KR102576231B1 - Natural language processing method for identification of counterfeit products and counterfeit product detection system - Google Patents

Abstract

The counterfeit detection system according to the embodiment proposes keyword detection logic suggesting counterfeit products to efficiently identify counterfeit products online. In the embodiment, the brand's online transaction products and prices are clustered, and keywords directly related to the product are indexed among product titles in the price category where counterfeit products are suspected. Afterwards, among keywords directly related to the product, words with characteristics similar to the brand name are detected. In the embodiment, when detecting a counterfeit product using a single condition, many inaccurate results may be included, so the probability of detecting a counterfeit product is increased by using logic that combines two or more conditions to increase accuracy.

Description

Natural language processing method and counterfeit detection system for identifying counterfeit products {NATURAL LANGUAGE PROCESSING METHOD FOR IDENTIFICATION OF COUNTERFEIT PRODUCTS AND COUNTERFEIT PRODUCT DETECTION SYSTEM}

This disclosure relates to a technology related to a natural language processing method for identifying counterfeit products online and a counterfeit product detection system using the same. Specifically, it relates to a system and method for efficiently identifying counterfeit products sold online through natural language processing specialized for counterfeit products.

Unless otherwise indicated herein, the material described in this section is not prior art to the claims of this application, and is not admitted to be prior art by inclusion in this section.

As of 2017, the domestic counterfeit product sales amounted to 26 trillion won and continues to grow every year. However, the actual crackdown performance is 68.2 billion won, which is only 0.26% of the total market. The size of the counterfeit product market is growing due to the reimportation of counterfeit products from domestic brands through overseas direct purchases. In addition, it is being organized in the form of a livelihood replica through an open market using SNS. In addition, guerrilla sales such as operating individual stores of large quantities of counterfeit products using the automatic registration function of open markets, selling during early morning hours, and operating closed stores are frequently occurring, and the intelligent sales behavior of counterfeit products is continuously expanding.

Recently, the production cycle of counterfeit products for popular domestic brands has been shortened due to the spread of Korean Wave content, and young people with weak purchasing power are becoming less aware of the problem of using counterfeit products. In addition, it takes an average of 125MM (20,000H) just to review online products, making it virtually impossible to crack down on the sale of complex and intelligent counterfeit products with human resources. For this reason, counterfeit product search programs have been used both domestically and internationally to crack down on the sale of counterfeit products online.

Conventional counterfeit search programs utilize the information extraction function of natural language processing technology. The information extraction function is divided into entity name recognition and relationship extraction. In order to extract words or keywords with a certain purpose, meaningful words are conventionally extracted through morphological analysis of entity name recognition. However, the keyword extraction method through morphological analysis is often not suitable for the purpose of detecting counterfeit products. For example, when a keyword is created by intentionally inserting a space in the brand name, such as “Chanel” or “Balenciaga,” a conventional counterfeit search program separates before and after the space and recognizes it as a different word and meaning, resulting in an expanded keyword. It is impossible to search for counterfeit products using .

1. Korean Patent Publication No. 10-1777981 (September 13, 2017) 2. Korean Patent Publication No. 10-1612291 (April 7, 2016)

The natural language processing method and counterfeit detection system for identifying counterfeit products according to an embodiment clusters products and prices traded online of a brand and indexes keywords directly related to the product among product titles in the price category where counterfeit products are suspected. do. Afterwards, words with similar characteristics to the brand name are detected among relevant keywords.

Since many inaccurate results are included when detecting a counterfeit product using a single condition, in the embodiment, the probability of detecting a counterfeit product is improved by using logic that combines two or more counterfeit identification conditions to increase the accuracy of counterfeit product identification. At this time, the elements for identifying counterfeit products include keywords extracted from the title, content or comments in the text, sellers with a history of selling counterfeit products (blacklist), characteristic prices of counterfeit products, and images with a characteristic tendency of counterfeit products. This may be included.

Additionally, the embodiment provides a platform that facilitates sanctions after the user identifies, confirms, and classifies the detected potential counterfeit products.

In addition, by applying trademark surveillance technology to extract a chunk of text that is related to the brand's appearance, title, concept, etc., patterns of keywords suggesting counterfeit products that are used irregularly are extracted.

In the natural language processing method for identifying counterfeit product sales webs according to an embodiment, (A) performing web crawling by setting brand name and product line as search conditions to extract samples of counterfeit web pages in a preprocessing module; (B) Setting the initial search term based on the search term related to the brand name searched in the search engine in the search term setting module; (C) The result value searched by the initial search term set in the counterfeit product judgment data generation module is divided into title, price, seller, and description. Step of storing each search result element data including phrases, images, comments, categories, and product classifications; (D) Clustering the data collected in the clustering module based on priority data including price, seller, and number step; (E) Extracting keywords from each clustered group in the keyword extraction module and identifying keywords suggestive of counterfeit products through the extracted keywords; (F) identifying counterfeit products on a web page through the identified counterfeit product suggestive keywords in the judgment module and classifying unique products according to product description text on the web page and comment analysis results; and (G) Group and manage suspected counterfeit products detected in the counterfeit sales web management module by product group and seller group. If the counterfeit seller is confirmed, an online warning letter is sent to the counterfeit seller, and only counterfeit products that are not removed even after platform warning measures are taken. Separate reporting steps; Includes.

Preferably, step (D); When the standard for clustering when creating a group in the clustering module is price, the step of distinguishing the upper and lower sections of the product sales price; and setting a genuine product section and a counterfeit section according to the number of sellers in the upper section and lower section; It includes step (F); Determining the tendency of counterfeit products by identifying patterns of specific images, including images containing counterfeit products and mosaic-processed images; Setting image and image pattern data associated with a counterfeit product as a condition for suspecting a counterfeit product; and learning an image set with region-specific tendencies in the same region; Includes.

The natural language processing method and counterfeit detection system for identifying counterfeit products as described above can detect a more comprehensive range of products that infringe intellectual property rights, while conventional counterfeit detection solutions aim only to detect “counterfeit products.” The comprehensive intellectual property rights infringement product detection function provided in the embodiment includes brand name theft promotional activities that promote other products by stealing the name of the brand, custom products that create third added value by remanufacturing the product, and use of the brand's parts. This includes selling the product as if it were the original product.

By way of example, by effectively eliminating counterfeit products prevalent in the online market, approximately 10% to 40% or more of the demand for purchasing counterfeit products can be converted to purchasing genuine products, thereby improving corporate profits and contributing to national economic development. In addition, by efficiently dealing with counterfeit issues, the company's resources can be focused on the company's unique core competencies that have higher added value than counterfeit response activities. In addition, by effectively eliminating counterfeit products, it contributes to creating a virtuous social atmosphere that values intellectual property rights and reinvests in R&D.

The effects of the present invention are not limited to the effects described above, and should be understood to include all effects that can be inferred from the configuration of the invention described in the detailed description or claims of the present invention.

1 is a diagram showing a data processing block of a counterfeit detection system according to an embodiment.
Figure 2 is a diagram showing a data processing block of the keyword extraction module 500 according to an embodiment.
Figure 3 is a diagram showing the natural language processing process of a counterfeit product detection system according to an embodiment
Figure 4 is a diagram showing a keyword extraction process for identifying counterfeit products according to an embodiment.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. The present embodiments are merely provided to ensure that the disclosure of the present invention is complete and to provide common knowledge in the technical field to which the present invention pertains. It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

In describing embodiments of the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. The terms described below are terms defined in consideration of functions in embodiments of the present invention, and may vary depending on the intention or custom of the user or operator. Therefore, the definition should be made based on the contents throughout this specification.

1 is a diagram showing a data processing block of a counterfeit product detection system according to an embodiment.

Referring to FIG. 1, the counterfeit detection system according to the embodiment includes a preprocessing module 100, a search word setting module 200, a counterfeit product judgment data generation module 300, a clustering module 400, a keyword extraction module 500, and a judgment It may be configured to include a module 600 and a counterfeit product sales web management module 700. The term 'module' used in this specification should be interpreted to include software, hardware, or a combination thereof, depending on the context in which the term is used. For example, software may be machine language, firmware, embedded code, and application software. As another example, hardware may be a circuit, processor, computer, integrated circuit, integrated circuit core, sensor, Micro-Electro-Mechanical System (MEMS), passive device, or a combination thereof.

The preprocessing module 100 performs web crawling to extract samples of web pages related to counterfeit products. In an embodiment, web crawling may extract specific information from a website using C#, Python, etc. software and an MS-SQL database.

In the embodiment, the preprocessing module 100 sets the brand name and product line as search conditions in more than 500 shopping malls and SNS, performs a search according to the set search conditions, and performs web crawling for sample extraction. In an embodiment, the preprocessing module 100 may sequentially expand search conditions to repeatedly search, and web crawl each search result. For example, after setting a plurality of search conditions, the preprocessing module 100 expands the search conditions by combining the set search conditions, and crawls the searched web pages according to the expanded search conditions. Specifically, the brand name is set as the first search condition, the product group is set as the second search condition, the price range is set as the third search condition, and the product name is set as the fourth search condition. In the embodiment, at least one of the set first to fourth search conditions is extracted, and the extracted search conditions are combined to create an expanded search condition. Afterwards, you can perform a search for each of the expanded search conditions and crawl the search results. In an embodiment, search conditions may be added based on brand name, product group, and product details.

The search term setting module 200 sets an initial search term based on search terms related to the brand name searched in major search engines according to web crawling results. For example, the search word setting module 200 can set an initial search term based on search terms related to the brand name that is most searched in major search engines. Specifically, in the case of Chanel, the initial search term is set to Chanel bag, Chanel card wallet, Chanel No. 5, Chanel clutch bag, Chanel Gabrielle hobo bag, Chanel hand cream, Chanel necklace, Chanel earrings, Chanel Classic Medium, etc.

The counterfeit product judgment data generation module 300 stores the search results with the set initial search term as search result element data including title, price, seller, description, image, comment, category, and product classification.

The clustering module 400 creates groups by clustering the collected data based on priority data including price and number of sellers. When the basis for clustering when creating a group is price, the clustering module 400 divides the upper and lower sections of the product sales price, and determines the section considered genuine and the section considered counterfeit according to the number of sellers in the divided upper section and lower section. You can set it. For example, the clustering module 400 first clusters the collected data based on price and number of sellers to create groups. Specifically, the standard for clustering is the section where a large number of sellers are gathered in the top 1~60% of the entire price range, the section considered genuine (15% above/below the reference point, Group H), and the section where the majority of sellers are gathered in the top 61~100% of the price range. The section where sellers are gathered can be set as the fake product section (15% above/below, L group). In addition, the clustering module 400 can set a counterfeit product consideration section based on the prices of products sold in the corresponding product group by a blacklist of counterfeit product sellers that has already been accumulated.

The keyword extraction module 500 extracts keywords from each clustered group and identifies keywords suggestive of counterfeit products through the extracted keywords.

The judgment module 600 identifies counterfeit products through the identified counterfeit product suggestive keywords. In an embodiment, the judgment module 600 may determine the tendency of counterfeit products by identifying patterns of specific images. Additionally, in the embodiment, the judgment module 600 may determine the tendency of counterfeit products based on images through Google Vision's AUTOML, etc. For example, the judgment module 600 may determine the tendency of counterfeit products based on patterns of specific images. Specifically, specific images include: a) a pixelated image of the model's face, b) a pixelated image of the product's logo, c) an image containing the model's tattoo, d) an image of the model wearing a mask, and e) a famous image. It includes images related to counterfeit products, such as street images of celebrities, f) images with faces covered to a certain level with a smartphone, and g) background images when a series of photos of counterfeit products are taken based on a specific background at a factory manufacturing counterfeit products. . The judgment module 600 may set image and image pattern data associated with a counterfeit product as conditions for suspecting a counterfeit product.

In addition, since counterfeit products used in the same area, such as clothing and bags, have similar tendencies, the judgment module 600 learns an image set with a tendency for each area through Google Vision's AUTOML. In the embodiment, the judgment module 600 performs machine learning on at least 300 images for each pattern, preferably about 2,000 images. In the embodiment, the tendency determination data set is centered on brand-neutral images and the basic data set is maintained to maintain at least 70% accuracy when determining the trend of a new image.

The judgment module 600 determines an image that has a similarity over a certain level (e.g., 70%) with the image set that tends to be a counterfeit product as an element of the counterfeit product. The judgment module 600 considers other factors such as price, seller, comments, keywords, etc. in overlap and finally selects counterfeit products.

Additionally, in the embodiment, the judgment module 600 can be used as a sole selection condition after user confirmation in the case of an image of a counterfeit product that is characteristically used only by the corresponding brand. At this time, the image of the counterfeit product includes images of the product taken in a separate studio to promote the counterfeit product at the factory that manufactures the counterfeit product.

In addition, the judgment module 600 analyzes web page product description text and comments based on preset text suggesting counterfeit products and custom products, and can identify counterfeit products according to the analysis results. For example, the judgment module 600 can distinguish unique products based on product description phrases and comment analysis results. Specifically, if the manufacturer and import source registered in the product description are different from the designated company and counterfeit product handling data managed in the counterfeit identification system database, if the manufacturer and import source are not designated, or if there is an inquiry about genuine products in the comments, the main text reads “ You can identify counterfeit products by distinguishing between explanatory phrases that imply counterfeit/custom products, such as “custom”, “Chinese OEM”, “replica”, “genuine inquiry”, “custom made”, etc.

In an embodiment, the determination module 600 executes crawling for detecting counterfeit products and collects posts confirming counterfeit products through web crawling according to the results of analyzing the execution results. The judgment module 600 determines whether the crawling execution result includes a keyword suggesting a counterfeit product, is included in a counterfeit section, is included in a blacklist, includes comments and body text suggesting a counterfeit product, or includes an image with a tendency to counterfeit product. If at least two conditions are met, the crawled post can be confirmed as counterfeit. Additionally, if the image analyzed on the web is determined to be an image exclusively for selling counterfeit products, the judgment module 600 may determine that the crawled post is a counterfeit product.

The counterfeit sales web management module 700 groups the initially detected suspected counterfeit products by product information and seller information, and determines and manages the counterfeit seller by item or seller. In addition, counterfeit product data can be managed through web-based counterfeit product management platforms such as Windows Server, .NET platform, and MS-SQL. For example, when the counterfeit product sales web management module 700 groups the initially detected suspected counterfeit products by product group and seller group, sellers are grouped based on name, business number, and phone number. Through an embodiment, the counterfeit product sales web management module 700 can manage sellers with one seller ID even if each seller sells a product in a separate shopping mall. Through this, brand managers can manage counterfeit sellers by item or seller. In the actual counterfeit product sales market, 50 to 80 sellers often sell more than 5,000 counterfeit products, so managing each counterfeit product seller provides the advantage of reducing the data processing required to individually check many counterfeit products.

In addition, the counterfeit product sales web management module 700 according to the embodiment sends a warning to the counterfeit product seller in the form of an online warning letter defined by the system when the counterfeit product seller is confirmed, or transfers the information to an affiliated law firm to take separate legal action. Additionally, counterfeit products can be reported through the automatic reporting function of the shopping mall platform. In addition, sellers of counterfeit products confirmed in the embodiment can be managed in a blacklist and used to detect counterfeit products of other brands.

The counterfeit product sales web management module 700 generates and outputs statistical data on the price, seller, and shopping mall of counterfeit products being sold, and generates statistics for each seller and item subject to crackdown to perform an automatic report. In addition, only counterfeit products that are not removed even after platform warning measures are reported separately.

Figure 2 is a diagram showing a data processing block of the keyword extraction module 500 according to an embodiment.

Referring to Figure 2, the keyword extraction module 500 includes a filtering unit 510, a conversion unit 520, a partial match keyword extraction unit 530, a keyword extraction unit 540, and a counterfeit product suggestive keyword generation unit 550. It can be configured to include. In the embodiment, the keyword extraction module 500 extracts the keywords of each clustered group (H group and L group) through similar trademark judgment logic.

The filtering unit 510 removes words excluding stop words and nouns from the titles of each clustered group. The conversion unit 520 converts the comparison target brand into Korean and English, transliterates the English text, and converts it into Korean.

The partial match keyword extraction unit 530 extracts continuous partial match keywords and extracts discontinuous partial match keywords. For example, the partial match keyword extraction unit 530 extracts words in which part of the brand name continuously matches part of the word to be compared. Specifically, the partial match keyword extractor 530 extracts 6 letters of the English/transliterated alphabet if at least 50% or more of them match inseparably from the first syllable in the case of less than 6 letters of the English/transliterated alphabet or 2 letters of the Korean alphabet. Cases where more than 60% of letters or more than 3 Korean characters match are extracted as consecutive partial match keywords. Specifically, the partial match keyword extractor 530 can extract 'blenciaga', etc. as a partial match keyword for 'Balenciaga'.

Additionally, the partial match keyword extractor 530 extracts discontinuous partial match keywords. In the embodiment, a discontinuous partial match keyword is a word in which part of a brand name is discontinuously matched with part of a word to be compared. For example, in the case of less than 6 letters in the English/transliterated alphabet and less than 2 letters in the Korean alphabet, words in which at least 60% or more of the syllables match discontinuously from the first syllable, 70% in the case of 6 or more letters in the English/transliterated alphabet and 3 or more Korean letters. The above extracts words with discontinuously matching syllables. Specifically, the partial match keyword extraction unit 530 can extract 'Valentino', 'Valentino', etc. as non-consecutive partial match keywords of 'Valentino'.

The keyword extraction unit 540 extracts keywords based on similar title substitution logic and extracts words combined with brand names and words characteristically repeated in counterfeit products.

In the embodiment, extracting keywords based on similar title substitution logic involves extracting words in which some syllables of a brand name are pronounced as similar titles. The keyword extraction unit 540 may extract keywords based on similar title substitution logic through a previously stored substitution table. For example, the keyword extraction unit 540 extracts keywords by applying Korean-English substitution, pseudo-vowel substitution, and pseudo-consonant substitution based on the Korean transliteration table. Specifically, in the case of 'Balenciaga', 'Balenciaga' can be extracted as a pronunciation keyword with a similar title, and in the case of 'Comme des Garçons', 'Comme des Garçons' can be extracted as a keyword.

The counterfeit product-suggesting keyword generation unit 550 deletes keywords that overlap between groups among the keywords extracted from the clustered groups and generates the remaining keywords in the counterfeit-deemed section as counterfeit product-suggesting keywords. For example, the counterfeit product suggestive keyword generator 550 extracts a word that is combined with a brand name and words that are characteristically repeated in counterfeit products. Specifically, the counterfeit product suggestive keyword generation unit 550 is “Valenst, Charles brooch, Chanel.cn, Comme des Garçons. Extract words such as “CDG”. Afterwards, the counterfeit product suggestive keyword generation unit 550 performs a continuous partial match keyword extraction process, a discontinuous partial match keyword extraction process, a keyword extraction process based on similar title substitution logic, and a word in the form of a combination of a brand name and a word characteristically repeated in counterfeit products. Among the keywords of each group extracted through the extraction process, duplicate keywords can be deleted and the remaining keywords in the counterfeit section can be proposed as keywords suggesting counterfeit products. For example, the counterfeit product suggestive keyword generator 550 deletes overlapping keywords among the L group keywords and H group keywords and suggests the remaining L group keywords as counterfeit product suggestive keywords. Through this, it is possible to extract keywords that are characteristically used in abnormally distributed products.

Below, we will sequentially explain the natural language processing method for identifying counterfeit products. Since the operation (function) of the natural language processing method for identifying counterfeit products according to the embodiment is essentially the same as the function of the counterfeit product detection system, descriptions overlapping with FIGS. 1 and 2 will be omitted.

Figure 3 is a diagram showing the natural language processing process of a counterfeit product detection system according to an embodiment.

In step S100, the preprocessing module 100 performs web crawling to extract samples of web pages related to counterfeit products. In an embodiment, in step S100, search conditions can be sequentially expanded to repeatedly search, and each search result can be web crawled. For example, after setting a plurality of search conditions, the search conditions are expanded by combining the set search conditions, and the searched web pages are crawled according to the expanded search conditions. Specifically, the brand name is set as the first search condition, the product group is set as the second search condition, the price range is set as the third search condition, and the product name is set as the fourth search condition. In the embodiment, at least one of the set first to fourth search conditions is extracted, and the extracted search conditions are combined to create an expanded search condition. Afterwards, you can perform a search for each of the expanded search conditions and crawl the search results. In an embodiment, search conditions may be added based on brand name, product group, and product details.

In step S200, the search term setting module 200 sets an initial search term based on a search word related to the brand name searched in major search engines according to web crawling results.

In step S300, the result values searched with the initial search term set in the counterfeit product judgment data generation module 300 are stored as search result element data including title, price, seller, description, image, comment, category, and product classification.

In step S400, the data collected from the clustering module 400 is combined into price and seller

Clustering is performed based on priority data containing numbers. In the embodiment, when the standard for clustering is price, the upper and lower sections of the product sales price are divided, and the genuine product section and counterfeit section are set according to the number of sellers in the divided upper section and lower section. At this time, the counterfeit product consideration section can be set according to the identified price by identifying the price of counterfeit products sold in the relevant product group through accumulated blacklist data of counterfeit sellers.

In step S500, keywords for each clustered group are extracted from the keyword extraction module 500, and keywords suggestive of counterfeit products are identified through the extracted keywords.

In step S600, counterfeit products are identified through counterfeit product suggestive keywords identified in the judgment module 600. Additionally, in step S600, specific images related to counterfeit products are machine-learned and the patterns of specific images can be identified to determine the tendency of counterfeit products. In addition, in step S600, web page product description text and comments are analyzed based on preset text suggesting counterfeit products and custom products, and counterfeit products, which are unique products, can be identified based on the analysis results. In addition, in step S600, crawling for detecting counterfeit products is performed and posts confirming counterfeit products can be collected through web crawling according to the results of analyzing the execution results.

In an embodiment, if the image analyzed on the web is determined to be an image exclusively for selling counterfeit products, the crawled post may be confirmed as a counterfeit product. In addition, in step S600, the crawling execution result contains at least a keyword suggesting a counterfeit product, is included in a counterfeit section, is included in a blacklist, includes comments and body text suggesting a counterfeit product, or contains an image with a tendency to counterfeit product. If two conditions are met, the crawled post is confirmed as counterfeit.

The natural language processing process for detecting counterfeit products according to the embodiment groups the initially detected suspected counterfeit products by product information and seller information, and determines and manages the counterfeit seller by item or seller. When a counterfeit seller is confirmed, an online warning letter is sent to the counterfeit seller, counterfeit seller information is delivered to an affiliated law firm, the counterfeit seller is reported to the shopping mall platform, and the confirmed counterfeit seller can be set to a blacklist. In addition, the initially detected suspected counterfeit products can be grouped by product information and seller information, and the counterfeit seller can be confirmed and managed by item or seller.

Figure 4 is a diagram showing a keyword extraction process for identifying counterfeit products according to an embodiment.

In step S510, words excluding stop words and nouns are removed from the titles of each clustered group in the filtering unit. In step S520, the comparison unit converts the brand to be compared into Korean and English, and English is transliterated and converted into Korean.

In step S530, continuous partial match keywords are extracted from the partial match keyword extraction unit, and non-contiguous partial match keywords are extracted. In step S540, keywords are extracted based on similar title substitution logic in the keyword extraction unit, and words combined with brand names and words characteristically repeated in counterfeit products are extracted as keywords.

In step S550, among the keywords extracted from each clustered group in the counterfeit product suggestive keyword generation unit, overlapping keywords between groups are deleted and the remaining keywords in the counterfeit deemed section are proposed as counterfeit product suggestive keywords.

The natural language processing method and counterfeit detection system for identifying counterfeit products as described above maximizes the efficiency of detecting counterfeit products when the word itself representing the counterfeit product is used as a search keyword. For example, you can easily search for counterfeit products by searching “Hermes,” “Louis x,” “Chanel.cn,” “Lenciaga,” etc. on Naver Shopping. Counterfeit product suggestive keywords are words intended to be easily searchable by customers who want to buy counterfeit products while maintaining the similarity in name with the original brand that is generally searched for. Keywords alluding to counterfeit products are words created by the intention of counterfeit product sellers and naturally by consumers in the market. In the embodiment, patterns of counterfeit product-suggestive keywords are identified, counterfeit product-suggestive keywords are extracted, and counterfeit products and counterfeit product sales webs can be easily detected online.

In addition, through an embodiment, keywords that can characterize counterfeit products can be extracted from most brands in which counterfeit products are actively traded among a series of product groups with a large price difference between genuine and counterfeit products, including luxury product product lines, and the extracted keywords Through this, online counterfeit products can be efficiently detected. In addition, in cases where the price difference between genuine and counterfeit products is not large, counterfeit product detection is possible using a combination of information that specifies the counterfeit product in addition to search keywords. In addition, through an embodiment, thousands of counterfeit products are grouped into counterfeit product sellers and organized into dozens of counterfeit product sellers, thereby improving counterfeit product data processing efficiency and enabling response to each counterfeit product seller. Additionally, in the embodiment, counterfeit product sellers are grouped using business registration numbers, business names, phone numbers, etc., and counterfeit product sellers included in the group can be efficiently responded to. In addition, the counterfeit detection system provides management functions such as sending automatic warning letters to illegal sellers, making it easy to eliminate a series of counterfeit sales activities. In addition, through the embodiment, it is possible to detect online trademark theft by trading other companies' products while using the brand name as an online search keyword.

The disclosed content is merely an example, and various modifications and implementations may be made by those skilled in the art without departing from the gist of the claims, so the scope of protection of the disclosed content is limited to the above-mentioned specific scope. It is not limited to the examples.

Claims (8)

In a natural language processing method for identifying counterfeit products on a web,
(A) performing web crawling by setting the brand name and product line as search conditions to extract samples of web pages related to counterfeit products in the preprocessing module;
(B) setting an initial search term in the search term setting module based on the search term related to the brand name searched in the search engine;
(C) Saving the search results with the initial search term set in the counterfeit product judgment data generation module as search result element data including title, price, seller, description, image, comment, category, and product classification;
(D) Clustering the data collected in the clustering module based on priority data including price, seller, and number;
(E) Extracting keywords from each clustered group in the keyword extraction module and identifying keywords suggestive of counterfeit products through the extracted keywords;
(F) identifying counterfeit products on a web page through the identified counterfeit product suggestive keywords in the judgment module and classifying unique products according to product description text on the web page and comment analysis results; and
(G) Suspicious counterfeit products detected in the counterfeit sales web management module are grouped and managed by product group and seller group, and if the counterfeit seller is confirmed, an online warning letter is sent to the counterfeit seller, and only counterfeit products that are not removed even after platform warning measures are separately managed. Reporting step; Including,
Step (D) above; Is
When the clustering standard is price when creating a group in the clustering module, dividing the upper and lower sections of the product sales price; and
Setting a genuine product section and a counterfeit section according to the number of sellers in the upper section and lower section; Including,
Step (F) above; Is
Determining the tendency of counterfeit products by identifying patterns of specific images, including images containing counterfeit products and mosaic-processed images;
Setting image and image pattern data associated with a counterfeit product as a condition for suspecting a counterfeit product; and
Learning an image set with region-specific tendencies in a homogeneous region; A natural language processing method comprising:
The method of claim 1, wherein step (F); Is
Analyzing product description text and comments registered on a web page based on preset text suggesting counterfeit and custom products;
Identifying counterfeit products according to the analysis results; A natural language processing method comprising:
The method of claim 1, wherein step (E); Is
(E-1) removing words excluding stop words and nouns from the titles of each clustered group;
(E-2) Converting the comparison target brand into Korean and English, and transliterating the English text into Korean after applying Korean-English substitution, pseudo-vowel substitution, and pseudo-consonant substitution based on the Korean transliteration table;
(E-3) extracting consecutive partial matching keywords;
(E-4) extracting non-contiguous partial matching keywords;
(E-5) extracting keywords based on similar name substitution logic;
(E-6) extracting words combined with brand names and words characteristically repeated in counterfeit products;
(E-7) Among the keywords extracted from the clustered groups, deleting overlapping keywords between groups and proposing the remaining keywords in the section considered counterfeit as keywords suggesting counterfeit products; A natural language processing method comprising:
The method of claim 1, wherein step (F); Is
If the crawling execution result satisfies at least one of the following cases: includes keywords suggestive of counterfeit products, is included in the counterfeit section, is included in the blacklist, contains comments and body text suggestive of counterfeit products, or contains images with a tendency to counterfeit products. A natural language processing method characterized by determining crawled posts as counterfeit.
In the counterfeit detection system,
A preprocessing module that performs web crawling by setting brand names and product groups as search conditions to extract samples of web pages related to counterfeit products;
a search word setting module that sets an initial search word based on search terms related to the brand name searched in major search engines according to the web crawling results;
a counterfeit product judgment data generation module that stores the result values searched with the initial search term set as search result element data including title, price, seller, description, image, comment, category, and product classification;
Clustering module that clusters the collected data based on priority data including price and number of sellers;
A keyword extraction module that extracts keywords from each clustered group and identifies keywords suggestive of counterfeit products through the extracted keywords;
a judgment module that identifies counterfeit products through the identified counterfeit product suggestive keywords and classifies unique products according to the product description text and comment analysis results on the web page; and
Suspicious counterfeit products detected initially are grouped by product information and seller information, and counterfeit sellers are identified and managed by item or seller. If a counterfeit seller is confirmed, an online warning letter is sent to the counterfeit seller, and if the counterfeit product is not removed even after platform warning measures, Counterfeit product sales web management module that reports only counterfeit products separately; Including,
the clustering module; silver
When creating a group in the clustering module, if the basis for clustering is price, the upper and lower sections of the product sales price are distinguished, and the genuine and counterfeit sections are set according to the number of sellers in the upper and lower sections.
The judgment module; silver
Determine the tendency of counterfeit products by identifying patterns in specific images, including images containing counterfeit products and mosaic-processed images, and set image and image pattern data related to counterfeit products as conditions for suspecting counterfeit products to identify trends by region in the same area. A counterfeit detection system characterized by learning an image set.
According to claim 5, wherein the decision module; silver
A counterfeit product detection system that analyzes web page product description text and comments based on preset text suggesting counterfeit products and custom products, and identifies counterfeit products based on the analysis results.
The method of claim 5, further comprising: the keyword extraction module; silver
A filtering unit that removes words excluding stop words and nouns from the titles of each clustered group;
A conversion unit that converts the comparison target brand into Korean and English, and converts the English text into Korean by transliterating it after applying Korean-English substitution, similar vowel substitution, and similar consonant substitution based on the Korean transliteration table;
A partial match keyword extraction unit that extracts continuous partial match keywords and extracts non-contiguous partial match keywords;
A keyword extraction unit that extracts keywords based on similar title substitution logic and extracts words combined with brand names and words characteristically repeated in counterfeit products;
Among the keywords extracted from the clustered groups, a counterfeit product suggestive keyword generator that deletes overlapping keywords between groups and generates the remaining keywords in the counterfeit product deemed section as counterfeit product suggestive keywords; A counterfeit detection system comprising:
According to claim 5, wherein the decision module; silver
If the crawling execution result satisfies at least one of the following cases: includes keywords suggestive of counterfeit products, is included in the counterfeit section, is included in the blacklist, contains comments and body text suggestive of counterfeit products, or contains images with a tendency to counterfeit products. A counterfeit detection system characterized by determining crawled posts as counterfeit.