Method for detecting specified products, system for detecting specified products, and program

Abstract

This invention provides a method for detecting specified products, a system for detecting specified products, and a program for doing so. [Solution] A method for detecting a specified product, comprising: at least one processor acquiring multiple product datasets, each of which includes product descriptions for multiple products, and each product description including a description of the ingredients contained in the corresponding product; acquiring one or more keywords, each of which indicates one or more specified ingredients; extracting one or more suspected products from multiple products by performing a keyword search on the multiple product datasets using one or more keywords; detecting one or more non-containing products from the one or more suspected products; and outputting a list of containing products by excluding one or more non-containing products from the one or more suspected products.

Description

Technical Field

[0001] The present disclosure relates to a method for detecting a designated product, a designated product detection system, and a program.

Background Art

[0002] In an EC (Electric Commerce) site such as an Internet auction, various products are traded. There is a possibility that illegal products prohibited by laws and regulations are listed on such EC sites. Detecting such inappropriate listings quickly and accurately is important for protecting consumer safety.

[0003] Patent Document 1 discloses a method of extracting a character string included in listing information and comparing the character string with an NG character string in order to check for illegal products at the time of listing. The NG character string is, for example, "Copy CD-ROM" or "User registration is not possible". By searching the listing information using keywords such as the NG character string, it is possible to detect illegal products such as illegally copied software products. As a result, it becomes possible to stop the listing of illegal products.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] For example, when a product is a health food, if the product contains a component corresponding to a pharmaceutical product, it becomes an illegal product as an unapproved and unauthorized pharmaceutical product. That is, although it is legal for a certain component to be included in a pharmaceutical product, a health food containing that component can become an illegal product.

[0006] Thus, when judging the suitability of a product, especially based on its ingredients, simply searching listing information using ingredient names as keywords makes it difficult to detect illegal products. This challenge can arise not only when detecting illegal products, but also when trying to detect any product containing any designated ingredient.

[0007] This disclosure aims to provide a designated product detection method, a designated product detection system, and a program that can detect designated products containing designated ingredients. [Means for solving the problem]

[0008] A method for detecting designated products according to one aspect of the present disclosure includes: at least one processor acquiring a plurality of product datasets, each of which includes product descriptions for a plurality of products, the product descriptions including descriptions of the ingredients contained in the corresponding products; acquiring one or more keywords, each of which indicates one or more designated ingredients; extracting one or more suspected products from the plurality of products by performing a keyword search on the plurality of product datasets using the one or more keywords, each of which has at least one of the one or more designated ingredients in its corresponding product description; detecting one or more non-containing products from the one or more suspected products that do not contain the one or more designated ingredients based on the product description corresponding to each of the one or more suspected products; and outputting a list of containing products that contain at least one of the plurality of designated ingredients by excluding the one or more non-containing products from the one or more suspected products.

[0009] A method for detecting designated products according to one aspect of the present disclosure comprises at least one processor acquiring a plurality of product datasets, each of which includes product descriptions for a plurality of products, the product descriptions including descriptions of the components contained in the corresponding products, acquiring one or more keywords, each of which indicates one or more designated components, and extracting one or more suspected products from the plurality of products by performing a keyword search on the plurality of product datasets using the one or more keywords, wherein each of the suspected products corresponds to the respective product The description includes: including that the description contains at least one of the one or more designated ingredients; detecting negative descriptions from the product descriptions of each of the one or more suspected products that describe non-contained ingredients not contained in the corresponding suspected product; detecting one or more non-contained ingredients described in each of the negative descriptions; and displaying a confirmation screen on the display for determining whether each suspected product contains the one or more designated ingredients, wherein the confirmation screen is configured to display the product description of the corresponding suspected product, and the one or more non-contained ingredients are highlighted in the product description on the confirmation screen.

[0010] A designated product detection system according to one aspect of the present disclosure comprises at least one memory for storing computer program code and at least one processor, wherein the at least one processor executes the computer program code to acquire a plurality of product datasets, each of which includes product descriptions for a plurality of products, and the product descriptions include descriptions of the ingredients contained in the corresponding products; acquires one or more keywords, each of which indicates one or more designated ingredients; extracts one or more suspected products from the plurality of products by performing a keyword search on the plurality of product datasets using the one or more keywords, each of which has at least one of the one or more designated ingredients in its corresponding product description; detects one or more non-containing products from the plurality of suspected products based on the product description corresponding to each of the one or more suspected products, and outputs a list of containing products that contain at least one of the plurality of designated ingredients by excluding the one or more non-containing products from the plurality of suspected products.

[0011] A designated product detection system according to one aspect of this disclosure comprises at least one memory for storing computer program code and at least one processor, wherein the at least one processor executes the computer program code to obtain a plurality of product datasets, each of which includes product descriptions for a plurality of products, and the product descriptions include descriptions of the ingredients contained in the corresponding products; and obtains one or more keywords, each of which indicates one or more designated ingredients; and performs a keyword search on the plurality of product datasets using the one or more keywords to find one of the plurality of products from among the plurality of products The method involves extracting the above-mentioned suspected products, wherein each of the suspected products contains at least one of the one or more designated ingredients in its corresponding product description, detecting negative descriptions from the product description of each of the one or more suspected products that describe ingredients not contained in the corresponding suspected product, detecting one or more of the ingredients not contained described in each negative description, and displaying a confirmation screen on the display for determining whether each of the suspected products contains the one or more designated ingredients, wherein the confirmation screen is configured to display the product description of the corresponding suspected product, and the one or more ingredients not contained in the product description on the confirmation screen are highlighted.

[0012] A program according to one aspect of the present disclosure causes at least one processor to perform the following actions: acquire a plurality of product datasets corresponding to a plurality of products, each of which product datasets includes a product description of the ingredients contained in the corresponding product; acquire one or more keywords, each of which indicates one or more designated ingredients; extract one or more suspected products from the plurality of products by performing a keyword search on the plurality of product datasets using the one or more keywords, each of which suspected products includes at least one of the plurality of designated ingredients in its corresponding product description; detect one or more non-containing products from the one or more suspected products based on the product description corresponding to each of the one or more suspected products, and output a list of containing products that include at least one of the plurality of designated ingredients by excluding the one or more non-containing products from the one or more suspected products.

[0013] A program according to one aspect of the present disclosure causes at least one processor to perform the following actions: acquire a plurality of product datasets, each of which includes product descriptions for a plurality of products, the product descriptions including descriptions of the ingredients contained in the corresponding products; acquire one or more keywords, each of which indicates one or more designated ingredients; extract one or more suspected products from the plurality of products by performing a keyword search on the plurality of product datasets using the one or more keywords, each of which has at least one of the one or more designated ingredients in its corresponding product description; detect negative descriptions from the product description of each of the one or more suspected products that describe ingredients not contained in the corresponding suspected product; detect one or more of the ingredients not contained described in each negative description; and display a confirmation screen on a display for determining whether each suspected product contains the one or more designated ingredients, the confirmation screen being configured to display the product description of the corresponding suspected product, and the one or more ingredients not contained being highlighted in the product description on the confirmation screen. [Brief explanation of the drawing]

[0014] [Figure 1] Figure 1 is a schematic diagram illustrating the designated product detection system according to an embodiment. [Figure 2] Figure 2 is an explanatory diagram illustrating a product sales page. [Figure 3] Figure 3 is a table illustrating some of the designated ingredients in health foods. [Figure 4] Figure 4 is a flowchart illustrating the method for generating a classification model. [Figure 5] Figure 5 is a flowchart showing the method for detecting the specified product. [Figure 6] Figure 6 is an explanatory diagram illustrating a confirmation screen used to detect a specified product. [Modes for carrying out the invention]

[0015] Examples of designated product detection methods, designated product detection systems 11, and programs will be described with reference to Figures 1 to 6. The present invention is not limited to these examples and is intended to include all modifications within the meaning and scope equivalent to the claims, as shown in the claims. Furthermore, the numerals such as "first," "second," etc., used in the following description are used to distinguish the components to which they are attached and do not rank them.

[0016] [Overview of the Designated Product Detection System] The designated product detection system (hereinafter simply referred to as the "detection system") 11 shown in Figure 1 includes a designated product detection device (hereinafter simply referred to as the "detection device") 20. The detection device 20 is configured to detect designated products containing one or more designated ingredients from among multiple products listed on an e-commerce site or multiple products sold on an e-commerce site. When one or more designated ingredients include multiple designated ingredients, the multiple designated ingredients are also called a group of designated ingredients. If a designated ingredient is an illegal ingredient, the detection device 20 detects illegal products containing the illegal ingredient from among the multiple products.

[0017] The detection device 20 may be implemented as a computer comprising at least one processor 21, at least one memory 22, and a communication IF 23. Below, an example is described in which the detection device 20 comprises one processor 21 and one memory 22. The communication IF 23 enables communication with other devices via a network.

[0018] Memory 22 stores a program 24 executed by the processor 21 and detection data 25. The program 24 includes an application and an operating system. The processor 21 implements various functions by executing processing based on the program 24. The detection data 25 includes various data used when detecting a specified product.

[0019] The detection system 11 may include a database 13. The database 13 may be stored in a web server 14 for providing an EC site. The web server 14 may be a computer including components similar to those of the detection device 20 (for example, at least one processor, at least one memory, and a communication IF). Alternatively, part or all of the database 13 may be stored in the memory 22 of the detection device 20.

[0020] The detection system 11 may include a plurality of purchaser terminals 15 that can access an EC site (web server 14) via a network. The EC site may be, for example, a mall-type or single-store-type shopping site, or an online auction site, but is not limited thereto. The detection system 11 may include a plurality of seller terminals 16 that provide products for sale on the EC site via a network.

[0021] The purchaser terminal 15 and the seller terminal 16 may be, for example, personal computers, or portable terminals such as smartphones or tablets. Each of the purchaser terminal 15 and the seller terminal 16 may include a display as an output device. Each display may include a touch panel as an input device.

[0022] The database 13 may include a seller data table, a purchaser data table, and a product data table. The seller data table stores a plurality of seller data sets for each of a plurality of sellers who provide products. The purchaser data table stores a plurality of purchaser data sets for each of a plurality of purchasers who have purchased products using the EC site and a plurality of registered users (potential purchasers) of the EC site. The product data table stores a plurality of product data sets for each of a plurality of products. One product data set includes one or more data items for the corresponding one product. Each of the plurality of product data sets corresponds to a plurality of products listed on the EC site or a plurality of products sold on the EC site.

[0023] The detection system 11 may include a learning device 30 that can communicate with the detection device 20 via a network. The learning device 30 may be implemented as a computer, similar to the detection device 20, comprising at least one processor 31, at least one memory 32, and a communication interface 33. Below, we will describe an example in which the learning device 30 comprises one processor 31 and one memory 32.

[0024] The memory 32 stores a program 34. The program 34 includes an application and an operating system. The processor 31 implements various functions by executing processing based on the program 34. The memory 32 may also store 35 for training one or more learning models. One or more generated training data 35 may be stored in the memory 22 of the detection device 20.

[0025] The detection system 11 may include a language model 17 that can communicate with the learning device 30 via a network. The language model 17 may be a large language model (LLM) or a small language model (SLM). A large language model is a language model trained using a large amount of text data. A small language model is a language model that is smaller in scale than a large language model (for example, with fewer parameters). The language model 17 may also be a general-purpose natural language processing (NLP) model that can adapt to various natural language processing tasks such as information extraction, text summarization, text generation, or question and answer in response to an input prompt.

[0026] The language model 17 is configured to receive a prompt containing some instruction, generate data corresponding to the instruction, and output it as a completion. In this example, the learning device 30 generates a prompt and inputs it to the language model 17, and then the learning device 30 acquires the completion generated by the language model 17. Alternatively, the detection device 20 may use a language model 17 that is not included in the detection system 11 via the network.

[0027] The detection system 11 may include, instead of the learning device 30, another information processing device for performing at least one of the following: creating a prompt, inputting the prompt into the language model 17, and obtaining completion. The other information processing device may be a computer having similar components to the detection device 20 (e.g., at least one processor, at least one memory, and a communication interface).

[0028] The detection system 11 may include a worker terminal 40 that can communicate with the detection device 20 via a network. The worker operating the worker terminal 40 may be, for example, an operator of an e-commerce site, or an administrator who manages products listed on the e-commerce site.

[0029] The worker terminal 40 may be implemented as a computer, similar to the detection device 20, comprising at least one processor 41, at least one memory 42, and a communication IF 43. The following describes an example where the worker terminal 40 comprises one processor 41 and one memory 42. The worker terminal 40 may also include a display 46. Alternatively, an external display 46 may be connected to the worker terminal 40.

[0030] [Product Dataset] Each of the multiple products may be classified into one of the multiple product categories. These product categories may include, but are not limited to, at least one of the following: health foods, pharmaceuticals, quasi-drugs, and cosmetics.

[0031] Each product dataset may include, but is not limited to, information about the corresponding product, such as the product name, product category, listing date, sales start date, sales price, and product description. The product category may be entered manually in advance by the seller or operator. The product description may include a description of the ingredients contained in the corresponding product.

[0032] As shown in Figure 2, the product description may include multiple descriptive phrases to be displayed on the product sales page 50 on the e-commerce site. The sales page 50 may include, for example, a tagline 51, a title 52, a detailed description 53, and one or more images 54. The one or more images 54 may be still images or videos.

[0033] If at least a portion of one or more images 54 contains text strings, the detection device 20 may be configured to extract text strings from one or more images 54. The detection device 20 may add the extracted text strings extracted from one or more images 54 to the product description of the product dataset. In this case, the product description includes text strings contained in the tagline 51, title 52, and detailed description 53, as well as the extracted text strings.

[0034] For at least some of the multiple product categories, one or more designated ingredients may be set for each product category. A product category in which one or more designated ingredients are set is called a designated product category. For example, if multiple designated ingredients (referred to as the "first designated ingredient group") are designated for a health food that is one of the multiple product categories, the first designated ingredient group may include illegal ingredients that are illegal in the health food product category. Illegal ingredients are, for example, ingredients that fall under pharmaceuticals and are prohibited under the Pharmaceutical Affairs Law. If another set of designated ingredients (referred to as the "second designated ingredient group") are designated for a pharmaceutical, the second designated ingredient group may include, for example, pharmaceutical ingredients that are not certified in the country or region where the pharmaceutical is sold.

[0035] Figure 3 illustrates a portion of the first designated ingredient group for health foods. The first designated ingredient group represents the names of pharmaceutical ingredients that, if contained in health foods, would constitute a violation of the Pharmaceutical Affairs Law. Each of the multiple designated ingredients can be set as multiple keywords. In this case, multiple keywords may be set for one designated ingredient. For example, two or more keywords may be set for one designated ingredient from among its Katakana name, English name, Kanji name, and alternative name. The set keywords may be included in the detection data 25 as a search keyword set.

[0036] One or more designated ingredients may be prohibited from being traded on e-commerce sites, even if they are not illegal. Products containing prohibited ingredients may also include illegal products. Furthermore, one or more designated ingredients may be set not only for the detection of illegal ingredients, but also for the purpose of detecting specific products. Specific products may, for example, be products containing specific allergens, or products containing ingredients included for a specific purpose. In this case, multiple groups of designated ingredients corresponding to multiple purposes may be designated for one designated product category (e.g., health foods).

[0037] [Machine learning models for classification] If the product dataset does not include data indicating product categories, the detection device 20 or the learning device 30 may generate data indicating product categories. The data indicating product categories may be generated using a classification machine learning model (hereinafter referred to as the "classification model"). The generated data indicating product categories may be added to the product description.

[0038] The learning device 30 may generate a classification model (a machine learning model for classification) by training a pre-trained learning model or a pre-trained learning model with training data. The classification model may be a binary classification model configured to output whether or not a product belongs to one of several product categories when a product description is input.

[0039] For example, if a product category includes one product category (health foods, which is the first category) and another product category (pharmaceuticals, which is the second category), the first classification model may be configured to output whether a product is a health food or not, and the second classification model may be configured to output whether a product is a pharmaceutical or not. Alternatively, the first classification model may be configured to classify which of the multiple product categories a product belongs to. Alternatively, the other classification model may be configured to classify which of the multiple product categories a product belongs to, or whether it does not belong to any of those product categories.

[0040] If multiple products include a set product which consists of multiple types of products, the classification model may be a multi-class classification model. The multi-class classification model may be configured to output which of the following three groups (classes) 1) to 3) a product belongs to when a product description is input.

[0041] 1) A group of target products belonging to product category 1 (target product group). 2) A group of non-eligible products belonging to product categories other than those specified in 1 (non-eligible product group).

[0042] 3) A group of products that include the target product (product set group). The set product group in 3) above may be a group of all set products, regardless of whether or not it includes the target product.

[0043] The learning device 30 may use multiple product datasets obtained from the database 13 as training data. Multiple product datasets used as training data are called multiple training product datasets (hereinafter simply referred to as "training datasets"), and product descriptions corresponding to the training datasets are also called training product descriptions (hereinafter simply referred to as "training descriptions"). The training descriptions include descriptions of the ingredients contained in the corresponding products.

[0044] The classification model may be, for example, a supervised learning model based on a decision tree algorithm that performs the classification task described above. For example, the classification model may use, but is not limited to, models employing Random Forest, GBDT, XGBoost, or lightGBM.

[0045] The training data includes multiple ground truth datasets for use as ground truth data. Each ground truth dataset may include training descriptions of the corresponding products and ground truth labels indicating whether the corresponding product is a target product, a non-target product, or a set product.

[0046] If there are insufficient ground truth datasets to be used as training data, the detection device 20 or the learning device 30 may generate ground truth datasets. For example, if a portion of the training dataset obtained from database 13 does not contain data indicating product categories that serve as ground truth labels, the detection device 20 or the learning device 30 may generate data indicating product categories.

[0047] The following describes an example in which the learning device 30 generates data indicating product categories using the language model 17. In this case, the learning device 30 generates prompts for input to the language model 17 in order to generate multiple ground truth datasets. The prompts in this example are generated in such a way that, as missing information from the ground truth dataset, the answer can be obtained as to whether a given product category corresponds to a target product, a non-target product, or a set product.

[0048] The prompt may include training instructions as information to obtain a response. In this case, the language model 17 will use the training instructions to determine whether the corresponding product belongs to a product category and whether it is part of a set of products that includes the product in question. The prompt may include multiple training instructions to obtain responses for multiple products at once, or it may include one training instruction to obtain a response for a single product.

[0049] In one example, a prompt includes multiple training descriptions for each of several products, and an instruction to prompt the language model 17 to output a response. The response that the instruction should output is whether the products described by each of the multiple training descriptions fall into one of the following categories: target products (health foods), non-target products (products other than health foods), or set products (products including health foods). The instruction could be written, for example, as follows: "Based on the product description, please answer whether the product falls into the category of health foods, set products including health foods, or other products."

[0050] When the learning device 30 inputs the prompt created in this way to the language model 17, the language model 17 outputs a completion that includes the answer to the instruction. The learning device 30 obtains the output result corresponding to the prompt from the language model 17. Subsequently, based on the output result of the language model 17, the learning device 30 extracts one or more products that have been determined to be target products from among multiple products. Then, the learning device 30 adds product category data (for example, "health food") to the training dataset of each extracted product to complete the generation of the ground truth dataset.

[0051] [Method for generating classification models] Referring to Figure 4, an example of how the learning device 30 generates a classification model is explained. In step S11, the learning device 30 obtains multiple product datasets from the database 13 to be used as training data. Next, in step S12, the learning device 30 prepares the training data to be input into the pre-training model.

[0052] Preparing the training data may include preprocessing the training data. Preprocessing may include checking for missing data in each training dataset. Checking for missing data may include, for example, checking for the presence or absence of correct labels, i.e., whether or not product categories are included.

[0053] Preparing the training data may include extracting multiple training datasets from the acquired product datasets that contain the correct product categories (ground truth labels). Furthermore, as a check for missing data, the operator may verify whether the ground truth labels in at least some of the training datasets are correct. In addition, the operator may create ground truth labels for some product datasets that do not contain them.

[0054] The preparation of training data may include splitting multiple training datasets into training data, validation data, and test data. In this case, at least one of the training datasets from which the operator generated the correct labels and the dataset from which the operator verified the correct labels may be used as test data.

[0055] In step S13, the learning device 30 determines whether the amount of training data (e.g., the number of training datasets) is above a threshold. The threshold can be set by statistically calculating the number of data required to train the learning model. If the amount of data is below the threshold, the learning device 30 determines that the amount of data is insufficient (step S13 = NO) and proceeds to step S14 to add training data. The additional training data can be obtained, for example, by the learning device 30 acquiring multiple new product datasets from the database 13.

[0056] Alternatively, or in addition to the above, additional training data can also be generated, as described above, by adding data indicating product categories to product descriptions that do not contain the correct product category labels in the acquired product datasets. For example, the learning device 30 can generate product category data using the language model 17, as described above.

[0057] The learning device 30 may acquire a dummy product dataset generated by an operator for training purposes as at least part of the training data. For example, an operator may generate a dummy product description by adding the description of one or more specified ingredients to an arbitrary product description. In particular, when the product dataset of infringing products is small, it is useful to generate correct data for infringing products using a dummy product dataset.

[0058] After adding training data in step S14, the learning device 30 may return to step S12 to prepare the training data again. Once the training data is ready, the learning device 30 proceeds to step S13. In step S13, if the amount of data is above the threshold, the learning device 30 determines that the amount of data is sufficient (step S13 = YES) and proceeds to step S15.

[0059] In step S15, the learning device 30 trains the learning model by inputting training data into the learning model. This updates the weights in the learning model. In the following step S16, the learning device 30 tunes the parameters of the learning model using validation data.

[0060] In step S17, the learning device 30 tests the learning model using test data. For example, the learning device 30 calculates the accuracy of the learning model as a result of the test. The accuracy metric may be, for example, AUC (Area Under the Curve), but is not limited to this. Subsequently, the learning device 30 may determine whether the accuracy of the learning model is equal to or greater than a predetermined accuracy threshold. The accuracy threshold may be stored in memory 32. If the accuracy of the learning model is less than the accuracy threshold, the learning device 30 may determine that the learning of the learning model is insufficient and proceed to step S14 to add training data.

[0061] When the accuracy of the learning model exceeds a threshold, the learning device 30 determines that the learning model has been properly trained. In this case, the learning device 30 proceeds to step S18, stores the trained learning model in memory 32 as a trained classification model, and terminates the process.

[0062] [Method for detecting specified products] Referring to Figure 5, an example is given in which the detection device 20 detects illegal products containing multiple designated ingredients specified for health foods from among multiple products belonging to multiple product categories.

[0063] In step S21, the detection device 20 obtains multiple product datasets from the database 13 or memory 22. The multiple product datasets may include data relating to products before listing, or data relating to products currently on sale.

[0064] As shown in Figure 6, the detection device 20 may extract textual information contained in each product dataset as a product description. The product description may include, for example, a string 61 extracted from the tagline 51, a string 62 extracted from the title 52, a string 63 extracted from the detailed description 53, and a string 64 extracted from the image 54, all of which are included in the sales page 50 shown in Figure 2.

[0065] In step S22, the detection device 20 obtains one or more keywords from memory 22. In this example, the detection device 20 obtains a set of search keywords containing multiple keywords that each represent a different specified component. The set of search keywords includes, for example, the names of various components that would be illegal if contained in health foods, as shown in Figure 3.

[0066] In step S23, the detection device 20 performs a keyword search on the multiple product datasets acquired in step S21 using the multiple keywords acquired in step S22. More specifically, the detection device 20 searches the multiple product descriptions extracted from each of the multiple product datasets using the multiple keywords and extracts product descriptions that contain at least one of the keywords. The products corresponding to the extracted product descriptions are called suspected products. Therefore, the keyword search in step S23 extracts one or more suspected products whose product descriptions contain at least one of the one or more specified ingredients.

[0067] In step S24, the detection device 20 classifies the extracted one or more suspected items into one of three groups based on the corresponding product description: the target product (health food) group, the non-target product (products other than health foods) group, or the set product group.

[0068] More specifically, the detection device 20 inputs the product description of each suspected product into a classification model trained to classify health foods generated in the manner shown in Figure 4. The classification model then outputs the classification result for the corresponding suspected product based on the input product description. The detection device 20 obtains a list of target products extracted as health foods by acquiring the classification result from the classification model for each of the one or more suspected products.

[0069] Note that the classification in step S24 may be performed before the keyword search in step S22. In this case, a keyword search will be performed on multiple target products classified as health foods to extract one or more suspected products that contain at least one designated ingredient among the health foods.

[0070] In step S25, the detection device 20 performs named entity recognition (NER) on the product descriptions of each target product, i.e., the suspected products classified as health foods, and extracts ingredient names from the extracted proper nouns. As a result, one or more ingredient names are detected from the product descriptions of each suspected product.

[0071] The detection device 20 performs the processes in steps S26 to S30 for each of the multiple suspected items. The following explanation uses an example in which the detection device 20 detects multiple ingredient names from the product description of one suspected item.

[0072] In step S26, the detection device 20 extracts one or more specified ingredients that match the keyword from among multiple ingredient names in the product description. In Figure 6, the shaded proper nouns in the product description are the extracted specified ingredients.

[0073] In step S27, the detection device 20 detects negative descriptions from the product description of product 1 that describe ingredients that product 1 does not contain. Negative descriptions are statements that explicitly state ingredients that the product does not contain, but may also include statements that imply the absence of such ingredients. In the example shown in Figure 6, the sentence "We do not use silicone, mineral oil, ethanol, petroleum-based surfactants, wheat, egg, or shellfish-derived ingredients, cetanol, chlorphenesin, methylisothiazolinones, UV absorbers, UV scattering agents, parabens, fragrances, colorants, etc." is negative description 66.

[0074] The detection of negative descriptions may be achieved, for example, by including a set of negative expressions such as "not used," "not used," and "does not contain" in the detection data 25, and by the detection device 20 performing a search using the negative expressions as keywords for the product description. Alternatively, the detection device 20 may detect negative descriptions using a language model for detecting negative sentences.

[0075] In step S28, the detection device 20 detects the non-present components described in the negative explanation 66 from among the multiple component names. In Figure 6, among the shaded designated components, the component names enclosed in square frames are the non-present components.

[0076] In step S29, the detection device 20 detects positive descriptions from the product description 1 that describe the ingredients contained in product 1. A positive description is a statement that explicitly states the ingredients contained in the product, but it may also include statements that imply the ingredients. In the example shown in Figure 6, the sentence "[Contains the latest popular ingredients] Carefully selected and blended with retinol, arbutin, placenta, fullerene, plant stem cells, human-type ceramide (booster), vitamin A derivative, collagen, centella asiatica extract, and glycyrrhizic acid licorice" is positive description 67.

[0077] The detection of positive descriptions may be achieved, for example, by including a set of multiple ingredient expression phrases such as "formulation" and "raw materials" in the detection data 25, and by the detection device 20 performing a search using the ingredient expression phrases as keywords against the product description. Alternatively, the detection device 20 may detect positive descriptions using a language model for detecting the ingredient information of a product.

[0078] In step S30, the detection device 20 detects the contained components described in the positive explanation 67 from among multiple component names. In Figure 6, the component names highlighted in the positive explanation 67 are the contained components.

[0079] In step S31, the detection device 20 detects one or more non-containing products from among the one or more suspected products based on the product descriptions corresponding to each of the one or more suspected products, which do not contain one or more designated ingredients. More specifically, based on the results of steps S28 and S30, the detection device 20 determines that a corresponding suspected product is a non-containing product if all of the one or more designated ingredients included in each product description are non-containing ingredients included in the negative description 66, and one or more designated ingredients included in that product description are not included as contained ingredients in the positive description 67.

[0080] Steps S29 and S30 may be omitted. In this case, the detection device 20 may determine that the corresponding suspected product is a non-containing product if all one or more designated components included in each product description are non-containing components that are only included in the negative description 66. However, if the designated component is not included in the negative description but is included in the positive description, the corresponding suspected product may be determined to be a containing product.

[0081] In step S32, the detection device 20 excludes the non-containing product detected in step S31 from one or more suspected products. Then, in step S33, the detection device 20 outputs a list of infringing products, designating the remaining suspected products (illegal products in this example) as infringing products, and terminates the process. In this example, the list of infringing products is a list of containing products that contain at least one of several designated ingredients.

[0082] The detection device 20 may be configured to display a confirmation screen 60, including the detection and extraction results from steps S25 to S30, on the display 46 of the worker terminal 40, depending on the worker's work. The confirmation screen 60 may be configured to display, for example, the product description of one of the product datasets among multiple product datasets.

[0083] On the confirmation screen 60, one or more non-contained components and contained components may be highlighted with different marks. These different marks may be color-coded, such as red and blue markings, or they may be different textual styles, such as underlining and bolding. The confirmation screen 60 may include a legend 65 explaining the display of marks for non-contained and contained components.

[0084] In Figure 6, ingredient names other than those listed as "non-contained" in the negative description are highlighted as contained ingredients. Alternatively, non-contained ingredients in the negative description and contained ingredients in the positive description may be highlighted with different marks. Or, different marks may be used for non-contained ingredients in the negative description, contained ingredients in the positive description, and other ingredient names.

[0085] On the confirmation screen 60 for checking the contained ingredients, or on a confirmation screen separate from the confirmation screen 60 for checking keywords, the ingredient names extracted by named entity recognition in step S25 and one or more specified ingredients extracted using keywords in step S26 may be displayed with different marks. In this case, for example, the operator can visually identify the names of specified ingredients that are not included in the keyword set, or ingredient names that did not match the keywords due to a typographical error in the description. In this way, by extracting ingredient names by named entity recognition, it is possible to check whether the keyword settings are appropriate and to detect any missing keywords.

[0086] If a designated product detected as described above is a violation or illegal product, the worker may notify the seller of the designated product and request that they cease listing it, or the worker may directly cease listing or selling the designated product. If the designated product is a product containing a specific ingredient, the worker may continue listing or selling it after adding information to the designated product that is appropriate for its intended use. For example, the product may be labeled as containing a specific ingredient and sold, or the specific ingredient may be added to the product description (e.g., tagline) as a contained ingredient so that the product can be found through product searches using that ingredient.

[0087] [Effects of this disclosure] When a product description includes information about the product's ingredients, that description may include both positive descriptions of the ingredients the product contains and negative descriptions of ingredients the product does not contain. When such product descriptions are searched using keywords, the keywords will match not only the ingredients the product contains, but also the ingredients not included in the negative descriptions. As a result, products that do not contain the specified ingredient may be mistakenly identified as products that do contain the specified ingredient.

[0088] In this disclosure, after extracting suspected products from multiple products using keyword search, non-contained products that do not contain the designated ingredient are detected based on the product description of the suspected product. By excluding non-contained products from the suspected products, it becomes possible to detect only products that contain the designated ingredient.

[0089] Product descriptions may include the names of other ingredients that are not actually contained in the product, even if they are included in the description of the product's efficacy. For example, as shown in Figure 6, "Ingredients effective against dullness include niacinamide, tretinoin, and retinol, but this product contains particularly high-purity pure hydroquinone from carefully selected ingredients." In this respect, detecting positive descriptions in addition to negative descriptions, and detecting only products that explicitly contain the ingredient as containing products, allows for more accurate detection of the designated product. On the other hand, if products that do not contain the ingredient are excluded from the suspected products based on negative descriptions without detecting positive descriptions, a wider range of products that may contain the designated ingredient can be extracted.

[0090] When a product is sold as a set, the product description for that set includes descriptions for each of the multiple products included in the set. Therefore, by classifying the set from among multiple products, the set can be detected separately from the target product. This makes it possible to improve the detection accuracy of the target product. For example, for a set, the operator may detect whether or not it contains a specified product.

[0091] [Effects of this disclosure] According to this disclosure, the following effects can be achieved. (1) By excluding non-contained products that do not contain the designated ingredient from among suspected products that contain the designated ingredient in their product descriptions, it is possible to more accurately detect products that do contain the designated ingredient.

[0092] (2) By detecting negative descriptions within the product description, it is possible to determine that the ingredient names included in the negative descriptions are ingredients that the product does not contain. Therefore, if all of the specified ingredients included in the product description are included only in the negative descriptions, the corresponding suspected product can be determined to be a product that does not contain these ingredients.

[0093] (3) By detecting negative and positive descriptions within the product description, it is possible to determine whether the designated ingredient included in the product description is an ingredient contained in the suspected product or an ingredient not contained in the suspected product. Therefore, if the designated ingredient included in the product description is only included in the negative description and is not included in the positive description, the corresponding suspected product can be determined to be a product that does not contain the designated ingredient. Furthermore, if the designated ingredient is not included in the negative description but is included in the positive description, the corresponding suspected product can be determined to be a product that contains the designated ingredient.

[0094] (4) By performing named entity recognition on the product description, one or more ingredient names can be detected from the product description. This makes it possible to detect a specified ingredient from among the ingredient names.

[0095] (5) By displaying the confirmation screen 60 on the display 46, the operator can confirm the detected negative explanations, positive explanations, contained components, and non-contained components. This allows the operator to confirm whether the detection by the detection device 20 is being performed appropriately.

[0096] (6) If a designated ingredient is an illegal ingredient in a product category, the illegal product can be detected from among multiple products by outputting a list of products containing the illegal ingredient from among the products belonging to that product category. As a result, the listing and sale of illegal products on the e-commerce site can be stopped.

[0097] (7) By setting designated ingredients for each product category, it is possible to detect products containing the designated ingredients for each product category. (7) By extracting text from image 54, it becomes possible to detect the names of the components contained in image 54.

[0098] (8) By classifying multiple products into either target products belonging to a certain product category or non-target products belonging to other product categories based on their corresponding product descriptions, target products can be detected for each product category. This allows non-target products to be excluded from keyword searches or from the extraction of ingredient names. Furthermore, by classifying bundled products from among multiple products, it becomes possible to perform separate detection processing for bundled products.

[0099] (9) By using a classification model, the classification process for products can be carried out more efficiently. (10) By using the language model 17, the dataset can be generated more efficiently than the ground truth dataset.

[0100] This embodiment can be implemented with the following modifications. This embodiment and the following modifications can be combined with each other to the extent that they do not contradict each other technically. [Example of change 1] In the method illustrated in Figure 5 and the system and program for implementing this method, steps S31 to S33 may be replaced with the display of the confirmation screen 60 illustrated in Figure 6. In this case, the determination of whether the suspected product is a non-contained product or a contained product may be made by an operator instead of the detection device 20. Alternatively, the operator may use the confirmation screen 60 to confirm at least some of the multiple suspected products.

[0101] [Example of change 2] In the method illustrated in Figure 5 and the system and program for implementing this method, steps S27, S28, and S32 may be omitted, and instead of step S31, the detection device 20 may detect only products containing the designated ingredient in the positive description as infringing products.

[0102] [Example of change 3] When the detection device 20 detects one or more suspected items by keyword search in step S23, it may mark the ingredient names that match the keywords in the product description of each suspected item. In this case, the extraction of ingredient names by named entity recognition in step S25 and the extraction of specified ingredients in ingredient names in step S26 may be omitted.

[0103] [Example of change 4] Products detected based on designated ingredients are not limited to those containing undesirable ingredients such as illegal ingredients or allergens. For example, the technology of this disclosure can also be used to detect products containing specific desirable ingredients or products containing selectively used ingredients (e.g., antipyretic analgesics containing specific antipyretic and analgesic ingredients, or antiallergic drugs containing specific anti-allergic ingredients).

[0104] [Example of change 5] When detecting positive or negative explanations, the detection may include other requirements in addition to the ingredient name, such as the part of the plant used. This is particularly important because, in the case of plant-derived substances, only specific parts of the plant may be used as pharmaceuticals.

[0105] [Example of change 6] The methods for detecting negative and positive descriptions may differ depending on the product category. For example, if there is a characteristic way of describing a product category, a detection method based on that description can be adopted.

[0106] [Other examples of changes] The flowcharts and configuration diagrams of this disclosure illustrate the architecture, functionality, and operation of the apparatus, system, method, and program according to embodiments of this disclosure. Each step included in these flowcharts and each component included in the configuration diagrams may correspond to a portion of computer program code containing one or more instructions for realizing a logical functional unit. In other embodiments, some of the illustrated steps may be omitted, other steps may be included, the order of the steps may be different, or some steps may be executed simultaneously. Furthermore, a flowchart described as a series of actions may be divided into several parts and executed, or multiple flowcharts may be executed sequentially or in relation to each other. Also in other embodiments, some of the illustrated components may be omitted, other components may be included, or the arrangement of components may be changed. Furthermore, the functions realized by these steps and components may be realized by hardware, software, or a combination of hardware and software.

[0107] The systems of this disclosure may be implemented as a single device or distributed across multiple devices or subsystems that collaborate to execute a program. Examples of processing systems include general-purpose central processing units, application-specific processors, and logic devices, as well as any other type of processing devices, combinations thereof, or variations thereof.

[0108] The method relating to this disclosure is embodied in software including various programs and data, and is executed by a computing system (information processing system) that runs the software. The computing system comprises a non-transitory computer-readable medium for storing various instructions included in the program, and one or more processors for executing these software instructions. The computing system may be implemented by a standalone computer, or by a client-server architecture in which the software is executed in a distributed manner by multiple computers.

[0109] Memory is a computer-readable storage medium, including non-transitory computer-readable medium. Non-transitory computer-readable medium may include, but is not limited to, ROM, hard disks, storage, removable media, flash memory, memory sticks, optical media, magneto-optical media, and CD-ROMs.

[0110] At least one processor may include, but is not limited to, other processors including a CPU (central processing unit), GPU (graphics processing unit), APU (accelerated processing unit), NPU (Neural network Processing Unit), microprocessor, microcontroller, DSP (digital signal processor), FPGA (field programmable gate array), CPLD (Complex Programmable Logic Device), application-specific integrated circuit (ASIC), general-purpose processor, or any combination thereof designed to perform the functions described herein.

[0111] Communication between multiple devices or systems may be carried out via one or more communication networks in accordance with well-known communication protocols. The communication network may be, but is not limited to, an intranet, the internet, a local area network, a wide area network, a wireless network, a wired network, a virtual network, a software-defined network, or any other type of network, or a combination thereof.

[0112] A communication interface (IF) enables one device to communicate with another device via a communication network. The communication interface may be, but is not limited to, a LAN (Local Area Network), Wi-Fi (registered trademark), Bluetooth (registered trademark), NFC (Near Field Communication), or other wireless communication interfaces.

[0113] The embodiments and modifications described above are listed below. [1] At least one processor, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. Based on the product description corresponding to each of the one or more suspected products, one or more non-containing products that do not contain the one or more designated ingredients are detected from among the one or more suspected products. By excluding the one or more non-containing products from the one or more suspected products, a list of containing products containing at least one of the multiple designated ingredients is output. A method for detecting specified products, including the method described above.

[0114] [2] Detecting one or more non-containing products means that for each of the one or more suspected products, From the aforementioned product descriptions, we will detect negative descriptions that explain non-contained ingredients not present in the corresponding product, To detect one or more of the non-contained components described in the negative explanation above, If all of the one or more designated ingredients included in the product description are the non-included ingredients that are included only in the negative description, then the corresponding suspected product will be determined to be the non-included product. The specified product detection method described in [1] above, including the method described in [1] above.

[0115] [3] Detecting one or more non-containing products means that for each of the one or more suspected products, From the aforementioned product descriptions, we will detect negative descriptions that explain non-contained ingredients not present in the corresponding product, From the aforementioned product descriptions, we will detect positive descriptions that explain the ingredients contained in the corresponding product, To detect one or more of the non-contained components described in the negative explanation above, To detect one or more of the aforementioned components described in the affirmative explanation, If all of the one or more designated ingredients included in the product description are the non-included ingredients included in the negative description, and the one or more designated ingredients included in the product description are not included as included ingredients in the positive description, then the corresponding suspected product will be determined to be the non-included product. including, The method for detecting the designated product as described in [1] above.

[0116] [4] This includes displaying a confirmation screen on the display, The confirmation screen is configured to display the product description of one of the multiple product datasets. In the product description on the confirmation screen, the one or more non-contained ingredients and the one or more contained ingredients are highlighted with different marks. A method for detecting designated products as described in any of the above [1] to [3].

[0117] [5] At least one processor, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. From the product descriptions of each of the one or more suspected products, detect negative descriptions that explain non-contained components that are not present in the corresponding suspected product. To detect one or more of the non-contained components described in each of the negative explanations above, Displaying a confirmation screen on the display for determining whether each of the suspected products contains the one or more designated components, wherein the confirmation screen is configured to display the product description of the corresponding suspected product, and the one or more non-contained components are highlighted in the product description on the confirmation screen, A method for detecting specified products, including the method described above.

[0118] [6] Detecting one or more non-containing products means that for each of the one or more suspected products, By performing named entity recognition on the aforementioned product description, one or more ingredient names are detected within the product description. Extracting one or more designated components from the one or more component names mentioned above, including, A method for detecting designated products as described in any of the above [1] to [5].

[0119] [7] Each of the aforementioned products is classified into one of the following product categories: The one or more designated ingredients mentioned above are designated for one of the product categories among the multiple product categories mentioned above. The one or more designated ingredients mentioned above are illegal ingredients that are illegal in the product category mentioned in item 1. A method for detecting designated products as described in any of the above [1] to [6].

[0120] [8] Each of the aforementioned products is classified into one of the following product categories: The aforementioned product categories include at least one of the following: health foods, pharmaceuticals, quasi-drugs, and cosmetics. The one or more specified ingredients mentioned above are set for each product category. A method for detecting designated products as described in any of the above [1] to [7].

[0121] [9] If one of the multiple product datasets contains one or more images, and at least a portion of those one or more images contains a string, the method further includes extracting the string from the one or more images. The product description of the product dataset mentioned above includes an extracted string, which is the string extracted from the one or more images mentioned above. A method for detecting designated products as described in any of the above [1] to [8].

[0122]

[10] Each of the aforementioned products is classified into one of the following product categories: The one or more designated ingredients mentioned above are designated for one of the product categories among the multiple product categories mentioned above. The process includes classifying the aforementioned multiple products into three groups based on the product description corresponding to each of the aforementioned multiple products, wherein the three groups are: The group of target products belonging to the product category 1 above, A group of non-applicable products belonging to product categories other than those specified in item 1 above, A group of set products including the aforementioned target product, A method for detecting designated products as described in any of the above [1] to [9].

[0123]

[11] Classifying into the three groups mentioned above is, The method involves inputting one product description into a multi-class classification model, wherein the multi-class classification model is configured to classify the product into one of the three groups based on the product description. This includes obtaining the classification result of product 1 from the multi-class classification model, A method for detecting designated products as described in any of the above [1] to

[10] .

[0124]

[12] The multi-class classification model is generated by training it with training data, To generate multiple ground truth datasets to be used as training data, Includes, The aforementioned training data includes multiple training product datasets, Each of the aforementioned training product datasets includes a training product description for the corresponding product. The aforementioned training product description includes a description of the ingredients contained in the corresponding product, Each of the aforementioned ground truth datasets is: The training product description for the corresponding product, The above includes a correct label indicating whether the corresponding product is the target product, the non-target product, or the set product, Generating the aforementioned multiple ground truth datasets is, Entering prompts into the language model, Obtaining output results corresponding to the prompt from the language model, Includes, The aforementioned prompt is, A plurality of training product descriptions for each of the plurality of products, The instructions include causing the language model to output a response indicating whether each of the multiple products described by each of the multiple training product descriptions falls under the category of target products, non-target products, and set products, A method for detecting designated products as described in any of the above [1] to

[11] .

[0125]

[13] At least one memory for storing computer program code, At least one processor, The system comprises, and the at least one processor executes the computer program code, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. Based on the product description corresponding to each of the one or more suspected products, one or more non-containing products that do not contain the one or more designated ingredients are detected from among the one or more suspected products. By excluding the one or more non-containing products from the one or more suspected products, a list of containing products containing at least one of the multiple designated ingredients is output. A specified product detection system configured to perform the following actions.

[0126]

[14] At least one memory for storing computer program code, At least one processor, The system comprises, and the at least one processor executes the computer program code, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. From the product descriptions of each of the one or more suspected products, detect negative descriptions that explain non-contained components that are not present in the corresponding suspected product. To detect one or more of the non-contained components described in each of the negative explanations above, Displaying a confirmation screen on the display for determining whether each of the suspected products contains the one or more designated components, wherein the confirmation screen is configured to display the product description of the corresponding suspected product, and the one or more non-contained components are highlighted in the product description on the confirmation screen, A specified product detection system configured to perform the following actions.

[0127]

[15] At least one processor, The process involves obtaining multiple product datasets, each corresponding to a specific product, wherein each product dataset includes a product description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the multiple designated ingredients in its corresponding product description. From among the one or more suspected products, one or more non-containing products that do not contain the multiple designated ingredients are detected based on the product description corresponding to each of the one or more suspected products. By excluding the one or more non-containing products from the one or more suspected products, a list of containing products that contain at least one of the multiple designated ingredients is output. A program to execute.

[0128]

[16] At least one processor, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. From the product descriptions of each of the one or more suspected products, detect negative descriptions that explain non-contained components that are not present in the corresponding suspected product. To detect one or more of the non-contained components described in each of the negative explanations above, Displaying a confirmation screen on the display for determining whether each of the suspected products contains the one or more designated components, wherein the confirmation screen is configured to display the product description of the corresponding suspected product, and the one or more non-contained components are highlighted in the product description on the confirmation screen, A program to execute. [Explanation of Symbols]

[0129] 11…Specified product detection system, 13…Database, 14…Web server, 15…Purchaser terminal, 16…Seller terminal, 17…Language model, 20…Specified product detection device, 21,31,41…Processor, 22,32,42…Memory, 23,33,43…Communication interface, 24,34…Program, 25…Detection data, 30…Learning device, 35…Learning data, 40…Worker terminal, 46…Display, 50…Sales page, 51…Tagline, 52…Title, 53…Detailed description, 54…Image, 60…Confirmation screen, 61,62,63,64…String, 66…Negative description, 67…Positive description.

Claims

1. At least one processor, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. Based on the product description corresponding to each of the one or more suspected products, one or more non-containing products that do not contain the one or more designated ingredients are detected from among the one or more suspected products. By excluding the one or more non-containing products from the one or more suspected products, a list of containing products containing at least one of the multiple designated ingredients is output. A method for detecting specified products, including the method described above.

2. Detecting one or more non-containing products means that for each of the one or more suspected products, From the aforementioned product descriptions, we will detect negative descriptions that explain non-contained ingredients not present in the corresponding product, To detect one or more of the non-contained components described in the negative explanation above, If all of the one or more designated ingredients included in the product description are the non-included ingredients that are included only in the negative description, then the corresponding suspected product will be determined to be the non-included product. A method for detecting a designated product according to claim 1, including the method described in claim 1.

3. Detecting one or more non-containing products means that for each of the one or more suspected products, From the aforementioned product descriptions, we will detect negative descriptions that explain non-contained ingredients not present in the corresponding product, From the aforementioned product descriptions, we will detect positive descriptions that explain the ingredients contained in the corresponding product, To detect one or more of the non-contained components described in the negative explanation above, To detect one or more of the aforementioned components as described in the affirmative explanation above, If all of the one or more designated ingredients included in the product description are the non-included ingredients included in the negative description, and the one or more designated ingredients included in the product description are not included as included ingredients in the positive description, then the corresponding suspected product will be determined to be the non-included product. A method for detecting a designated product according to claim 1, including the method described in claim 1.

4. This includes displaying a confirmation screen on the display, The confirmation screen is configured to display the product description of one of the multiple product datasets. In the product description on the confirmation screen, the one or more non-contained ingredients and the one or more contained ingredients are highlighted with different marks. The method for detecting a designated product as described in claim 3.

5. At least one processor, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. From the description of each of the one or more suspected products, detect negative descriptions that explain non-contained components that are not present in the corresponding suspected product. To detect one or more of the non-contained components described in each of the negative explanations above, Displaying a confirmation screen on the display for determining whether each of the suspected products contains the one or more designated components, wherein the confirmation screen is configured to display the product description of the corresponding suspected product, and the one or more non-contained components are highlighted in the product description on the confirmation screen. A method for detecting specified products, including the method described above.

6. Detecting one or more non-containing products means that for each of the one or more suspected products, By performing named entity recognition on the aforementioned product description, one or more ingredient names are detected within the product description. Extracting one or more designated components from the one or more component names mentioned above, including, A method for detecting a designated product according to any one of claims 1 to 4.

7. Each of the aforementioned products is classified into one of the following product categories: The one or more designated ingredients mentioned above are designated for one of the multiple product categories mentioned above. The one or more designated ingredients mentioned above are illegal ingredients that are illegal in the product category mentioned in item 1. A method for detecting a designated product according to any one of claims 1 to 5.

8. Each of the aforementioned products is classified into one of the following product categories: The aforementioned multiple product categories include at least one of the following: health foods, pharmaceuticals, quasi-drugs, and cosmetics. The one or more specified ingredients mentioned above are set for each product category. A method for detecting a designated product according to any one of claims 1 to 5.

9. If one of the product datasets among the plurality of product datasets contains one or more images, and at least a portion of the one or more images contains a string, the method further includes extracting the string from the one or more images. The product description of the product dataset mentioned above includes an extracted string, which is the string extracted from the one or more images mentioned above. A method for detecting a designated product according to any one of claims 1 to 5.

10. Each of the aforementioned products is classified into one of the following product categories: The one or more designated ingredients mentioned above are designated for one of the multiple product categories mentioned above. The process includes classifying the aforementioned multiple products into three groups based on the product description corresponding to each of the aforementioned multiple products, wherein the three groups are: The group of target products belonging to the product category 1 above, A group of non-applicable products belonging to product categories other than those specified in item 1 above, A group of set products including the aforementioned target product, A method for detecting a designated product according to any one of claims 1 to 5.

11. Classifying into the three groups mentioned above is The method involves inputting one product description into a multi-class classification model, wherein the multi-class classification model is configured to classify the product into one of the three groups based on the product description. This includes obtaining the classification result of product 1 from the multi-class classification model, The method for detecting a designated product according to claim 10.

12. The multi-class classification model is generated by training it with training data, To generate multiple ground truth datasets to be used as training data, Includes, The aforementioned training data includes multiple training product datasets, Each of the aforementioned training product datasets includes a training product description for the corresponding product. The aforementioned training product description includes a description of the ingredients contained in the corresponding product, Each of the aforementioned ground truth datasets is: The training product description for the corresponding product, The above includes a correct label indicating whether the corresponding product is the target product, the non-target product, or the set product, Generating the aforementioned multiple ground truth datasets is, Entering prompts into the language model, Obtaining output results corresponding to the prompt from the language model, Includes, The aforementioned prompt is, A plurality of training product descriptions for each of the plurality of products, The instructions include causing the language model to output a response indicating whether each of the multiple products described by each of the multiple training product descriptions falls under the category of target products, non-target products, and set products, The method for detecting a designated product according to claim 11.

13. At least one memory for storing computer program code, At least one processor, The system comprises, and the at least one processor executes the computer program code, The method involves obtaining multiple product datasets, each of which includes product descriptions for multiple products, and each product description includes a description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the one or more designated ingredients in its corresponding product description. Based on the product description corresponding to each of the one or more suspected products, one or more non-containing products that do not contain the one or more designated ingredients are detected from among the one or more suspected products. By excluding the one or more non-containing products from the one or more suspected products, a list of containing products containing at least one of the multiple designated ingredients is output. A specified product detection system configured to perform the following actions.

14. At least one memory for storing computer program code, At least one processor, The system comprises, and the at least one processor executes the computer program code, The process involves obtaining a product dataset, wherein the product dataset includes a product description, and the product description includes a description of the product's ingredients. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The method involves performing a keyword search on the product description using the one or more keywords mentioned above to determine whether the product is a suspected product, wherein the suspected product contains at least one of the one or more designated ingredients mentioned above in the product description. From the description of the suspected product, detect any negative descriptions that explain components not contained in the suspected product. To detect one or more of the non-contained components described in the negative explanation above, Displaying a confirmation screen on a display for determining whether the suspected product contains the one or more designated components, wherein the confirmation screen is configured to display the product description of the suspected product, and the one or more non-contained components are highlighted in the product description on the confirmation screen. A specified product detection system configured to perform the following actions.

15. At least one processor, The process involves obtaining multiple product datasets, each corresponding to a specific product, wherein each product dataset includes a product description of the ingredients contained in the corresponding product. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The process involves extracting one or more suspected products from the multiple product datasets by performing a keyword search using one or more keywords, wherein each suspected product contains at least one of the multiple designated ingredients in its corresponding product description. From among the one or more suspected products, one or more non-containing products that do not contain the multiple designated ingredients are detected based on the product description corresponding to each of the one or more suspected products. By excluding the one or more non-containing products from the one or more suspected products, a list of containing products that contain at least one of the multiple designated ingredients is output. A program to execute.

16. At least one processor, The process involves obtaining a product dataset, wherein the product dataset includes a product description, and the product description includes a description of the product's ingredients. The means of obtaining one or more keywords, wherein each of the one or more keywords represents one or more specified components, The method involves performing a keyword search on the product description using the one or more keywords mentioned above to determine whether the product is a suspected product, wherein the suspected product contains at least one of the one or more designated ingredients mentioned above in the product description. From the description of the suspected product, detect any negative descriptions that explain components not contained in the suspected product. To detect one or more of the non-contained components described in the negative explanation above, Displaying a confirmation screen on a display for determining whether the suspected product contains the one or more designated components, wherein the confirmation screen is configured to display the product description of the suspected product, and the one or more non-contained components are highlighted in the product description on the confirmation screen. A program to execute.

Images