KR20240131244A - artificial intelligence-based brand value evaluation electronic device and method therof - Google Patents

Abstract

In an electronic device for evaluating brand value, the device comprises a memory for storing product information and brand information, and a processor for processing the product information and the brand information, wherein the processor, based on processing item information and review information included in the product information, inputs the product information into a sales prediction model to output a predicted sales amount, and inputs the predicted sales amount and the brand information into a brand value evaluation model to output at least one value index of the brand.

Description

{Artificial intelligence-based brand value evaluation electronic device and method therof}

The present disclosure relates to an electronic device for preprocessing data for web crawling-based learning and a method thereof, and more specifically, to an electronic device for preprocessing data for web crawling-based learning and a method thereof, which can collect unstructured data into structured data according to certain criteria and refine the data so that an artificial intelligence model can learn.

A brand aggregator is a company that has a business model that grows through economies of scale by continuously acquiring (or gathering or cooperating with) brands of small and medium-sized businesses with online marketability and marketability. These companies generally sell or market various products on one platform through the Internet. These platforms often provide consumers with a convenient shopping experience and have the advantage of being able to compare and select various products, and they can also cooperate with small and medium-sized (SME) brands that have their own online and offline brands for overseas expansion and brand growth. Therefore, since brand aggregators require analysis of SME brands from various perspectives, they need data to analyze these SME brands from various perspectives.

Recently, with the rapid growth of the Web, a huge amount of information is being provided through the Web. When the Web first appeared, the amount of information it contained was relatively small, so these early Webs were expressed in the form of static pages. The URLs (Uniform Resource Locators) of these static pages usually existed continuously, so they could be visited by following hyperlinks. The Web with these characteristics is called the Surface Web.

Many search engines have been developed to search this surface web. These search engines use programs called crawlers to visit web pages following hyperlinks and create an index of the visited web pages. The created index is then used to find pages that satisfy the user's query.

However, the current web has much more information than the early web. In order to efficiently manage and search this information, the current web stores information in its back-end database. In general, users use the search function of the website to retrieve information from the database at the back of the web. Here, websites provide this search function using the <FORM> tag of HTML (Hypertext Markup Language). The <FORM> tag of HTML plays a role in transmitting the user's input to the web server. In other words, when a user enters a query in the search form, the website searches its back-end database for information related to the query. The website dynamically generates a page containing the search results and shows this page to the user.

Currently, many search engine companies use a huge amount of distributed computing resources for web crawling. For example, Google (http://www.google.com) has hundreds of thousands of computers installed directly around the world and uses them for web crawling.

Accordingly, brand aggregators are trying to secure data using such web crawling in order to obtain data for analyzing small and medium-sized (SME) brands from various perspectives. However, data obtained using the above-mentioned web crawling is unstructured data without a unified and consistent standard, and there is a problem that there is no discrimination for analysis from various perspectives. In particular, in a situation where the same product or various product names are traded online (e-commerce), there is a problem that the same product is treated as different products and stored as data even if data is obtained through web crawling.

On the other hand, reviews of products and ratings of said products include people's various emotions and subjective ratings based on them. Accordingly, when some people value service, they may give high scores even if they are not satisfied with the product, if they receive high quality service. In other words, the ratings given in this way may be meaningless and indiscriminate if given to people who prioritize satisfaction with the product.

Therefore, product reviews need to be objective by applying uniform standards to specific areas such as products and services through analysis of the reviews in question.

Meanwhile, evaluating the value of small and medium-sized (SME) brands requires consideration of various indices. Other examples of electronic devices for evaluating the value of such brands include, for example, according to Korean Patent Publication No. 10-2488653, a financial statement DB storing financial statements of an analyzed company for a preset analysis period; a financial understanding module for understanding financial information including net income, net assets, sales, and operating profit for each year of the analyzed company based on the financial statements; a value calculation module for calculating an overvalued value, which is the highest point of the stock value, and an undervalued value, which is the lowest point of the stock value, for each year of the analyzed company based on the financial information; a corporate classification module for classifying multiple companies into corporate groups based on corporate information including industry, business type, and corporate size; A value prediction module including a change understanding unit that graphically understands changes in net income, net assets, sales, and operating profit over time, a financial prediction unit that calculates projected financial information including projected net income, projected net assets, projected sales, and projected operating profit for a specific point in the future based on the changes in the net income, net assets, sales, and operating profit, a value prediction unit that calculates an overvalued value and an undervalued value of the analyzed company for a specific point in the future based on the projected financial information, a trend comparison unit that compares change trends including rising, sideways, and falling graphs for each of the net income, sales, and operating profit, and a projected net income correction unit that corrects and processes the projected net income for the specific point in the future based on the comparison results of the trend comparison unit; A system has been presented which includes a chart display module that displays changes in the overvalued and undervalued values over time as well as the overvalued and undervalued values for a specific point in the future of the analyzed company in a chart; and wherein the expected net income correction unit further reflects a corporate weight differentially set for each corporate group to which the analyzed company belongs in the comparison result of the trend comparison unit, and corrects the expected net income for the specific point in the future according to the high and low of the correction value calculated through the following mathematical expression 1.

That is, as mentioned above, various technologies for evaluating the value of conventional brands have been presented, but the above conventional technologies had the following problems.

Simply evaluating the value of a small and medium-sized (SME) brand based solely on its financial statements is only possible for SME brands that properly manage their financial statements. For brands that do not properly manage their financial statements, it is very difficult to accurately evaluate the value of the brand. In addition, since the value of a brand cannot be represented solely by financial criteria, there is a problem in that the accurate value of the brand cannot be determined.

In addition, while existing valuations focus on small and medium-sized companies and only consider financial, physical and human resources such as assets, liabilities, human resources and brand sales of the company, brand aggregators only consider the value of the company's brand.

Accordingly, there is a need to develop technology to evaluate the value of a brand based on not only the brand's financial data but also non-financial big data.

Evaluating the value of a brand based on such non-financial data is problematic in that there are approximately 20,000 detailed categories of e-commerce products sold, each with different attributes, and it is realistically difficult to objectively separate the evaluations of various customers on products.

In addition, there is a problem of lack of discrimination when evaluating each product with a consistent standard and rating based on customers' personal evaluations.

One aspect of the disclosed invention is to provide an electronic device and method for preprocessing data for web crawling-based learning suitable for training an artificial intelligence model by obtaining structured data having a unified and consistent standard.

In addition, one aspect of the disclosed invention is to provide a web crawling-based review analysis electronic device and method capable of objectifying ratings through sentiment analysis for a specific area such as a product, service, etc. through analysis of product reviews.

In addition, one aspect of the disclosed invention is to provide an artificial intelligence-based brand value evaluation electronic device and method capable of evaluating the comprehensive value of a brand by considering evaluations other than financial criteria, such as awareness and expected sales.

In an electronic device for brand value assessment according to one aspect of the disclosed invention, the electronic device includes a memory for storing product information, review information and brand information, and a processor for processing the product information, the review information and the financial information, wherein the processor, based on the processed product information, inputs the product information into a sales volume prediction model to output a predicted sales volume, and inputs the predicted sales volume, the review information and the brand information into a brand value assessment model to output at least one value index of the brand.

In addition, the processor can input the product information into a first sales prediction model to output a first predicted sales amount, input the product information into a second sales prediction model to output a second predicted sales amount, and output a predicted sales amount based on the first predicted sales amount and the second predicted sales amount.

In addition, the first sales prediction model may be a model learned based on a Temporal Convolutional Network (TCN) algorithm, and the second sales prediction model may be a model learned based on a Long Short Term Memory (LSTM) algorithm.

In addition, the predicted sales volume is time series data regarding the predicted sales volume by predicted date, and outputting the predicted sales volume may include assigning weights to the first predicted sales volume and the second predicted sales volume, and outputting the predicted sales volume based on the first predicted sales volume and the second predicted sales volume to which the weights are assigned.

Additionally, the processor may assign a weight to the first predicted sales amount inversely proportional to the prediction period, and may assign a weight to the second predicted sales amount in proportional to the prediction period.

In addition, the above sales volume prediction model can be learned based on multiple product names corresponding to a category and first item information corresponding to each of the product names.

Additionally, the above sales prediction model can be trained by further including an embedding layer.

In addition, the processor can identify the contribution of data included in the product information input to the sales prediction model through an integrated gradient, and transmit data having a contribution greater than a predetermined value to the user terminal.

In addition, the brand value evaluation model can be trained to output a review index score, a rating index score, a ranking index score, and a profit index score by taking the predicted sales volume, the review information, and the brand information as inputs.

Meanwhile, in a brand value assessment method according to one aspect of the disclosed invention, the method may include a step of storing product information and brand information, a step of inputting the product information into a sales volume prediction model and outputting a predicted sales volume based on processing item information and review information included in the product information, and a step of inputting the predicted sales volume and the brand information into a brand value assessment model and outputting at least one value indicator of the brand.

According to one aspect of the disclosed invention, a data preprocessing electronic device and method for web crawling-based learning suitable for training an artificial intelligence model by obtaining structured data having a unified and consistent standard can be provided.

In addition, according to one aspect of the disclosed invention, a web crawling-based review analysis electronic device and method can be provided that can derive influence on predicting future sales by objectifying ratings through specific areas such as product, service, category-specific attributes, etc. through analysis of product reviews.

In addition, according to one aspect of the disclosed invention, an artificial intelligence-based brand value evaluation electronic device and method can be provided that can evaluate the comprehensive value of a brand by considering evaluations other than financial criteria, such as recognition and expected sales, by utilizing non-financial big data.

Figure 1 is a conceptual diagram explaining an artificial intelligence-based brand value assessment system according to one embodiment.
Figure 2 is a block diagram showing the configuration of an electronic device according to one embodiment.
FIG. 3 is a drawing for explaining product information stored in an electronic device according to one embodiment.
FIG. 4 is a conceptual diagram for explaining time-series data of product information stored by an electronic device according to one embodiment.
FIG. 5 is a diagram for explaining collection information collected by an electronic device according to one embodiment.
FIG. 6 is a drawing for explaining collection information collected by an electronic device according to one embodiment.
FIG. 7 is a diagram for explaining missing values of collected information collected by an electronic device according to one embodiment.
FIG. 8 is a diagram for explaining a sales volume prediction model of an electronic device according to one embodiment.
FIG. 9 is a diagram for explaining review information collected by an electronic device according to one embodiment.
FIG. 10 is a drawing for explaining identification of representative sentences of review information of an electronic device according to one embodiment.
FIG. 11 is a diagram for explaining learning data for learning a rating extraction model of an electronic device according to one embodiment.
FIG. 12 is a drawing for explaining an embodiment of rating extraction of an electronic device according to one embodiment.
Figure 13 is a flowchart for explaining a method for preprocessing learning data according to one embodiment.
Figure 14 is a flowchart for explaining a method for preprocessing learning data according to one embodiment.
Figure 15 is a flowchart for explaining a review analysis method according to one embodiment.
Figure 16 is a flowchart for explaining a brand value evaluation method according to one embodiment.
Figure 17 is a flowchart for explaining a brand value evaluation method according to one embodiment.

Like reference numerals refer to like elements throughout the specification. This specification does not describe all elements of the embodiments, and any content that is general in the technical field to which the disclosed invention belongs or that overlaps between the embodiments is omitted. The terms 'part, module, element, block' used in the specification can be implemented in software or hardware, and according to the embodiments, a plurality of 'parts, modules, elements, blocks' can be implemented as a single element, or a single 'part, module, element, block' can include a plurality of elements.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only a direct connection but also an indirect connection, and an indirect connection includes a connection via a wireless communications network.

Additionally, when a part is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise specifically stated.

Throughout the specification, when it is said that an element is "on" another element, this includes not only cases where the element is in contact with the other element, but also cases where there is another element between the two elements.

The terms first, second, etc. are used to distinguish one component from another, and the components are not limited by the aforementioned terms.

Singular expressions include plural expressions unless the context clearly indicates otherwise.

The identification codes in each step are used for convenience of explanation and do not describe the order of each step. Each step may be performed in a different order than specified unless the context clearly indicates a specific order.

The working principle and embodiments of the disclosed invention will be described with reference to the attached drawings below.

Brand aggregators are trying to secure data using web crawling to obtain data for analyzing small and medium-sized (SME) brands from various perspectives. However, data obtained using the above-mentioned web crawling is unstructured data without a unified and consistent standard, and there is a problem that there is no discrimination for analysis from various perspectives. In particular, in a situation where the same product or various product names are traded online (e-commerce), there is a problem that the same product is treated as different products and stored as data even if data is obtained through web crawling.

The present invention may be intended to solve the problem of performing data analysis without discrimination from unstructured data by obtaining structured data so that data can be analyzed from various viewpoints by performing preprocessing in a form suitable for learning an artificial intelligence model from a web page through web crawling.

In addition, reviews of products and ratings of said products include people's various emotions and subjective ratings based on them. Accordingly, when some people value service, they may give high scores even if they are not satisfied with the product, if they receive high quality service. In other words, the ratings given in this way may be meaningless and indiscriminate if given to a person who prioritizes satisfaction with the product.

The present invention may be intended to solve the problem of meaningless ratings without discrimination by analyzing information about reviews acquired from web pages through web crawling, extracting emotions about products, emotions about services, and overall emotions included in the reviews, thereby revising ratings through more objective criteria that make ratings that are only valid or useless to specific people more objective, and thereby obtaining more objective rating data.

In addition, simply evaluating the value of a brand based solely on the financial statements of small and medium-sized (SME) brands is only possible for SME brands that properly manage their financial statements. In the case of brands that do not properly manage their financial statements, it is very difficult to accurately evaluate the value of the brand. In addition, since the value of a brand cannot be represented solely by financial criteria, there is a problem in that the accurate value of the brand cannot be determined.

The present invention may be intended to solve the above-described problem by performing a multi-faceted brand value evaluation through various indicators such as predicted sales volume predicted through more objective review data and time-series sales volume data.

Figure 1 is a conceptual diagram for explaining an artificial intelligence-based brand value assessment system (1000) (hereinafter, referred to as “this system” for convenience of explanation) according to one embodiment.

Referring to FIG. 1, the system (1000) according to one embodiment of the present disclosure may include an electronic device (100) (hereinafter referred to as “the device” for convenience of explanation) that receives webpage data from a server (200) and processes the webpage data, a server (200) that transmits the webpage data to the electronic device (100), and a user terminal (300) that can transmit/receive data from the server (200) and/or the device (100).

The system (1000) according to one embodiment of the present disclosure can obtain and process data existing in the server (200). That is, the system (1000) can request data from the server (200) and obtain and process the requested data so as to obtain data that is continuously and/or periodically updated in the server (200).

Meanwhile, the system (1000) can store product information including at least one product name and first item information corresponding to the product name. That is, the product information can be in the form of metadata consisting of data corresponding to the product name, and as time-series data, can have a data form accumulated according to the time at which the data acquired by the system (1000) from the server (200) is updated. Meanwhile, the product name here can be named, for example, as a product ID. That is, the product name can be understood as a unique identification code for a specific product.

Meanwhile, the product information may include at least one product name and first item information corresponding to the product name, as will be described in detail below. More specifically, the product information may be divided into groups according to multiple categories and may have information about at least one group. For example, the product information may include multiple products such as peanuts, chestnuts, walnuts, almonds, pine nuts, ginkgo nuts, pistachios, and cashew nuts in the category of nuts. That is, the product information may include a group for at least one category, product names for products belonging to the group, and first item information corresponding to the product name.

For example, the first item information may include at least one of the sales volume, cumulative number of reviews, rank, price, link, rating, shipping fee, category, brand name or manufacturer name, number of likes, release date, etc. for the product name being distributed. Meanwhile, here, the category included in the first item information or the second item information corresponding to the product name may mean, for example, an identification value regarding the category to which the product name belongs. For example, if the first product belongs to the first category, it can be understood that the first item information for the first product may include an attribute (identification value or group) called the first category. Accordingly, the category group to which the product name belongs can be identified through the product name and the category attribute (identification value or group) included in the first item information or the second item information corresponding to the product name.

The system (1000) according to one embodiment of the present disclosure can identify collected information including at least one product name and second item information corresponding to the product name based on processing the acquired data. Meanwhile, the second item information is named only to be distinguished from the first item information of the product information, for example, and is not limited thereto, and can include items for the same information as the first item information. That is, the first item information and the second item information can be named as item information.

For example, if the first item information includes items about sales volume, cumulative number of reviews, and price, the second item information may include items about sales volume, cumulative number of reviews, and price, similar to the first carrier information. Meanwhile, more specifically, the product information may be, for example, time-series data in which the collection information is accumulated. In other words, information accumulated and stored according to the time when the collection information is collected may be the product information.

The system (1000) according to one embodiment of the present disclosure can delete data on duplicate product names based on processing of identified collected information. That is, this can be used to prevent the cumulative generation of duplicate data by filtering data on duplicate product names within web page data.

More specifically, web page data may include multiple product names and item information corresponding to the product names. However, the web page data may include multiple item information for the same product name due to various reasons, such as a seller's advertisement registration. Accordingly, the system (1000) may process duplicate data by deleting item information for the same product name in the collected information based on processing the collected information identified based on processing the web page data, thereby deleting duplicate product names and item information corresponding to the product names.

For example, in a case where a first product and item information corresponding to the first product are included in a first page of web page data, and a first product and item information corresponding to the first product are included in a second page of the web page data, the system (1000) may identify collection information including two first products and item information corresponding to the first product based on the web page data processed. Accordingly, the system (1000) may delete one of the first product names and the item information corresponding to the product names among the two first product names and the item information corresponding to the product names. Accordingly, duplicate data may be filtered and/or deleted.

In addition, it can be understood based on the above description that the present system (1000) can organize duplicate review information by deleting duplicate data based on the author ID among the review information included in the collected information. For example, the present system (1000) can delete review data corresponding to a duplicate author ID among multiple author IDs included in the review information based on processing product information including item information and review information corresponding to a specific product name. However, the present invention is not limited thereto.

The system (1000) according to one embodiment of the present disclosure can determine whether the product information and the collection information are the same product based on processing the pre-stored product information and the collection information. More specifically, determining whether the product information and the collection information are the same product can be determining whether the product information and the collection information are the same product based on processing at least one product name included in the product information and at least one product name included in the collection information.

For example, if the product information includes a product name called a first product and first item information corresponding to the first product, the system (1000) may, based on processing web page data received from the server (200), identify collection information including a product name called a second product and second item information corresponding to the second product, and determine whether the first product and the first product are the same product. Meanwhile, since the product information and the collection information may each include at least one product name and item information corresponding thereto, they may include a plurality of product names and item information corresponding thereto.

In the case of the above-described embodiment, the system (1000) can individually determine whether multiple product names included in the collected information are the same product based on processing multiple product names included in the product information and web page data.

Accordingly, the system (1000) can update the product information in response to the product information and the collection information being determined to be the same product. More specifically, in response to the product name included in the product information and the product name included in the collection information being determined to be the same product, the product information corresponding to the product name can be updated (in time series) and stored based on the collection information.

Meanwhile, the system (1000) may determine whether or not it is the same product, for example, to prevent the use of different product names (such as using additional descriptive words) depending on the supplier (or vendor) even in the case of the same product, thereby preventing the creation of non-standard data.

According to one embodiment of the present disclosure, the system (1000) can identify collection information including at least one product name and at least one review information corresponding to the product name based on processing acquired data. More specifically, the system (1000) can identify collection information including at least one product name and second item information corresponding to the product name, included in the webpage data, based on processing webpage data received from the server (200). That is, the item information is data in the form of metadata corresponding to each product name, and the second item information can include, for example, review information including information regarding a review written by a user of a product corresponding to the product name.

Accordingly, the second item information included in the collection information acquired by the system (1000) may further include review information. More specifically, the review information may include, for example, a plurality of author IDs corresponding to a specific product name and customer review text, product photos, ratings, number of likes, number of posts, posting dates, categories, sales options, and reference ratings corresponding to the IDs. Meanwhile, the reference rating here may be a rating output by the system (1000) inputting review information excluding the reference rating into a rating extraction model, and the output rating may be updated in the review information as the reference rating.

More specifically, the system (1000) can identify at least one representative sentence of the review information based on the processing of the collected information, and input the representative sentence into a rating extraction model described in detail below to extract a rating (reference rating) of the review information.

Meanwhile, the system (1000) according to one embodiment of the present disclosure can input product information into a sales volume prediction model, which will be described in detail below, based on processing pre-stored product information and brand information, and output a predicted sales volume. Accordingly, the system (1000) can input the output (extracted) predicted sales volume and brand information into a brand value evaluation model, which will be described in detail below, and output (extract) at least one value indicator of the brand.

The main device (100), the server (200), and the user terminal (300) of the system (1000) according to one embodiment of the present disclosure can be connected to a network (1). Here, being connected to a network (1) can be understood to mean being electrically or communicatively connected.

Examples of networks (1) include, but are not limited to, 3rd Generation Partnership Project (3GPP) networks (such as 3G networks, 4G or LTE (Long Term Evolution) networks, 5G or NR (New Radio) networks, or 6G networks), WIMAX (World Interoperability for Microwave Access) networks, the Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), Bluetooth networks, satellite broadcasting networks, analog broadcasting networks, and DMB (Digital Multimedia Broadcasting) networks.

The present device (100) can transmit and/or receive data to the server (200). More specifically, the present device (100) can transmit a request message requesting data transmission to the server (200) and receive data corresponding to the request message from the server (200). Accordingly, the present device (100) can identify collection information based on processing the received data. More specifically, the present device (100) can identify collection information including at least one product name and second item information corresponding to the product name based on processing the received webpage data.

Meanwhile, the data that the device (100) requests from the server (200) may include, for example, web page data and/or image data and/or text data. In another embodiment, the data may be, for example, metadata including search results corresponding to a specific keyword. However, the present invention is not limited thereto.

Meanwhile, the request data including web page data that the device (100) acquires from the server (200) may be acquired using, for example, a multi-threading method that uses multiple threads to collect web page data about multiple web pages, a multi-processing method that uses at least two processes to collect web page data about separate web pages, or a distributed crawling method that has multiple devices (100) and independently collects web page data about web pages. However, the present invention is not limited thereto. That is, it can be understood that the web page data that the device (100) acquires from the server (200) may use a known web crawling method and/or a web crawling method to be developed in the future.

The server (200) according to one embodiment of the present disclosure can process and store various types of data, and transmit and/or receive the processed data. More specifically, the server (200) can be mutually connected to the present device (100) and the user terminal (300) via the network (1). Accordingly, the server (2000) can transmit and/or receive data for linking data between each other based on the network (1). The server (200) can be, for example, a platform and/or a web page cloud server, but is not limited thereto.

The server (200) may be configured as an integral part of a database, for example, and may search for data corresponding to a request according to a request of the present device (100), generate requested specific information based on the searched information, and transmit the generated specific information back to the present device (100).

The server (200) may exist as a single server or may be composed of multiple servers. In addition, the server (200) may include at least one processor that processes various types of data.

A user terminal (300) according to one embodiment of the present disclosure may include, but is not limited to, all types of input/output capable wired/wireless communication devices, such as, for example, a PCS (Personal Communication System), a GSM (Global System for Mobile communication), a PDC (Personal Digital Cellular), a PHS (Personal Handyphone System), a PDA (Personal Digital Assistant), an IMT (International Mobile Telecommunication)-2000, a CDMA (Code Division Multiple Access)-2000, a W-CDMA (WCode Division Multiple Access), a Wibro (Wireless Broadband Internet) terminal, a Smartphone, a SmartPad, a tablet PC, a laptop, a wearable device, a digital signage, etc.

In the above, the device (100), the server (200), and the user terminal (300) that constitute the system (1000) have been briefly described. Hereinafter, the configuration of the system (1000) will be described in detail with reference to the drawings. Meanwhile, the embodiments performed by the device (100) described below may also be understood as embodiments performed by the system (1000) including the device (100).

Figure 2 is a block diagram showing the configuration of an electronic device according to one embodiment.

Referring to FIG. 2, the device (100) according to one embodiment of the present disclosure may include a communication unit (110), a control unit (120) including a processor (121) and a memory (122).

A communication unit (110) according to one embodiment of the present disclosure can communicate with a server (200) via a network (1), request data including web page data from the server (200), and receive the requested data from the server (200). In addition, the communication unit (110) according to one embodiment of the present disclosure can be configured with a wireless communication module of a known type.

The control unit (120) may include a processor (121) and a memory (122).

The processor (121) can process data including web page data obtained from the communication unit (110) and product information stored in the memory (122). For example, the processor (121) can include a digital signal processor (DSP) and/or a micro control unit (MCU).

FIG. 3 is a diagram for explaining product information stored in an electronic device according to one embodiment. FIG. 4 is a conceptual diagram for explaining time-series data of product information stored in an electronic device according to one embodiment.

Referring to FIG. 3, a processor (121) according to one embodiment of the present disclosure can process product information (21) pre-stored in a memory (122). More specifically, the product information (21) can include at least one product name and item information (31) corresponding to the product name. The item information (31) can mean detailed information on a product for the product name, for example, in the form of metadata corresponding to the product name.

For example, product information (21) may include item information (31) including a product name and a category corresponding to the product name, a ranking, a sales volume, a price (41), a link, a rating, a shipping fee, a brand name, a manufacturer, a cumulative number of reviews, an acquisition time (a time of acquiring web page data), review information (51), etc.

Product information (21) may be big data in the form of accumulated metadata of collected information including at least one product name and second item information corresponding to the product name, which is identified based on the processor (121) processing web page data acquired from the server (200). That is, product information (21) may be data accumulated in time series according to the time at which the processor (121) acquired the collected information.

Referring to FIG. 4, FIG. 4 is a diagram for explaining time-series data of first item information for at least one product name included in product information. Information on price (41) among product information pre-stored in memory (122) according to one embodiment of the present disclosure may be information accumulated according to the time at which the processor (121) acquires webpage data from the server (200). However, it is not limited thereto, and in another embodiment, information on ranking, sales volume, rating, and accumulated number of reviews included in item information (31) may be information accumulated according to the time at which they were acquired, like the information on price (41) described above. That is, it is not limited to FIG. 4.

FIG. 5 is a diagram for explaining collection information collected by an electronic device according to one embodiment.

Referring to FIG. 5, a processor (121) according to one embodiment of the present disclosure may process webpage data (61) in response to receiving the webpage data (61) from a server (200) through a communication unit (110). Accordingly, the processor (121) may identify collection information including at least one product name and second item information corresponding to the product name based on the processed webpage data (61).

Meanwhile, the collected information is information obtained based on processing web page data received from, for example, at least one server (200), and when web page data obtained from different servers (200) and/or web page data obtained from the same server (200) are received, it may include different product names and item information corresponding to each of the product names for the same product.

For example, referring to (a) of FIG. 5, the processor (121) according to one embodiment of the present disclosure can identify a first product name, 'Lactofit Gold 2gX50', based on processing web page data (61). In addition, referring to (b) of FIG. 5, the processor (121) can identify a second product name, 'Lactofit Gold Probiotics 2gX50', based on processing web page data (61). In this case, the first product name and the second product name have different product names, but are the same product. That is, the web page data (61) may include collection information having different product names for the same product.

Accordingly, the processor (121) can determine whether the product information and the collection information are the same product based on the processing of the stored product information and the identified collection information. More specifically, the processor (121) can identify the similarity between at least one product name included in the product information and at least one product name included in the collection information, and determine whether the product name included in the product information and the product name included in the collection information are the same product based on the similarity. That is, as illustrated in FIG. 5, the processor (121) can identify the similarity between at least one product name included in the product information and a product name included in the identified collection information (for example, the first product name or the second product name).

In more detail, the processor (121) can calculate a sentence embedding of a product name included in product information or collection information.

Sentence embedding refers to converting text into a numeric vector in the field of natural language processing (NLP) technology. For example, the processor (121) may calculate sentence embedding by using at least one of the following: an average pooling method that generates a sentence vector by calculating the average of all word vectors included in a sentence; a max pooling method that generates a sentence vector by using a vector having the largest value among all word vectors included in a sentence; a method that considers the word order in a sentence by using a recurrent neural network architecture such as a recurrent neural network (RNN) or a long short-term memory (LSTM); and a method that utilizes an algorithm such as BERT or GPT, which is a transformer-based model. However, the present invention is not limited thereto, and a publicly known sentence embedding calculation method or a sentence embedding calculation method to be developed in the future may be applied.

According to one embodiment of the present disclosure, the processor (121) can calculate a sentence embedding of at least one product name included in the product information based on the product information processed. That is, the processor (121) can calculate sentence embeddings for each of a plurality of pre-stored product names and store them in the memory (122).

Additionally, the processor (121) can calculate a sentence embedding of at least one product name included in the collected information based on the processing of the collected information.

Accordingly, the processor (121) can calculate cosine similarity between the calculated sentence embedding for at least one product name included in the product information and the calculated sentence embedding for at least one product name included in the collection information.

Cosine similarity is one of the methods of measuring the similarity between vectors based on the angle between the two vectors. The closer it is to 0, the larger the angle between the vectors, and the closer it is to 1, the smaller the angle between the vectors. That is, the processor (121) can identify the similarity between two product names by calculating the cosine similarity between the sentence embeddings of at least one product name included in the product information and at least one product name included in the collection information.

The processor (121) may determine that the product name included in the product information and the product name included in the collection information are the same product in response to the identified similarity (e.g., cosine similarity value) being greater than a predetermined value. Conversely, the processor (121) may determine that the product name included in the product information and the product name included in the collection information are different products in response to the identified similarity being less than a predetermined value.

Meanwhile, the above predetermined value is a value obtained experimentally or empirically, and may have different predetermined values depending on the product or product category. Accordingly, the above predetermined value may be set to determine that the same product with a similar product name is the same product.

Accordingly, the processor (121) calculates sentence embeddings for at least one product name included in the product information and at least one product name included in the collection information, and calculates cosine similarity between the sentence embeddings of the product name included in the product information and the character embeddings of the product name included in the collection information, thereby determining whether the products are the same.

According to one embodiment of the present disclosure, the processor (121) may, in response to determining that the cosine similarity between the sentence embeddings of at least one product name included in the product information and at least one product name included in the collection information is greater than a predetermined value, update the first item information corresponding to the product name included in the product information based on the second item information corresponding to the product name included in the collection information.

For example, the processor (121) can update the price information, which is the first item information corresponding to the first product included in the product information, to the price information, which is the second item information corresponding to the first product included in the collection information. That is, the processor (121) can change (update) the price information of the first item information included in the product information related to the product name determined to be the same product to the price information of the second item information included in the collection information.

In another embodiment, the processor (121) may cumulatively store information on the price, which is the second item information corresponding to the first product included in the collection information in the product information. That is, as illustrated in FIG. 4, the processor (121) may cumulatively store collection information on product information for product names that are determined to be identical to the product, depending on the time of acquisition of the collection information. However, the present invention is not limited thereto.

FIG. 6 is a diagram for explaining collection information collected by an electronic device according to one embodiment. In addition, FIG. 7 is a diagram for explaining missing values of collection information collected by an electronic device according to one embodiment. In addition, FIG. 8 is a diagram for explaining a sales volume prediction model of an electronic device according to one embodiment.

The processor (121) according to one embodiment of the present disclosure can identify the collection information based on processing the web page data received from the server (200). Accordingly, the processor (121) can identify at least one missing value among the second item information included in the collection information based on the processing of the collection information. That is, the web page data may include at least one product name and collection information corresponding to the product name, but if specific information is missing in the web page data, the processor (121) can identify a missing value (null) in the collection information based on processing the identified collection information.

The missing value correction process may be a process to prevent the reliability of an artificial intelligence model from decreasing when training an artificial intelligence model based on collected information that contains missing values, for example. Accordingly, the missing value correction process can produce a highly reliable correction value for a missing value and compensate for the missing value based on the correction value to supplement (supplement) non-existent data.

Meanwhile, the processor (121) may perform a missing value correction process in response to identifying at least one missing value among second items of the collected information by processing the collected information, and identify a correction value based on the processing of first item information corresponding to the missing value. More specifically, the processor (121) may identify a correction value based on the processing of a value corresponding to the missing value among first item information for the specific product name in response to identifying a missing value among second item information for the specific product name based on the processing of the collected information. That is, the processor (121) may identify a correction value based on the processing of first item information of product information regarding a product name of product information that is determined to be identical to a product so as to correspond to an item in which a missing value is identified among second item information included in the collected information.

The processor (121) according to one embodiment may identify the first item information of product information corresponding to the identified missing value based on the processing of the collected information. That is, the first item information of product information corresponding to the identified missing value may identify the value of the first item information of product information corresponding to the specific product name (determined to be the same product) for the item in which the missing value occurs when a missing value is identified among the second item information corresponding to a specific product name in the collected information.

In this case, the processor (121) may identify the value of the most recent item corresponding to the missing value of the collected information among the first item information as the correction value. For example, when the processor (121) identifies a missing value for the rating (one of the aircraft carrier information) based on processing the collected information, in response, the processor may identify the value of the rating among the first item information corresponding to the most recent acquisition time as the correction value, and correct the missing value of the collected information with the correction value. Accordingly, the missing value of the collected information may be corrected with the most recent first item information of the item corresponding to the missing value as the correction value.

In another embodiment, the processor (121) may identify an average value of a predetermined period of first item information corresponding to a missing value as a correction value. More specifically, the processor (121) may identify an average value of values of a predetermined period of shipping cost among the second item information of the collected information as a correction value in response to identifying a missing value regarding a shipping cost among the second item information of the collected information. Accordingly, the processor (121) may correct the missing value of the collected information by using an average value of a predetermined period of shipping cost for an item corresponding to a missing value among the accumulated product information as a correction value.

Meanwhile, the processor (121) according to one embodiment may, in response to a case where there is no value for a predetermined period of first item information corresponding to a missing value of the collected information, identify an average value for a predetermined period of first item information corresponding to the missing value of a category included in the product name as a correction value and correct the missing value based on the processed product information.

More specifically, for example, in the case of a recently launched product, there may be cases where the product information for which pre-stored and accumulated collection information is insufficient. In order to prevent the occurrence of missing values due to this, the processor (121) according to one embodiment may identify the average value of the first item information for a predetermined period corresponding to the missing value of the identified collection information of the category to which the product name belongs based on the processed product information as a correction value.

More specifically, the product information may include a group according to at least one category, a plurality of product names corresponding to the group, and first item information corresponding to each of the product names. For example, the product information may include a plurality of product names included in the nuts category, such as Daily Nuts, Dr. Nuts, and Today Nuts.

That is, according to one embodiment of the present disclosure, the processor (121), in response to identifying a missing value for a specific product name of the collected information, identifies a correction value based on first item information of product information corresponding to the missing value, and in response to the absence of first item information of product information corresponding to the missing value, identifies the correction value based on first item information corresponding to the missing value of a group of categories to which the specific product name belongs.

For example, in response to the price of the first product name being identified as a missing value among the collected information, the processor (121) may identify, based on the processed product information, an average value of the price of at least one of the second product, the third product, or the fourth product of the group to which the first product name belongs during a predetermined period as a correction value, and correct the missing value with the correction value.

In another embodiment, in response to the fact that the price of the first product name among the collected information is identified as a missing value, the processor (121) may identify, based on the processed product information, a value related to the price among the first item information corresponding to at least one of the most recent acquisition points, such as the second product name, the third product name, or the fourth product name of the group to which the first product name belongs, as a correction value, and correct the missing value of the collected information with the correction value.

Referring to FIGS. 6 and 7, a processor (121) according to one embodiment of the present disclosure, in response to an item regarding sales volume (71) being identified as a missing value based on processing of collected information, may calculate a correction value based on data regarding sales volume of a first item of product information determined to be the same product as the product name corresponding to the collected information.

For example, if data on sales volume among the identified collected information is identified as a missing value based on the processor (121) processing web page data, a missing value correction process different from the above-described missing value correction process may be applied. However, it is not limited thereto, and it can be understood that the above-described missing value correction process may be applied.

Meanwhile, the processor (121) according to one embodiment of the present disclosure can identify missing values among the identified collection information based on processing web page data, as well as identify duplicate values and delete the duplicate values (duplicate data). For example, in the web page data, if a specific product name and item information corresponding to the product name are identified as being duplicated (for example, if multiple pieces of data for the same product name are identified), this can be used to prevent duplicate storage of data.

More specifically, in a case where the webpage data includes a first product and item information corresponding to the first product within a first page, and the webpage data includes a first product and item information corresponding to the first product within a second page, the processor (121) may identify, based on the webpage data processed, collection information including two first products and item information corresponding to the first product. Accordingly, the processor (121) may delete one of the two first product names and the item information corresponding to the product names, the first product name and the item information corresponding to the product name. That is, the processor (121) may filter and/or delete duplicate data.

Meanwhile, the processor (121) according to one embodiment of the present disclosure can identify an outlier based on collected information identified based on processing web page data.

Here, an outlier means a value in which, for example, the product name included in the collected information and the item information corresponding to the product name, such as rank, price, and review information (e.g., number of reviews/day, etc.), have an unusual deviation from the average, and may mean data with high reliability and errors in data that are highly correlated with the predicted sales volume among the data input to the sales volume prediction model described below. However, it is not limited thereto.

For example, the processor (121) may identify an outlier based on the processing of collected information identified based on the processing of product information and web page data including a stored product name and item information corresponding to the product name. More specifically, the processor (121) may extract (identify or output) the IQR (interquartile range) of each item of item information corresponding to each product name based on the processing of the product information. That is, the processor (121) may identify the 25% point (Q1) and the 75% point (Q3) in order of the size of the data (value) corresponding to each item among the product name of the product information which is time-series data and the item information corresponding to the product name based on the processing of the product information, and may identify the IQR based on the identified Q1 and Q3 (for example, through the formula Q3 - Q1 = IQR).

For example, when the processor (121) identifies the IQR for the price among the item information of the first product (a specific product) among the product information, the IQR can be identified based on the price at the 25% point (Q1) and the 75% point (Q3) in the order of the size for the price by date. However, the present invention is not limited thereto, and the IQR for each item included in the item information can be identified.

Accordingly, the processor (121) can identify an outlier in the collected information identified based on processing the web page data based on the identified IQR.

More specifically, the processor (121) can identify an upper limit value (e.g., Q3 + 1.5*IQR = upper limit value) and a lower limit value (e.g., Q1 - 1.5*IQR = lower limit value) for each item of item information included in the product information. It can be understood that the above-described formula is exemplary, and the upper limit value and the lower limit value can be determined based on a predetermined weight for the IQR.

In the above-described embodiment, the processor (121) can identify an abnormal value for an item based on an upper limit and/or a lower limit for each item of item information corresponding to a specific product name identified in advance based on processing product information and item information corresponding to a specific product name among a plurality of product names and corresponding item information included in the collected information identified based on processing web page data.

For example, the processor (121) may identify an abnormal value of the first item in response to the fact that, based on processing the product information, an upper limit value of the first item among the item information for the first product is identified as the first upper limit and a lower limit value is identified as the first lower limit, and based on processing the collected information, in response to the fact that, based on processing the collected information, data of the first item for the first product included in the collected information is greater than the first upper limit or less than the first lower limit. Conversely, the processor (121) may identify a normal value in response to the fact that the data of the first item included in the collected information is within the first upper limit and the first lower limit.

Meanwhile, a processor (121) according to one embodiment of the present disclosure may correct data of a specific item corresponding to a specific product in response to data of the specific item being identified as an abnormal value among the collected information identified (acquired) based on processing web page data.

In more detail, the processor (121) can perform K-means clustering on values identified as outliers corresponding to a specific product. For example, k-means clustering can be performed on values identified as outliers with K=2 (divided into two clusters). In this case, since k is fixed to 2, the cluster with a higher average value of the two clusters can be named a major cluster, and the remaining clusters can be named a minor cluster.

Accordingly, the processor (121) can identify a correction value by dividing an abnormal value belonging to a major cluster by the maximum value (max) of the major cluster and then multiplying the result by Q3, and correct the data of an item corresponding to a specific product for which the abnormal value has been identified with the correction value. In addition, the processor (121) can identify a correction value by dividing an abnormal value belonging to a minor cluster by the maximum value (max) of the minor cluster and then multiplying the result by Q3, and correct the data of an item corresponding to a specific product for which the abnormal value has been identified with the correction value. However, the present invention is not limited thereto.

In another embodiment, the processor (121) may correct the data of an item identified as an abnormal value of a specific product included in the collection information to a specific value (e.g., Q3, Q2 (50% point), Q1). In another embodiment, the processor (121) may correct the data of an item identified as an abnormal value of a specific product included in the collection information based on a value generated through random variable generation within the range of Q1 to Q3. However, the present invention is not limited thereto.

FIG. 8 is a diagram for explaining a sales volume prediction model of an electronic device according to one embodiment.

Referring to FIG. 8, a processor (121) according to an embodiment of the present disclosure may perform a missing value correction process in response to identifying at least one missing value among second item information of collected information identified based on processing web page data. More specifically, in response to the identified collected missing value being an item regarding sales volume, the processor (121) may input first item information among product information (21) corresponding to the product name for which the missing value is identified into a sales volume prediction model (81, 82), thereby outputting a predicted sales volume. Accordingly, a correction value may be identified based on the output predicted sales volume (83), and the missing value of the second item information may be corrected based on the correction value.

That is, the processor (121) inputs product information, which is chronological data according to the acquisition time of the product name corresponding to the collection information where the missing value occurs, into a sales prediction model, predicts the sales amount at the acquisition time when the collection information including the missing value is acquired, identifies the predicted sales amount as a correction value, and corrects the missing value based on the correction value.

In more detail, the processor (121) inputs item information and review information included in product information, which are time-series data according to the acquisition time of the product name corresponding to the collected information where the missing value occurs, into a sales prediction model, predicts the sales volume at the acquisition time when the collected information including the missing value is acquired, identifies the predicted sales volume as a correction value, and corrects the missing value based on the correction value.

The processor (121) may include at least one processor, for example, a plurality of processors, for processing various types of data. For example, the processor (121) may include, for example, a learning processor for machine learning, and may train a model composed of an artificial neural network using learning data. Here, the trained artificial neural network may be referred to as a learning model. The learning model may be used to infer a result value for new input data other than learning data, and the inferred value may be used as a basis for judgment for performing a certain operation.

The processor (121) according to one embodiment can train a learning model through a deep learning algorithm. More specifically, the processor (121) can train a learning model through a Temporarily Convolutional Network (TCN) or a Long Short-Term Memory (LSTM) algorithm or an XGBoost algorithm.

Deep learning algorithm is a type of machine learning algorithm, and refers to a modeling technique developed from an artificial neural network modeled after a human neural network. An artificial neural network can be structured into a multi-layer hierarchical structure.

An artificial neural network (ANN) can be composed of a hierarchical structure including an input layer, an output layer, and at least one intermediate layer (or hidden layer) (e.g., a kernel) between the input layer and the output layer. A deep learning algorithm can produce reliable results as a result of learning to optimize the weights of the activation function between layers based on this multi-layer structure.

Deep learning algorithms applicable to the processor (121) according to one embodiment of the present disclosure may include, but are not limited to, a convolutional neural network (CNN) or a recurrent neural network (RNN). It can be understood that different deep learning algorithms may be applied depending on the embodiment.

Convolutional Neural Network (CNN) is characterized by having a structure that extracts data features and identifies patterns of features, unlike techniques that extract knowledge from existing data and perform a learning process. The convolutional neural network (CNN) can be performed through a convolution process and a pooling process. In other words, the convolutional neural network (CNN) can include an algorithm that is configured by combining a convolution layer and a pooling layer. Here, in the convolution layer, a process of extracting data features (e.g., a convolution process (convolution process)) is performed. The convolution process is a process of investigating adjacent components of each component in the data to identify features and deriving the identified features in one sheet, which can effectively reduce the number of parameters as a compression process. In the pooling layer, a process of reducing the size of a layer that has undergone a convolution process (e.g., a pooling process) is performed. The above pooling process can reduce the size of data, offset noise, and provide consistent features in fine details. For example, the above convolutional neural network (CNN) can be used in various fields such as information extraction, sentence classification, and face recognition. Meanwhile, since the convolutional neural network (CNN) is a publicly known technology, a detailed description will be omitted below.

Accordingly, the processor (121) inputs product information corresponding to a product name in which a missing value occurs among the second item information into a sales prediction model, outputs a predicted sales amount at the time of acquisition when the collected information including the missing value is acquired, and identifies the output predicted sales amount as a correction value.

According to one embodiment, the sales prediction model may be an artificial intelligence model learned for each group according to a category, for example. That is, the sales prediction model may be an artificial intelligence model learned for each group according to a category, and may be learned based on a plurality of product names included in each group of each category and first item information corresponding to the product names. That is, the sales prediction model may be a model learned to output a predicted sales volume specialized for a category, by taking as input a plurality of product names corresponding to a category and first item information corresponding to each of the product names.

In another embodiment, the sales volume prediction model may be an artificial intelligence model learned by product name. That is, it learns based on each product name and the first item information corresponding to the product name, learns a characteristic correlation according to a specific product (product name), and can predict sales volume specialized for a specific product.

In more detail, the sales volume prediction model is learned based on the first item information for each product name and a specific period corresponding to the product name, and learns a characteristic correlation according to a specific product (product name, product) ID, so that it can predict sales volume specialized for a specific product and a specific period. For example, the specific period may mean product information for 28 days, but is not limited thereto.

Meanwhile, a sales volume prediction model according to one embodiment of the present disclosure may be a model trained to output a predicted sales volume by taking as input product information including review information and brand information, which will be described in detail below.

Meanwhile, the sales volume prediction model according to one embodiment of the present disclosure may further include an embedding layer in order to learn by distinguishing different trends according to the attributes of a specific product. That is, in addition to having high prediction accuracy for a specific product by learning based on product information for a specific product and/or brand information including inventory information for a specific product, the sales volume prediction model further includes an embedding layer to distinguish and learn product information and/or brand information for a specific product, so that when product information and/or brand information for a specific product is input to a single sales volume prediction model, the result value can be output by distinguishing it from the trends (attributes) of other products. Accordingly, the sales volume prediction model can learn meaning by capturing data-semantic similarities for a specific product well by including an embedding layer.

That is, the sales volume prediction model can output a predicted sales volume for a specific product with higher accuracy when product information and/or brand information for a specific product is input by learning the characteristics of a specific product or the characteristics according to a period. However, it is not limited thereto.

Meanwhile, the sales prediction model according to one embodiment of the present disclosure may be an ensemble model, which is a model learned to output predicted sales by taking as input the output values of multiple models. That is, the sales prediction model may be composed of at least two sales prediction models that are learned by inputting learning data (raw data) so that models utilizing different algorithms have the same input and output. Accordingly, the sales prediction model may include, for example, a third sales prediction model learned to take as input the outputs of the first sales prediction model and the second sales prediction model and output the predicted sales. However, the present invention is not limited thereto.

In another embodiment, the fourth sales prediction model may be trained to take as input the outputs of the first sales prediction model, the second sales prediction model, and the third sales prediction model and output a predicted sales amount.

In another embodiment, referring to FIG. 8, the processor (121) according to one embodiment of the present disclosure may input product information (21) into a first sales prediction model (81) to output a first predicted sales amount, and input the product information (21) into a second sales prediction model (82) to output a second predicted sales amount.

Accordingly, the processor (121) can identify (output) the predicted sales amount based on the output first predicted sales amount and second predicted sales amount.

According to one embodiment, the first sales prediction model (81) is a model learned based on a Temporal Convolutional Network (TCN) algorithm, and the second sales prediction model (82) is a model learned based on a Long Short Term Memory (LSTM) algorithm. However, the present invention is not limited thereto. Meanwhile, in another embodiment, any one of the first to third sales prediction models may be a model learned based on, for example, an Extreme Gradient Boosting (XGBoost) algorithm. The TCN algorithm and the LSTM algorithm according to one embodiment of the present invention are deep learning models for processing continuous (time-series) data. More specifically, the TCN is a model that applies a convolutional neural network (CNN) to sequence data, and is an algorithm designed to learn temporal dependency. The TCN can process sequences of arbitrary length, effectively learn long temporal dependencies, and generally provide faster learning and parallel processing than the LSTM. The main feature of TCN is that it learns the patterns of the input sequence using convolutional layers, and by increasing the depth, it can identify long sequence dependencies, that is, correlations between time and data.

In addition, LSTM is a type of recurrent neural network (RNN) and is mainly used to process sequence data and time series data. More specifically, LSTM has an internal state that manages "short-term memory" and "long-term memory", so it can easily learn long-term dependencies, that is, correlations between time and data.

The processor (121) according to one embodiment of the present disclosure can train the first sales prediction model (81), the second sales prediction model (82), and the third sales prediction model (not shown) to have the same input and output as the above-described sales prediction models. In another embodiment, the third sales prediction model trained based on the XGBoost algorithm may be trained with only the sales amount according to time (period) of a specific product as input, unlike the first and second sales prediction models.

In another embodiment, a sales prediction model learned based on the XGBoost algorithm is learned based on the price, average rating (or average of standard ratings), cumulative number of reviews, number of likes, ranking, and sales included in the product information of a specific product, and can output a predicted sales volume by taking the price, average rating (or average of standard ratings), cumulative number of reviews, and number of likes of a specific product as input.

More specifically, the processor (121) can analyze and process data based on raw data regarding the first item information of the product information (here, the raw data can be understood as learning data) to build a first sales prediction model, a second sales prediction model, and a third sales prediction model. That is, the processor (121) can train the first sales prediction model, the second sales prediction model, or the third sales prediction model to output time series data regarding the predicted sales for each predicted date when the first item information of the product information is input. Meanwhile, the first sales prediction model learned through TCN is capable of precise prediction for each predicted date, and in the case of LSTM, prediction is focused on predicting trends (e.g., rising, maintaining, falling) rather than precise prediction for each day. The third sales prediction model learned through XGBoost makes predictions with high accuracy even with a small amount of raw data (learning data).

Based on these characteristics, the processor (121) according to one embodiment of the present disclosure can input product information regarding a specific product into a first sales prediction model and a second sales prediction model, assign weights to the output first predicted sales amount and second predicted sales amount, and output (identify) a (final) predicted sales amount (83) based on the first predicted sales amount and second predicted sales amount to which the weights are assigned.

More specifically, the processor (121) can assign a weight to the first predicted sales amount inversely proportional to the prediction period, which is the period between the predicted date to be predicted and the current date, and can assign a weight to the second predicted sales amount in proportional to the prediction period. Accordingly, the processor (121) can output (identify) the (final) predicted sales amount (83) based on the first predicted sales amount and the second predicted sales amount to which the weights are assigned.

In another embodiment, when the sales prediction model further includes a third sales prediction model, the processor (121) may input product information about a specific product into the first sales prediction model, the second sales prediction model, and the third sales prediction model, assign weights to the output first predicted sales, second predicted sales, and third predicted sales, and output (identify) a (final) predicted sales based on the first to third predicted sales to which the weights are assigned.

Meanwhile, the processor (121) according to one embodiment of the present disclosure can train a sales prediction model to output a predicted sales amount by inputting product information (21) and/or brand information. That is, the processor (121) can train a correlation between items included in the first item information and inventory information (for example, a relationship between a standard rating included in the item information and review information and inventory information) and a correlation between time and sales amount by inputting product information and inventory information included in the brand information regarding a specific product name, so that a predicted sales amount is output.

More specifically, the processor (121) may calculate (produce) the lost sales for each day that would have been sold if there had been actual inventory based on the quantity that could have been sold if there had been inventory for each day during the period, based on the inventory information processed, if there is no inventory for a certain period of time. In this case, the processor (121) may learn the correlation between the items included in the first item information and the inventory information (for example, the relationship between the standard rating included in the item information and the review information and the inventory information) and the correlation between time and the sales amount so that the predicted sales amount is output by inputting product information including item information and review information and the lost sales that would have been sold if there had been actual inventory. However, the present invention is not limited thereto. Meanwhile, the processor (121) according to one embodiment of the present disclosure may identify the contribution of data included in the product information input to the sales amount prediction model through an integrated gradient. That is, the processor (121) can identify the degree (contribution) to which each data value of the first item information included in the product information contributes to the sales volume prediction model through the integrated gradient.

Here, the integrated gradient is one of the methods for interpreting the contribution of input features in a machine learning model, and is a method for determining how important a certain input plays a role in the prediction of an artificial intelligence model for the output. More specifically, the integrated gradient may include a step of setting model input data and a baseline value, which determine the input data and the input configuration of the model and set the baseline value as the starting point of the input data, a step of dividing the path from the baseline to the actual input into equal sections to determine important points, a step of calculating the gradient of the model at the midpoint of each section, and a step of generating an importance score indicating the importance of the input data by integrating the calculated gradient.

That is, the processor (121) according to one embodiment of the present disclosure can identify the contribution (importance) of the first item information data included in the product information to the predicted sales amount through the integrated gradient for the sales amount prediction model. Accordingly, it is possible to determine which data (number of reviews, rating, etc.) among the product information contributes to the predicted sales amount.

Meanwhile, the processor (121) according to one embodiment of the present disclosure can control the communication unit (110) to obtain the contribution of each of the first item information included in the product information obtained through the integrated gradient of the sales volume prediction model, and transmit the item and the value of the item whose contribution is greater than a predetermined value to the user terminal (300). Accordingly, the user can identify whether a specific product or a specific element (any one of the first item information) of a specific category has a large influence on the sales volume (sales). However, the present invention is not limited thereto.

FIG. 9 is a diagram for explaining review information collected by an electronic device according to one embodiment. In addition, FIG. 10 is a diagram for explaining identifying representative sentences of review information of an electronic device according to one embodiment.

Referring to FIGS. 6 and 9, the processor (121) according to one embodiment of the present disclosure may identify collection information including at least one product name and at least one review information (51) corresponding to the product name based on processing web page data acquired from the server (200). The review information (51) may be metadata including, for example, an author ID for the product name and a rating corresponding to the ID, customer review text, a product photo, the number of likes, the number of likes, the posting date, the category, the sales option, etc., corresponding to the product name. However, this is merely exemplary, and in another embodiment, the review information (51) may include a plurality of author IDs for the product name and ratings and reviews corresponding to the ID. However, for convenience of explanation, the following description will assume that the processor (121) identifies collection information including one product name and one review information corresponding to the product name based on processing web page data, as illustrated in FIG. 9.

According to one embodiment of the present disclosure, the processor (121) may, in response to identifying the collection information, identify at least one representative sentence of the review information based on processing the collection information. For example, when the review information includes a plurality of reviews, the processor (121) may identify at least one representative sentence for each review.

More specifically, the processor (121) may identify at least one representative sentence by decomposing the review information into a plurality of sentences through n-grams, identifying the similarity between each of the decomposed plurality of sentences and the review information, and identifying a predetermined number of sentences among the plurality of sentences having a high degree of identified similarity.

Referring to FIG. 10, an n-gram according to one embodiment is a method of dividing a text into n consecutive words or characters, where n is a parameter that determines the size of the n-gram and may be a predetermined value. Accordingly, the processor (121) according to one embodiment may decompose the review information into a plurality of sentences into n consecutive words through the n-gram based on the review information processed.

Thereafter, the processor (121) can identify the similarity between each of the decomposed sentences and the reviews of the entire review information (51). Here, identifying the similarity can be implemented through the sentence embedding described above. More specifically, the processor (121) according to one embodiment can calculate sentence embedding for each of the decomposed sentences and the reviews of the review information (51). Accordingly, the processor (121) can identify the cosine similarity between each of the decomposed sentences and the review information (51).

For example, as illustrated in FIG. 10, the processor (121) may decompose the review information (51) into the first to fifth sentences through n-grams. In this case, the processor (121) may calculate the sentence embeddings of the first to fifth sentences and the reviews included in the review information (51). Thereafter, the processor (121) may identify the cosine similarity between the sentence embeddings of the first to fifth sentences and the sentence embeddings of the review information (51). Accordingly, the processor (121) may identify the similarity between each of the decomposed first to fifth sentences and the review information (51), and identify a predetermined number of sentences in the order of high similarity among the plurality of sentences, thereby identifying at least one representative sentence. As illustrated, it can be understood that the predetermined number is set to three.

Meanwhile, the processor (121) according to one embodiment of the present disclosure may input at least one identified representative sentence into a rating extraction model to extract (output) a rating of review information. That is, since a review and a rating corresponding to the review include various emotions of people and subjective ratings accordingly, when some people value a specific area (e.g., quality of a product or quality of a service), even if the satisfaction level for an area other than the specific area is low, the person may give a high score if the specific area is given a high rating. That is, the rating given in this way may become a meaningless rating without discrimination, so in order to solve this problem, the processor (121) may newly extract (output) a rating for the review information (51). The rating newly output in this way is named a reference rating. That is, the reference rating may mean a rating output by inputting review information into the rating extraction model.

FIG. 11 is a diagram for explaining learning data for learning a rating extraction model of an electronic device according to one embodiment. Also, FIG. 12 is a diagram for explaining an embodiment of rating extraction of an electronic device according to one embodiment.

Referring to FIGS. 11 and 12, a processor (121) according to an embodiment of the present disclosure may input a plurality of review information (51) and overall review emotion (overall emotion), emotion about a product (product emotion), emotion about a service (service emotion), review type, and rating corresponding to each of the review information (51) to train a rating extraction model (52). That is, the rating extraction model (52) is trained based on review information (51) and raw data (learning data) obtained by a person directly evaluating and labeling the review information, and when review information including a sentence about a review is input, it may output at least one of overall emotion, product emotion, service emotion, review type, or rating corresponding to the sentence (review). Meanwhile, overall emotion, product emotion, and service emotion may be labeled as an evaluation of any one of positive, negative, or neutral emotions, for example. Additionally, for review types, they can be labeled with an evaluation of either a simple review based on the length of the review, an authentic review, or an artificial review containing redundant content.

In this case, the processor (121) identifies at least one representative sentence of the review information based on processing the collected information including the review information (51) as described above, and inputs the at least one representative sentence into the rating extraction model (52) to output at least one of the overall sentiment, product sentiment, sentiment for the service, review type, or rating.

In one embodiment, the processor (121) may extract a rating of review information based on a first score output by inputting the first sentence into the rating extraction model (52) and a second score output by inputting the second sentence into the rating extraction model, for example, if the representative sentence includes a first sentence and a second sentence. More specifically, the processor (121) may extract (output) an average of the first score and the second score as a rating of the review score.

For example, the rating extraction model (52) may be an ensemble model including a plurality of rating extraction models. The rating extraction model (52) according to one embodiment may include a first model and a second model. More specifically, the rating extraction model may include at least one of a first model or a second model trained with a learning data set including sentences about reviews and ratings corresponding to the sentences as input so that when review information (51) including sentences about reviews is input, a rating corresponding to the sentences is output.

The first model and the second model are artificial intelligence models learned based on different algorithms. For example, the first model or the second model may be one of the KeyBERT algorithm, the KoBERT algorithm, the DeBERTa algorithm, or the XLNet algorithm.

In the above-described example, the processor (121) can identify review information based on processing the collected information, and input at least one representative sentence identified based on processing the review information into the first model and the second model to obtain (extract or output) a first output and a second output, respectively. In this case, the processor (121) can extract a rating of the review information based on the first output and the second output.

More specifically, the processor (121) can input at least one representative sentence into the first model to output (obtain) a first output regarding the rating. In addition, the processor (121) can input at least one representative sentence into the second model to output (obtain) a second output regarding the rating. Accordingly, the processor (121) according to one embodiment can identify (extract) an average value of the first output and the second output as the rating.

Meanwhile, in another embodiment, the processor (121) may input at least one representative sentence into the first model, the second model, and the third model learned through an algorithm different from the first model and the second model, and obtain a first output, a second output, and a third output each including at least one of an overall sentiment, a product sentiment, a service sentiment, a review type, or a rating for the at least one representative sentence. In this case, the processor (121) may extract (output) the final overall sentiment based on the first output, the second output, and the third output corresponding to the overall sentiment. Similarly, the processor (121) may output the final product sentiment, the final service sentiment, the final review type, or the final rating based on the first to third models described above with respect to the product sentiment, the service sentiment, the review type, or the rating.

That is, based on the first to third outputs described above, outputting the final overall sentiment means that, for example, in response to the first, second, and third outputs being positive, positive, and negative for the overall sentiment, the absolute majority sentiment is output, so that the final overall sentiment can be output as positive. Similarly, based on the absolute majority sentiment or value (data) of the first to third outputs, the final product sentiment, the final service sentiment, or the final review type can be extracted. However, the present invention is not limited thereto, and in the case of the rating, the final rating can be output according to the average value of the first to third outputs.

Meanwhile, in the above-described embodiment, in response to the absence of an absolute majority, the processor (121) may extract (output) the result of the first output as the final result. For example, in response to the input of a representative sentence and the first output, the second output, and the third output regarding the overall sentiment being positive, neutral, and negative, respectively, the processor (121) may extract (output) the positive result of the first output, which is the output of the first model, as the overall sentiment of the input representative sentence.

In another embodiment, when the representative sentence includes, for example, a first sentence, a second sentence, and a third sentence, the processor (121) can input the first sentence into the first model, the second model, and the third model learned through an algorithm different from the first and second models described above, and obtain the first output to the third output obtained. In addition, the processor (121) can input the second sentence into the third model learned through an algorithm different from the first and second models described above, and obtain the first output to the third output obtained. In addition, the processor (121) can input the third sentence into the first model, the second model, and the third model learned through an algorithm different from the first and second models described above, and obtain the first output to the third output obtained.

In this case, the processor (121) can extract (output) the final overall sentiment, the final product sentiment, the final service sentiment, the final review type, or the final rating based on the first output to the third output obtained by inputting the first sentence to the first model to the third model. In addition, the processor (121) can extract (output) the final overall sentiment, the final product sentiment, the final service sentiment, the final review type, or the final rating based on the first output to the third output obtained by inputting the second sentence to the first model to the third model. That is, the processor (121) can extract (output) the final overall sentiment, the final product sentiment, the final service sentiment, the final review type, or the final rating based on the first output to the third output obtained by inputting the first model to the third model for each of at least one identified representative sentence of the review included in the review information.

Accordingly, the processor (121) can extract (output) the absolute majority of emotions or data of the final overall emotion, final product emotion, final service emotion, or final review type for each representative sentence as the overall emotion, product emotion, service emotion, or review type of the review information. However, it is not limited thereto, and in the case of the rating, the rating can be output according to the average value of the final rating for each representative sentence.

Meanwhile, it can be understood that the above-described rating extraction model (52) can be applied with a known natural language processing algorithm or a natural language processing algorithm to be developed in the future in addition to the KeyBERT algorithm, the KoBERT algorithm, the DeBERTa algorithm, or the XLNet algorithm, and it can be understood that a plurality of models can be applied in addition to the first to third models to output the overall sentiment, product sentiment, service sentiment, review type, or rating of the review information based on the output of each model.

According to one embodiment of the present disclosure, the processor (121) can identify a product name included in the collection information, at least one piece of review information corresponding to the product name, as described in the above-described embodiment, and extract a rating of the review information based on the processed review information. That is, the processor (121) can extract an objective standard rating separately from the ratings of customers included in the review information based on the text regarding the review included in the review information.

In more detail, the processor (121) inputs review information including an author ID for a product name and a rating corresponding to the ID, a customer review text, a product photo, the number of likes, the number of likes, the posting date, the category, the sales options, etc., included in the identified collection information, into a rating extraction model to extract a reference rating, and by adding a reference rating item to the review information, the reference rating for the review information can be updated. That is, the processor (121) can update the review information by adding a reference rating item to the review information, based on the processing of the review information included in the collection information. According to one embodiment of the present disclosure, in the collection information including an acquired product name and first item information corresponding to the product name, the first item information can include review information. However, the review information herein only includes rating information given by customers, and does not include a reference rating acquired through the rating extraction model. In this case, when updating pre-stored product information, the processor (121) may, for example, update a standard score obtained by inputting review information regarding a specific product name by newly creating a standard score item as an additional item to the first item information regarding the specific product name in the form of metadata.

Meanwhile, the processor (121) according to one embodiment of the present disclosure may input product information about at least one product name of a specific brand into the aforementioned sales prediction model, input the output predicted sales, review information corresponding to the at least one product name, and brand information pre-stored in the memory (122) or acquired through the communication unit (110) into the brand value evaluation model, thereby outputting at least one value index of the brand. In more detail, the processor (121) may input product information about at least one product name of a specific brand into the aforementioned sales prediction model, input the output predicted sales, input review information corresponding to the at least one product name into the aforementioned rating extraction model, and input the output overall sentiment, product sentiment, service sentiment, review type or standard rating, and brand information into the brand value evaluation model, thereby outputting at least one value index of the brand.

Here, brand information includes, for example, transaction information including a transaction period, sales information including weekly average sales, weekly sales volatility and weekly sales ratio volatility compared to annual sales, sales channel information including the number of sales channels and sales distribution by sales channel, customer information including the number of daily transaction customers, the number of loyal customers and new customers, reputation information including return rate and number of claims, financial information including cost of sales rate, advertising cost rate, logistics cost rate, service cost rate, inventory amount compared to sales and accounts receivable collection lead time, customer rating, total number of customer reviews, number of customer reviews by type, overall sentiment by review, number of SNS reviews, number of SNS reviews by type, overall sentiment by SNS review and eWOM information including seller ranking in the corresponding category, inventory information including lost GMV, out-of-stock rate, number of days in stock, turnover rate and order delivery rate, product information including number of SKUs, sales per SKU, number of new SKU launches, SKU life cycle, number of SKU clicks, dwell time on SKU detail page, order cancellation rate by SKU and delivery lead time by SKU, weekly average price, weekly average price It may include at least one of price information including volatility and promotional events and other information including seasonality.

Meanwhile, the processor (121) according to one embodiment of the present disclosure may train a brand value evaluation model based on learning data obtained by labeling predicted sales volume, the entire sentiment of review information corresponding to at least one product name of a specific brand, product sentiment, service sentiment, review type or rating, and brand information and a person directly evaluating and labeling the specific brand. Meanwhile, the learning data may include, for example, a review index score, a rating index score, a ranking index score, and a profit index score. That is, the brand value evaluation model may be an artificial intelligence model trained to output a review index score, a rating index score, a ranking index score, and a profit index score by inputting the predicted sales volume, the review information, and the brand information. Meanwhile, the above-described index scores may be labeled with scores corresponding to 0 to 10, respectively.

Meanwhile, the processor (121) according to one embodiment of the present disclosure can identify, through an integrated gradient for a brand value evaluation model, a predicted sales volume for at least one product name of a specific brand, an overall sentiment of review information corresponding to at least one product name of a specific brand, a product sentiment, a service sentiment, a review type or rating, and a contribution (importance) for each indicator (review indicator, rating indicator, ranking indicator, profit indicator) among brand information. Accordingly, a user can identify which improvement can increase the value of the brand. However, the present invention is not limited thereto.

In another embodiment, the processor (121) may control the communication unit (110) to obtain the contribution of each input piece of information through the integrated gradient of the brand value evaluation model and transmit items with a contribution greater than a predetermined value and the values of the items to the user terminal (300).

The memory (122) can store a program that performs the operations described above and the operations described below, and the processor (121) can execute the stored program. In the case where there are a plurality of memories (562) and processors (121), they can be integrated into one chip or provided in physically separate locations. The memory (562) can include volatile memory such as S-RAM (Static Random Access Memory) and D-RAM (Dynamic Random Access Memory) for temporarily storing data. In addition, the memory (562) can include nonvolatile memory such as ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory), and EEPROM (Electrically Erasable Programmable Read Only Memory) for long-term storing a control program and control data. The processor (121) can include various logic circuits and operation circuits, process data according to a program provided from the memory (562), and generate a control signal according to the processing result.

The memory (122) according to one embodiment of the present disclosure can store the product information, collection information, review information, and brand information described above.

Figures 13 and 14 are flowcharts for explaining a learning data preprocessing method according to one embodiment. Also, Figure 15 is a flowchart for explaining a review analysis method according to one embodiment. Figures 16 and 17 are flowcharts for explaining a brand value evaluation method according to one embodiment.

The methods illustrated in FIGS. 13 to 17 can be performed by the system (1000) or the device (100) described above. Therefore, even if the contents are omitted below, the contents described for the system (1000) and the device (100) can be equally applied to the methods described below.

Referring to FIG. 13, the device (100) can receive webpage data (S110).

Additionally, the device (100) can identify collected information based on processing web page data (S120).

In addition, the device (100) can determine whether the product information and the collection information are the same product based on the product information and the collection information processed (S130).

Additionally, the device (100) can update product information in response to being determined to be the same product (S140).

Referring to FIG. 14, the device (100) can calculate sentence embeddings of each product name included in the product information and collection information based on processing the product information and collection information (S210).

Additionally, the device (100) can identify cosine similarity between each sentence embedding (S220).

Additionally, the device (100) can determine whether the identified cosine similarity is greater than a predetermined value (S230).

In this case, the device (100) can determine that the product is the same in response to the identified cosine similarity being greater than a predetermined value (S240).

Meanwhile, the device (100) can determine that the product is a different product in response to the identified cosine similarity being smaller than a predetermined value (S250).

Referring to FIG. 15, the device (100) can receive webpage data (S310).

Additionally, the device (100) can identify collection information including review information based on processing web page data (S320).

Additionally, the device (100) can identify at least one representative sentence of the review information based on the processed collected information (S330).

In addition, the device (100) can extract a rating of review information by inputting at least one identified representative sentence into a rating extraction model (S340).

Referring to FIG. 16, the device (100) can store product information, review information, and brand information (S410).

In addition, the device (100) can input the product information into a sales volume prediction model and output a predicted sales volume based on the processed product information (S420).

In addition, the device (100) can input predicted sales volume, review information, and brand information into a brand value evaluation model to output at least one value indicator of the brand (S430).

Referring to FIG. 17, the device (100) can store product information (S510).

In addition, the device (100) can input product information into the first sales prediction model and output the first predicted sales amount (S520).

In this case, the device (100) can assign a weight to the first predicted sales amount inversely proportional to the prediction period (S530).

Meanwhile, the device (100) can input product information into a second sales prediction model and output a second predicted sales amount (S540).

In this case, the device (100) can assign a weight to the second predicted sales amount in proportion to the prediction period (S550).

In addition, the device (100) can output a predicted sales amount based on the weighted first predicted sales amount and the second predicted sales amount (S560).

Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium storing instructions executable by a computer. The instructions may be stored in the form of program codes, and when executed by a processor, may generate program modules to perform the operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

Computer-readable storage media include all types of storage media that store instructions that can be deciphered by a computer. Examples include ROM (Read Only Memory), RAM (Random Access Memory), magnetic tape, magnetic disk, flash memory, and optical data storage devices.

As described above, the disclosed embodiments have been described with reference to the attached drawings. Those skilled in the art to which the present invention pertains will understand that the present invention can be implemented in forms other than the disclosed embodiments without changing the technical idea or essential features of the present invention. The disclosed embodiments are exemplary and should not be construed as limiting.

1000: This system
100: Electronic devices
110: Communications Department
120: Control Unit
200: Server 300: User Terminal

Claims (19)

In brand value assessment electronic devices,
Memory for storing product information and brand information; and
A processor for processing the above product information and the above brand information; including:
The above processor,
Based on the processing of the item information and review information included in the above product information, the above product information is input into a sales volume prediction model to output the predicted sales volume.
Inputting the above predicted sales volume and the above brand information into a brand value evaluation model to output at least one value indicator of the brand.
Electronic devices. In the first paragraph,
The above processor,
Input the above product information into the first sales prediction model to output the first predicted sales volume,
Input the above product information into the second sales prediction model to output the second predicted sales volume,
Based on the first predicted sales amount and the second predicted sales amount, outputting the predicted sales amount.
Electronic devices. In the second paragraph,
The above first sales prediction model is a model learned based on the TCN (Temporal Convolutional Network) algorithm, and the above second sales prediction model is a model learned based on the LSTM (Long Short Term Memory) algorithm.
Electronic devices. In the third paragraph,
The above predicted sales volume is,
This is time series data on predicted sales by predicted date,
Outputting the above predicted sales volume is:
A method of assigning weights to the first predicted sales amount and the second predicted sales amount, and outputting the predicted sales amount based on the first predicted sales amount and the second predicted sales amount to which the weights are assigned.
Electronic devices. In paragraph 4,
The above processor,
The weight of the first predicted sales amount is given inversely to the prediction period, and the weight of the second predicted sales amount is given in proportion to the prediction period.
Electronic devices. In the first paragraph,
The above sales prediction model is,
Learned based on multiple product names corresponding to the category and the first item information corresponding to each of the product names
Electronic devices. In the first paragraph,
The above sales prediction model is,
Learned by including an embedding layer,
Electronic devices. In the first paragraph,
The above processor,
Identify the contribution of data included in the product information input to the sales prediction model through the integrated gradient,
Transmitting data having a contribution greater than a predetermined value to the user terminal;
Electronic devices. In the first paragraph,
The above brand value evaluation model is,
Trained to output review index scores, rating index scores, ranking index scores, and revenue index scores by inputting the above predicted sales volume, the above review information, and the above brand information.
Electronic devices. In terms of brand value evaluation methods,
Step for storing product information and brand information;
A step of inputting the product information into a sales volume prediction model and outputting the predicted sales volume based on processing the item information and review information included in the product information; and
A step of inputting the predicted sales volume and the brand information into a brand value evaluation model to output at least one value indicator of the brand;
method. In accordance with Article 10,
Outputting the above predicted sales volume is:
Input the above product information into the first sales prediction model to output the first predicted sales volume,
Input the above product information into the second sales prediction model to output the second predicted sales volume,
Based on the first predicted sales amount and the second predicted sales amount, outputting the predicted sales amount.
method. In Article 11,
The above first sales prediction model is a model learned based on the TCN (Temporal Convolutional Network) algorithm, and the above second sales prediction model is a model learned based on the LSTM (Long Short Term Memory) algorithm.
method. In Article 12,
The above predicted sales volume is,
This is time series data on predicted sales by predicted date,
Outputting the above predicted sales volume is:
A method of assigning weights to the first predicted sales amount and the second predicted sales amount, and outputting the predicted sales amount based on the first predicted sales amount and the second predicted sales amount to which the weights are assigned.
method. In Article 13,
The above method,
A step of further including: assigning a weight to the first predicted sales amount inversely proportional to the prediction period, and assigning a weight to the second predicted sales amount in proportional to the prediction period;
method. In Article 10,
The above sales prediction model is,
Learned based on multiple product names corresponding to a category and first item information corresponding to each of the product names.
method. In Article 10,
The above sales prediction model is,
Learned by including an embedding layer,
method. In Article 10,
The above method,
A step for identifying the contribution of data included in the product information input to the sales prediction model through an integrated gradient; and
Further comprising a step of transmitting data having a contribution greater than a predetermined value to the user terminal;
method. In Article 10,
The above brand value evaluation model is,
Trained to output review index scores, rating index scores, ranking index scores, and revenue index scores by inputting the above predicted sales volume, the above review information, and the above brand information.
method. A computer-readable recording medium having recorded thereon a program capable of executing the big data-based intelligent inflow sewage treatment method of Article 10.

Images