An online cosmetic review method, device, equipment and storage medium

Abstract

The application discloses an online auditing method, device and equipment for cosmetics and a storage medium. The method comprises the following steps: performing image recognition on a label information image of a cosmetic to be audited to obtain all feature information of the cosmetic to be audited, and classifying the all feature information to obtain first feature information corresponding to each auditing category; inputting the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with target auditing feature information of a preset target auditing point to obtain a first matching degree corresponding to each first feature information; and based on the first matching degree, auditing the cosmetic to be audited to obtain an auditing result. Compared with the prior art, the technical scheme of the application can improve the auditing efficiency of the cosmetic.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition processing, and in particular to an online auditing method and device for cosmetics, equipment and a storage medium. BACKGROUND

[0002] Nowadays, people's life has changed in many aspects, such as shopping habits. People used to buy products they like or niche products through online shopping, but they still prefer offline shopping for products with social attributes and emotional experiences, such as cosmetics.

[0003] For cosmetics, there are laws and regulations about cosmetics in different countries and regions, and there are strict requirements for the ingredients, labels, advertisements, etc. of cosmetics. Moreover, since cosmetics are directly in contact with people's skin, in order to ensure that the cosmetics meet the corresponding laws and regulations, avoid illegal behavior, and ensure that the cosmetics do not contain ingredients harmful to human health, it is necessary to audit the sold cosmetics.

[0004] When auditing cosmetics, it is generally done by manual experience-based auditing. However, online shopping is becoming more and more frequent, especially the online share of cosmetics is increasing, and the pressure of product auditing through e-commerce channels is increasing. Manual auditing has the problem of low auditing efficiency. SUMMARY

[0005] The technical problem to be solved by the present application is to provide an online auditing method, device, equipment and storage medium for cosmetics, which improves the auditing efficiency of cosmetics.

[0006] To solve the above technical problems, the present application provides an online auditing method for cosmetics, comprising:

[0007] performing image recognition on the label information image of the cosmetic to be audited to obtain all feature information of the cosmetic to be audited, and classifying the all feature information to obtain first feature information corresponding to each auditing category;

[0008] inputting the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with target auditing feature information of a preset target auditing point to obtain a first matching degree corresponding to each first feature information;

[0009] based on the first matching degree, auditing the cosmetic to be audited to obtain an auditing result.

[0010] In a possible implementation, the image recognition of the label information image of the cosmetic to be audited obtains all feature information of the cosmetic to be audited, and specifically includes:

[0011] The label information image is preprocessed to obtain a preprocessed label information image. The image preprocessing includes image resizing, noise removal, and contrast enhancement.

[0012] The preprocessed label information image is subjected to text recognition based on OCR technology to obtain label text data;

[0013] The label text data is subjected to data filtering processing to obtain standard label text data, wherein the data filtering processing includes removing erroneous characters and invalid characters;

[0014] Based on natural language processing technology, feature information is extracted from the standard label text data to obtain all feature information of the cosmetic product to be reviewed. The feature information includes the manufacturer's name, production license, registrant's name, registrant certificate number, product name, claims, trademarks, patent information, net content, shelf life, usage method, safety warnings, other labels and ingredient names.

[0015] In one possible implementation, before classifying all the feature information, the following steps are also included:

[0016] The review categories are set, including production information category, filing information category, product label category, and product full cost category;

[0017] Based on web crawling technology, multiple drug supervision and management regulations corresponding to the review category are obtained, and keywords are extracted from the multiple drug supervision and management regulations to obtain the first keyword corresponding to each review category.

[0018] Based on the first keyword, the target feature information for each review category is determined.

[0019] In one possible implementation, all the feature information is classified to obtain the first feature information corresponding to each review category, specifically including:

[0020] Based on the target feature information of each review category, all feature information is traversed, and the first feature information that is the same as the target feature information is extracted to obtain the first feature information corresponding to each review category;

[0021] The first feature information corresponding to the production information category includes the manufacturer's name and the production license;

[0022] The first feature information corresponding to the filing information category includes the name of the filer and the filing certificate number;

[0023] The first feature information corresponding to the product information category includes the product name, the claim terms, the trademark, the patent information, the net content, the shelf life, the usage method, the safety warning, and the other identifiers;

[0024] The first feature information corresponding to the full cost category of the product includes the ingredient name.

[0025] In one possible implementation, before inputting the first feature information into the preset rule recognition model, the method further includes:

[0026] Historical drug supervision and management penalty data corresponding to each review category is obtained. Based on natural language processing technology, key penalty information is extracted from the historical drug supervision and management penalty data to obtain historical penalty key data corresponding to each review category. The historical penalty key data includes penalty amount, penalty category, and penalty feature information corresponding to the penalty category.

[0027] Obtain all the key points for each audit category, and the audit feature value corresponding to each key point;

[0028] The penalty feature information corresponding to each audit category is compared with the audit feature value corresponding to all audit points to obtain the first similarity;

[0029] Based on the first similarity and the penalty amount, a weight value is set for the audit feature value corresponding to each audit point. Based on the audit feature value corresponding to each audit point and the weight value of the audit feature value, an initial rule recognition model is constructed.

[0030] Using the sample feature information of cosmetics as the model input and the sample matching degree corresponding to the sample feature information as the model output, the initial rule recognition model is trained to obtain the rule recognition model.

[0031] In one possible implementation, the first feature information is input into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain a first matching degree corresponding to each first feature information, specifically including:

[0032] The first feature information is input into a preset rule recognition model so that the rule recognition model calculates the first similarity between the first feature information and the target review feature information of the preset target review points. Based on the first similarity and the target weight value corresponding to the target review feature information, the first matching degree corresponding to each first feature information is obtained.

[0033] In one possible implementation, the cosmetic product to be reviewed is reviewed based on the first matching degree to obtain the review result, specifically including:

[0034] Based on the first matching degree, the number of completely mismatched items and the number of mismatched items at each level are obtained. The number of completely mismatched items and the number of mismatched items at each level are substituted into the total mismatch calculation formula to obtain the total mismatch degree of the cosmetic product to be reviewed.

[0035] Based on the total mismatch, the review results of the cosmetics to be reviewed are obtained, wherein the review results include cosmetics that are not allowed to be marketed and cosmetics that have been rectified before being marketed.

[0036] The present invention also provides an online review device for cosmetics, comprising: a cosmetic feature information extraction module, a matching degree calculation module, and a review result generation module;

[0037] The cosmetic feature information extraction module is used to perform image recognition on the label information image of the cosmetic to be reviewed, obtain all feature information of the cosmetic to be reviewed, and classify all feature information to obtain the first feature information corresponding to each review category.

[0038] The matching degree calculation module is used to input the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain the first matching degree corresponding to each first feature information;

[0039] The audit result generation module is used to audit the cosmetic product to be audited based on the first matching degree and obtain the audit result.

[0040] In one possible implementation, the cosmetic feature information extraction module is used to perform image recognition on the label information image of the cosmetic to be reviewed, to obtain all feature information of the cosmetic to be reviewed, specifically including:

[0041] The label information image is preprocessed to obtain a preprocessed label information image. The image preprocessing includes image resizing, noise removal, and contrast enhancement.

[0042] The preprocessed label information image is subjected to text recognition based on OCR technology to obtain label text data;

[0043] The label text data is subjected to data filtering processing to obtain standard label text data, wherein the data filtering processing includes removing erroneous characters and invalid characters;

[0044] Based on natural language processing technology, feature information is extracted from the standard label text data to obtain all feature information of the cosmetic product to be reviewed. The feature information includes the manufacturer's name, production license, registrant's name, registrant certificate number, product name, claims, trademarks, patent information, net content, shelf life, usage method, safety warnings, other labels and ingredient names.

[0045] In one possible implementation, the cosmetic feature information extraction module, before classifying all the feature information, further includes:

[0046] The review categories are set, including production information category, filing information category, product label category, and product full cost category;

[0047] Based on web crawling technology, multiple drug supervision and management regulations corresponding to the review category are obtained, and keywords are extracted from the multiple drug supervision and management regulations to obtain the first keyword corresponding to each review category.

[0048] Based on the first keyword, the target feature information for each review category is determined.

[0049] In one possible implementation, the cosmetic feature information extraction module is used to classify all the feature information to obtain the first feature information corresponding to each review category, specifically including:

[0050] Based on the target feature information of each review category, all feature information is traversed, and the first feature information that is the same as the target feature information is extracted to obtain the first feature information corresponding to each review category;

[0051] The first feature information corresponding to the production information category includes the manufacturer's name and the production license;

[0052] The first feature information corresponding to the filing information category includes the name of the filer and the filing certificate number;

[0053] The first feature information corresponding to the product information category includes the product name, the claim terms, the trademark, the patent information, the net content, the shelf life, the usage method, the safety warning, and the other identifiers;

[0054] The first feature information corresponding to the full cost category of the product includes the ingredient name.

[0055] The present invention provides an online review device for cosmetics, which further includes: a rule recognition model generation module;

[0056] The rule recognition model generation module is used to obtain historical drug supervision and management penalty data corresponding to each review category. Based on natural language processing technology, it extracts key penalty information from the historical drug supervision and management penalty data to obtain historical penalty key data corresponding to each review category. The historical penalty key data includes penalty amount, penalty category, and penalty feature information corresponding to the penalty category.

[0057] Obtain all the key points for each audit category, and the audit feature value corresponding to each key point;

[0058] The penalty feature information corresponding to each audit category is compared with the audit feature value corresponding to all audit points to obtain the first similarity;

[0059] Based on the first similarity and the penalty amount, a weight value is set for the audit feature value corresponding to each audit point. Based on the audit feature value corresponding to each audit point and the weight value of the audit feature value, an initial rule recognition model is constructed.

[0060] Using the sample feature information of cosmetics as the model input and the sample matching degree corresponding to the sample feature information as the model output, the initial rule recognition model is trained to obtain the rule recognition model.

[0061] In one possible implementation, the matching degree calculation module is used to input the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain a first matching degree corresponding to each first feature information, specifically including:

[0062] The first feature information is input into a preset rule recognition model so that the rule recognition model calculates the first similarity between the first feature information and the target review feature information of the preset target review points. Based on the first similarity and the target weight value corresponding to the target review feature information, the first matching degree corresponding to each first feature information is obtained.

[0063] In one possible implementation, the review result generation module is used to review the cosmetic product to be reviewed based on the first matching degree to obtain a review result, specifically including:

[0064] Based on the first matching degree, the number of completely mismatched items and the number of mismatched items at each level are obtained. The number of completely mismatched items and the number of mismatched items at each level are substituted into the total mismatch calculation formula to obtain the total mismatch degree of the cosmetic product to be reviewed.

[0065] Based on the total mismatch, the review results of the cosmetics to be reviewed are obtained, wherein the review results include cosmetics that are not allowed to be marketed and cosmetics that have been rectified before being marketed.

[0066] The present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the online verification method for cosmetics as described in any of the preceding claims.

[0067] The present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the online review method for cosmetics as described in any of the preceding claims.

[0068] This invention provides an online review method, apparatus, device, and storage medium for cosmetics, which, compared with the prior art, have the following advantages:

[0069] By performing image recognition on the label information image of the cosmetics to be reviewed, all feature information of the cosmetics to be reviewed can be obtained, which can effectively avoid errors or tediousness of manual input or copy and paste, and improve the efficiency and accuracy of subsequent review. All feature information is then classified to obtain the first feature information corresponding to each review category. This classification allows for more detailed organization and categorization of the feature information of the cosmetics to be reviewed, facilitating subsequent rule matching and review processes. The first feature information is input into a preset rule recognition model, which matches the first feature information with the target review feature information of preset target review points to obtain the first matching degree corresponding to each first feature information. Through the matching of the rule recognition model, it is possible to more accurately assess whether the cosmetics to be reviewed meet the review requirements. Based on the first matching degree, the cosmetics to be reviewed are reviewed to obtain the review result. By using the first matching degree as an evaluation indicator, it is possible to automatically and quickly determine whether the cosmetics to be reviewed pass the review, thereby improving review efficiency. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating an embodiment of an online review method for cosmetics provided by the present invention;

[0071] Figure 2 This is a schematic diagram of one embodiment of an online cosmetics review device provided by the present invention;

[0072] Figure 3 This is a schematic diagram of target feature information for various review categories according to an embodiment of the present invention;

[0073] Figure 4 This is a schematic diagram of the key points of the audit and its audit feature values ​​according to an embodiment of the present invention. Detailed Implementation

[0074] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] Example 1, see Figure 1 , Figure 1 This is a schematic flowchart of an embodiment of an online cosmetics review device provided by the present invention, as shown below. Figure 1 As shown, the method includes steps 101-103, as detailed below:

[0076] Step 101: Perform image recognition on the label information image of the cosmetic to be reviewed to obtain all feature information of the cosmetic to be reviewed, and classify all feature information to obtain the first feature information corresponding to each review category.

[0077] In one embodiment, a camera device is used to photograph the cosmetic product to be reviewed, thereby obtaining an image of the label information of the cosmetic product to be reviewed. The label information image is an image of the packaging information of the cosmetic product to be reviewed, and the packaging information image includes an image of the outer packaging information and an image of the inner packaging information.

[0078] Preferably, for cosmetics to be reviewed that do not have outer packaging, such as shampoo, since their information is mainly on the inner packaging, the inner packaging information image of the cosmetic to be reviewed is obtained; for cosmetics to be reviewed that have outer packaging, the outer packaging information image of the cosmetic to be reviewed is obtained.

[0079] In one embodiment, when performing image recognition on the label information image of the cosmetic product to be reviewed, it is necessary to first perform image preprocessing on the label information image to obtain a preprocessed label information image. The image preprocessing includes image resizing, image noise removal, and contrast enhancement. Based on the above image preprocessing, the label information image can be adjusted to a uniform size, which facilitates subsequent operations. At the same time, the image noise removal process can avoid the label information image from being affected by noise, resulting in image blurring, color shift, or other interference. Furthermore, the contrast enhancement process can also avoid the problem that the text and patterns in the label information image may appear dim or blurry due to lighting conditions or image quality, thereby improving the readability of the image.

[0080] Specifically, the label information image is scaled according to a preset image size to obtain a first label information image. The first label information image is then filtered using an image filtering algorithm to obtain a first filtered label information image. Finally, the first filtered label information image is enhanced using a contrast enhancement algorithm to obtain a preprocessed label information image. The image filtering algorithm includes mean filtering, median filtering, and Gaussian filtering. The contrast enhancement algorithm includes histogram equalization and adaptive histogram equalization.

[0081] In one embodiment, text recognition is performed on the preprocessed label information image based on OCR technology to obtain label text data.

[0082] Specifically, the text is located in the preprocessed label information image based on the edge detection algorithm to obtain the text region in the preprocessed label image, and the text region is segmented into multiple character regions; the multiple character regions are then recognized based on OCR technology to obtain the label text data.

[0083] In one embodiment, the label text data is subjected to data filtering processing to obtain standard label text data, wherein the data filtering processing includes removing erroneous characters and invalid characters.

[0084] In one embodiment, feature information is extracted from the standard label text data based on natural language processing technology to obtain all feature information of the cosmetic product to be reviewed.

[0085] Specifically, the standard label text data is segmented using natural language processing technology to divide it into multiple word groups, and the word groups are tagged with part-of-speech tags. Pre-selected cosmetic feature words and their corresponding part-of-speech tags are used as extracted features. Regular expressions are used to extract feature information from the standard label text data based on the extracted features to obtain all feature information of the cosmetic product to be reviewed.

[0086] In one embodiment, all the feature information includes the manufacturer's name, production license, registrant's name, registrant certificate number, product name, claims, trademarks, patent information, net content, shelf life, usage method, safety warnings, other identifiers, and ingredient names.

[0087] In one embodiment, since cosmetics are strictly regulated products, various countries have relevant regulations and standards that stipulate the requirements for the production, registration, labeling and ingredients of cosmetics. Classifying the extracted feature information according to the focus of the regulations helps to ensure that the product complies with the regulations and avoids violations of laws and regulations, thereby improving compliance. Furthermore, by classifying all feature information of the cosmetics to be reviewed, the relevant regulatory authorities can more easily conduct inspections and supervision, thereby improving the efficiency of subsequent reviews.

[0088] In one embodiment, review categories are set, including production information category, filing information category, product label category, and product full cost category.

[0089] In one embodiment, multiple drug supervision and management regulations corresponding to the review category are obtained based on web crawling technology; keywords are extracted from the multiple drug supervision and management regulations to obtain a first keyword corresponding to each review category; based on the first keyword, target feature information for each review category is determined; such as... Figure 3 As shown, Figure 3 This is a diagram illustrating the target feature information for each audit category.

[0090] In one embodiment, when determining the target feature information of each review category based on the first keyword corresponding to each review category, the first keyword corresponding to each review category is compared with all the feature information. Based on the comparison results, the feature information with the highest similarity is selected as the target feature information corresponding to the first keyword, so as to determine the target feature information of each review category.

[0091] Specifically, the process involves identifying websites containing drug regulatory regulations and using Python technology to obtain the regulations issued by the National Medical Products Administration (NMPA) over the past 20 years for each audit category. Based on the requirements for cosmetic information in these regulations, key information is extracted. For example, for the production information category, the regulations require manufacturers to possess a cosmetic production license, which is the primary keyword for this category. The manufacturer information on the cosmetic packaging is verified through a cosmetic production enterprise information platform. For product label information, the regulations on label information are extracted, and label-related features are identified. These include requirements for the cosmetic name (which is the primary keyword for this category), font requirements, and unit of measurement requirements. The regulatory information is then compared with the product label information, and based on the primary keywords extracted from each audit category, target feature information for each audit category is generated.

[0092] In one embodiment, based on the target feature information based on each review category, all feature information is traversed, and first feature information that is the same as the target feature information is extracted to obtain the first feature information corresponding to each review category; so as to realize the classification of all feature information of cosmetics to be reviewed.

[0093] In one embodiment, the first feature information corresponding to the production information category includes the manufacturer's name and the production license.

[0094] In one embodiment, the first feature information corresponding to the filing information category includes the name of the filer and the filing certificate number.

[0095] In one embodiment, the first feature information corresponding to the product information category includes the product name, the claim terms, the trademark, the patent information, the net content, the shelf life, the usage method, the safety warning, and the other identifiers;

[0096] In one embodiment, the first feature information corresponding to the full cost category of the product includes the ingredient name.

[0097] Step 102: Input the first feature information into a preset rule recognition model so that the rule recognition model matches the first feature information with the preset target review feature information of the target review points to obtain the first matching degree corresponding to each first feature information.

[0098] In one embodiment, historical drug supervision and management penalty data corresponding to each review category is obtained. Based on natural language processing technology, key penalty information is extracted from the historical drug supervision and management penalty data to obtain historical penalty key data corresponding to each review category. The historical penalty key data includes the penalty amount, penalty category, and penalty feature information corresponding to the penalty category. Preferably, the penalty feature information is the feature information of the review category that was penalized.

[0099] In one embodiment, all audit points corresponding to each audit category and the audit feature value corresponding to each audit point are obtained.

[0100] Specifically, based on NLP technology, deep learning can be used to obtain the latest global regulations, lists of prohibited and restricted raw materials, and hazardous substances from relevant websites, thereby acquiring all the key audit points for each audit category and the audit feature values ​​corresponding to each key audit point. For example... Figure 4 As shown, Figure 4 This is a diagram illustrating the key points of the audit and their characteristic values. Figure 4 The key points are sourced from various relevant websites.

[0101] Preferably, relevant websites also include ChemLinked, which provides real-time updates on global regulations.

[0102] Preferably, historical manual review information data is obtained, and NLP technology is used to classify and learn all review points corresponding to each preset review category, so as to obtain each review point and its corresponding review feature value; wherein, the review feature value can be understood as specific review details; for example, for the review points in the production information category, including production qualifications, the review feature value corresponding to the production qualifications is that the production qualifications of eye products need to be marked with "#".

[0103] Preferably, the target audit feature information of the target audit points refers to the content of the audit requirements to be achieved as defined in regulations or standards.

[0104] In one embodiment, the penalty feature information corresponding to each audit category is compared with the audit feature values ​​corresponding to all audit points to obtain a first similarity; based on the first similarity and the penalty amount, a weight value is set for the audit feature value corresponding to each audit point; based on the audit feature value corresponding to each audit point and the weight value of the audit feature value, an initial rule recognition model is constructed.

[0105] In one embodiment, the sample feature information of cosmetics is used as the model input, and the sample matching degree corresponding to the sample feature information is used as the model output to train the initial rule recognition model to obtain the rule recognition model.

[0106] Preferably, after obtaining all the audit points corresponding to each audit category and the audit feature values ​​corresponding to each audit point, the weights of each audit feature value can be balanced based on industry professional regulations and senior technical experts, such as technical personnel who have been engaged in relevant positions for more than 15 years, to obtain a preset rule recognition model.

[0107] In one embodiment, the system also reacquires the first historical drug supervision and management penalty data corresponding to each review category within a preset time period. Based on the first historical drug supervision and management penalty data, the weight value of the review feature value is updated to update the weight value in the rule recognition model. This enables the system to update the review content according to changes in global laws and regulations, thereby improving the effectiveness of the review.

[0108] In one embodiment, the first feature information is input into a preset rule recognition model so that the rule recognition model calculates the first similarity between the first feature information and the target review feature information of the preset target review points. Based on the first similarity and the target weight value corresponding to the target review feature information, the first matching degree corresponding to each first feature information is obtained.

[0109] Specifically, when calculating the first similarity between the first feature information and the target review feature information of the preset target review points, it is based on the preset source of the key points to determine whether the first feature information meets the target review feature information based on the preset target review points, and the degree of matching between the two is measured by calculating the similarity.

[0110] Specifically, using text similarity calculation methods, such as cosine similarity or edit distance, the first feature information and the target review feature information of the preset target review points are represented as vectors. Then, the similarity between these two vectors is calculated. The similarity calculation result can be used to evaluate the degree of matching between the first feature information and the preset target review points. If the similarity is high, it means that the first feature information satisfies the target review feature information of the preset target review points well.

[0111] Preferably, when the audit category is production information and full component category, when the similarity calculation between the corresponding first feature information and the target audit feature information corresponding to the audit key point and the first feature information is not 100%, the similarity is set to 0, that is, the mismatch is 100%.

[0112] In one embodiment, since the rule recognition model is set with target weight values ​​corresponding to each target review feature information, the first matching degree corresponding to each first feature information is calculated based on the target weight value and the first similarity.

[0113] Step 103: Based on the first matching degree, review the cosmetic product to be reviewed and obtain the review result.

[0114] In one embodiment, based on the first matching degree, the number of completely mismatched items and the number of mismatched items at each level are obtained. The number of completely mismatched items and the number of mismatched items at each level are substituted into the total mismatch calculation formula to obtain the total mismatch degree of the cosmetic product to be reviewed.

[0115] In one embodiment, the formula for calculating the total mismatch is as follows:

[0116] P = X * W 100 +∑Y a *W a ;

[0117] In the formula, P represents the total mismatch degree, X represents the number of completely mismatched items, and W... 100 For the term of complete mismatch, W 100 =100%, Y a The number of mismatch items for level a, 0 <a<100%,W a This is a level of mismatch.

[0118] In one embodiment, the term "complete mismatch" is set to "no relevant product manufacturing qualifications" or "more than two penalties related to product quality and safety".

[0119] In one embodiment, the mismatch degree items at each level are set as label error, missing label, etc., and the range of the mismatch degree at each level is 0. <a<100%。

[0120] In one embodiment, the review result of the cosmetic product to be reviewed is obtained based on the total mismatch degree, wherein the review result includes cosmetic products that are not allowed to be marketed and cosmetic products that have been rectified before being marketed.

[0121] Specifically, if the total mismatch is greater than 100%, the audit result will be that the cosmetics are not allowed to be marketed. If the total mismatch is 100% > total mismatch > 0, the audit result will be that the cosmetics can be marketed after rectification.

[0122] In summary, this embodiment provides an online cosmetics review method. By performing image recognition on the label information image of the cosmetics to be reviewed, all feature information of the cosmetics to be reviewed is obtained. This effectively avoids errors or tediousness from manual input or copy-pasting, improving the efficiency and accuracy of subsequent reviews. All feature information is then categorized to obtain first feature information corresponding to each review category. This categorization allows for more detailed organization and classification of the feature information of the cosmetics to be reviewed, facilitating subsequent rule matching and review processes. The first feature information is input into a preset rule recognition model, which matches the first feature information with the target review feature information of preset target review points, obtaining a first matching degree for each first feature information. Through the matching of the rule recognition model, the compliance of the cosmetics to be reviewed can be more accurately assessed. Based on the first matching degree, the cosmetics to be reviewed are reviewed to obtain the review result. By using the first matching degree as an evaluation indicator, the review of the cosmetics to be reviewed can be automated and quickly determined, improving review efficiency.

[0123] Example 2

[0124] See Figure 2 , Figure 2 This is a schematic diagram of one embodiment of an online cosmetics review device provided by the present invention, as shown below. Figure 2As shown, the device includes a cosmetic feature information extraction module 201, a matching degree calculation module 202, and an audit result generation module 203, as detailed below:

[0125] The cosmetic feature information extraction module 201 is used to perform image recognition on the label information image of the cosmetic to be reviewed, obtain all feature information of the cosmetic to be reviewed, and classify all feature information to obtain the first feature information corresponding to each review category.

[0126] The matching degree calculation module 202 is used to input the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with the preset target review feature information of the target review points to obtain the first matching degree corresponding to each first feature information.

[0127] The audit result generation module 203 is used to audit the cosmetic product to be audited based on the first matching degree and obtain the audit result.

[0128] In one embodiment, the cosmetic feature information extraction module 201 is used to perform image recognition on the label information image of the cosmetic to be reviewed, and obtain all feature information of the cosmetic to be reviewed. Specifically, it includes: performing image preprocessing on the label information image to obtain a preprocessed label information image, wherein the image preprocessing includes image resizing, image noise removal, and contrast enhancement; performing text recognition on the preprocessed label information image based on OCR technology to obtain label text data; performing data filtering on the label text data to obtain standard label text data, wherein the data filtering includes removing erroneous characters and invalid characters; and performing feature information extraction on the standard label text data based on natural language processing technology to obtain all feature information of the cosmetic to be reviewed, wherein all feature information includes manufacturer name, production license, registrant name, registration certificate number, product name, claims, trademarks, patent information, net content, shelf life, usage method, safety warnings, other labels, and ingredient names.

[0129] In one embodiment, the cosmetic feature information extraction module 201, before classifying all the feature information, further includes: setting review categories, wherein the review categories include production information categories, filing information categories, product label categories, and product full cost categories; obtaining multiple drug supervision and management regulations corresponding to the review categories based on web crawling technology, extracting keywords from the multiple drug supervision and management regulations to obtain a first keyword corresponding to each review category; and determining the target feature information for each review category based on the first keyword.

[0130] In one embodiment, the cosmetic feature information extraction module 201 is used to classify all feature information to obtain first feature information corresponding to each review category. Specifically, it includes: based on the target feature information of each review category, traversing all feature information, extracting first feature information that is the same as the target feature information, and obtaining first feature information corresponding to each review category; wherein, the first feature information corresponding to the production information category includes the manufacturer's name and the production license; the first feature information corresponding to the filing information category includes the filer's name and the filing certificate number; the first feature information corresponding to the product information category includes the product name, the claim terms, the trademark, the patent information, the net content, the shelf life, the usage method, the safety warning, and the other markings; the first feature information corresponding to the product full cost category includes the ingredient name.

[0131] The online review device for cosmetics provided in this embodiment also includes a rule recognition model generation module.

[0132] In one embodiment, the rule recognition model generation module is used to acquire historical drug supervision and management penalty data corresponding to each review category, extract key penalty information from the historical drug supervision and management penalty data based on natural language processing technology, and obtain historical penalty key data corresponding to each review category. The historical penalty key data includes penalty amount, penalty category, and penalty feature information corresponding to the penalty category. It also acquires all review points corresponding to each review category and review feature values ​​corresponding to each review point. The module compares the similarity of the penalty feature information corresponding to each review category with the review feature values ​​corresponding to all review points to obtain a first similarity. Based on the first similarity and the penalty amount, it sets weight values ​​for the review feature values ​​corresponding to each review point. Based on the review feature values ​​corresponding to each review point and the weight values ​​of the review feature values, it constructs an initial rule recognition model. Using sample feature information of cosmetics as model input and the sample matching degree corresponding to the sample feature information as model output, it trains the initial rule recognition model to obtain a rule recognition model.

[0133] In one embodiment, the matching degree calculation module 202 is used to input the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain a first matching degree corresponding to each first feature information. Specifically, it includes: inputting the first feature information into the preset rule recognition model, so that the rule recognition model calculates the first similarity between the first feature information and the target review feature information of the preset target review points, and obtaining the first matching degree corresponding to each first feature information based on the first similarity and the target weight value corresponding to the target review feature information.

[0134] In one embodiment, the audit result generation module 203 is used to audit the cosmetic product to be audited based on the first matching degree and obtain an audit result. Specifically, it includes: obtaining the number of completely mismatched items and the number of mismatched items at each level based on the first matching degree; substituting the number of completely mismatched items and the number of mismatched items at each level into the total mismatch calculation formula to obtain the total mismatch of the cosmetic product to be audited; and obtaining the audit result of the cosmetic product to be audited based on the total mismatch. The audit result includes cosmetic products that are not allowed to be marketed and cosmetic products that have been rectified before being marketed.

[0135] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0136] It should be noted that the above-described embodiment of the online cosmetics review device is merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0137] Based on the above-described embodiments of the online cosmetics review method, another embodiment of the present invention provides an online cosmetics review terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the online cosmetics review method of any embodiment of the present invention.

[0138] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the online verification terminal device for cosmetics.

[0139] The online verification terminal for cosmetics can be a desktop computer, laptop, handheld computer, or cloud server, etc. The online verification terminal for cosmetics may include, but is not limited to, processors and memory.

[0140] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the online verification terminal equipment for cosmetics, connecting all parts of the online verification terminal equipment via various interfaces and lines.

[0141] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the online review terminal device for cosmetics. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0142] Based on the above-described embodiments of the online review method for cosmetics, another embodiment of the present invention provides a storage medium comprising a stored computer program, wherein, when the computer program is running, the device containing the storage medium controls the execution of the online review method for cosmetics according to any embodiment of the present invention.

[0143] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0144] In summary, the online review method, apparatus, device, and storage medium provided by this invention involves image recognition of the label information image of the cosmetic to be reviewed, obtaining all feature information of the cosmetic to be reviewed, classifying all feature information to obtain first feature information corresponding to each review category, inputting the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of preset target review points, obtaining a first matching degree corresponding to each first feature information, and reviewing the cosmetic to be reviewed based on the first matching degree to obtain the review result. Compared with the prior art, the technical solution of this invention can improve the review efficiency of cosmetics.

[0145] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. An online verification method for cosmetics, characterized in that, include: Image recognition is performed on the label information image of the cosmetic to be reviewed to obtain all feature information of the cosmetic to be reviewed, and all feature information is classified to obtain the first feature information corresponding to each review category; The first feature information is input into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain a first matching degree corresponding to each first feature information. The construction process of the rule recognition model includes: acquiring historical drug supervision and management penalty data corresponding to each review category; extracting key penalty information from the historical drug supervision and management penalty data based on natural language processing technology to obtain historical penalty key data corresponding to each review category; wherein the historical penalty key data includes penalty amount, penalty category, and penalty feature information corresponding to the penalty category; acquiring each review... The system identifies all review points corresponding to a category and the review feature value corresponding to each review point. It then compares the similarity of the penalty feature information corresponding to each review category with the review feature values ​​corresponding to all review points to obtain a first similarity. Based on the first similarity and the penalty amount, it sets a weight value for the review feature value corresponding to each review point. Based on the review feature value corresponding to each review point and the weight value of the review feature value, it constructs an initial rule recognition model. Using the sample feature information of cosmetics as the model input and the sample matching degree corresponding to the sample feature information as the model output, it trains the initial rule recognition model to obtain a rule recognition model. Based on the first matching degree, the cosmetic product to be reviewed is reviewed to obtain a review result. The review result includes: based on the first matching degree, obtaining the number of completely mismatched items and the number of items at each level of mismatch degree; substituting the number of completely mismatched items and the number of items at each level of mismatch degree into the total mismatch degree calculation formula to obtain the total mismatch degree of the cosmetic product to be reviewed; based on the total mismatch degree, obtaining the review result of the cosmetic product to be reviewed, wherein the review result includes cosmetic products that are not allowed to be marketed and cosmetic products that have undergone rectification before being marketed. The total mismatch degree calculation formula is as follows: ； In the formula, Total mismatch, The number of completely mismatched items. For items with a complete mismatch, , The number of mismatch items at level a. , This is a level of mismatch.

2. The online verification method for cosmetics as described in claim 1, characterized in that, Image recognition is performed on the label information image of the cosmetic product to be reviewed to obtain all feature information of the cosmetic product to be reviewed, specifically including: The label information image is preprocessed to obtain a preprocessed label information image. The image preprocessing includes image resizing, noise removal, and contrast enhancement. The preprocessed label information image is subjected to text recognition based on OCR technology to obtain label text data; The label text data is subjected to data filtering processing to obtain standard label text data, wherein the data filtering processing includes removing erroneous characters and invalid characters; Based on natural language processing technology, feature information is extracted from the standard label text data to obtain all feature information of the cosmetic product to be reviewed. The feature information includes the manufacturer's name, production license, registrant's name, registrant certificate number, product name, claims, trademarks, patent information, net content, shelf life, usage method, safety warnings, other labels and ingredient names.

3. The online verification method for cosmetics as described in claim 2, characterized in that, Before classifying all the aforementioned feature information, the following steps are also included: The review categories are set, including production information category, filing information category, product label category, and product full cost category; Based on web crawling technology, multiple drug supervision and management regulations corresponding to the review category are obtained, and keywords are extracted from the multiple drug supervision and management regulations to obtain the first keyword corresponding to each review category. Based on the first keyword, the target feature information for each review category is determined.

4. The online verification method for cosmetics as described in claim 3, characterized in that, All the aforementioned feature information is classified to obtain the first feature information corresponding to each review category, specifically including: Based on the target feature information of each review category, all feature information is traversed, and the first feature information that is the same as the target feature information is extracted to obtain the first feature information corresponding to each review category; The first feature information corresponding to the production information category includes the manufacturer's name and the production license; The first feature information corresponding to the filing information category includes the name of the filer and the filing certificate number; The first feature information corresponding to the product label category includes the product name, the claim terms, the trademark, the patent information, the net content, the shelf life, the usage method, the safety warning, and the other identifiers; The first feature information corresponding to the full cost category of the product includes the ingredient name.

5. The online verification method for cosmetics as described in claim 1, characterized in that, The first feature information is input into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain the first matching degree corresponding to each first feature information, specifically including: The first feature information is input into a preset rule recognition model so that the rule recognition model calculates the first similarity between the first feature information and the target review feature information of the preset target review points. Based on the first similarity and the target weight value corresponding to the target review feature information, the first matching degree corresponding to each first feature information is obtained.

6. An online verification device for cosmetics, characterized in that, include: The module includes a cosmetics feature information extraction module, a matching degree calculation module, and an audit result generation module. The cosmetic feature information extraction module is used to perform image recognition on the label information image of the cosmetic to be reviewed, obtain all feature information of the cosmetic to be reviewed, and classify all feature information to obtain the first feature information corresponding to each review category. The matching degree calculation module is used to input the first feature information into a preset rule recognition model, so that the rule recognition model matches the first feature information with the target review feature information of the preset target review points to obtain a first matching degree corresponding to each first feature information. The construction process of the rule recognition model includes: acquiring historical drug supervision and management penalty data corresponding to each review category; extracting key penalty information from the historical drug supervision and management penalty data based on natural language processing technology to obtain historical penalty key data corresponding to each review category, wherein the historical penalty key data includes penalty amount, penalty category, and penalty feature information corresponding to the penalty category; acquiring all review points corresponding to each review category and the review feature value corresponding to each review point; comparing the similarity of the penalty feature information corresponding to each review category with the review feature values ​​corresponding to all review points to obtain a first similarity; and based on the first similarity and the penalty amount, performing a matching degree calculation on the target review feature information of each review category. Each audit feature value corresponding to each audit point is assigned a weight value. Based on the audit feature value corresponding to each audit point and the weight value of the audit feature value, an initial rule recognition model is constructed. The initial rule recognition model is trained using sample feature information of the cosmetic as model input and sample matching degree corresponding to the sample feature information as model output to obtain a rule recognition model. The audit result generation module is used to audit the cosmetic to be audited based on the first matching degree to obtain an audit result. The audit result includes: based on the first matching degree, obtaining the number of completely mismatched items and the number of mismatched items at each level; substituting the number of completely mismatched items and the number of mismatched items at each level into the total mismatch calculation formula to obtain the total mismatch of the cosmetic to be audited; based on the total mismatch, obtaining the audit result of the cosmetic to be audited, wherein the audit result includes cosmetics that are not allowed to be marketed and cosmetics that have been rectified before being marketed. The total mismatch calculation formula is as follows: ； In the formula, Total mismatch, The number of completely mismatched items. For items with a complete mismatch, , The number of mismatch items at level a. , This is a level of mismatch.

7. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the online review method for cosmetics as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the online review method for cosmetics as described in any one of claims 1 to 5.

Images