spread) exceeds a preset threshold value; the condition that the network centrality index of the user node is included within the top 15% of the entire potential customer candidate group; and the condition that the user has an interaction history of at least 3 times within the last 7 days; wherein the threshold value is set based on the average diffusion rate of past serial purchase groups at the beginning of system operation and is subsequently automatically adjusted through feedback from a sales performance analysis unit (150). Claim 2 In claim 1, the potential buyer comment collection unit (120) classifies the priority of potential customers by utilizing a quantitative indicator called a 'purchase intention score' through an NLP engine and provides this data as analysis input data to the sales performance analysis unit (150), wherein the NLP engine (Natural Language Processing Engine) is a software module that converts unstructured data in text form into structured information that a computer can understand and analyze, and is characterized by separating Korean comments into word (token) units, identifying the part of speech and semantic unit of each word, classifying the overall purpose of the comments into categories, and determining whether the tone of the comments is positive, negative, or neutral. Claim 3 In paragraph 2, the content providing unit (110) inserts a unique traceable link (URL_content_ID) generated for each content into at least one of the description box of the content, a pinned comment, or a caption within the video, and when a user clicks the link and enters, the group purchasing server (100) performs a redirection and collects inflow data including the time of click and the inflow platform path in real time and records it in the DB, characterized by a low-aging recipe content-based health food and related product group purchasing system. Claim 4 A joint purchasing system for health foods and related products based on slow aging recipe content, characterized in that, in paragraph 2, the sales performance analysis unit (150) integrates and analyzes collected content response data and sales data to calculate the purchase conversion rate by content type, and transmits feedback data to the content provision unit (110) to adjust the content production strategy so as to increase the production proportion of content types identified as having a relatively high purchase conversion rate. Claim 5 A joint purchasing system for health foods and related products based on slow-aging recipe content, characterized in that, in paragraph 2, the sales performance analysis unit (150) generates analyzed indicators in the form of a visualized report so that a manager can understand them and provides them to a manager dashboard, and the content produced by the content provision unit (110) is classified into thematic categories including consumption methods, usage recipes, and user challenge reviews. Claim 6 The system includes a group buying server (100) for uploading certain content and sending notification messages to potential buyers based on comments of the uploaded content, wherein the group buying server (100) is intended to produce and distribute content regarding 'slow aging' recipes and consumption methods, and includes a content provision unit (110) for uploading content regarding 'slow aging' recipes and consumption methods to a content platform and managing views, reactions, and group buying connection links (URLs) included within the content, a potential buyer comment collection unit (120) for automatically detecting and collecting specific keywords among video content comments and identifying author IDs to generate a list of potential buyers, a group buying resumption time determination unit (130) for loading information on the final sales date (D_last), sales quantity (Q_sold), and average user consumption cycle (Cycle_avg) of the previous group buying from the DB of the group buying server (100) and providing a group buying start date (D_GO), and is implemented within the group buying server (100). A group purchase notification unit (140) that receives the group purchase start date (D_GO) confirmed by the group purchase resumption time determination unit (130) and automatically sends personalized notification messages and purchase links to targets based on the list of potential buyers generated by the potential buyer comment collection unit (120); a sales performance analysis unit (150) that collects all data generated throughout the entire process of the system, analyzes the purchase conversion rate by content type, analyzes the repurchase rate and purchase frequency of potential customers, generates the analysis results in the form of a visualized report, and transmits the report results to the content provision unit (110) and the group purchase resumption time determination unit (130) to induce strategy modification; and constructs a heterogeneous graph between users, products, and interaction events including the customer ID obtained from the potential buyer comment collection unit (120), and applies a graph neural network (GNN) model to the heterogeneous graph to calculate the group unit diffusion probability (P_spread) for each user node.The system includes a social graph-based group purchase amplification unit (160) that identifies a group invitation initiator to induce chain purchases based on the calculated group unit diffusion probability (P_spread) and transmits it to a group purchase notification unit (140); the potential buyer comment collection unit (120) primarily receives only comments filtered by a keyword filter, identifies the sentence structure centered on core keywords in the comments, extracts the context of 3 to 5 words before and after the keywords to determine whether the comment is a request related to group purchase, determines whether the comment is a positive purchase opportunity or a complaint or negative feedback, and quantifies the final purchase intention of the comment by synthesizing the analyzed results; and the social graph-based group purchase amplification unit (160) identifies the group invitation initiator only when all of the following conditions are met: the condition that the group unit diffusion probability (P_spread) exceeds a preset threshold value, the condition that the network centrality index of the user node is included within the top 15% of the entire potential customer candidate group, and the condition that the user has an interaction history of at least 3 times within the last 7 days. A low-speed aging recipe content-based health food and related product group purchasing system, characterized in that the above standard value is set based on the average diffusion rate of past serial purchasing groups at the beginning of system operation and is subsequently automatically adjusted through feedback from the sales performance analysis unit (150). Claim 7 In claim 6, the joint purchase resumption time determination unit (130) loads the final sales date (D_last) and average consumption cycle (Cycle_avg) for product P from the DB of the joint purchase server (100), adds the average consumption cycle to the date the final sale was completed to calculate the time point (D_pred) at which the buyer is expected to have used up the product using the formula (D_pred = D_last + Cycle_avg), compares the current inventory (S_current) with the expected demand (Q_pred), determines whether new items arrive from the inventory system by the time of D_pred and the total inventory (S_total) exceeds Q_pred, and if the inventory is sufficient or is scheduled to be secured at the time of D_pred, sets D_GO = D_pred, and if the inventory is insufficient at the time of D_pred and is to be secured only after the delivery date (LT), sets [D_GO = D_last + Cycle_avg + delivery delay period], characterized by a slow aging recipe. Content-based group purchasing system for health foods and related products. Claim 8 A system for purchasing health foods and related products based on slow-aging recipe content according to claim 7, wherein the content provider (110) inserts a unique traceable link (URL_content_ID) generated for each content into at least one of the description box of the content, a pinned comment, or a caption within the video, and when a user clicks the link and enters, the joint purchasing server (100) performs a redirection and collects inflow data including the time of click and the inflow platform path in real time and records it in a DB. Claim 9 A joint purchasing system for health foods and related products based on slow aging recipe content, characterized in that, in claim 7, the sales performance analysis unit (150) integrates and analyzes collected content response data and sales data to calculate the purchase conversion rate by content type, and transmits feedback data to the content provision unit (110) to adjust the content production strategy so as to increase the production proportion of content types identified as having a relatively high purchase conversion rate. Claim 10 A joint purchasing system for health foods and related products based on slow-aging recipe content, characterized in that, in claim 7, the sales performance analysis unit (150) generates analyzed indicators in the form of a visualized report so that a manager can understand them and provides them to a manager dashboard, and the content produced by the content provision unit (110) is classified into thematic categories including consumption methods, usage recipes, and user challenge reviews. Claim 11 The system includes a group buying server (100) for uploading certain content and sending notification messages to potential buyers based on comments of the uploaded content, wherein the group buying server (100) is intended to produce and distribute content regarding 'slow aging' recipes and consumption methods, and includes a content provision unit (110) for uploading content regarding 'slow aging' recipes and consumption methods to a content platform and managing views, reactions, and group buying connection links (URLs) included within the content, a potential buyer comment collection unit (120) for automatically detecting and collecting specific keywords among video content comments and identifying author IDs to generate a list of potential buyers, a group buying resumption time determination unit (130) for loading information on the final sales date (D_last), sales quantity (Q_sold), and average user consumption cycle (Cycle_avg) of the previous group buying from the DB of the group buying server (100) and providing a group buying start date (D_GO), and is implemented within the group buying server (100). A group purchase notification unit (140) that receives the group purchase start date (D_GO) confirmed by the group purchase resumption time determination unit (130) and automatically sends personalized notification messages and purchase links to targets based on the list of potential buyers generated by the potential buyer comment collection unit (120); a sales performance analysis unit (150) that collects all data generated throughout the entire process of the system, analyzes the purchase conversion rate by content type, analyzes the repurchase rate and purchase frequency of potential customers, generates the analysis results in the form of a visualized report, and transmits the report results to the content provision unit (110) and the group purchase resumption time determination unit (130) to induce strategy modification; and constructs a heterogeneous graph between users, products, and interaction events including the customer ID obtained from the potential buyer comment collection unit (120), and applies a graph neural network (GNN) model to the heterogeneous graph to calculate the group unit diffusion probability (P_spread) for each user node.The system includes a social graph-based group purchase amplification unit (160) that identifies a group invitation initiator to induce chain purchases based on the calculated group unit diffusion probability (P_spread) and transmits it to a group purchase notification unit (140); the potential buyer comment collection unit (120) primarily receives only comments filtered by a keyword filter, identifies the sentence structure centered on core keywords in the comments, extracts the context of 3 to 5 words before and after the keywords to determine whether the comment is a request related to group purchase, determines whether the comment is a positive purchase opportunity or a complaint or negative feedback, and quantifies the final purchase intention of the comment by synthesizing the analyzed results; and the social graph-based group purchase amplification unit (160) identifies the group invitation initiator only when all of the following conditions are met: the condition that the group unit diffusion probability (P_spread) exceeds a preset threshold value, the condition that the network centrality index of the user node is included within the top 15% of the entire potential customer candidate group, and the condition that the user has an interaction history of at least 3 times within the last 7 days. A method for joint purchasing health food and related products based on slow aging recipe content using a system for joint purchasing health food and related products based on slow aging recipe content, characterized in that the above standard value is set based on the average diffusion rate of past serial purchase groups at the beginning of system operation and is subsequently automatically adjusted through feedback from the sales performance analysis unit (150), wherein the content provision unit (110) of the joint purchasing server (100) produces 'slow aging recipe' content and distributes it to an external content platform by inserting a traceable link (URL_tracking) to generate content response data, and the potential buyer comment collection unit (120) of the joint purchasing server (100) periodically collects comment data from the content platform, andA method for joint purchasing health food and related products based on slow aging recipe content, comprising a third step of determining whether the joint purchasing server (100) has reached the optimal joint purchasing resumption date (D_GO) determined by the joint purchasing resumption time determination unit (130). Claim 12 A method for joint purchasing health food and related products based on slow-aging recipe content according to claim 11, comprising: a fourth step in which the joint purchasing notification unit (140) of the joint purchasing server (100) automatically sends a personalized notification message and a purchase tracking link to a list of potential buyers on the confirmed optimal joint purchasing resumption date (D_GO); a fifth step in which the joint purchasing server (100) collects data including sales records, buyer IDs, and repurchase status during the joint purchasing period from a sales / payment system and stores it in a DB; and a sixth step in which the sales performance analysis unit (150) of the joint purchasing server (100) analyzes the purchase conversion rate (CR), repurchase rate, and notification effect by content type using the collected data; wherein, in the second step, the potential buyer comment collection unit (120) applies an NLP engine to the collected comments to analyze purchase intent keywords and positive sentiment scores, and registers user IDs whose scores are above a threshold in the list of potential buyers DB. Claim 13 The system includes a group buying server (100) for uploading certain content and sending notification messages to potential buyers based on comments of the uploaded content, wherein the group buying server (100) is intended to produce and distribute content regarding 'slow aging' recipes and consumption methods, and includes a content provision unit (110) for uploading content regarding 'slow aging' recipes and consumption methods to a content platform and managing views, reactions, and group buying connection links (URLs) included within the content, a potential buyer comment collection unit (120) for automatically detecting and collecting specific keywords among video content comments and identifying author IDs to generate a list of potential buyers, a group buying resumption time determination unit (130) for loading information on the final sales date (D_last), sales quantity (Q_sold), and average user consumption cycle (Cycle_avg) of the previous group buying from the DB of the group buying server (100) and providing a group buying start date (D_GO), and is implemented within the group buying server (100). A group purchase notification unit (140) that receives the group purchase start date (D_GO) confirmed by the group purchase resumption time determination unit (130) and automatically sends personalized notification messages and purchase links to targets based on the list of potential buyers generated by the potential buyer comment collection unit (120); a sales performance analysis unit (150) that collects all data generated throughout the entire process of the system, analyzes the purchase conversion rate by content type, analyzes the repurchase rate and purchase frequency of potential customers, generates the analysis results in the form of a visualized report, and transmits the report results to the content provision unit (110) and the group purchase resumption time determination unit (130) to induce strategy modification; and constructs a heterogeneous graph between users, products, and interaction events including the customer ID obtained from the potential buyer comment collection unit (120), and applies a graph neural network (GNN) model to the heterogeneous graph to calculate the group unit diffusion probability (P_spread) for each user node.A method for joint purchasing health foods and related products based on slow-aging recipe content using a slow-aging recipe content-based health food and related product joint purchasing system, wherein the social graph-based group purchasing amplification unit (160) includes a group invitation initiator to induce chain purchasing based on the calculated group unit diffusion probability (P_spread) and transmits it to a joint purchasing notification unit (140); wherein the social graph-based group purchasing amplification unit (160) classifies the group invitation initiator only when all of the following conditions are satisfied: the condition that the group unit diffusion probability (P_spread) exceeds a preset threshold value; the condition that the network centrality index of the user node is included within the top 15% of the entire potential customer candidate group; and the condition that the user has an interaction history of at least 3 times within the last 7 days; and wherein the threshold value is set based on the average diffusion rate of past chain purchasing groups at the beginning of system operation and is subsequently automatically adjusted through feedback from a sales performance analysis unit (150). In this method, the content provision unit (110) of the joint purchasing server (100) produces 'slow-aging recipe' content, and traceable A method for joint purchasing health food and related products based on slow-aging recipe content, characterized by comprising: a first step of generating content response data by distributing to an external content platform by inserting a link (URL_tracking); a second step in which a potential buyer comment collection unit (120) of the joint purchasing server (100) periodically collects comment data from the content platform; a third step in which the joint purchasing server (100) determines whether the optimal joint purchasing resumption date (D_GO) determined by the joint purchasing resumption time determination unit (130) has been reached; and a fourth step in which a joint purchasing notification unit (140) of the joint purchasing server (100) automatically sends a personalized notification message and a purchase tracking link to a list of potential buyers on the determined optimal joint purchasing resumption date (D_GO). Claim 14 In claim 13, the group buying server (100) collects data including sales records, buyer IDs, and repurchase status during the group buying period from a sales / payment system and stores it in a DB in step 5, and the sales performance analysis unit (150) of the group buying server (100) analyzes the purchase conversion rate (CR), repurchase rate, and notification effect by content type using the collected data in step 6, and the group buying resumption time determination unit (130) of step 3 loads the final sales date (D_last) and average consumption cycle (Cycle_avg) for product P from the DB of the group buying server (100), adds the average consumption cycle to the date the final sale was completed to primarily calculate the time point (D_pred) at which the buyer is expected to have finished using the product through the formula (D_pred = D_last + Cycle_avg), compares the current inventory (S_current) with the expected demand (Q_pred), and new items from the inventory system up to the D_pred time A method for joint purchasing of health foods and related products based on slow aging recipe content, characterized by determining whether the total inventory (S_total) exceeds Q_pred upon arrival, and if the inventory is sufficient or scheduled to be secured at time D_pred, setting D_GO = D_pred, and if the inventory is insufficient at time D_pred and is to be secured only after the delivery date (LT), setting [D_GO = D_last + Cycle_avg + delivery delay period].
Citation Information
Patent Citations
Marketing topic key user identification method and device and computer equipment
CN118069936A
System for mediating group purchase using moving image ads
KR101286730B1
Recipe contents providing system and method
KR1020230142865A
Method, device and computer-readable recording medium for analyzing store reviews through automatically generated marketing content
KR1020240178819A
Method, device and system for automating order processing and inventory management for food distribution
KR102630807B1
