KR102736311B1 - Method for artificial intelligence-based machine false listing detection - Google Patents

Abstract

A method for detecting false machine listings according to one embodiment of the present invention comprises the steps of: obtaining business information of a machine seller, a machine purchase contract regarding the machine, and machine image data of the machine; an owner information verification step of comparing the business information and the text of the machine purchase contract to determine whether the business registration number, business name, or business address match; a machine information verification step of extracting text from the machine image and comparing it with the text of the machine purchase contract to determine whether the manufacturer, model name, serial number, or model year of the machine match; A step of calculating a reliability score of the machine is included according to the owner information verification status and the machine information verification status.

Description

{METHOD FOR ARTIFICIAL INTELLIGENCE-BASED MACHINE FALSE LISTING DETECTION}

The present invention relates to a method for detecting machine false listings, and more specifically, to a method for detecting machine false listings using artificial intelligence technology.

In the existing used machinery trading market, buyers manually review and evaluate the list of machines uploaded by sellers to find machines that meet their requirements. At this time, buyers spent a lot of time manually reviewing various machine specifications, and in particular, it was difficult to detect duplicate and false listings, which caused a lot of time and cost to purchase machines.

Korean Patent Publication No. 10-1935177 (Registered on January 11, 2019)

The purpose of the present invention is to provide a method for detecting false listings in a machine transaction process by utilizing artificial intelligence technology.

According to one embodiment of the present invention, a method for detecting false machine listings includes: a step of acquiring business information of a machine seller, a machine purchase contract regarding the machine, and machine image data of the machine taken by photographing the machine; an owner information verification step of comparing the business information with a text of the machine purchase contract to determine whether a business registration number, a business name, or a business address match; a machine information verification step of extracting text from the machine image and comparing it with the text of the machine purchase contract to determine whether a manufacturer, model name, serial number, or model year of the machine match; and a step of calculating a reliability score of the machine based on the owner information verification status and the machine information verification status.

According to one embodiment, the owner information verification step may include a step of extracting a first candidate for a business registration number from the business registration information, a step of generating a second candidate for a business registration number from a text extracted from the machine purchase contract, and a step of determining that the business registration numbers match if there is an element of the first candidate group and an element of the second candidate group that is the same.

According to one embodiment, the owner information verification step may further include a step of obtaining location information from metadata of the machine image, comparing the location information with address information on the business registration certificate, and determining whether the location and the address are within a specific distance.

According to one embodiment, the machine information verification step may include a step of extracting a primary serial number candidate from the machine purchase contract, a step of generating a regular expression encompassing the primary serial number candidate, and a step of extracting a secondary serial number candidate group from text extracted from the machine image using the regular expression.

According to one embodiment, the machine information verification step may include a step of extracting a model name of the machine from the machine purchase contract, a step of generating a regular expression that encompasses serial number information of other machines that match the model name from a machine information database that is linked to the processor and contains model name and serial number information of other machines, and a step of extracting a secondary serial number candidate group from text extracted from the machine image using the regular expression.

According to one embodiment, the reliability score can be defined by the following [Mathematical Formula 1].

[Mathematical formula 1]

Here, t is the reliability score, n is the business registration number, the business name, the business address, the manufacturer, the model name, the serial number, and the year, m _n is the value of whether the variable (n) matches, and w _n is the weight for the variable (n).

According to one embodiment, the step of calculating the reliability score may further include the step of obtaining information on the time of shooting of the image from metadata of the machine nameplate image, comparing the information on the time of shooting with the contract date information of the machine purchase contract, and determining whether the time of shooting is before the contract date.

According to an embodiment of the present invention, the reliability of a machine is evaluated by verifying the owner and machine information of the machine through various documents related to the sale/purchase of the machine. This allows false/duplicate products to be detected in advance, thereby increasing the reliability of the transaction.

Figure 1 is a flowchart showing a method for detecting machine false listings.
Figure 2 is a conceptual diagram showing a machine false listing detection device.
Figure 3 is an example of a business registration certificate image.
Figure 4 is an example of a machinery purchase contract that shows business information.
Figure 5 is an example of a machine nameplate image showing manufacturer, model name, serial number, and year information.
Figure 6 is an example of a machine purchase contract showing machine information.
Figure 7 is an example of a machine nameplate image where it is unclear whether a serial number is written or not.
Figure 8 is a block diagram of a computing system.

Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. The advantages and features of the present disclosure and the methods for achieving them will become apparent with reference to the embodiments described in detail below together with the accompanying drawings. However, the technical idea of the present disclosure is not limited to the following embodiments but may be implemented in various different forms, and the following embodiments are provided only to make the technical idea of the present disclosure complete and to fully inform a person having ordinary skill in the art to which the present disclosure belongs of the scope of the present disclosure, and the technical idea of the present disclosure is defined only by the scope of the claims.

When adding reference signs to components of each drawing, it should be noted that the same components are given the same signs as much as possible even if they are shown on different drawings. In addition, when describing the present disclosure, if it is determined that a specific description of a related known configuration or function may obscure the gist of the present disclosure, the detailed description is omitted.

Unless otherwise defined, all terms (including technical and scientific terms) used herein can be used in a commonly understood meaning by a person of ordinary skill in the art to which this disclosure belongs. In addition, terms defined in commonly used dictionaries are not to be ideally or excessively interpreted unless explicitly specifically defined. The terminology used herein is for the purpose of describing embodiments and is not intended to limit this disclosure. In this specification, the singular also includes the plural unless specifically stated in the phrase. In this specification, "or" can mean and/or.

Also, in describing components of the present disclosure, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only intended to distinguish the components from other components, and the nature, order, or sequence of the components are not limited by the terms. When it is described that a component is "connected," "coupled," or "connected" to another component, it should be understood that the component may be directly connected or connected to the other component, but another component may also be "connected," "coupled," or "connected" between each component.

As used herein, the terms “comprises” and/or “comprising” do not exclude the presence or addition of one or more other components, steps, operations and/or elements.

Components included in one embodiment and components that have common functions may be described using the same names in other embodiments. Unless otherwise stated, descriptions made in one embodiment may be applied to other embodiments, and specific descriptions may be omitted within the overlapping scope or within the scope that can be clearly understood by a person skilled in the art.

Hereinafter, the present invention will be described in detail with reference to preferred embodiments of the present invention and the attached drawings.

Figure 1 is a flowchart showing a method for detecting machine false listings, and Figure 2 is a conceptual diagram showing a machine false listing detection device.

According to one embodiment of the present invention, a method for detecting false machine listings includes a step (S100) of acquiring business information of a machine seller, a machine purchase contract regarding the machine, and machine image data of the machine taken by photographing the machine, an owner information verification step (S200) of comparing the business information with the text of the machine purchase contract to determine whether a business registration number, a business name, or a business address match, a machine information verification step (S300) of extracting text from the machine image and comparing it with the text of the machine purchase contract to determine whether a manufacturer, model name, serial number, or model year of the machine match, and a step (S400) of calculating a reliability score of the machine based on the owner information verification status and the machine information verification status.

The above-described machine false listing detection method can be performed by a machine false listing detection device (10) including at least one processor.

A machine false listing detection device (10) according to one embodiment of the present invention includes a data acquisition unit (100) that acquires business information of a machine seller, a machine purchase contract regarding the machine, and machine image data photographing the machine, an owner information verification unit (200) that compares the business information with the text of the machine purchase contract to confirm whether the business registration number, business name, or business address match, a machine information verification unit (300) that extracts text from the machine image and compares it with the text of the machine purchase contract to confirm whether the manufacturer, model name, serial number, or model year of the machine match, and a reliability score calculation unit (400) that calculates a reliability score for the machine according to the owner information verification status and the machine information verification status.

The machine false listing detection device (10) can perform the role of a server. The device (10) can obtain data from a terminal device or client device used by a machine seller or a machine demander (buyer) to be described later. After performing a machine false listing detection method such as an embodiment of the present invention, the device (10) can transmit appropriate data to a terminal or client.

First, a step (S100) of acquiring a business registration certificate of a machine seller, a machine purchase contract for the machine, and machine image data taken of the machine is performed. The step (S100) can be performed by a data acquisition unit (100).

The machine that the seller is trying to sell may be a 'used' machine that he or she purchased and used somewhere else in the past. The machine seller can send information about the machine that he or she is trying to sell and his or her business information to the device (10) or server through the client terminal.

Business information may include data such as the seller's business registration number, business name, business address, opening date, and type of business. Business information may include a business registration certificate image or file data.

A machine purchase contract may be a contract that states the terms of the machine purchase at the time the seller purchased the machine in the past. A machine purchase contract may refer to documents that were exchanged when trading machines, such as estimates, appraisals, specifications, and receipts, in addition to the document titled 'contract'.

The machine purchase contract may include various data such as the machine type, manufacturer, year, serial number, model name, capacity, voltage, frequency, power consumption, weight, size, precision, maximum speed, RPM, output, pressure, temperature range, material, color, power usage, noise level, power usage, noise level, fuel type, control method, installation environment (e.g. indoor, outdoor), maintenance requirements, certification (e.g. CE, ISO), warranty period, operating hours, cooling method, operating temperature, load capacity, list of compatible parts, storage method, vibration level, internal memory capacity (for recording machine data), network connectivity, maximum allowable load, emergency stop function, safety device, ambient temperature tolerance, machine delivery conditions, packaging method, shipping method, price, and the specifications, installation status or current performance status of the machine. The machine purchase contract may be in the form of an image or file.

Machine image data may be an image taken of the actual machine. Machine image data may include an image taken of the nameplate portion of the machine in particular. Machine image data may include an image taken of the machine to express the operating status, repair status, etc. of the machine.

When a seller uploads his/her business information and machine information through a client terminal, the device (10) receives the data and can extract necessary business information, machine specifications, etc. through text parsing, etc.

Afterwards, an owner information verification step (S200) is performed to compare the business information with the text of the machine purchase contract to check whether the business number, business name, or business address matches. The above step (S200) can be performed by the owner information verification unit (200).

In one embodiment of the present invention, various documents are compared to check whether the machine uploaded by the seller is a false listing or a duplicate listing. For example, in one embodiment, a business registration certificate and a machine purchase contract are compared to determine whether the uploaded machine listing is actually owned by the seller.

The owner information verification unit (200) can extract the business number (business registration number) text from the business information.

FIG. 3 is an example of a business registration certificate image. When the R31 area of FIG. 3 is analyzed using optical character recognition (OCR) technology, text such as 'Registration Number: 000-00-00000' is extracted. In one embodiment, text presumed to be the business registration number is extracted from the machine purchase contract using a regular expression such as "/([0-9]{3})-?([0-9]{2})-?([0-9]{5})/". The regular expression above is only an example for extracting text in the format of xxx-xx-xxxxx (x is a number), and regular expressions are not limited thereto.

When the text matching the above regular expression is found in the entire text acquired through OCR of the image in Fig. 3, '000-00-00000' is extracted.

In the case of a business registration certificate in the Republic of Korea, the format is fixed. Therefore, if the entire text is extracted correctly, only one 'business registration number 1st candidate' can be extracted with the above regular expression. However, depending on the state of the document being photographed or scanned, there may be an error in text recognition. Accordingly, the text extracted with the above regular expression may not yet be confirmed as the correct business registration number. Therefore, the text extracted with the regular expression, etc. is referred to as 'business registration number 1st candidate' or '1st candidate'. The 1st candidate extracted from the business registration certificate in this way can be verified by the text extracted from the machine purchase contract.

The owner information verification unit (200) can extract the second group of candidates for business registration numbers from the machine purchase contract.

Figure 4 is an example of a machine purchase contract that shows business information. Some parts of the contract are grayed out to protect personal information. Referring to the bottom right of Figure 4, in the case of "乙", you can see that the company's nameplate was printed instead of signing according to the address, business registration number, business name, and representative. The OCR results of the signature area (R41) in this contract are as shown in Table 1 below. In the text below, some numbers were replaced with x and some texts were replaced with 0 to protect personal information.

"Gap" Address: Seoul Yeongdeungpo-gu Dorim-ro xxx (Daerim-dong) Business registration number: 118-00-03063 Company name : (주) xxx Representative: Yu xx "Eul" Address: 314-00-88777 Shin xxx Park xxx Daejeon Metropolitan City xxxxx

As seen in the text extraction result above, a text in the format of a business registration number was extracted next to the 'Address' text by the nameplate stamp. In this way, depending on the method of signing/stamping the contract, there is a possibility that other content may not be incorrectly entered in the "Business Registration Number", "Address", etc. fields. Accordingly, an error may occur when extracting a business registration number based on the distance between words with the string called a business (registration) number. Therefore, in one embodiment of the present invention, a group of business registration number candidates is extracted using a regular expression regardless of the text surrounding the actual business (registration) number. In the case of a contract, since two or more people sign, when extracting text using the regular expression described above, two business registration numbers are extracted, such as ['118-00-03063', '314-00-88777'].

The multiple business registration numbers extracted from the contract below are referred to as the ‘secondary candidate group of business registration numbers’ or ‘secondary candidate group’.

Thereafter, the owner information verification unit (200) can determine that the business registration numbers match if there is an element in the second candidate group whose similarity to the first candidate is greater than a certain value.

For example, let's assume that the first candidate for the business registration number extracted from the business registration certificate is '314-00- 66 777', and the second candidate for the business registration number extracted from the contract is ['100-00-03063', '314-00- 88 777']. The second candidate extracted from the contract image is '88', not the actual text '66', which is 'incorrectly' extracted.

The owner information verification unit (200) can find a business registration number from the second candidate group whose similarity with the first candidate is greater than a certain value. For example, when calculating string similarity using Python's difflib built-in module, the similarities of '314-00-66777' and ['100-00-03063', '314-00- 88 777'] are 0.36 and 0.64, respectively. When the threshold is set to 0.6, the owner information verification unit (200) can determine that '314-00- 66 777' and '314-00- 88 777''match' each other. The owner information verification unit (200) can confirm the first candidate, '314-00- 66 777', as the final business registration number. It goes without saying that the above schedule value is not limited to 0.6. If there are multiple items with a similarity level higher than a certain value, the owner information verification unit (200) can confirm the item with the highest similarity level as the final business registration number.

Similarly, the owner information verification unit (200) can compare the business information with the text of the machine purchase contract to check whether the business name or address matches. For example, the owner information verification unit (200) can first extract the corporate name, business name, address, or location information from the business registration certificate. Then, the owner information verification unit (200) can find text in the machine purchase contract that has a similarity level higher than a certain value with the extracted corporate name, business name, or address. If a text with a similarity level higher than a certain value is found, the owner information verification unit (200) can determine that the corporate name, business name, address, or location matches.

Meanwhile, in one embodiment, the reliability of owner information can be additionally determined through machine image data.

According to one embodiment, the owner information verification step (S200) may further include a step of obtaining location information from metadata of the machine image, comparing the location information with address information on the business registration certificate, and determining whether the location and the address are within a specific distance.

In particular, in the case of images taken using a smartphone, GPS information, i.e., location information of the place where the photo was taken, may be included in the metadata of the image. In one embodiment, if the image data includes GPS information, the address shown in the business registration certificate may be converted into GPS coordinates, and then the distance between the coordinates of the image data and the address coordinates of the business registration certificate may be calculated. If the distance is less than a specific value, the owner information verification unit (200) may determine that the address information matches. For example, the specific value may be set to 100 m, 1 km, etc., but the present invention is not limited thereto.

When converting address information, you can use coordinate conversion (geocoder) algorithms and external APIs.

The owner information verification unit (200) described above can verify whether the business registration number, business name, or business address match through the artificial intelligence module. The artificial intelligence module can extract texts that are likely to be 'business registration number', 'business name', or 'address' from the entire text through the language model (LLM). At this time, in one embodiment, information on regular expressions can be input or learned into the language model (LLM). The artificial intelligence module can compare the extracted key:value to determine whether the business information and the owner information in the machine purchase contract match.

Meanwhile, the device (10) can verify the extracted business registration number through an external database. According to one embodiment, the device (10) can include a business registration information verification unit (not shown). The business registration information verification unit can access a website such as the National Tax Service that can search for business registration numbers and whether business activities are in progress. The business registration information verification unit can search for business information through the extracted business registration number. The business registration information verification unit can check whether the extracted business registration number is an actual business registration number through the search result. In addition, detailed information such as the business name and address can be confirmed through the corresponding business registration number.

Thereafter, a machine information verification step (S300) is performed to extract text from the machine image, compare it with the text of the machine purchase contract, and check whether the manufacturer, model name, serial number, or model year information of the machine matches. The step (S300) can be performed by a machine information verification unit (300).

The machine information verification unit (300) can extract text data from machine image data through a linked OCR module. For example, the machine information verification unit (300) can extract text from an image of a machine nameplate and then obtain data such as the manufacturer, model name, serial number, and year.

The machine information verification unit (300) can be linked with an artificial intelligence module. The artificial intelligence module can extract text that is likely to be manufacturer, model name, serial number, and model year information from the entire text extracted from the machine image.

The manufacturer can be expressed in the extracted text as, for example, 'manufacturer', 'producer', etc.

The model name can be expressed as, for example, 'Model number', 'Model No.', 'Model', etc.

The serial number can be expressed as, for example, 'Serial number', 'Serial No.', 'Item No.', 'Item Id', etc.

The year can be expressed as, for example, 'manufacturing year and month' or 'manufacturing year, month, and day'.

The machine information verification unit (300) first extracts synonyms/similar words corresponding to the key as above, and can extract the value of the key from the text in the neighborhood of the extracted text. For example, the machine information verification unit (300) can extract text that is within 2 'words' from the extracted key text as the value of the key.

Fig. 5 is an example of a machine nameplate image showing manufacturer, model name, serial number, and year information. For example, in the case of a serial number, the machine information verification unit (300) can first extract texts such as 'Serial number', 'Serial No.', and 'Item No.', which are synonyms/similar words to the serial number. In the example of Fig. 5, the text 'SERIAL NO' is first extracted as a key, and the text '0039-006535', which is a distance of 1 from it, can be extracted as a value (key). Through this, the machine information verification unit (300) can finally extract '0039-006535' as the serial number.

In the example of Fig. 5, the model name, serial number, and year (manufacturing date) information are clearly displayed in a key: value structure. However, there may be cases where only the value is present without information about the key. For example, Fig. 5 includes the text 'San Infracore'. 'Infracore' is the name of the machine manufacturer. In this case, the machine nameplate of Fig. 5 does not explicitly state 'manufacturer'. Therefore, the device (10) cannot determine whether 'Infracore' is the manufacturer name based solely on the extracted text itself.

In cases where the key of the variable to be extracted is not explicitly described, the device (10) can utilize the machine information database.

In one embodiment, the device (10) can compare text extracted from a machine image with a machine information database to determine whether the manufacturer matches. The device (10) may be linked to a database or file storing the names of various machine equipment companies.

The machine information verification unit (300) can determine which text among the entire text corresponds to the 'manufacturer' through comparison with the machine information database. The machine information database regarding the 'manufacturer' can include several manufacturer names such as 'Infracore'. The machine information verification unit (300) can compare all manufacturer names to determine whether there is a matching text. Accordingly, the machine information verification unit (300) can determine that 'Infracore' is the value of the variable (key) called 'manufacturer'.

Similarly, the device (10) can compare the text extracted from the machine image with the machine information database to check whether the model name matches. In the machine field, the 'model name' corresponds to the 'product name', and the device (10) can access the manufacturer's homepage, etc., to build a database of various model names. Even if the machine nameplate image does not explicitly state 'model name' or 'Model', the machine information verification unit (300) can determine which part of the extracted text corresponds to the 'model name' information through comparison with the model name database.

On the other hand, unlike the manufacturer or model name, the serial number of the machine is a unique value assigned to each machine, so it is often difficult to create a database. In this case, if the text 'Serial Number' or 'Serial No' is not shown in the machine nameplate image, it can be difficult to determine which text is the 'Serial Number'.

In one embodiment of the present invention, a regular expression of a serial number is generated using information extracted from a ‘contract’ to confirm serial number information.

According to one embodiment, the machine information verification step (S300) includes a step of extracting a primary candidate for a serial number from a machine purchase contract, a step of generating a regular expression encompassing the primary candidate for the serial number, and a step of extracting a secondary candidate for the serial number from the machine image through the regular expression.

Figure 6 is an example of a machine purchase contract showing machine information.

In the case of a machine purchase contract, in most cases, phrases indicating the model name and serial number are clearly written to specify the purchase target. In fact, when OCRing the R61 area in the image of Fig. 6, the text "Model: SDPO-7500 Serial No : 180276 " can be extracted. In this way, when utilizing a machine purchase contract, as described above, a value can be extracted through the distance of words based on the key text called 'Serial No'. Through this, the machine information verification unit (300) can extract the primary candidate text of the serial number.

Thereafter, the machine information verification unit (300) can generate a regular expression that encompasses the first serial number candidate.

For example, in the text of Fig. 6, the serial number was extracted as '180276'. The machine information verification unit (300) recognizes that the serial number is a 6-digit number and can generate a regular expression related to the serial number as '/\b\d{6}\b/g'.

Meanwhile, in the R62 area of Fig. 6, the serial number was extracted as '180800001C'. At this time, the machine information verification unit (300) recognizes that the serial number is a 9-digit number + 1 alphabet digit, and can generate a regular expression related to the serial number as '/\b\d{9}[A-Za-z]\b/g'. It goes without saying that the above-described regular expression is only an expression expressing the serial number shown in Fig. 6, and does not limit the present invention. In one embodiment, an artificial intelligence module linked with the machine information verification unit (300) can generate a regular expression.

Thereafter, a step of extracting a secondary candidate serial number from the text extracted from the machine image is performed using the above regular expression.

Figure 7 is an example of a machine nameplate image where it is unclear whether a serial number is written. The R71 area is assumed to be the serial number, but since there is no word such as 'Serial No' in the nameplate image, it is not certain whether the text in the R71 area means the serial number. In addition, since the color of the nameplate has faded, it is not clear whether the extracted numeric text is accurate.

At this time, if a regular expression called '/^\d{4}-\d{2}-\d{3}-\d{1}$/g' is created through the first candidate for the serial number extracted from the machine purchase contract, the text '16 2 0-80-502-1' can be extracted from the entire text obtained by OCRing the image in Fig. 7. Depending on the regular expression, the number of second candidates can be 0, 1, or multiple. The text extracted in this way is called the second candidate for the serial number.

The number of secondary candidates being 0 means that no text matching the generated regular expression was found. At this time, the machine information verification unit (300) can postpone the determination of whether the serial number matches and update the regular expression.

If the number of secondary candidates is 1 or more, and there is a text similarity between the elements of the primary serial number candidate and the secondary serial number candidate that is higher than a certain value (threshold), the machine information verification unit (300) can determine that the serial numbers match. For example, in the case of Fig. 7, the serial number of the actual nameplate is '16 8 0-80-502-1', and the secondary candidate extracted through OCR and regular expression is '16 2 0-80-502-1'. Since the text similarity is higher than a certain value, the machine information verification unit (300) can determine that the serial numbers match.

Below, an embodiment of determining a serial number by generating a regular expression using a model name and serial number database is described.

In one embodiment, the machine information verification step (S300) may include a step of extracting the model name of the machine from the machine purchase contract, a step of generating a regular expression that includes serial number information of other machines that match the model name in a machine information database that is linked to the processor and contains model name and serial number information of other machines, and a step of extracting a secondary serial number candidate group from text extracted from the machine image through the regular expression.

First, a step of extracting the model name of the machine from the machine purchase contract is performed. As described above, in most cases, the machine purchase contract clearly states the phrase indicating the model name. In fact, when OCRing the R61 area in the image of Fig. 6, the text " Model: SDPO-7500 Serial No: 180276" can be extracted. In this case, the value can be extracted through the distance of the word based on the key text called 'Model'. Through this, the machine information verification unit (300) can extract the first candidate (text) of the model name.

The machine information verification unit (300) can compare the model name primary candidate with the machine information database containing the 'model name' information to check whether there is data with a matching model name. Afterwards, a step of generating a regular expression that includes serial number information of other machines with matching model names in the database is performed.

For example, let us assume that there is a machine information database as shown in [Table 2] below.

number model name Serial number 1 SDPO-6500 180124-A 2 SDPO-6500 180207-B 3 SDPO-7500 180276 4 SDPO-7500 1802761

Among these, the machine numbers that match the model name ('SDPO-7500') extracted from the contract of Fig. 6 are 3 and 4. The serial numbers of the two machines are '180276' and '1802761', respectively. The machine information verification unit (300) generates a regular expression that can encompass the two serial numbers. For example, the machine information verification unit (300) can recognize that the serial number is a 6 or 7-digit number and generate a regular expression such as "/\b\d{6,7}\b/g". As the number of machines with the same model name in the database increases, the commonalities and differences of each serial number can be discovered, and thus the pattern of the regular expression can become more sophisticated. Thereafter, a step of extracting a secondary candidate serial number from the text extracted from the machine image is performed using the generated regular expression.

If a serial number matching the above regular expression is found in the text data obtained by OCRing the machine nameplate, the text belongs to the second candidate group of serial numbers. If there is an element of the first candidate serial number extracted from the contract and the second candidate serial number extracted from the machine image whose text similarity is above a certain value, the machine information verification unit (300) can determine that the serial numbers match.

Similarly, information such as the year of the machine can also be verified by the machine information verification unit (300). For example, if the regular expression for verifying the year of the machine is '/\b\d{4}([-./])?(0?[1-9]|1[0-2])?([-./])?(0?[1-9]|[12][0-9]|3[01])?\b/g', the regular expression can extract texts related to year/month/day such as '2011', '2011.3', and 2011-03-08'. The machine information verification unit (300) can compare the texts obtained through the regular expression in the machine purchase contract and the machine (nameplate) image to verify whether the years match.

Thereafter, a step (S400) of calculating a reliability score for the machine is performed according to the owner information verification status and the machine information verification status. The step (S400) can be performed by a reliability score calculation unit (400).

The reliability score calculation unit (400) calculates/assigns a reliability score based on whether the business registration number, business name, address, manufacturer, model name, serial number, model year, etc. extracted from at least two different documents match. In addition, the reliability score calculation unit (400) can calculate/modify the reliability score using other values or data.

For example, if a variable such as a business registration number matches, the reliability score calculation unit (400) can assign a value of '1' to the match value. If they do not match, a value of '0' can be assigned. If a match cannot be determined because there is no document or the variable is not found, a preset value such as '0.5' can be assigned.

The reliability score calculation unit (400) can determine the importance of each variable and assign different weights to it.

For example, since the business registration number and serial number are 'unique' values for each business and machine, if the information matches, it can be determined that the reliability of the information is very high. For example, the reliability score calculation unit (400) can assign a weight of '10' to the business registration number and serial number.

For example, the business name and business address are values that can change due to changes in the company name, address transfer, etc. In this case, even if the information matches, the reliability may be lower than that of the business registration number. For example, the reliability score calculation unit (400) may assign a weight of '5' to the business registration number and serial number.

For example, in the case of the manufacturer and model name of a machine, depending on the nature/type of the document from which the text is extracted, a 'different' manufacturer/model name may be extracted. Also, in the case of the 'year of manufacture', other date information unrelated to the 'month and year of manufacture' may be extracted. Accordingly, even if the information is inconsistent, the penalty may be set smaller than other variables. For example, the reliability score calculation unit (400) may assign a weight of '1' to the manufacturer, model name, and serial number.

That is, the reliability score can be calculated as shown in [Mathematical Formula 1] below.

Here, t is the reliability score, n is the business registration number, the business name, the business address, the manufacturer, the model name, the serial number, and the year, m _n is the value of whether the variable (n) matches, and w _n is the weight for the variable (n). It should be understood that the above-described reliability score calculation method is exemplary and does not limit the present invention.

Meanwhile, in one embodiment, the reliability of machine information can be additionally determined through metadata of the machine image.

According to one embodiment, the step of calculating a reliability score (S400) may further include the step of obtaining information on the time at which the image was taken from metadata of the machine nameplate image, comparing the information on the time at which the image was taken with the contract date information of the machine purchase contract, and determining whether the time at which the image was taken is before the contract date.

The metadata of the image may include information on the time when the photo was taken. In this case, the reliability score calculation unit (400) may compare the contract conclusion date extracted by the machine information verification unit (300) with the corresponding photographing time information. The seller may have taken a photo of the machine to sell the used machine he or she used, and the time when the photo was taken is likely to be after the manufacturing year of the machine and the date when the machine was first purchased, that is, after the time when the machine purchase contract was written. Accordingly, if the time of the photo was taken before the contract date, the reliability score calculation unit (400) may lower the overall reliability score because it is a situation that is not logically correct.

According to one embodiment, the reliability score calculation unit (400) can determine whether the machine is a 'duplicate' listing and, if determined to be a duplicate listing, can lower the reliability score.

The device (10) may be linked to a database containing serial numbers of machines whose sales/purchases have ended. If the serial number of a machine uploaded by a seller as an item for sale is included in the serial number database of machines whose sales/purchases have ended, the reliability score calculation unit (400) may determine the machine as a 'duplicate' item and lower the reliability score.

The device (10) can transmit this reliability score to the buyer's terminal. For example, if the reliability score is higher than a certain value, marks such as 'owner authentication', 'real machine authentication', etc. can be displayed on the seller's or buyer's client. The device (10) can transmit information and a message regarding which information is judged to be inconsistent to the seller's or buyer's terminal, and in particular, can request the seller's terminal to update business registration certificate, contract, machine image, etc.

Meanwhile, in one embodiment, the reliability of machine information can be additionally determined through analysis of machine images.

For example, the device (10) can analyze color information in a machine image. If the machine is a 'monochromatic' machine, and the image analysis result shows a part where the color value differs by a certain value (threshold) or more, the device (10) can determine that the machine has been 'repaired'. In this case, if there is data such as 'no repair fact' in the data uploaded by the seller or the machine information data, the reliability score calculation unit (400) can lower the reliability score of the machine.

According to an embodiment of the present invention, the reliability of a machine is evaluated by verifying the owner and machine information of the machine through various documents related to the sale/purchase of the machine. This allows false/duplicate products to be detected in advance, thereby increasing the reliability of the transaction.

FIG. 8 is a block diagram illustrating a computing system according to an exemplary embodiment of the present disclosure.

Referring to FIG. 5, the computing system (1000) may configure a device (10) or a user terminal (20), and may include a processor (1100), a memory device (1200), a storage device (1300), a power supply (1400), and a display device (1500). Meanwhile, although not shown in FIG. 5, the computing system (1000) may further include ports that may communicate with a video card, a sound card, a memory card, a USB device, etc., or communicate with other electronic devices.

In this way, the processor (1100), the memory device (1200), the storage device (1300), the power supply (1400), and the display device (1500) included in the computing system (1000) can perform the artificial intelligence-based machine false listing detection method by configuring any one of the systems (1000) according to the embodiments of the technical idea of the present invention. Specifically, the processor (1100) can perform the artificial intelligence-based machine false listing detection method described above with reference to FIGS. 1 to 7 by controlling the memory device (1200), the storage device (1300), the power supply (1400), and the display device (1500).

The processor (1100) can perform specific calculations or tasks. According to an embodiment, the processor (1100) can be a microprocessor, a central processing unit (CPU). The processor (1100) can communicate with a memory device (1200), a storage device (1300), and a display device (1500) through a bus (1600), such as an address bus, a control bus, and a data bus. According to an embodiment, the processor (1100) can also be connected to an expansion bus, such as a Peripheral Component Interconnect (PCI) bus.

The memory device (1200) can store data required for the operation of the computing system (1000). For example, the memory device (1200) can be implemented as DRAM, mobile DRAM, SRAM, PRAM, FRAM, RRAM, and/or MRAM. The storage device (1300) can include a solid state drive, a hard disk drive, a CD-ROM, etc. The storage device (1300) can store a program, application data, system data, operating system data, etc. related to the artificial intelligence-based machine false listing detection method described above with reference to FIGS. 1 to 7.

The display device (1500) can be used as an output means for notifying a user, and can display information on an artificial intelligence-based machine false listing detection method to the user, etc. The power supply device (1400) can supply an operating voltage necessary for the operation of the computing system (1000).

As described above, exemplary embodiments have been disclosed in the drawings and the specification. Although specific terms have been used in the specification to describe the embodiments, these have been used only for the purpose of explaining the technical idea of the present disclosure and have not been used to limit the meaning or the scope of the present disclosure described in the claims. Therefore, those skilled in the art will understand that various modifications and equivalent other embodiments are possible from this. Accordingly, the true technical protection scope of the present disclosure should be determined by the technical idea of the appended claims.

Claims (15)

A method for detecting machine false listings, performed by at least one processor,
A step of obtaining business information of a machine seller, a machine purchase contract regarding the machine, and machine image data taken of the machine;
Owner information verification step of comparing the above business information and the text of the above machinery purchase contract to confirm whether the business registration number, business name or business address matches;
A machine information verification step of extracting text from the machine image and comparing it with the text of the machine purchase contract to check whether the manufacturer, model name, serial number, or year of the machine match; and
A step of calculating a reliability score of the machine according to the owner information verification status and the machine information verification status;
The above machine information verification step is:
In the above machine purchase contract, a step of extracting the model name of the machine through the distance of words based on the key text corresponding to the model name;
In the above machine purchase contract, a step of extracting a primary candidate group for the serial number of the machine through the distance of words based on the key text corresponding to the serial number;
A step of checking whether there is data matching the extracted model name in a machine information database that is linked to the above processor and contains model name and serial number information of other machines;
A step of generating a regular expression encompassing multiple serial numbers corresponding to the model name, if there is data matching the above model name;
A step of extracting a second set of serial number candidates from the text extracted from the machine image using the above regular expression;
If the number of the above secondary candidates is 0, the step of postponing the determination of whether the above serial numbers match and updating the regular expression by utilizing the above machine purchase contract and the above machine information database;
A step of determining that the serial numbers of the machine match when there is at least one of the above secondary candidates and the text similarity between the elements of the above primary candidates and the above secondary candidates is greater than a certain value;
A step of determining a regular expression for checking the model year of the above machine;
A step of comparing texts obtained from the machine purchase contract and the machine nameplate image using the above-determined regular expression to check whether the model year matches; and
A step of reflecting whether the serial number matches and whether the model year matches in the reliability score;
The steps for calculating the above reliability score are:
A method for detecting false machine listings, further comprising the step of obtaining information on the shooting time of the image from metadata of the machine image, comparing the shooting time information with the contract date information of the machine purchase contract, and determining whether the shooting time is before the contract date.
In the first paragraph,
The above owner information verification steps are:
Step of extracting the first candidate for business registration number from the above business information;
Step of generating a second group of candidates for business registration numbers from the text extracted from the above machinery purchase contract;
A method for detecting false machine listings, comprising: a step of determining that the business registration numbers match when there is an element of the first candidate group that is the same as an element of the second candidate group;
In the first paragraph,
The above owner information verification steps are:
A method for detecting machine false listings, further comprising the step of obtaining location information from metadata of the machine image, comparing the location information with the business address information, and determining whether the location and the address are within a specific distance.
In the first paragraph,
The above machine information verification step is:
In the above machine purchase contract, a step of extracting the first candidate for the serial number;
A step of generating a regular expression encompassing the first candidate for the above serial number;
A method for detecting machine false listings, comprising: a step of extracting a second candidate serial number from text extracted from the machine image using the above regular expression;
In the first paragraph,
The above machine information verification step is:
In the above machine purchase contract, a step of extracting the model name of the machine;
A step of generating a regular expression that includes serial number information of other machines that matches the model name in a machine information database that is linked to the above processor and contains model name and serial number information of other machines; and
A method for detecting machine false listings, comprising: a step of extracting a second candidate serial number from text extracted from the machine image using the above regular expression;
In the first paragraph,
The above reliability score is defined by the following [Mathematical Formula 1], a method for detecting machine false listings.
[Mathematical formula 1]

(Here, t is the reliability score, n is the business registration number, the business name, the business address, the manufacturer, the model name, the serial number, and the year, m _n is the value of whether the variable (n) matches, and w _n is the weight for the variable (n).)
delete In a machine counterfeit product detection system comprising at least one processor,
A data acquisition unit that acquires business information of a machine seller, a machine purchase contract regarding the machine, and machine image data taken of the machine;
Owner information verification unit that compares the above business information with the text of the above machine purchase contract to check whether the business registration number, business name, or business address matches;
A machine information verification unit that extracts text from the machine image and compares it with the text of the machine purchase contract to check whether the manufacturer, model name, serial number, or year of the machine match; and
A reliability score calculation unit for calculating a reliability score of the machine according to the owner information verification status and the machine information verification status;
The above machine information verification section,
In the above machine purchase contract, the model name of the machine is extracted through the distance of words based on the key text corresponding to the model name,
In the above machine purchase contract, the first candidate group for the serial number of the machine is extracted through the distance of words based on the key text corresponding to the serial number,
In the machine information database that is linked to the above processor and contains information on the model name and serial number of other machines, it is checked whether there is data matching the extracted model name.
If there is data matching the above model name, a regular expression is created that covers multiple serial numbers corresponding to the above model name,
Through the above regular expression, the second candidate serial number is extracted from the text extracted from the machine image,
If the number of the above secondary candidates is 0, the determination of whether the above serial numbers match is suspended, and the above regular expression is updated using the above machine purchase contract and the above machine information database.
If there is one or more of the above second candidate groups, and the text similarity between the elements of the above first candidate group and the above second candidate group is greater than a certain value, it is determined that the serial numbers of the above machines match,
Determine a regular expression to determine the year of the above machine,
Using the above-determined regular expression, compare the text obtained from the machine purchase contract and the machine nameplate image to check whether the model year matches.
Whether the above serial number matches or not and whether the above year matches are reflected in the above reliability score,
A machine false listing detection system characterized in that the reliability score calculation unit obtains information on the shooting time of the image from the metadata of the machine image, compares the shooting time information with the contract date information of the machine purchase contract, and determines whether the shooting time is before the contract date.
In Article 8,
The above owner information verification section is,
From the above business information, extract the first candidate for business registration number,
From the text extracted from the above machine purchase contract, a second group of candidates for business registration numbers is created,
A machine false listing detection system characterized in that if there is an element of the first candidate group that is the same as an element of the second candidate group, the business registration number is determined to match.
In Article 8,
The above owner information verification section is,
A machine false listing detection system characterized by obtaining location information from metadata of the machine image, comparing the location information with the business address information, and determining whether the location and the address are within a specific distance.
In Article 8,
The above machine information verification section,
In the above machine purchase contract, extract the first candidate for serial number,
Generate a regular expression that encompasses the first candidate for the above serial number,
A machine fake listing detection system characterized by extracting a second set of serial number candidates from text extracted from the machine image using the above regular expression.
In Article 8,
The above machine information verification section,
From the above machine purchase contract, extract the model name of the machine,
In a machine information database that is linked to the above processor and contains model name and serial number information of other machines, a regular expression is created that includes serial number information of other machines that match the above model name,
A machine fake listing detection system characterized by extracting a second set of serial number candidates from text extracted from the machine image using the above regular expression.
In Article 8,
The above reliability score is defined by the following [Mathematical Formula 1], a machine false listing detection system.
[Mathematical formula 1]

(Here, t is the reliability score, n is the business registration number, the business name, the business address, the manufacturer, the model name, the serial number, and the year, m _n is the value of whether the variable (n) matches, and w _n is the weight for the variable (n).)
delete In a storage device storing a method for detecting machine false listings,
The above method of detecting false machine listings is as follows:
A step of obtaining business information of a machine seller, a machine purchase contract regarding the machine, and machine image data taken of the machine;
Owner information verification step of comparing the above business information with the text of the above machine purchase contract to check whether the business registration number, business name, or business address match;
A machine information verification step of extracting text from the machine image and comparing it with the text of the machine purchase contract to check whether the manufacturer, model name, serial number, or year of the machine match; and
A step of calculating a reliability score of the machine according to the owner information verification status and the machine information verification status;
The above machine information verification step is:
In the above machine purchase contract, a step of extracting the model name of the machine through the distance of words based on the key text corresponding to the model name;
In the above machine purchase contract, a step of extracting a primary candidate group for the serial number of the machine through the distance of words based on the key text corresponding to the serial number;
A step of checking whether there is data matching the extracted model name in a machine information database that is linked to the processor and contains model name and serial number information of other machines;
A step of generating a regular expression encompassing multiple serial numbers corresponding to the model name, if there is data matching the above model name;
A step of extracting a second set of serial number candidates from the text extracted from the machine image using the above regular expression;
If the number of the above secondary candidates is 0, the step of postponing the determination of whether the above serial numbers match and updating the regular expression by utilizing the above machine purchase contract and the above machine information database;
A step of determining that the serial numbers of the machine match when there is at least one of the above secondary candidates and the text similarity between the elements of the above primary candidates and the above secondary candidates is greater than a certain value;
A step of determining a regular expression for checking the model year of the above machine;
A step of comparing texts obtained from the machine purchase contract and the machine nameplate image using the above-determined regular expression to check whether the model year matches; and
A step of reflecting whether the serial number matches and whether the model year matches in the reliability score;
The steps for calculating the above reliability score are:
A storage device further comprising a step of obtaining shooting time information of the image from metadata of the machine image, comparing the shooting time information with the contract date information of the machine purchase contract, and determining whether the shooting time is before the contract date.