Method and apparatus for predicting sales volume on basis of multimodal data preprocessing

Abstract

A method and an apparatus for predicting a sales volume on the basis of multimodal data preprocessing are provided. The apparatus for predicting a sales volume on the basis of multimodal data preprocessing, according to an embodiment, comprises: a data collection unit for collecting first data related to a social media posting of a target influencer, second data related to a product sales history through a market of the target influencer and a sales history through a market of a target product, and third data related to activities of a plurality of purchasers in the market of the target influencer; a first data preprocessing unit for preprocessing the first data; a second data preprocessing unit for preprocessing the second data; a third data preprocessing unit for preprocessing the third data; and a prediction model for outputting a sales volume of the target product for the target influencer in correspondence to an input dataset configured on the basis of a variable extracted by the first data preprocessing unit, a variable extracted by the second data preprocessing unit, and a variable extracted by the third data preprocessing unit.

Description

Method and device for predicting sales volume based on multimodal data preprocessing

[0001] The present invention relates to a method and device for predicting sales volume based on multimodal data preprocessing. Specifically, the present invention relates to a method and device for predicting sales volume based on multimodal data preprocessing, which collects various data related to an influencer, analyzes and processes the collected data, and predicts the influencer's product sales volume.

[0002] The content described in this section merely provides background information for the present embodiment and does not constitute prior art.

[0003] This research was supported by the Ministry of SMEs and Startups, Small and Medium Business Technology Innovation Development (SME Accounting) Project [Small Business Owner-Social Seller Matching and Sales Platform, Cami Partner Center, Project Identification Number: 1425176712, Subproject Number: 00255887].

[0004] Influencer marketing is a marketing strategy that leverages influential social media influencers to promote products or services. Influencers often have large followings based on their expertise or interests in a specific field. Their content can inform consumers about products or services, influencing their purchasing decisions. Companies that sell or manufacture products can also effectively leverage influencer marketing to increase brand awareness, boost conversion rates, and strengthen customer loyalty.

[0005] Traditionally, influencer marketing involved companies selecting specific influencers and commissioning them to write advertisements or reviews. However, recent trends have seen influencers establish their own brands and sell products through group purchases, or collaborate with companies to develop new brands and products and then sell them through group purchases. Influencers' influence is maintained and expanded based on their followers (consumers). Consumers are more influenced by the expertise and relevance of the information provided by influencers. Consumers tend to trust, accept, and rely on information provided by influencers when they perceive them to be highly professional and relevant.

[0006] Therefore, in recent influencer marketing, selecting products that align with the influencer's expertise and background has become the most crucial factor for successful influencer marketing and group buying. Furthermore, when influencers directly promote group buying, they are responsible for managing inventory, requiring them to predict sales volumes and manage product production and inventory.

[0007] Here, predicting influencer sales requires comprehensive analysis of various influencer-related data from various perspectives. For example, the influencer's past product sales, their posting data, and activity data from other users related to the influencer can serve as basic data that can influence the influencer's sales for a specific product. Thus, there is a need for methods and devices that can efficiently analyze multimodal data collected from various types, formats, and servers and utilize them for sales prediction.

[0008] The purpose of the present invention is to provide a method and device capable of predicting the sales volume of a target product of a target influencer by considering the influencer's sales history, the activity history of buyers related to the influencer, and the influencer's posting information.

[0009] In addition, an object of the present invention is to provide a sales volume prediction method and device that generates a plurality of variables judged in a prediction model through preprocessing of multimodal data in various ways, and includes a prediction model that uses these plurality of variables as input values.

[0010] In addition, an object of the present invention is to provide a sales volume prediction method and device that support more accurate and faster prediction by applying appropriate weights to multiple variables extracted from multimodal data.

[0011] The purposes of the present invention are not limited to those mentioned above. Other purposes and advantages of the present invention not mentioned above can be understood through the following description and will be more clearly understood through the embodiments of the present invention. Furthermore, it will be readily apparent that the purposes and advantages of the present invention can be realized by the means and combinations thereof set forth in the claims.

[0012] A device for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention is a server for predicting sales volume for a target product of a target influencer, the server comprising: a data collection unit for collecting first data related to SNS posting of the target influencer, second data related to product sales history through a market of the target influencer and sales history of the target product through a market, and third data related to activities of a plurality of buyers for the market of the target influencer; a first data preprocessing unit for performing preprocessing on the first data; a second data preprocessing unit for performing preprocessing on the second data; a third data preprocessing unit for performing preprocessing on the third data; A prediction model that outputs the sales volume of the target product of the target influencer in response to an input data set configured based on the variables extracted from the first data preprocessing unit, the variables extracted from the second data preprocessing unit, and the variables extracted from the third data preprocessing unit, wherein the first data includes posting content, comment content, and media data, and the first data preprocessing unit analyzes keywords of the posting content to determine the purpose of the posting and generates a first variable, analyzes the frequency of use of the keywords to determine keyword concentration and generates a second variable, analyzes the posting content and the comment content to determine the sentiment of the target influencer and generates a third variable, determines whether the target influencer has been exposed and the frequency of exposure from the media data to generate a fourth variable, extracts main color information of the media data, and determines a standard deviation of the extracted main color information to generate a fifth variable.

[0013] In addition, the first data preprocessing unit can generate a sixth variable based on quantitative data of comments in the comment content, and can generate a seventh variable based on quantitative data of comments exchanged between the target influencer and other influencers in the comment content.

[0014] In addition, the first data preprocessing unit may include text mining that extracts keywords from the content of the posting and creates a keyword cloud by extracting only nouns from the extracted keywords; a first keyword analysis model that determines the commercial purpose of the posting based on the degree to which keywords included in the keyword cloud are confirmed in a commercial keyword dictionary; and a second keyword analysis model that selects keywords with a high frequency of use among keywords included in the keyword cloud and determines the keyword concentration by determining the similarity between the selected keywords.

[0015] In addition, the second data preprocessing unit may analyze the second data to generate an eighth variable indicating sales history information of a market recently conducted by the target influencer; a ninth variable indicating sales history information according to the temporal characteristics of a market to be conducted by the target influencer; a tenth variable indicating a preference characteristic of a buyer for the target product; and an eleventh variable indicating the activity of the target influencer.

[0016] In addition, the third data preprocessing unit may analyze the third data to generate a twelfth variable representing activity information prior to product purchase of the plurality of purchasers; and a thirteenth variable representing activity information related to product purchase of the plurality of purchasers.

[0017] In addition, the learning unit may further include a learning unit that learns the prediction model, and the learning unit may include a variable importance calculation unit that determines variable importance of variables constituting the input data set; and a learning execution unit that learns the prediction model.

[0018] In addition, the prediction model can configure the input data set by applying variable weights according to the variable importance to the variables extracted from the first data preprocessing unit, the variables extracted from the second data preprocessing unit, and the variables extracted from the third data preprocessing unit.

[0019] In addition, the prediction model is an ensemble model based on a Gradient Boosting Decision Tree (GBDT), and includes first to Nth prediction models, and each of the first to Nth prediction models is a decision tree model based on a regression tree, and classifies the input data set according to a constructed branch and outputs a corresponding sales amount, and a sales amount for the target product of the target influencer can be calculated by applying a preset correction coefficient to the sales amount output from the first to Nth prediction models.

[0020] According to some embodiments of the present invention, a method for predicting sales volume of a target product of a target influencer, which is performed on a server, comprises the steps of: collecting first data related to SNS postings of the target influencer, second data related to product sales history through a market of the target influencer and sales history of the target product through the market, and third data related to activities of a plurality of buyers for the market of the target influencer; performing preprocessing on the first data, preprocessing on the second data, and preprocessing on the third data; configuring an input dataset based on variables extracted through preprocessing on the first data, variables extracted through preprocessing on the second data, and variables extracted through preprocessing on the third data; A method for preprocessing the input data set, comprising: applying the input data set to a prediction module, and outputting the sales volume of the target product of the target influencer as an output thereof; wherein the first data includes posting content, comment content, and media data; and preprocessing the first data includes: analyzing keywords of the posting content to determine the purpose of the posting and thereby generate a first variable; analyzing the frequency of use of the keywords to determine keyword concentration and thereby generate a second variable; analyzing the posting content and the comment content to determine the sentiment of the target influencer and thereby generate a third variable; determining whether the target influencer has been exposed and the frequency of exposure in the media data to generate a fourth variable; extracting main color information of the media data and determining a standard deviation of the extracted main color information to generate a fifth variable.

[0021] A computer program according to some embodiments of the present invention is stored in a medium to execute the multimodal data preprocessing-based sales volume prediction method in combination with hardware.

[0022] A method and device for predicting sales volume based on multimodal data preprocessing, according to some embodiments of the present invention, can predict sales volume for a target product by a target influencer based on the influencer's sales history, the activity history of buyers associated with the influencer, and the influencer's posting information. The predicted sales volume can then be provided to the target influencer or a brand associated with the target product. Accordingly, the target influencer or brand can determine whether to proceed with a market based on the predicted sales volume, determine inventory levels, and facilitate more successful market operations.

[0023] In addition, a method and device for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention extracts multiple variables judged in a prediction model through preprocessing of multimodal data in various ways, and predicts the sales volume of a target product through a prediction model that uses multiple variables as input values, and can determine the sales volume by considering various variables and factors.

[0024] In addition, a method and device for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention can determine the importance of a plurality of variables extracted from multimodal data and apply appropriate weights to the variables based on the determined importance, thereby providing more accurate and faster prediction results.

[0025] In addition to the above-described contents, the specific effects of the present invention are described together with the specific matters for carrying out the invention below.

[0026] FIG. 1 is an exemplary diagram illustrating a brokerage system for matching influencers and brands according to some embodiments of the present invention.

[0027] FIG. 2 illustrates an exemplary seller search interface according to some embodiments of the present invention.

[0028] FIG. 3 illustrates an example of a product search interface according to some embodiments of the present invention.

[0029] FIG. 4 illustrates an example of a market management interface for managing multiple markets run on the seller side according to some embodiments of the present invention.

[0030] FIG. 5 is a block diagram illustrating the main configuration of a server according to some embodiments of the present invention.

[0031] FIG. 6 is an exemplary diagram illustrating a process of collecting first data from multiple SNS servers according to some embodiments of the present invention.

[0032] FIG. 7 is an exemplary diagram illustrating the operation of a first data preprocessing unit according to some embodiments of the present invention.

[0033] FIG. 8 is an exemplary diagram illustrating detailed operations of a first data preprocessing unit according to some embodiments of the present invention.

[0034] FIG. 9 is an exemplary diagram illustrating a process of collecting second data from multiple SNS servers according to some embodiments of the present invention.

[0035] FIG. 10 is an exemplary diagram illustrating the operation of a second data preprocessing unit according to some embodiments of the present invention.

[0036] FIG. 11 is an exemplary diagram illustrating a process of collecting third data from multiple SNS servers according to some embodiments of the present invention.

[0037] FIG. 12 is an exemplary diagram illustrating the operation of a third data preprocessing unit according to some embodiments of the present invention.

[0038] FIG. 13 is an exemplary diagram illustrating a process of learning a prediction model according to some embodiments of the present invention.

[0039] FIG. 14 is an exemplary diagram illustrating a process for predicting sales volume for a target product of a target influencer according to some embodiments of the present invention.

[0040] FIG. 15 is a flowchart of a sales volume prediction method based on multimodal data preprocessing according to some embodiments of the present invention.

[0041] FIG. 16 is a diagram illustrating a hardware implementation of a device for performing a sales volume prediction method according to some embodiments of the present invention.

[0042] The terms and words used in this specification and claims should not be interpreted based on their general or dictionary meanings. In accordance with the principle that inventors can define the concepts of terms and words to best describe their inventions, they should be interpreted in a way that is consistent with the technical concept of the present invention. Furthermore, the embodiments described in this specification and the configurations depicted in the drawings are merely examples of how the present invention can be realized and do not fully represent the technical concept of the present invention. Therefore, it should be understood that various equivalents, modifications, and applicable examples may exist as of the time of filing.

[0043] The terms first, second, A, B, etc. used in this specification and claims may be used to describe various components, but the components should not be limited by these terms. These terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. The term "and / or" includes any combination of a plurality of related listed items or any item among a plurality of related listed items.

[0044] The terminology used in this specification and claims is for the purpose of describing specific embodiments only and is not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. It should be understood that terms such as "comprise" or "have" in this application do not preclude the presence or addition of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification.

[0045] In addition, each configuration, process, procedure or method included in each embodiment of the present invention may be shared within a scope that is not technically inconsistent with each other.

[0046] Additionally, unless otherwise defined herein, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention pertains.

[0047] Terms defined in commonly used dictionaries should be interpreted to have a meaning consistent with their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless expressly defined in this application.

[0048] Hereinafter, with reference to FIGS. 1 to 16, a method and device for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention will be described.

[0049] Figure 1 is an exemplary diagram illustrating a brokerage system that matches influencers and brands according to some embodiments of the present invention. Figure 2 illustrates a seller search interface according to some embodiments of the present invention. Figure 3 illustrates a product search interface according to some embodiments of the present invention. Figure 4 illustrates a market management interface for managing multiple markets run by sellers according to some embodiments of the present invention.

[0050] A brokerage system (10) according to some embodiments of the present invention can support brokering and matching between influencers and brands.

[0051] Referring to FIG. 1, a brokerage system (10) according to an embodiment includes a server (100), a plurality of influencer terminals (200), a plurality of brand terminals (300), a plurality of SNS servers (400), and a market server (500).

[0052] In the present invention, an influencer refers to a user who has a large number of followers on a social network service (hereinafter, "SNS"), including Facebook, Twitter, Instagram, YouTube, and internet blogs, and exerts influence on other users. In the present invention, an influencer may be an entity that sells a brand's products on behalf of the brand or sells products in collaboration with the brand, and may also be referred to as a "seller."

[0053] In the present invention, the "market" refers to the overall process by which matched brands and influencers sell products. For example, the "market status" refers to the status of product orders, cancellation management, return management, exchange management, and settlement for product sales.

[0054] Additionally, in the present invention, a brand may refer to an entity (company or individual) that develops, produces, and sells a product. Brands may need to connect with influencers to promote and market their products. Influencers may also need to be matched with brands that possess suitable products for group purchases, auctions, and other initiatives.

[0055] The server (100) can provide a brokerage service that matches multiple influencers with multiple brands and supports product sales and joint purchases through influencer marketing between the matched influencers and brands. Furthermore, the server (100) can easily monitor the progress of established markets between matched brands and influencers, and can provide influencers and brands with a user environment that allows for overall market management. Furthermore, the server (100) can recommend products from a brand to sell to influencers and provide estimated sales volumes for the recommended products, thereby supporting influencers in selecting products to sell and brands in managing product inventory.

[0056] The server (100) may be configured to exchange data with the server (100), multiple influencer terminals (200), multiple brand terminals (300), multiple SNS servers (400), and multiple market servers (500) via a network. Here, the network may include a network based on wired Internet technology, wireless Internet technology, and short-range communication technology. For example, the wired Internet technology may include at least one of a local area network (LAN) and a wide area network (WAN).

[0057] Wireless Internet technologies may include, for example, at least one of Wireless LAN (WLAN), Digital Living Network Alliance (DLNA), Wireless Broadband (Wibro), World Interoperability for Microwave Access (Wimax), High Speed ​​Downlink Packet Access (HSDPA), High Speed ​​Uplink Packet Access (HSUPA), IEEE 802.16, Long Term Evolution (LTE), Long Term Evolution-Advanced (LTE-A), Wireless Mobile Broadband Service (WMBS), and 5G NR (New Radio) technologies. However, the present embodiment is not limited thereto.

[0058] Short-range communication technologies may include, for example, at least one of Bluetooth, Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra-Wideband (UWB), ZigBee, Near Field Communication (NFC), Ultra Sound Communication (USC), Visible Light Communication (VLC), Wi-Fi, Wi-Fi Direct, and 5G NR (New Radio). However, the present embodiment is not limited thereto.

[0059] The server (100) can provide a user environment that can use a brokerage service that matches multiple influencers and multiple brands to multiple influencer terminals (200) and multiple brand terminals (300).

[0060] The server (100) can provide a service environment to users through a service application installed on each of the influencer terminal (200) and the brand terminal (300). Here, the service application may be a dedicated application for providing a service environment or a web browsing application for providing the service environment through a web page.

[0061] The influencer terminal (200) and the brand terminal (300) may be electronic devices capable of providing a user experience. For example, the influencer terminal (200) and the brand terminal (300) may be a smartphone, a type of portable terminal, but the present invention is not limited thereto. The influencer terminal (200) and the brand terminal (300) may include various types of electronic devices, such as a personal computer (PC), a laptop, a tablet, a mobile phone, a smartphone, or a wearable device (e.g., a watch-type terminal).

[0062] The plurality of SNS servers (400) may be servers that provide social network services (hereinafter, “SNS”), including Facebook, Twitter, Instagram, YouTube, and Internet blogs. The plurality of influencers may be registered in the SNS provided by the plurality of SNS servers (400), and may post various information on these SNSs to deliver and provide various information to followers. In addition, the sales and advertising activities of the market performed by the influencers may be

[0063] The server (100) can collect information from multiple SNS servers (400) and provide a matching service that matches influencers with brands based on the collected information. The server (100) can provide an environment where users of the brokerage service can check information about multiple influencers based on the information collected from multiple SNS servers (400).

[0064] The market server (500) can provide a market service where products are sold and purchased online. Here, the market server (500) may be an e-commerce site affiliated with the server (100) and may provide a market service linked to a market matched on the server (100). For example, the operating entity of the market server (500) and the operating entity of the server (100) may be the same entity. Markets and joint purchases concluded between influencers and brands matched on the server (100) can be sold through the market server (500). The server (100) and the market server (500) can operate in conjunction with each other, and the server (100) can easily collect necessary information from the market server (500).

[0065] However, the embodiments of the present invention are not limited thereto, and the market service of the market server (500) may be an open market corresponding to an e-commerce site open to both sellers and buyers, and the operating entities of the market server (500) and the operating entities of the server (100) may be different entities. Brands or influencers may freely participate in the market service in the form of such an open market, and may sell manufactured products, products purchased through joint purchases, or products for which a market agreement has been concluded, through the market service of the market server (500).

[0066] The server (100) can collect product-related information from the market server (500). The server (100) can collect sales history data related to products sold on the market server (500). In addition, the server (100) can collect activity data of general consumers and buyers using the market server (500). In addition, if the market server (500) is a general open market service, the server (100) can collect price information related to specific products from the market server (500). The server (100) can provide users of the brokerage service with a user environment that provides price information for products corresponding to products registered by a brand. The user environment can assist influencers in searching for products they wish to sell.

[0067] A brand can access the user environment provided by the server (100) through a brand terminal (300), and search for and select influencers that match the brand's image, product type, marketing direction, and purpose.

[0068] Figure 2 exemplarily illustrates a state in which Brand A utilizes a user interface to search for influencers (sellers). Referring to Figure 2, a seller search interface (I1) may be provided to the brand, allowing the brand to check the status of previous market progress, follower counts, etc., of multiple influencers. If the brand selects a specific influencer from the seller search interface (I1), a detailed information interface (I2) may be provided, providing detailed information about the selected influencer. Additional information may be provided, such as the influencer's main field, market progress with the influencer, and expected sales.

[0069] Influencers can access the user environment provided by the server (100) through the influencer terminal (200) and search for and select products of brands that are suitable for their expertise, competitiveness, and suitability.

[0070] Figure 3 exemplarily illustrates a state in which influencer A (seller A) utilizes a user interface to search for products to sell. Referring to Figure 3, the influencer can check information on multiple products registered by multiple brands through a product search interface (I3). The product search interface (I3) can provide price information (seller supply price, group purchase price) registered by the brand for each registered product, along with the price at which the product is sold on other open markets. Specifically, the seller supply price is the price the brand provides to the seller, and the group purchase price is the price at which the product is sold during the market process. The influencer can select a specific product and check detailed information through a detailed information interface (I4). Specifically, the detailed information interface (I4) can provide information such as the lowest online price, group purchase price, seller margin, the brand's group purchase history, and sample delivery conditions.

[0071] Influencers and brands can request market progress from their desired counterparts through the aforementioned search interfaces (I3, I1), and if mutual conditions are met, the market progress will be completed.

[0072] The server (100) can provide a market management interface (I5) as illustrated in FIG. 4. The market management interface (I5) can provide the progress of product sales between influencers and brands that have completed a market, by influencer or brand. FIG. 4 shows a state in which the ongoing market status is confirmed based on Brand A. The progress and related events of each influencer who has completed a market with Brand A can be confirmed through the market management interface (I5).

[0073]

[0074] According to an embodiment of the present invention, a server (100) may collect SNS posting data (hereinafter, “first data”) related to SNS activities of a plurality of influencers from a plurality of SNS servers (400). In addition, the server (100) may manage history data of products sold by influencers or brands matched through a brokerage service through a market server (500). Through data exchange with the market server (500), the server (100) may support sales management for products sold by influencers or brands and collect sales history data (hereinafter, “second data”). Here, the second data may include products sold, sales volume, sales date, sales price, etc. In addition, the server (100) may collect activity data (hereinafter, “third data”) of a plurality of buyers using the market server (500).

[0075] In some embodiments, an influencer may need information on the expected sales volume of a specific product when marketing it. Furthermore, a brand may need information on the expected sales volume of a specific influencer selling its product. The server (100) can utilize the first, second, and third data to predict the sales volume of a target influencer's target product.

[0076] Here, the target influencer may refer to an influencer whose product sales are being predicted. Furthermore, the target product may refer to a product whose sales are being predicted, and may be a product of a specific brand. Here, the first, second, and third data may be collected from different servers or may be multimodal data collected through different formats and methods.

[0077] The server (100) performs appropriate preprocessing on multimodal data, extracts multiple variables from the preprocessed data, and applies appropriate weights to the variables to predict sales volume for a product. The sales volume prediction may be key information for influencers to reference before selecting a specific product. Influencers can consider this information to select a product that is more suitable and appropriate for them. In some examples, the server (100) may provide sales volume for a target product of a target influencer through the detailed information interface (I2) or the detailed information interface (I4), thereby supporting influencers and brands in developing appropriate market plans and managing product inventory.

[0078]

[0079] Hereinafter, with reference to FIGS. 5 to 14, the main configuration of the server (100) and the sales volume prediction process based on multimodal data preprocessing performed in the server (100) will be described in more detail.

[0080] FIG. 5 is a block diagram illustrating the main configuration of a server according to some embodiments of the present invention. FIG. 6 is an exemplary diagram illustrating a process of collecting first data from multiple SNS servers according to some embodiments of the present invention. FIG. 7 is an exemplary diagram illustrating the operation of a first data preprocessing unit according to some embodiments of the present invention. FIG. 8 is an exemplary diagram illustrating a detailed operation of a first data preprocessing unit according to some embodiments of the present invention. FIG. 9 is an exemplary diagram illustrating a process of collecting second data from multiple SNS servers according to some embodiments of the present invention. FIG. 10 is an exemplary diagram illustrating the operation of a second data preprocessing unit according to some embodiments of the present invention. FIG. 11 is an exemplary diagram illustrating a process of collecting third data from multiple SNS servers according to some embodiments of the present invention. FIG. 12 is an exemplary diagram illustrating the operation of a third data preprocessing unit according to some embodiments of the present invention. FIG. 13 is an exemplary diagram illustrating a process of learning a predictive model according to some embodiments of the present invention. FIG. 14 is an exemplary diagram illustrating a process of predicting sales volume for a target product of a target influencer according to some embodiments of the present invention.

[0081]

[0082] Referring to FIG. 5, a server (100) according to some embodiments of the present invention includes a data collection unit (110), a first data preprocessing unit (120), a second data preprocessing unit (130), a third data preprocessing unit (140), a data storage unit (150), a learning unit (160), and a prediction model (170).

[0083] The server (100) according to the embodiments may be entirely hardware, or may have aspects that are partially hardware and partially software. For example, the server (100) and each component included therein in the present specification are intended to refer to a combination of hardware and software driven by the hardware. The hardware may be a data processing device including a Central Processing Unit (CPU) or other processor. Furthermore, the software driven by the hardware may refer to a running process, object, executable, thread of execution, program, etc.

[0084] In addition, although the data collection unit (110), the first data preprocessing unit (120), the second data preprocessing unit (130), the third data preprocessing unit (140), the data storage unit (150), the learning unit (160), and the prediction model (170) in FIG. 5 are illustrated as distinct blocks, each unit is not necessarily intended to refer to a physically distinct, separate component. Each unit constituting the system or terminal is merely functionally distinguished by the operations performed by the corresponding device, and some or all of them may be integrated into the same device, one or more may be implemented as a separate device that is physically distinct from other units, and may be components that are communicatively connected to each other under a distributed computing environment.

[0085] Referring to FIG. 6, the data collection unit (110) can collect first data from multiple SNS servers (400). The data collection unit (110) can crawl first data corresponding to each of multiple influencers from multiple SNS servers (400) via a network.

[0086] The data collection unit (110) may collect first data by crawling posts uploaded to the SNS accounts of multiple influencers at regular intervals, but the embodiments of the present invention are not limited thereto. In some embodiments, the data collection unit (110) may collect first data by crawling posts uploaded to the SNS accounts of multiple influencers in response to a command from an administrator of the server (100). If an influencer uploads multiple SNS postings during a regular interval, the data collection unit (110) may crawl posts corresponding to each SNS posting to collect first data.

[0087] The first data collected by the data collection unit (110) may include at least the posting date, posting content, comment content, and media data. Here, the posting date refers to the date the posting was uploaded to the SNS account.

[0088] Post content can be data crawled from the title and body of posts uploaded by influencers. Additionally, post content can include hashtag information related to the post.

[0089] Comment content may be information related to reactions from at least one of a follower, the influencer, and other influencers in response to a post. In some embodiments, the comment content may include positive response information (e.g., "likes"), negative response information (e.g., "dislikes"), and reply information for the post. The reply information may be data crawled from comments written by at least one of a follower, the influencer, and other influencers in response to the post.

[0090] Media data may include images and videos included in a post. Images and videos of the media data may be filmed to include the influencer, but are not limited to this. Media data may also be filmed and posted without exposing the influencer. Furthermore, media data may be composed using a variety of colors, but is not limited to this. Media data may also be composed using several colors to emphasize specific points. In other words, the frequency of use, composition method, and form of media data use may vary depending on the purpose of the post, the influencer's preferences, etc.

[0091] The data collection unit (110) may provide the collected first data to the first data preprocessing unit (120). The first data preprocessing unit (120) may perform preprocessing on the first data and store the preprocessed first data for each corresponding influencer account in the data storage unit (150). Here, the preprocessed first data may refer to a state in which each variable is derived through the process described below, and the first data of an influencer collected for a specific period may be preprocessed and stored together with the collected period information for the corresponding influencer account.

[0092] Referring to FIG. 7, the first data preprocessing unit (120) may perform an analysis on the first data to extract first to seventh variables. The first variable may be a variable related to the posting purpose of the first data. The second variable may be a variable related to the keyword concentration of an influencer corresponding to the first data. The third variable may be a variable related to the emotion determined from the first data. The fourth variable may be a variable related to whether or not the influencer corresponding to the first data has been exposed or the degree of exposure. The fifth variable may be a variable related to the media color of the first data. The sixth variable may be a variable related to the result of a quantitative analysis of comments. The seventh variable may be a variable related to the result of an analysis of the relationship between the influencer who wrote the comment.

[0093] The first data preprocessing unit (120) includes a text mining (120A), a first keyword analysis model (120B), a commercial keyword dictionary (120C), a second keyword analysis model (120D), a sentiment analysis model (120E), a facial recognition model (120F), a color recognition model (120G), a quantitative analysis model (120H), and a relationship analysis model (120I).

[0094] Referring to (a) of FIG. 8, text mining (120A) can analyze the posting content of the first data to extract main keywords. Specifically, text mining (120A) can divide the posting content into morpheme units of nouns, verbs, and adjectives. For example, text mining (120A) can divide the posting content of the first data into morpheme units using the Okt class of KoNLPy, but the embodiment of the present invention is not limited thereto. Text mining (120A) can extract only nouns from the divided morpheme units to generate a keyword cloud. The keyword cloud generated by text mining (120A) can be provided to the first keyword analysis model (120B).

[0095] The first keyword analysis model (120B) can determine the purpose of a posting by analyzing keywords included in the keyword cloud. Posts uploaded by influencers can be categorized into commercial posts intended to sell products and non-commercial posts sharing their daily lives. The first keyword analysis model (120B) can determine the purpose of keywords included in the keyword cloud through a comparative analysis and filtering process with multiple keywords included in the keyword cloud and a commercial keyword dictionary (120C) or a non-commercial keyword dictionary (not shown), thereby determining the purpose of posting the first data.

[0096] In some embodiments, the first keyword analysis model (120B) may determine the purpose of keywords included in the keyword cloud based on the extent to which the plurality of keywords included in the keyword cloud are identified in the commercial keyword dictionary (120C). For example, if the proportion of keywords determined to be commercial keywords among the keywords included in the keyword cloud exceeds a preset proportion, the first keyword analysis model (120B) may determine that the keyword cloud corresponds to a commercial purpose. Furthermore, in some embodiments, the first keyword analysis model (120B) may determine the purpose of keywords included in the keyword cloud based on the extent to which the plurality of keywords included in the keyword cloud are identified in the non-commercial keyword dictionary. For example, if the proportion of keywords determined to be non-commercial keywords among the keywords included in the keyword cloud exceeds a preset proportion, the first keyword analysis model (120B) may determine that the keyword cloud corresponds to a non-commercial purpose.

[0097] Analysis of the first keyword analysis model (120B) can classify the first data as either a commercial posting or a non-commercial posting. At least the first variable may include information on whether the posting content of the first data is for commercial or non-commercial purposes.

[0098] Referring to (b) of Fig. 8, the second keyword analysis model (120D) can determine the keyword concentration of the keyword cloud. The second keyword analysis model (120D) can extract multiple major keywords with a relatively higher frequency of use than other keywords among the multiple keywords included in the keyword cloud, and determine the similarity of the extracted major keywords to determine the keyword concentration.

[0099] Here, keyword concentration can be interpreted differently depending on the purpose of the primary data posting. If the similarity of the main keywords corresponding to commercial posts is measured high, the influencer may be judged to have a tendency to specialize in sales by focusing on a specific field, product, or category. If the similarity of the main keywords corresponding to non-commercial posts is measured high, the influencer may be judged to have a tendency to upload posts focused on a specific topic.

[0100] The second keyword analysis model (120D) can calculate the similarity for main keywords using cosine similarity. The second keyword analysis model (120D) can convert main keywords into vectors, calculate the cosine similarity between two keywords, and output the averaged result. For example, if there are 10 main keywords, 45 cosine similarities can be calculated, and the averaged result can be output. Here, the cosine similarity value has a value between -1 and 1. When it is close to -1, it means that the two keywords have an opposite relationship, and when it is close to 1, it means that the two keywords have a synonymous relationship. If the cosine similarity value for the main keywords is large, the main keywords have a similar, synonymous relationship with each other, and the influencer can be determined to exhibit characteristics focused on a specific keyword.

[0101] Analysis of the second keyword analysis model (120D) can determine the keyword concentration of an influencer corresponding to the first data. This keyword concentration can be determined through the cosine similarity value between keywords. The second variable can include information about the keyword concentration of an influencer corresponding to the first data.

[0102] Referring to (c) of FIG. 8, the sentiment analysis model (120E) can analyze the content of a post and comments, and comprehensively analyze the influencer's sentiment and the influencer's sentiment regarding the post and comments to output a third variable. The sentiment analysis model (120E) can analyze the content of a post and comments to determine a sentiment corresponding to positive, negative, or neutral. In other words, the third variable can be a positive sentiment, negative sentiment, or neutral sentiment. In an exemplary embodiment, the sentiment analysis model (120E) may include, but is not limited to, Kobert as a text mining technique.

[0103] Referring to (d) of FIG. 8, the facial recognition model (120F) can identify a person in media data and identify the person's face. In addition, the facial recognition model (120F) can compare the facial image of each influencer stored in the data storage unit (150) of the server (100) with the facial image of the person identified in the media data, and determine whether the influencer corresponding to the first data is included in the media data. In other words, the facial recognition model (120F) can determine whether the face identified in the media data corresponds to the corresponding influencer, and determine whether the influencer is exposed to the media data. Here, the media data may be a plurality of images or a video in which a plurality of images are continuously played. The facial recognition model (120F) can determine whether the influencer is exposed and the exposure frequency in the media data, and can configure a fourth variable based on this information.

[0104] Referring to (e) of FIG. 8, the color recognition model (120G) can extract color information of media data. Specifically, the color recognition model (120G) can determine the main color information of images included in the media data. That is, the color recognition model (120G) can determine the R, G, B values ​​of the colors mainly used in each image. Here, the media data may be a plurality of images or a video in which a plurality of images are played continuously. The face recognition model (120F) can extract main color information from each of the plurality of images included in the media data and calculate the standard deviation of the plurality of extracted main color information. That is, whether an influencer uses various colors when using the media data can be confirmed through the standard deviation. The color recognition model (120G) can output the standard deviation of the main color information of the plurality of images constituting the media data as a fifth variable.

[0105] Referring to (f) of Figure 8, the quantitative analysis model (120G) can output a sixth variable by quantitatively counting the number of comments exchanged between the influencer and other users in the comment content. Here, the number of comments can refer to the total number of comments. Here, the "other users" can refer to users of the SNS, and can also include general users and other influencers. In other words, the quantitative analysis model (120G) can generate a sixth variable based on quantitative data from the comments in the comment content.

[0106] Referring to (g) of FIG. 8, the relationship analysis model (120I) can analyze the comment content to recognize comments exchanged between an influencer and other influencers, and can quantitatively count the number of comments exchanged between the influencer and other influencers to output a seventh variable. The relationship analysis model (120I) can generate the seventh variable based on quantitative data of comments exchanged between the target influencer and other influencers in the comment content. In other words, the relationship analysis model (120I) can determine how much interaction an influencer corresponding to the first data had with other influencers. The relationship analysis model (120I) can identify the ID of another user who wrote a comment, determine whether the ID matches the SNS ID of an influencer stored in the data storage unit (150), determine the ID of the other user who matches as another influencer ID, and output a seventh variable by counting the number of comments between the influencer corresponding to the first data and another influencer.

[0107]

[0108] Referring to FIG. 9, the market server (500) can manage the sales history of products sold to buyers. The market server (500) can manage the sales history for markets conducted by each influencer. That is, the market server (500) can conduct a market established between an influencer and a brand, sell multiple products to multiple buyers, and record information related to product sales and sales history information during the market period. The sales history information can include at least one of the following: sales, sales quantity, products sold, number of orders, total number of purchasing customers, number of new purchasing customers, number of repeat purchasing customers, number of regular customers, and repeat purchasing rate for each influencer's market.

[0109] The data collection unit (110) can collect second data (sales history data) from the market server (500). Here, the second data may be collected based on sales history by influencer, but is not limited thereto. The data collection unit (110) can also collect information on the influencer who sold a specific product.

[0110] The data collection unit (110) may collect second data from the market server (500) periodically, but is not limited thereto. The data collection unit (110) may also collect second data based on markets conducted by matching influencers and brands. Specifically, the data collection unit (110) may be configured to collect second data related to a market after it concludes. The collected second data may be stored in the data storage unit (150) for each influencer account.

[0111] The data storage unit (150) can store sales history data (second data) for markets hosted by influencers. For example, information on sales, sales volume, products sold, number of orders, total number of customers purchasing, number of new customers purchasing, number of repeat customers purchasing, number of regular customers purchasing, and repeat purchase rate for markets hosted by a specific influencer can be stored in correspondence with the influencer's account. Furthermore, the second data can be sorted and retrieved based on the products sold in the data storage unit (150). Furthermore, the second data can be sorted and retrieved based on the market.

[0112] Referring to FIG. 10, the second data preprocessing unit (130) may perform preprocessing on the second data for each influencer stored in the data storage unit (150) to extract multiple variables. The second data preprocessing unit (130) may extract the eighth to eleventh variables related to the market conducted by the influencer based on the second data.

[0113] Here, the eighth variable corresponds to an indicator reflecting the sales history information of the most recent market in which the influencer participated. The second data preprocessing unit (130) extracts values ​​for multiple items related to the most recent market in which the influencer participated from the data storage unit (150) where multiple second data are stored, and calculates the eighth variable by applying a predetermined weight to the extracted values. For example, the second data preprocessing unit (130) extracts the values ​​of each item for the most recent market, including sales, sales quantity, number of orders, total number of purchasing customers, number of new purchasing customers, number of repeat purchasing customers, number of regular customers, and repeat purchasing rate, and applies preset weights to the extracted item values ​​to calculate the eighth variable representing the characteristics of the most recent market in which the influencer participated. Here, the weights may be set differently for each variable described below, but are not limited thereto. In addition, the weights may be set through regression analysis in consideration of the characteristics of the variables.

[0114] Additionally, the ninth variable may be an indicator reflecting the temporal characteristics of the market. Extracting the ninth variable may require specifying schedule information corresponding to the market's progress period. In some embodiments, if explicit schedule information is not provided, the ninth variable may be extracted based on the current time. The second data preprocessing unit (130) may extract the ninth variable that may reflect the temporal characteristics by reflecting information on past markets that were held in the same month or season as the schedule information. For example, the second data preprocessing unit (130) may check information on markets that were held 9 to 15 months prior to the start date according to the schedule information and extract the ninth variable that may reflect the temporal characteristics. The second data preprocessing unit (130) extracts the respective item values ​​for the market sales, sales quantity, number of orders, total number of purchasing customers, number of new purchasing customers, number of repeat purchasing customers, number of regular customers, and repeat purchasing rate of a past market (for example, a market held one year ago) that was held in the same month or season as the period according to the schedule information, and applies a preset weight to the extracted item values ​​to calculate a ninth variable that represents the characteristics of the time period in which the market is held.

[0115] In addition, the tenth variable may be an indicator reflecting the current buyer's preference characteristics for the target product. Extracting the tenth variable may require specifying the target product. The second data preprocessing unit (130) may extract the tenth variable that may reflect the current buyer preference characteristics of the product by reflecting information on the most recently conducted market for products belonging to the same category as the specified target product. For example, the second data preprocessing unit (130) may extract the respective item values ​​for sales, sales quantity, number of orders, total number of purchasing customers, number of new purchasing customers, number of repeat purchasing customers, number of regular customers, and repurchase rate of the recently conducted market for products belonging to the same category as the target product, and may calculate the tenth variable representing the current buyer's preference characteristics for the product by applying preset weights to the extracted item values.

[0116] Additionally, the eleventh variable may be an indicator extracted based on the number of markets recently hosted by the influencer and the number of identical products recently hosted by the influencer. In other words, whether the influencer has been actively engaged in recent activities and whether the influencer has recently focused on a specific product can be confirmed through the eleventh variable. Here, "recent" may refer to three months prior to the current date, but the embodiments of the present invention are not limited thereto. The second data preprocessing unit (130) may extract the number of markets hosted by a specific influencer in the past three months and the number of markets hosted for the same category of products, and apply preset weights to the extracted numbers to calculate the eleventh variable representing the influencer's recent activity characteristics.

[0117]

[0118] Referring to Figure 11, the market server (500) can collect buyer activity information identified during the product sales process and manage the collected activity information. The market server (500) can provide a buyer environment (P) for purchasing products related to a market run by each influencer. Buyers can purchase products from the market through the buyer environment. While utilizing this buyer environment, buyers can perform various activities.

[0119] Buyers can enter the product detail page of the buyer environment and decide whether to purchase. The market server (500) can collect various activity information on these buyers before making a purchase within the service. For example, key activities related to the product detail page before a purchase can include the number of times a buyer enters the product detail page, the number of times a shopping cart containing a product is temporarily clicked, the number of times a notification banner related to the influencer's product sales is displayed corresponding to the number of times a notification banner related to the influencer's product sales is selected, and the number of times a notification notification announcing the launch of the influencer's market is requested. For example, if a buyer selects a notification banner, they may be directed to the influencer's product detail page, and the buyer can request to receive this notification banner. The market server (500) can record key activity information on multiple buyers' activities related to the product detail page for each influencer.

[0120] Additionally, buyers can perform activities to purchase products identified through the buyer environment. The market server (500) can collect key activities related to the buyer's product purchases. For example, key activities related to product purchases may include the number of completed payments for products sold by influencers through the market, the purchase conversion rate (the ratio of completed payments to the number of buyers entering the detail page), the shopping cart purchase conversion rate (the ratio of completed payments to the number of products in the shopping cart), and the notification subscription purchase conversion rate (the ratio of completed payments to the number of notification subscriptions).

[0121] The market server (500) can collect and manage information on the buyer's activities prior to purchasing a product and information on activities related to purchasing a product in response to the influencer's market.

[0122] The data collection unit (110) can collect third data (activity information) from the market server (500). The data collection unit (110) can collect third data from the market server (500) at regular intervals, but is not limited thereto. The data collection unit (110) can also collect third data based on a market conducted by matching influencers and brands. In other words, the data collection unit (110) can be configured to collect third data related to a market after it has ended. The collected third data can be stored in the data storage unit (150) for each influencer account.

[0123] The data storage unit (150) can store third-party data (activity data) collected in response to a market hosted by an influencer. For example, pre-purchase activity information and activity information related to product purchases of multiple buyers in response to a market hosted by a specific influencer can be stored in the data storage unit (130).

[0124] Referring to FIG. 12, the third data preprocessing unit (140) may perform preprocessing on third data for each influencer stored in the data storage unit (150) to extract multiple variables. The third data preprocessing unit (140) may extract the 12th and 13th variables based on the activities of multiple buyers in relation to a market conducted by the influencer, based on the third data. Here, the first activity indicator may refer to activity information of multiple buyers before purchasing a product, and the second activity indicator may refer to activity information related to the purchase of a product by multiple buyers.

[0125] Here, the twelfth variable may represent the pre-purchase activity information of multiple buyers for a market where the influencer has participated, based on an indicator extracted from the first activity data. The third data preprocessing unit (140) may extract values ​​for multiple items of the first activity data and apply a predetermined weight to the extracted values ​​to calculate the twelfth variable. For example, the third data preprocessing unit (140) may extract the values ​​of each item for the number of entries by buyers to the product detail page, the number of clicks on the shopping cart, the number of notification banner impressions, and the number of notification subscriptions, and may apply preset weights to the extracted item values ​​to calculate the twelfth variable representing the pre-purchase activity characteristics of multiple buyers related to the influencer. Here, the weights may be set through regression analysis in consideration of the characteristics of the variables.

[0126] In addition, the 13th variable may represent activity information related to product purchases by multiple buyers in a market where the influencer has participated, based on an indicator extracted from the 12th activity data. The third data preprocessing unit (140) may extract values ​​for multiple items of the second activity data and apply a predetermined weight to the extracted values ​​to calculate the 13th variable. For example, the third data preprocessing unit (140) may extract values ​​for each item of the number of completed product payments, purchase conversion rate, shopping cart purchase conversion rate, and notification subscription purchase conversion rate, and may apply preset weights to the extracted item values ​​to calculate the 13th variable representing activity characteristics related to product purchases by multiple buyers related to the influencer. Here, the weights may be set through regression analysis in consideration of the characteristics of the variables.

[0127] The data storage unit (150) can store data collected by the data collection unit (110) and data preprocessed by the first to third data preprocessing units (110, 120, 130). In addition, the data storage unit (150) can temporarily or semi-permanently store data required for the operation of the learning unit (160) and the prediction model (170) and data generated according to the operation.

[0128]

[0129] The learning unit (160) can form a learning dataset based on the data preprocessed in the first to third data preprocessing units (110, 120, 130) and learn the prediction model (170). In some embodiments, the learning unit (160) can form a learning dataset based on the data preprocessed in the first data preprocessing unit (110), that is, data corresponding to at least one of the first to seventh variables, and learn the prediction model (170). In addition, the learning unit (160) can form a learning dataset based on the data preprocessed in the first data preprocessing unit (110), the data preprocessed in the second data preprocessing unit (120), or the data preprocessed in the third data preprocessing unit (130), and learn the prediction model (170). That is, the prediction model (170) is formed based on the first data, but may also be formed by taking into consideration at least one of the second data and the third data. In the following description, a prediction model (170) is constructed using the first to thirteenth variables and the importance of each variable is determined. However, the prediction model (170) may be constructed with at least some of the first to thirteenth variables omitted.

[0130] The prediction model (170) can predict the sales volume of a target product when the target influencer conducts a market campaign for the target product. The prediction model (170) can be input with multidimensional variables related to the target influencer, and the prediction model (170) can predict the sales volume of the target influencer for the target product by applying weights to each multidimensional variable.

[0131] The learning unit (160) can perform learning on the prediction model (170) based on the first to thirteenth variables extracted through the preprocessing process for the first data, the second data, and the third data. The learning unit (160) can perform learning on the prediction model (170) by configuring the sales volume for a specific product sold by an actual influencer according to the first to thirteenth variables extracted from the first to third data of each of a plurality of influencers into one learning dataset.

[0132] The predictive model (170) can be implemented as a machine learning module. The input values ​​to the predictive model (170) can be multiple variables, and the output values ​​can be sales volumes based on the multiple variables. The sales volumes can be variable and continuous results. In an embodiment of the present invention, the predictive model (170) can be configured as a decision tree model based on a regression tree to predict sales volumes, which are variable and continuous results, based on the characteristics of the multiple variables.

[0133] The prediction model (170) can create child nodes from a parent node by setting a split that can minimize the sum of squares of errors based on at least one variable. Each branch can be divided into child nodes or leaf nodes, and the leaf nodes can output results according to the variable, i.e., sales volume. Here, the sum of squares of errors can be the mean squared error (MSE) or the mean absolute error (MAE). The mean squared error refers to the average of the squared errors between the model predicted value and the actual value, and the mean absolute error can refer to the average of the absolute errors between the model predicted value and the actual value. The child nodes can be created based on the variable and the split value that most reduces the mean squared error or the mean absolute error of the parent node. The regression tree is modeled in a top-down manner, and the method of dividing the branches based on the variable by finding the best branch can be repeated. For example, if the first variable minimizes the sum of squared errors the most, the first branch can be generated through the first variable, and then the next branch can be generated through the fourth variable that can minimize the sum of squared errors. The prediction model (170) can be configured to sequentially generate at least one branch through at least one variable, and can be configured to classify input data based on the variable so that sales volume can be predicted.

[0134] In an embodiment of the present invention, the prediction model (170) may be configured as an ensemble model based on a Gradient Boosting Decision Tree (GBDT). That is, the prediction model (170) may have a structure in which multiple regression tree-based decision tree models are linked to each other.

[0135] Referring to FIG. 13, the prediction model (170) may include a first prediction model to an Nth prediction model. Here, each of the first prediction model to an Nth prediction model may be a decision tree model based on an individual regression tree, and may be constructed to classify input data according to branches and output corresponding sales volumes. Here, correction may be performed on the learning data of the subsequent prediction model according to the classification result of the preceding prediction model, and the subsequent prediction model may proceed with learning through the corrected data, thereby constructing a more improved model.

[0136] The learning unit (160) may include a variable importance calculation unit (160A) and a learning execution unit (160B). The variable importance calculation unit (160A) may calculate the importance of each variable based on the prediction model (170) for which learning has been completed. The learning execution unit (160B) may control the learning of the prediction model (170) to proceed through the process described below.

[0137] The learning execution unit (160B) performs the above-described prediction process for each of the first to Nth prediction models, but sequential learning can be performed considering the results of the preceding prediction model. Through this sequential learning, the subsequent prediction model can be trained to have improved prediction performance compared to the preceding prediction model. Here, N corresponds to a natural number greater than or equal to 2, and can be determined as an optimal number considering the prediction performance of the prediction model (170).

[0138] The first prediction model may be the initial model (base model), and sequential learning of the first to Nth prediction models may be performed based on the first prediction model. New learning data may be constructed by applying weights to the learning data reflecting the prediction errors of the preceding prediction model, and learning may be performed for the subsequent prediction model. Accordingly, the prediction accuracy of the subsequent prediction model may be further improved, and the overall performance of the prediction model (170) may be improved.

[0139] First, the learning execution unit (160B) can perform learning on the first prediction model through the first learning dataset. The first learning dataset can include a plurality of input datasets each comprising first to thirteenth variables extracted from first to third data of a plurality of influencers and a target product. The first prediction model predicts the sales volume of the target product for each input dataset. The first prediction model can determine a branch for at least one variable corresponding to the first learning dataset, and can predict the sales volume for the input dataset through classification according to the determined branch.

[0140] Here, if the predicted result output from the first prediction model matches the actual result, the corresponding input dataset is classified into the first group (S). Furthermore, if the predicted result output from the first prediction model differs from the actual result, the corresponding input dataset is classified into the second group (W). The learning execution unit (160B) can reconstruct the second learning dataset by applying weights to the input dataset in which an error occurs between the predicted result and the actual result.

[0141] In some embodiments, the learning execution unit (160B) may calculate a prediction error between a predicted result and an actual result, and may calculate learning data weights based on the prediction error. Here, the learning data weights may mean performing corrections on the input data set corresponding to the second group (W) so that the input data set corresponding to the second group (W) is included in the first group (S) in the subsequent prediction model. The learning execution unit (160B) may perform corrections on the misclassified input data set using the calculated learning data weights, and reconstruct the second learning data set.

[0142] The learning execution unit (160B) can perform learning on a second prediction model using a second learning dataset. The second learning dataset may have a different data structure from the first learning dataset due to the above-described correction. The second prediction model may determine a branch for at least one variable to correspond to the second learning dataset, and may be capable of predicting sales for the input dataset through classification according to the determined branch. The main variables and branches configured in the second prediction model may be configured in a structure different from that of the first prediction model, but are not limited thereto.

[0143] Based on the results output by the second prediction model, the second learning dataset can be classified again into the first group (S) and the second group (W), and a third learning dataset can be reconstructed by applying weights to the input datasets in which an error occurred between the predicted results and the actual results, and learning for the third prediction model can be performed through the third learning dataset.

[0144] The learning execution unit (160B) can build a subsequent prediction model by weighting misclassified data to increase its size and reducing the size of correctly classified data using the weights to compensate for errors. The learning execution unit (160B) performs learning on a sequential prediction model using a boosting method.

[0145] When a preset number of prediction models are sequentially constructed, the learning execution unit (160B) may combine the constructed prediction models to terminate learning for the prediction model (170). In some embodiments, when the prediction error of the current prediction model is below a preset threshold, the learning execution unit (160B) may terminate generating the subsequent prediction model, combine the prediction models learned up to this point, and terminate learning for the prediction model (170).

[0146] The prediction model (170) for which learning has been completed is composed of a plurality of prediction models (the first to Nth prediction models), and the plurality of prediction models can each output results. The results output from each are combined and output as a final result. The plurality of prediction models each output results, and a correction coefficient can be applied to the output results and each result to configure the combined value as a final prediction result. Here, each correction coefficient applied to the results of the first to Nth prediction models can be determined in consideration of the learning data weight applied to each prediction model, but the embodiments of the present invention are not limited thereto. In some embodiments, the correction coefficient may be applied equally to each prediction model, or may be set to different values ​​through regression analysis, etc.

[0147] After the learning of the prediction model (170) performed in the learning execution unit (160B) is completed, the variable importance calculation unit (160A) can determine the variable importance of the first to thirteenth variables.

[0148] The variable importance calculation unit (160A) can calculate a weight, which is the number of times each variable is used to divide data in the first to Nth prediction models. In addition, the variable importance calculation unit (160A) can calculate a cover corresponding to the number of data separated through each variable in the first to Nth prediction models. In addition, the variable importance calculation unit (160A) can calculate a gain, which is an average information gain that decreases when each variable is used in the first to Nth prediction models. Here, the information gain refers to the information gain when a data set is divided and classified using a variable. The information gain corresponds to the result of subtracting the overall entropy of a plurality of lower nodes (child nodes) divided through a variable from the entropy of the data set of the upper node (parent node) before division. Since the details of the calculation process of information gain and entropy have already been disclosed, the details are omitted here.

[0149] The variable importance calculation unit (160A) can determine the importance of the first to thirteenth variables by considering the weight, cover, and information gain of each variable calculated in the first to Nth prediction models. Based on the weight, cover, and information gain of each variable calculated in the first to Nth prediction models, the variable importance calculation unit (160A) can determine the rank of the first to thirteenth variables, and can differentially determine the importance according to the determined rank.

[0150] In some embodiments, when the prediction model (170) considers all of the multiple variables, the output results may be accurate, but this may increase the amount of data processing, resulting in processing delays and resource loss. Retraining of the prediction model (170) may be performed by applying the variable importance calculated by the variable importance calculation unit (160A). For example, retraining of the prediction model (170) may be performed by excluding at least one variable whose variable importance is evaluated as low.

[0151] Additionally, in some embodiments, variable importance calculated by the variable importance calculation unit (160A) may be applied to generate variable weights for input values ​​(for variables) to the prediction model (170). That is, the variable weights may be configured to have a lower influence on high variables with relatively lower ranks of variable importance compared to other variables. The variable weights may perform corrections for multiple variables input to the prediction model (170) so that high-ranking variables exhibit higher influence, and the prediction results of the prediction model (170) may be supported to be output more accurately.

[0152] In some embodiments, the completion of training for the prediction model (170) in the learning unit (160) may mean that retraining of the prediction model (170) based on variable importance or setting variable weights based on variable importance has been completed. Using the trained prediction model (170), sales volume prediction for the target product of the target influencer can be performed.

[0153] Referring to Figure 14, an exemplary process for predicting when a target influencer will sell a target product is illustrated using a trained prediction model. The target influencer can request a sales volume prediction for the target product from the server (100). However, this is not limited to this, and the brand of the target product can request the predicted sales volume from the server (100) when the target influencer sells the target product.

[0154] These requests may be provided through an influencer terminal or brand terminal via a user environment provided by the server (100). Additionally, in some embodiments, the requests may include schedule information regarding the timing of the market.

[0155] The server (100) may collect first data, second data, and third data related to the target influencer from the SNS server (400) and the market server (500), or may utilize data stored in the data storage unit (150). In addition, preprocessing of the first data, second data, and third data may be performed in the first data preprocessing unit, the second data preprocessing unit, and the third data preprocessing unit, respectively, to extract a plurality of variables. The prediction model may predict the sales volume of the target product of the target influencer in response to an input dataset configured based on the variables extracted by the first data preprocessing unit, the variables extracted by the second data preprocessing unit, and the variables extracted by the third data preprocessing unit.

[0156] Here, the plurality of variables may be first to thirteenth variables. The first data includes posting content, comment content, and media data, and the first data preprocessing unit analyzes keywords of the posting content to determine the purpose of the posting and generates a first variable, analyzes the frequency of use of the keywords to determine keyword concentration and generates a second variable, analyzes the posting content and the comment content to determine the sentiment of the target influencer and generates a third variable, determines whether the target influencer has been exposed and the frequency of exposure from the media data to generate a fourth variable, extracts main color information from the media data, and determines the standard deviation of the extracted main color information to generate a fifth variable. The first data preprocessing unit may generate a sixth variable based on quantitative data of comments from the comment content, and generate a seventh variable based on quantitative data of comments exchanged between the target influencer and other influencers from the comment content.

[0157] The second data preprocessing unit may analyze the second data to generate an eighth variable indicating sales history information of a market recently conducted by the target influencer; a ninth variable indicating sales history information according to the temporal characteristics of a market to be conducted by the target influencer; a tenth variable indicating a buyer's preference characteristics for the target product; and an eleventh variable indicating the activity of the target influencer.

[0158] The third data preprocessing unit may analyze the third data to generate a twelfth variable representing activity information prior to product purchase by the plurality of purchasers; and a thirteenth variable representing activity information related to product purchase by the plurality of purchasers.

[0159] Here, the first to eighth variables and the eleventh variable may be extracted based on the influencer's social media posts or sales history. The ninth variable may be extracted based on market schedule information and the sales history of the target product. The tenth variable may be extracted based on recent sales history of the target product. The twelfth and thirteenth variables may be extracted based on behavioral data of buyers who used the market hosted by the influencer.

[0160] An input dataset may be constructed based on at least one of the first to thirteenth variables extracted in relation to the target influencer, the target product, and the market schedule information. In some embodiments, the prediction model (170) may correct the values ​​of the first to thirteenth variables by applying variable weights, and the input dataset may be constructed using the corrected plurality of variables. In response to the input dataset, the prediction model performs a prediction on product sales for the target product. Here, the prediction model may be an ensemble model based on a Gradient Boosting Decision Tree (GBDT) composed of the first to Nth prediction models, each of which may predict and output the product sales. The predicted sales output from the first to Nth prediction models may be calculated by applying a correction coefficient, and the final sales of the target influencer for the target product may be predicted and provided to the target influencer or the target brand.

[0161] A method and device for predicting sales volume based on multimodal data preprocessing, according to some embodiments of the present invention, can predict sales volume for a target product by a target influencer based on the influencer's sales history, the activity history of buyers associated with the influencer, and the influencer's posting information. The predicted sales volume can then be provided to the target influencer or a brand associated with the target product. Accordingly, the target influencer or brand can determine whether to proceed with a market based on the predicted sales volume, determine inventory levels, and facilitate more successful market operations.

[0162] In addition, a method and device for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention extracts multiple variables judged in a prediction model through preprocessing of multimodal data in various ways, and predicts the sales volume of a target product through a prediction model that uses multiple variables as input values, and can determine the sales volume by considering various variables and factors.

[0163] In addition, a method and device for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention can determine the importance of a plurality of variables extracted from multimodal data and apply appropriate weights to the variables based on the determined importance, thereby providing more accurate and faster prediction results.

[0164]

[0165] Hereinafter, a sales volume prediction method based on multimodal data preprocessing according to an embodiment of the present invention will be described. The sales volume prediction method based on multimodal data preprocessing according to the following embodiment is performed by the server (100) described above. Reference may be made to FIGS. 1 to 14 and their related descriptions.

[0166]

[0167] FIG. 15 is a flowchart of a sales volume prediction method based on multimodal data preprocessing according to some embodiments of the present invention.

[0168] Referring to FIG. 15, a method for predicting sales volume based on multimodal data preprocessing according to some embodiments of the present invention includes a step of collecting first data related to SNS postings of the target influencer, second data related to product sales history through the market of the target influencer and sales history of the target product through the market, and third data related to activities of multiple buyers for the market of the target influencer (S100); a step of performing preprocessing on the first data, preprocessing on the second data, and preprocessing on the third data (S110); a step of configuring an input dataset based on variables extracted through preprocessing on the first data, variables extracted through preprocessing on the second data, and variables extracted through preprocessing on the third data (S120); and a step of applying the input dataset to a prediction module and outputting the sales volume of the target product of the target influencer as an output thereof (S130).

[0169] In some embodiments, the first data includes posting content, comment content, and media data, and in step (S110), preprocessing for the first data may include analyzing keywords of the posting content to determine the purpose of the posting and thereby create a first variable, analyzing the frequency of use of the keywords to determine keyword concentration and thereby create a second variable, analyzing the posting content and the comment content to determine the sentiment of the target influencer and thereby create a third variable, determining whether the target influencer has been exposed and the frequency of exposure in the media data to create a fourth variable, extracting main color information of the media data, and determining the standard deviation of the extracted main color information to create a fifth variable.

[0170] In some embodiments, in step S110, the preprocessing of the first data may include generating a sixth variable based on quantitative data of comments in the comment content, and generating a seventh variable based on quantitative data of comments exchanged between the target influencer and other influencers in the comment content.

[0171] In some embodiments, in step S110, the preprocessing of the second data may include analyzing the second data to generate an eighth variable representing sales history information of a market recently conducted by the target influencer; a ninth variable representing sales history information according to the timing characteristics of a market to be conducted by the target influencer; a tenth variable representing a preference characteristic of a buyer for the target product; and an eleventh variable representing the activity of the target influencer.

[0172] In some embodiments, in step (S110), the preprocessing of the third data may include analyzing the third data to generate a twelfth variable representing activity information prior to the purchase of goods by the plurality of buyers; and a thirteenth variable representing activity information related to the purchase of goods by the plurality of buyers.

[0173] In some embodiments, the method further comprises a step of learning the prediction model, and the step of learning the prediction model may further comprise a step of determining variable importance of variables constituting the input dataset.

[0174] In some embodiments, step (S130) may include configuring the input dataset by applying variable weights according to the variable importance to the variables extracted from the first data preprocessing unit, the variables extracted from the second data preprocessing unit, and the variables extracted from the third data preprocessing unit.

[0175] In some embodiments, the prediction model is an ensemble model based on a Gradient Boosting Decision Tree (GBDT), and includes first to Nth prediction models, each of the first to Nth prediction models is a decision tree model based on a regression tree, and each classifies the input dataset according to a constructed branch and outputs a corresponding sales amount, and step (S130) may include calculating the sales amount for the target product of the target influencer by applying a preset correction coefficient to the sales amount output from the first to Nth prediction models.

[0176]

[0177] FIG. 16 is a diagram illustrating a hardware implementation of a device that performs a sales volume prediction method based on multimodal data preprocessing according to some embodiments of the present invention.

[0178] Referring to FIG. 16, a server (100) according to some embodiments of the present invention may be implemented as an electronic device (1000). The electronic device (1000) may include a processor (1010), an input / output device (1020), a memory (1030), an interface (1040), storage (1050), and a bus (1060). The processor (1010), the input / output device (1020), the memory (1030), the interface (1040), and / or the storage (1050) may be coupled to each other via a bus (1060). The bus (1060) corresponds to a path through which data is transferred.

[0179] Specifically, the processor (1010) may include at least one of a Central Processing Unit (CPU), a Micro Processor Unit (MPU), a Micro Controller Unit (MCU), a Graphic Processing Unit (GPU), a microprocessor, a digital signal processor, a microcontroller, an application processor (AP), and logic elements capable of performing functions similar thereto.

[0180] The input / output device (1020) may include at least one of a keypad, a keyboard, a touchscreen, and a display device.

[0181] The memory (1030) can load data and / or programs, etc. At this time, the memory (1030) is an operating memory for improving the operation of the processor (1010) and may include high-speed DRAM and / or SRAM. The memory (1030) may include one or more volatile memory devices such as DDR SDRAM (Double Data Rate Static DRAM) and SDR SDRAM (Single Data Rate SDRAM) and / or one or more non-volatile memory devices such as EEPROM (Electrically Erasable Programmable ROM) and flash memory.

[0182] The interface (1040) may perform a function of transmitting data to or receiving data from a communication network. The interface (1040) may be wired or wireless. For example, the interface (1040) may include an antenna or a wired or wireless transceiver.

[0183] Storage (1050) can store and preserve data and / or programs. Storage (1050) can include one or more non-volatile memory devices, such as a solid state drive (SSD), a hard drive, or flash memory. In the present invention, storage (1050) can store a computer program composed of instructions for performing a sales volume prediction method based on multimodal data preprocessing.

[0184] The server (100) according to embodiments of the present invention may be a system formed by connecting multiple electronic devices (1000) to each other via a network. In this case, each module or combination of modules may be implemented as an electronic device (1000). However, the present embodiment is not limited thereto.

[0185] Additionally, the server (100) may be implemented as at least one of a workstation, a data center, an internet data center (IDC), a direct attached storage (DAS) system, a storage area network (SAN) system, a network attached storage (NAS) system, a redundant array of inexpensive disks (RAID) system, and an electronic document management (EDMS) system, but the present embodiment is not limited thereto.

[0186]

[0187] The sales volume prediction method based on multimodal data preprocessing according to the embodiment can also be implemented in the form of a computer-readable medium that stores computer-executable commands and data. In this case, the commands and data can be stored in the form of program code, and when executed by a processor, a predetermined program module can be generated to perform a predetermined operation. In addition, the computer-readable medium can be any available medium that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium can be a computer recording medium, and the computer recording medium can include both volatile and nonvolatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable commands, data structures, program modules, or other data. For example, the computer recording medium can be a magnetic storage medium such as an HDD or SSD, an optical recording medium such as a CD, DVD, or Blu-ray disc, or a memory included in a server accessible via a network.

[0188] In addition, the sales volume prediction method based on multimodal data preprocessing according to the embodiment may be implemented as a computer program (or computer program product) including computer-executable instructions. The computer program includes programmable machine instructions processed by the processor and may be implemented in a high-level programming language, an object-oriented programming language, assembly language, or machine language. In addition, the computer program may be recorded on a tangible computer-readable recording medium (e.g., memory, a hard disk, a magnetic / optical medium, or a solid-state drive (SSD), etc.).

[0189] Accordingly, the sales volume prediction method based on multimodal data preprocessing according to the embodiment can be implemented by executing the above-described computer program on a computing device. The computing device may include at least some of a processor, memory, a storage device, a high-speed interface connecting the memory and a high-speed expansion port, and a low-speed interface connecting a low-speed bus and the storage device. Each of these components is interconnected using various buses and may be mounted on a common motherboard or in another suitable manner.

[0190] Here, the processor can process instructions within the computing device, such as instructions stored in a memory or storage device to display graphical information for providing a graphical user interface (GUI) on an external input / output device, such as a display connected to a high-speed interface. In another embodiment, multiple processors and / or multiple buses may be utilized, as appropriate, together with multiple memories and memory types. The processor may also be implemented as a chipset comprising multiple independent analog and / or digital processors.

[0191] Memory also stores information within a computing device. For example, memory may consist of volatile memory units or a collection of volatile memory units. For another example, memory may consist of nonvolatile memory units or a collection of nonvolatile memory units. Memory may also be another form of computer-readable media, such as magnetic or optical disks.

[0192] A storage device can provide a large amount of storage space to a computing device. The storage device can be a computer-readable medium or a configuration including such a medium, and can include, for example, devices within a storage area network (SAN) or other configurations, and can be a floppy disk device, a hard disk device, an optical disk device, a tape device, flash memory, or other similar semiconductor memory device or device array.

[0193] The above description is merely an example of the technical idea of ​​the present embodiment, and those skilled in the art will appreciate that various modifications and variations can be made without departing from the essential characteristics of the present embodiment. Therefore, the present embodiments are not intended to limit the technical idea of ​​the present embodiment, but rather to explain it, and the scope of the technical idea of ​​the present embodiment is not limited by these embodiments. The scope of protection of the present embodiment should be interpreted by the claims below, and all technical ideas within a scope equivalent thereto should be interpreted as being included in the scope of rights of the present embodiment.

Claims

1. A server that predicts the sales volume of a target product of a target influencer, wherein the server: A data collection unit that collects first data related to the SNS posting of the target influencer, second data related to the sales history of products through the market of the target influencer and the sales history of the target product through the market, and third data related to the activities of multiple buyers for the market of the target influencer; A first data preprocessing unit that performs preprocessing on the first data; A second data preprocessing unit that performs preprocessing on the second data; A third data preprocessing unit that performs preprocessing on the third data; A prediction model that outputs the sales volume of the target product of the target influencer in response to an input data set composed based on the variables extracted from the first data preprocessing unit, the variables extracted from the second data preprocessing unit, and the variables extracted from the third data preprocessing unit, The above first data includes posting content, comment content, and media data, The above first data preprocessing unit, By analyzing the keywords in the above posting content, the purpose of the above posting is determined and the first variable is created. By analyzing the frequency of use of the above keywords, the keyword concentration is determined and a second variable is created. By analyzing the above posting content and the above comment content, the sentiment of the target influencer is judged and a third variable is created. A fourth variable is created by determining whether the target influencer is exposed and the frequency of exposure from the above media data, Extracting the main color information of the above media data and determining the standard deviation of the extracted main color information to create a fifth variable. Server.

2. In paragraph 1, The above first data preprocessing unit, In the above comment content, a sixth variable is created based on the quantitative data of the comment, In the above comment content, the seventh variable is created based on quantitative data of comments exchanged between the target influencer and other influencers. Server.

3. In paragraph 2, The above first data preprocessing unit, Text mining that extracts keywords from the above posting content and creates a keyword cloud by extracting only nouns from the extracted keywords; A first keyword analysis model that determines the commercial purpose of the posting based on the extent to which keywords included in the keyword cloud are confirmed in a commercial keyword dictionary; and A second keyword analysis model that selects keywords with high frequency of use among keywords included in the keyword cloud and determines the keyword concentration by determining the similarity between the selected keywords. Server.

4. In paragraph 1, The second data preprocessing unit analyzes the second data, An eighth variable indicating the sales history information of the market recently conducted by the above target influencer; A ninth variable indicating sales history information according to the temporal characteristics of the market to be operated by the above target influencer; A tenth variable representing the buyer's preference characteristics for the above target product; and Creating an 11th variable representing the activity of the above target influencer, Server.

5. In paragraph 1, The third data preprocessing unit analyzes the third data, A twelfth variable representing the activity information of the multiple purchasers before purchasing the product; and Generating a 13th variable representing activity information related to the purchase of products by the above multiple buyers, Server.

6. In paragraph 1, It further includes a learning unit for learning the above prediction model, The above learning department, A variable importance calculation unit that determines the variable importance of the variables constituting the above input dataset; and including a learning execution unit that learns the above prediction model, Server.

7. In paragraph 6, The above prediction model is, The input dataset is configured by applying variable weights according to the variable importance to the variables extracted from the first data preprocessing unit, the variables extracted from the second data preprocessing unit, and the variables extracted from the third data preprocessing unit. Server.

8. In paragraph 1, The above prediction model is an ensemble model based on GBDT (Gradient Boosting Decision Tree), and includes the first to Nth prediction models. Each of the above first to Nth prediction models is a decision tree model based on a regression tree, and each classifies the input dataset according to the constructed branch and outputs the corresponding sales amount. The sales volume of the target product of the target influencer is calculated by applying a preset correction coefficient to the sales volume output from the first to Nth prediction models. Server.

9. A method for predicting the sales volume of a target product by a target influencer performed on a server. A step of collecting first data related to the SNS posting of the target influencer, second data related to the sales history of products through the market of the target influencer and the sales history of the target product through the market, and third data related to the activities of multiple buyers for the market of the target influencer; A step of performing preprocessing on the first data, preprocessing on the second data, and preprocessing on the third data; A step of configuring an input dataset based on variables extracted through preprocessing of the first data, variables extracted through preprocessing of the second data, and variables extracted through preprocessing of the third data; Including a step of applying the above input dataset to a prediction module and outputting the sales volume of the target product of the target influencer as an output thereof. The above first data includes posting content, comment content, and media data, Preprocessing for the above first data is as follows: By analyzing the keywords in the above posting content, the purpose of the above posting is determined and the first variable is created. By analyzing the frequency of use of the above keywords, the keyword concentration is determined and a second variable is created. By analyzing the above posting content and the above comment content, the sentiment of the target influencer is judged and a third variable is created. A fourth variable is created by determining whether the target influencer is exposed and the frequency of exposure from the above media data, Extracting the main color information of the above media data, and generating a fifth variable by judging the standard deviation of the extracted main color information, A sales volume prediction method based on multimodal data preprocessing.

10. A computer program stored on a medium for executing a sales volume prediction method based on multimodal data preprocessing according to Article 9 in combination with hardware.

Images