Fraudulent commodity detection method, fraudulent commodity detection device, equipment and medium

Abstract

The invention relates to a fraudulent commodity detection method and device, equipment and a medium, and the method comprises the steps: determining a target detection commodity and related information based on the category information of a to-be-detected commodity; performing similarity matching on the target detection commodity and commodities in a preset market price library, and recalling part of commodities matched with the target detection commodity in the market price library to form a matched commodity set; on the basis of brand information, category information and multiple key attributes of corresponding commodities in the matched commodity set, calculating factor matching scores of the commodities in the matched commodity set, and determining the commodity with the highest factor matching score; and determining a price threshold based on the plurality of commodity prices corresponding to the commodity with the highest score, and comparing the commodity price of the target detection commodity with the price threshold to judge whether the target detection commodity is a fraudulent commodity. According to the invention, through a multi-stage and multi-dimensional detection and matching mechanism, accurate detection of fraudulent commodities is carried out for standard commodities.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a fraudulent product detection method and its device, equipment, and medium. Background Art

[0002] With the rapid development of e-commerce, online commodity transactions have become one of the main ways for consumers to shop. However, with the continuous expansion of the market, the problem of fraudulent goods has become increasingly prominent, becoming a difficult problem that needs to be solved urgently in the field of e-commerce technology. Fraudulent goods usually attract consumers at extremely low prices, but the actual goods received by consumers after purchase are often significantly different from the descriptions on the e-commerce platform. Some merchants even lose contact directly after receiving the payment, causing consumers to not only suffer economic losses, but also face a serious shopping trust crisis.

[0003] In the prior art, only routine compliance checks are conducted when products are put on the shelves. This test mainly focuses on the surface compliance of product information, and no special tests are conducted on fraudulent products. This detection method cannot identify fraudulent behaviors that circumvent the rules by disguising or tampering with product information. For example, fraudulent merchants may conceal the true attributes of products by modifying product titles, descriptions, or pictures, or mislead consumers through false brand information and category information. These behaviors are difficult to detect in routine compliance tests, resulting in fraudulent products being circulated on the platform, further exacerbating the consumer trust crisis. Summary of the invention

[0004] The primary purpose of the present application is to solve at least one of the above problems and provide a fraudulent product detection method and its device, equipment, and medium.

[0005] In order to meet the various objectives of this application, this application adopts the following technical solutions:

[0006] A fraudulent product detection method provided for one of the purposes of this application comprises the following steps:

[0007] Based on the category information of the product to be detected, determine the target product and its related information, wherein the related information includes brand information, category information, and multiple key attributes;

[0008] Perform similarity matching between the target detected product and the products in the preset market price library, and recall some products in the market price library that match the target detected product to form a matching product set;

[0009] Calculating factor matching scores of products in the matching product set based on brand information, category information, and multiple key attributes of corresponding products in the matching product set, and determining a product with the highest factor matching score;

[0010] Based on the multiple commodity prices corresponding to the commodity with the highest score, a price threshold is determined, and the commodity price of the target detected commodity is compared with the price threshold to determine whether the target detected commodity is a fraudulent commodity.

[0011] On the other hand, a fraudulent commodity detection device provided for one of the purposes of the present application includes:

[0012] A target detection commodity determination module is used to determine the target detection commodity and related information based on the category information of the commodity to be detected, wherein the related information includes brand information, category information and multiple key attributes;

[0013] A similarity matching module is used to perform similarity matching between the target detected product and the products in the preset market price library, that is, to recall some products in the market price library that match the target detected product to form a matching product set;

[0014] A matching score calculation module, used to calculate the factor matching scores of the products in the matching product set based on the brand information, category information and multiple key attributes of the corresponding products in the matching product set, and determine the product with the highest factor matching score;

[0015] The fraudulent product judgment module is used to determine a price threshold based on multiple product prices corresponding to the product with the highest score, and compare the product price of the target detected product with the price threshold to determine whether the target detected product is a fraudulent product.

[0016] On the other hand, a computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the fraudulent product detection method described in the present application.

[0017] On the other hand, a computer program product provided to meet another purpose of the present application includes a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any embodiment of the present application.

[0018] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0019] On the one hand, this application uses a multi-stage screening mechanism to quickly filter out irrelevant goods and determine the goods that best match the target detection goods. Among them, in the primary stage of screening, based on the category information of the goods to be detected, the goods to be detected whose category information is characterized as standard goods are determined as target detection goods; in the first stage of product screening in the market price library, the similarity matching method is used to screen out goods that match the target detection goods to form a matching product set, which effectively narrows the detection range and improves the detection efficiency; in the second stage of product screening in the market price library, the goods in the narrowed matching product set are subjected to multi-dimensional factor matching score calculations to determine the goods that are most similar to the target detection goods. This multi-stage detection mechanism not only improves the efficiency of detection, but also significantly improves the accuracy of detection.

[0020] On the other hand, this application uses a multi-dimensional matching mechanism to comprehensively evaluate products from multiple angles to ensure the comprehensiveness and robustness of the test results. In the factor matching score calculation, the factor matching score of each product is calculated based on multi-dimensional information such as brand information, category information, and multiple key attributes. Through this multi-dimensional matching scoring mechanism, the similarity between products can be fully reflected, avoiding the deviation that may be caused by single-dimensional scoring.

[0021] Through a multi-stage and multi-dimensional detection and matching mechanism, this application significantly improves the efficiency, accuracy, comprehensiveness and robustness of fraudulent product detection, provides the platform with a more effective means of fraudulent product detection, and thus ensures the transaction security and user experience of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0023] Figure 1 A flowchart of a typical embodiment of the fraudulent product detection method of the present application;

[0024] Figure 2 A schematic diagram of a process for determining a target detection commodity based on category information in an embodiment of the present application;

[0025] Figure 3 A schematic diagram of a similarity matching process in an embodiment of the present application;

[0026] Figure 4 This is a schematic diagram of a process for selecting commodities using the common subsequence recall rate in an embodiment of the present application;

[0027] Figure 5 A schematic diagram of a process for calculating factor matching scores in an embodiment of the present application;

[0028] Figure 6 It is a flowchart of determining a price threshold based on an average discount margin value in an embodiment of the present application;

[0029] Figure 7 This is a schematic diagram of a process for further marking abnormal commodities according to the deviation ratio in an embodiment of the present application;

[0030] Figure 8 This is a principle block diagram of the fraudulent product detection device of the present application;

[0031] Fig. 9 A schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[0032] A fraudulent commodity detection method of the present application can be programmed as a computer program product, deployed in a server and implemented. For example, in the exemplary application scenario of the present application, it can be deployed in the server of an e-commerce platform, wherein the e-commerce platform can be an e-commerce platform that opens independent station services. An independent station refers to a new type of official website (website) established on the basis of a SaaS technology platform, which has an independent domain name, private content, data, and rights and interests, independent business sovereignty and business entity responsibilities, supported by social cloud computing capabilities, and can independently and freely connect to third-party software tools, publicity and promotion media and channels. Usually, the seller of goods builds and operates an online store on an independent station, so that the online store can display its goods on the independent station, and provide e-commerce services such as searching for goods, adding shopping carts, settlement, logistics delivery, placing orders, and after-sales service on the online store to the buyers of goods (that is, users). The buyers of goods (users) perform various interactive operations with the e-commerce platform on the terminal device.

[0033] In some embodiments of the present application, the relevant information of the goods is stored in the database (relational database or non-relational database) of the e-commerce platform, and the e-commerce platform usually has one or more databases to store the relevant information of the goods. The present application first obtains the category information of the goods to be detected from the database, and judges whether the goods to be detected are applicable to the fraudulent goods detection method of the present application according to the category information. It should be noted that the fraudulent goods detection method of the present application is only applicable to standard goods (goods with clear standards and rules). For example, if the category information characterizes that the goods are home appliances, it is determined that the goods to be detected are applicable to the fraudulent goods detection method of the present application, and the goods to be detected are determined as the target detection goods; if the category information characterizes that the goods are clothing, it is determined that the goods to be detected are not applicable to the fraudulent goods detection method of the present application. When it is determined that the goods to be detected are applicable to the method of the present application, the goods to be detected are determined as target detection goods, and the relevant information of the target detection goods is obtained, and the relevant information includes brand information, category information and multiple key attributes, and the key attributes can be models, specifications, etc.

[0034] After determining the target detection commodity and its related information, the target detection commodity is matched with the commodities in the preset market price library for similarity, and the commodities in the market price library are recalled, and some commodities that match the target detection commodity are recalled. Among them, the preset market price library is a database that contains standard market price information of a variety of standard commodities. The database is used to provide a reference point so that when detecting fraudulent commodities, the price of the target detection commodity can be compared with it, so as to identify commodities with abnormal prices. For the specific steps of similarity matching between the target detection commodity and the commodities in the preset market price library, please refer to the subsequent specific implementation method, which will not be repeated here.

[0035] Filter out some of the commodities in the market price library that match the target detected commodities to form a set of matched commodities. Based on the set of matched commodities, calculate the brand information, category information and respective matching scores (factor matching scores) of the commodities in the set of matched commodities. The respective matching scores of the brand information, category information and multiple key attributes can be added to determine the commodity with the highest factor matching score, and then obtain multiple commodity prices of the highest commodity, wherein the commodity has multiple commodity prices due to specification differences, configuration differences, version differences and the like. Based on multiple commodity prices, determine a price threshold, which is used to determine whether the target detected commodity is a fraudulent commodity. For example, when the price of the target detected commodity is lower than the price threshold, the target detected commodity is determined to be a fraudulent commodity.

[0036] A fraudulent product detection method of the present application can be programmed as a computer program product and deployed in a server for execution. For example, in the exemplary application scenario of the present application, it can be deployed and implemented in a server of an e-commerce customer service platform, whereby the method can be executed by accessing an interface opened after the computer program product is run and performing human-computer interaction with the process of the computer program product through a graphical user interface.

[0037] See also Figure 1 The fraudulent commodity detection method of the present application, in its typical embodiment, comprises the following steps:

[0038] Step S5100: Determine the target detected product and related information thereof based on the category information of the product to be detected, wherein the related information includes brand information, category information, and multiple key attributes;

[0039] In one embodiment, the category information of the commodity to be detected can be directly obtained from a database storing relevant information of the commodity. In another embodiment, the product title text of the commodity to be detected is obtained, and the product title text is input into a pre-trained category recognition model. The category recognition model is trained based on a large amount of commodity data, and can automatically identify the category information to which the commodity belongs based on the product title text. For example, for the product title text "Apple iPhone14Pro Max 512GB Deep Space Black", the category recognition model will recognize that the product belongs to the "smartphone" category.

[0040] In another embodiment, after obtaining the category information of the commodity to be detected, the category information is matched with multiple category information in a preset standard product category library. The standard product category library is pre-defined based on standard products. If the category information of the commodity to be detected successfully matches a certain category information in the standard product category library, the commodity to be detected is determined as a target detection commodity. In another embodiment, the commodity title text and commodity description text of the commodity to be detected are obtained, and the commodity title text and commodity description text are input into a pre-trained commodity recognition model. The standard product recognition model can automatically identify whether the commodity is a standard product based on the commodity title text and the commodity description text. For example, for the commodity title text "Apple iPhone 14Pro Max 512GB Deep Space Black", the standard product recognition model will identify the commodity as a standard product because it has clear attributes such as model, color and storage capacity. The commodity to be detected identified as the identification commodity is determined to be a target detection commodity. And obtain the relevant information of the target detection commodity from the database storing the relevant information of the commodity, including the brand information, category information and multiple key attributes of the target detection commodity. Key attributes may include model, specification, etc.

[0041] Step S5200: perform similarity matching between the target detected product and the products in the preset market price library, and recall some products in the market price library that match the target detected product to form a matching product set;

[0042] The preset market price library contains the market price information of various standard products. After determining the target detection product and its related information in the previous step, the target detection product is matched with the products in the preset market price library for similarity, aiming to quickly screen out products similar to the target detection product through similarity matching.

[0043] In one embodiment, the relevant information of the target detected commodity is matched with the commodities in the market price library for similarity. The market price library is pre-constructed and contains relevant information and price data of a large number of commodities. The data in the market price library is constructed based on historical transaction records, market research, industry standards and other sources. The similarity matching process is divided into two stages. In the first stage, the brand information, category information and multiple key attributes of the target detected commodity are matched with the commodities in the market price library for cosine similarity. Cosine similarity is a commonly used similarity calculation method that can measure the similarity between two vectors. The cosine similarity between the target detected commodity and each commodity in the market price library is calculated, and based on a preset first similarity score threshold, the commodities that are preliminarily matched with the target detected commodity are screened out to form a preliminarily matched commodity set. In the second stage, the longest common subsequence recall rate of the relevant information of the target detected commodity and each commodity in the preliminarily matched commodity set is further calculated. The longest common subsequence recall rate is an indicator for measuring the similarity of two text sequences, which can capture the degree of matching of key information in the text. The longest common subsequence recall rate of the target detected product and each preliminary matching product is calculated, and based on the preset recall rate threshold, the products that are secondary matched with the target detected product are further screened out to form the final set of matching products.

[0044] Through this two-stage similarity matching process, it is possible to efficiently screen out products similar to the target product from the market price database to form a set of matching products. The matching products obtained will provide an important basis for subsequent price comparison and fraud judgment, ensuring the accuracy and reliability of fraud product detection.

[0045] Step S5300: Calculate the factor matching scores of the products in the matching product set based on the brand information, category information, and multiple key attributes of the corresponding products in the matching product set, and determine the product with the highest factor matching score;

[0046] After completing the similarity matching between the target detected product and the products in the market price database, this step calculates the factor matching score of each product based on the brand information, category information and multiple key attributes of the corresponding products in the matching product set, and determines the product with the highest factor matching score. The purpose is to screen out the products that are most similar to the target detected product through multi-dimensional matching scores, providing a more accurate basis for subsequent price comparison and fraud judgment.

[0047] Specifically, the brand information, category information and multiple key attributes of each product in the matching product set are obtained. The key attributes may include product title text, description text and other important attributes, such as model, color, storage capacity, etc. Then, the corresponding matching score is preset based on the brand information, category information and multiple key attributes. Among them, the matching score is preset according to the degree of influence or experience of the brand information, category information and multiple key attributes on the price. For example, if the brand information of the matching product is completely consistent with the brand information of the target detection product, a higher brand matching score (for example, 3 points) is obtained; if the category information is consistent, a category matching score (for example, 2 points) is obtained; if the key attributes in the product title text or description text match, a corresponding text matching score (for example, 1 point) is obtained. The matching scores of the brand information, category information and multiple key attributes of each product are added together to obtain the factor matching score of the corresponding product. The factor matching score reflects the degree of matching between the product and the target detection product in multiple dimensions. The higher the score, the higher the matching degree. The factor matching score of each product in the matching product set is calculated one by one, and the product with the highest score is determined.

[0048] In this way, the products that are most similar to the target product can be screened out from the set of matching products, ensuring that subsequent price comparisons and fraud judgments are based on the most relevant market data. This not only improves the accuracy of fraudulent product detection, but also provides a reliable basis for subsequent price threshold calculations.

[0049] Step S5400: determine a price threshold based on the multiple product prices corresponding to the product with the highest score, and compare the product price of the target detected product with the price threshold to determine whether the target detected product is a fraudulent product.

[0050] Based on the multiple commodity prices corresponding to the commodity, a price threshold is calculated, and the commodity price of the target detection commodity is compared with the price threshold to determine whether the target detection commodity is a fraudulent commodity. This step identifies abnormally low-priced commodities through price comparison to determine whether there is fraud. Specifically, multiple commodity prices corresponding to the highest-scoring commodity are obtained, wherein, during the fraudulent commodity detection process, the commodity prices of standard commodities usually fluctuate according to their key attributes (such as size, specifications, models, etc.). For example, for a standard commodity such as a television, the prices of an 85-inch television and a 32-inch television will differ significantly due to their different sizes.

[0051] In one embodiment, after obtaining multiple commodity prices corresponding to the commodity with the highest score, the average discount value of the commodity is calculated based on the multiple commodity prices and the historical prices corresponding to each commodity price. For the specific calculation steps, please refer to the subsequent specific implementation method, which will not be repeated here. Based on the calculated average discount value and the preset mapping relationship, the corresponding detection ratio is determined. The mapping relationship is predefined and can be expressed in the form of a mapping table. For example, if the average discount value is 10%, find the detection ratio corresponding to the average discount value of 10% in the mapping table, assuming it is 0.8. Then multiply 0.8 by the lowest price among the prices corresponding to the highest-scoring commodity to obtain the price threshold. If the lowest price of the highest-scoring commodity is 5,000 yuan and the detection ratio is 0.8, the price threshold is 4,000 yuan.

[0052] After determining the price threshold, the price of the target detected product is compared with the price threshold. If the price of the target detected product is lower than the price threshold, the product is judged to be a fraudulent product. Since its price is significantly lower than the lowest price of similar products in the market price database, there may be fraud. If the price of the target detected product is higher than or equal to the price threshold, the product is judged to be a normal product and its price is within a reasonable range.

[0053] In this way, the price threshold can be accurately calculated based on the data in the market price database, and abnormally low-priced goods can be identified through price comparison to determine whether they are fraudulent goods. This step not only improves the accuracy of fraudulent goods detection, but also provides the platform with an effective means of fraudulent goods detection, ensuring the transaction security and user experience of the platform.

[0054] It can be known from the typical embodiments of the present application that the technical solution of the present application has many advantages, including but not limited to the following aspects:

[0055] On the one hand, the present application uses a multi-stage screening mechanism to quickly filter out irrelevant goods and determine the goods that best match the target detection goods. Among them, in the screening of the primary stage, based on the category information of the goods to be detected, the goods to be detected whose category information is characterized as standard goods are determined as target detection goods; in the screening of goods in the market price library in the first stage, through the similarity matching method, the cosine similarity matching method is first used to quickly screen out the preliminary matching goods set, and then the longest common subsequence recall rate is used to further screen the preliminary matching goods set, and the goods that match the target detection goods are screened out to form a matching goods set, which effectively narrows the detection range and improves the detection efficiency; in the screening of goods in the market price library in the second stage, the goods in the narrowed matching goods set are subjected to multi-dimensional factor matching score calculations to determine the goods that are most similar to the target detection goods. This multi-stage detection mechanism not only improves the efficiency of detection, but also significantly improves the accuracy of detection.

[0056] On the other hand, this application uses a multi-dimensional matching mechanism to comprehensively evaluate products from multiple angles to ensure the comprehensiveness and robustness of the test results. In the factor matching score calculation, the factor matching score of each product is calculated based on multi-dimensional information such as brand information, category information, and multiple key attributes. Through this multi-dimensional matching scoring mechanism, the similarity between products can be fully reflected, avoiding the deviation that may be caused by single-dimensional scoring.

[0057] Through a multi-stage and multi-dimensional detection and matching mechanism, this application significantly improves the efficiency, accuracy, comprehensiveness and robustness of fraudulent product detection, provides the platform with a more effective means of fraudulent product detection, and thus ensures the transaction security and user experience of the platform.

[0058] For further examples, see Figure 2 , based on the category information of the commodity to be detected, determining the target detection commodity and its related information, wherein the related information includes brand information, category information and multiple key attributes, comprises the following steps:

[0059] Step S5110, obtaining the product title text of the product to be detected;

[0060] The product title text contains the basic description information of the product, such as brand, model, function, etc. Specifically, the product title text of the product to be tested is directly obtained from the database that stores product-related information. The product title text is the title used when the product is displayed on the platform, and is usually edited by the seller or platform operator according to the actual attributes of the product. For example, for a smartphone, its product title text may be "Apple iPhone 14Pro Max 512GB Space Black", which contains key information such as brand (Apple), model (iPhone 14Pro Max), storage capacity (512GB) and color (Space Black).

[0061] Step S5120: input the product title text into a pre-trained category recognition model to obtain category information of the product to be detected;

[0062] After obtaining the product title text of the product to be detected, the product title text is input into the pre-trained category recognition model to automatically identify the category information of the product to be detected. The category recognition model is trained based on a large number of product title texts and their corresponding category information. The trained category recognition model can accurately identify the category to which the product belongs based on the key information in the product title text.

[0063] In one embodiment, the category recognition model analyzes key information such as brand, model, and function in the product title text to determine the category information of the product. For example, for the product title text "Apple iPhone 14 Pro Max 512GB Space Black", the category recognition model will recognize that the product belongs to the "smartphone" category because it contains key information such as brand (Apple) and model (iPhone 14Pro Max), which are highly matched with the characteristics of the smartphone category.

[0064] The training data of the category recognition model can come from the historical product data of the platform, covering the product title text and corresponding category information of various products. Through a large number of training samples, the category recognition model can learn the title text features of products in different categories, thereby achieving high-precision category recognition in practical applications. After completing category recognition, the category information of the product to be detected is obtained, such as "smartphone", "television", "laptop", etc.

[0065] Step S5130: Match the category information of the commodity to be detected with multiple category information in a preset category rule library. If the match is successful, the commodity to be detected is determined as a target detection commodity.

[0066] Match the category information of the product to be detected determined in the previous step with multiple category information in the preset category rule library. The category rule library is pre-defined and contains category information of multiple standard products. Specifically, match the category information of the product to be detected with the multiple category information in the category rule library one by one. If the category information of the product to be detected is "smartphone", match "smartphone" with all category information in the category rule library. If the category information of the product to be detected successfully matches a certain category information in the category rule library, the product to be detected is determined as the target detection product.

[0067] In this embodiment, the category recognition model can accurately identify the category to which the product belongs based on the key information in the product title text (such as brand, model, function, etc.). This automated category recognition mechanism not only improves the efficiency of recognition, but also reduces the errors caused by manual intervention, ensuring the accuracy of the category information. After determining the category information of the product to be detected, the category information is matched with multiple category information in the preset category rule library. By matching one by one, it is possible to quickly determine whether the product to be detected belongs to a standard product and determine it as a target detection product.

[0068] For further examples, see Figure 3 , the target detected product is matched with the products in the preset market price library by similarity, and some products in the market price library that match the target detected product are roughly recalled to form a matching product set, including the following steps:

[0069] Step S5210: In the first stage, the relevant information of the commodities in the preset market price library is preliminarily matched with the relevant information of the target detection commodity, and based on a preset first similarity score threshold, commodities that are preliminarily matched with the target detection commodity are screened out from the market price library to form a preliminarily matched commodity set;

[0070] After determining the target detection product and its related information, the relevant information of the target detection product is matched with the products in the preset market price library by cosine similarity. Specifically, the brand information, category information and multiple key attributes of the target detection product are matched with the brand information, category information and multiple key attributes of the products in the market price library by cosine similarity. Cosine similarity is a commonly used similarity calculation method that can measure the similarity between two vectors. The cosine similarity between the target detection product and each product in the market price library is calculated, and based on the preset first similarity score threshold, the products that are preliminarily matched with the target detection product are screened out to form a preliminary matching product set.

[0071] For example, if the target detection product is an "Apple iPhone 14Pro Max 512GB Space Black" smartphone, its brand information (Apple), category information (smartphone) and key attributes (model, storage capacity, color, etc.) will be calculated by cosine similarity with the relevant information of all standard products in the market price library, and products with a similarity higher than the preset similarity score threshold (for example, 0.8) will be screened out to form a preliminary matching product set.

[0072] In this way, products similar to the target detection product can be quickly screened out in the first stage.

[0073] Step S5220, in the second stage, calculate the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set, and based on a preset recall rate threshold, further screen out products that are secondarily matched with the target detected product from the preliminary matching product set to form a final product matching set.

[0074] In the second stage, the longest common subsequence recall rate of the relevant information of the target detection product and each product in the preliminary matching product set is further calculated. The longest common subsequence recall rate is an indicator to measure the similarity of two text sequences. It can capture the degree of matching of key information in the text and is particularly suitable for processing complex product description information. Specifically, the longest common subsequence (LCS) of the product title text, description text and multiple key attributes of the target detection product and the corresponding information of each product in the preliminary matching product set is calculated. LCS refers to the longest identical subsequence in two text sequences, and its length reflects the similarity between the two texts. The LCS length of the target detection product and each preliminary matching product is calculated, and the recall rate is calculated based on the length. The recall rate refers to the ratio of the LCS length to the length of the relevant information of the target detection product, which is used to measure the degree of matching between the matching product and the target detection product in key information. For example, if the product title of the target detection product is "Apple iPhone 14Pro Max 512GB Space Black", and there is a product in the preliminary matching product set with the title "Apple iPhone 14Pro 256GB Silver", the longest common subsequence of the two titles (such as "Apple iPhone 14Pro") is calculated, and the recall rate is calculated based on the length of the subsequence. Assuming that the title length of the target detection product is 25 characters and the LCS length is 15 characters, the recall rate is 60%.

[0075] In one embodiment, the calculated recall rate is compared with a preset recall rate threshold. If the recall rate of a preliminary matching product is higher than the preset recall rate threshold (e.g., 80%), it is further determined whether the category information of the product is the same as the category information of the target detection product. If the category information is the same, the product is used to form the final product matching set. If the recall rate is lower than the preset recall rate threshold, it is further determined whether the cosine similarity of the product is higher than a preset second similarity threshold (e.g., 0.7). If the cosine similarity is higher than the second similarity threshold, the product is also used to form the final product matching set. Please refer to the subsequent specific implementation method for the specific steps of this embodiment, which will not be repeated here.

[0076] In this way, products that are highly matched with the target detected products can be further screened out in the second stage to form the final product matching set.

[0077] In this embodiment, through the two-stage similarity matching mechanism, efficient screening and accurate positioning of matching products are achieved, which significantly improves the efficiency and accuracy of fraudulent product detection. The similarity matching mechanism of this embodiment combines cosine similarity matching and accurate LCS recall calculation, which can quickly screen out preliminary matching products in the first stage, and further accurately screen out highly matching products in the second stage, thereby achieving efficient screening and accurate positioning of matching products.

[0078] For further examples, see Figure 4 In the second stage, the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set is calculated, and based on a preset recall rate threshold, products that are secondarily matched with the target detected product are further screened out from the preliminary matching product set to form a final product matching set, including the following steps:

[0079] Step S5221, calculating and obtaining the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set;

[0080] In the similarity matching process of the second stage, the longest common subsequence (LCS) recall rate of the relevant information of the target detected product and each product in the preliminary matching product set is calculated. Specifically, the LCS is calculated for the product title text, description text, and multiple key attributes of the target detected product and the corresponding information of each product in the preliminary matching product set. In this way, the degree of matching between the target detected product and the preliminary matching product in terms of key information can be quantified.

[0081] Step S5222: compare the longest common subsequence recall rate with a preset recall rate threshold. If the longest common subsequence recall rate is higher than the preset recall rate threshold, further determine whether the category information of the corresponding product is the same as the category information of the target detection product. If they are the same, the product is used to form a final product matching set.

[0082] After calculating the longest common subsequence recall rate of the target detection product and each product in the preliminary matching product set, these recall rates are compared with the preset recall rate threshold. The recall rate threshold is pre-set and is used to determine whether the degree of matching between the matching product and the target detection product in key information meets the expected standard. If the recall rate of a preliminary matching product is higher than the preset recall rate threshold (for example, 80%), it is further determined whether the category information of the product is the same as the category information of the target detection product. Matching of category information is one of the important conditions to ensure the similarity of products. Even if the key attributes of products in different categories are similar, there may be essential differences.

[0083] For example, if the target detected product is an "Apple iPhone 14Pro Max 512GB Space Black" smartphone, and there is a product in the preliminary matching product set with a recall rate of 85% (higher than the recall rate threshold of 80%), further check whether the category information of the product is "smartphone". If the category information is the same, it means that the product is highly matched with the target detected product, and the product is used to form the final product matching set.

[0084] In this way, when the recall rate is higher than the threshold, the category information of the matching products can be further verified to ensure that the products in the final product matching set are highly consistent with the target detection products in multiple dimensions, thereby providing more reliable data support for subsequent price comparison and fraudulent product detection.

[0085] Step S5223: If the longest common subsequence recall rate is lower than the preset recall rate threshold, further determine whether the cosine similarity of the corresponding product is higher than the preset second similarity threshold; if it is higher than the second similarity threshold, use the product to form a final product matching set.

[0086] If the longest common subsequence recall rate of the goods in the preliminary matching goods set is lower than the preset recall rate threshold (e.g., 80%), it will not be directly excluded, but further judged whether the cosine similarity of the goods is higher than the preset second similarity threshold. The second similarity threshold is preset and is used to measure the similarity of goods. Specifically, by comparing the cosine similarity of the goods with the target detection goods. If the cosine similarity of the goods is higher than the preset second similarity threshold (e.g., 0.7), it means that although the similarity of the goods with the target detection goods is low when calculated by the longest common subsequence recall rate, the similarity is high when calculated by the cosine similarity, indicating that the goods still have a high degree of matching with the target detection goods in some key attributes. Therefore, although its recall rate is lower than the threshold, it can still be included in the final set of matching goods. In this way, when the recall rate is low, the heterogeneous similarity can be further used as a supplementary judgment criterion to ensure that those goods that are highly matched in key attributes but may have slight differences in description details can also be included in the final set of matching goods. This helps to improve the comprehensiveness and accuracy of matches, especially when dealing with complex or diverse product descriptions.

[0087] In this embodiment, through a multi-level similarity matching and verification mechanism, accurate screening and comprehensive coverage of matching products are achieved, significantly improving the accuracy and robustness of fraudulent product detection. First, when the recall rate is higher than the threshold, the category information of the matching products can be further verified to ensure that the products in the final product matching set are highly consistent with the target detection products in multiple dimensions, thereby providing more reliable data support for subsequent price comparisons and fraudulent product detection. Secondly, when the recall rate is low, cosine similarity can be further used as a supplementary judgment criterion to ensure the comprehensiveness and accuracy of the match. This multi-level similarity matching and verification mechanism not only improves the accuracy of detection, but also enhances the robustness and comprehensiveness of detection.

[0088] For further examples, see Figure 5 , based on the brand information, category information and multiple key attributes of the corresponding products in the matching product set, calculating the factor matching scores of the products in the matching product set, and determining the product with the highest factor matching score, including the following steps:

[0089] Step S5310: Obtain brand information, category information, and multiple key attributes of the corresponding product in the matching product set and the target detection product, wherein the multiple key attributes include product title text and product description text;

[0090] Obtain relevant information about each product in the matching product set and the target detection product in order to perform multi-dimensional matching scoring. Specifically, extract the brand information, category information, and multiple key attributes of each product from the matching product set. These key attributes include product title text and product description text. The product title text contains basic description information of the product, such as brand, model, function, etc. For example, for a smartphone, its product title text may be "Apple iPhone 14Pro Max 512GB Deep Space Black", which contains key information such as brand (Apple), model (iPhone14Pro Max), storage capacity (512GB), and color (Deep Space Black). The product description text is a detailed description of the product, which may include more attribute information, functional features, usage scenarios, etc. For example, for the above smartphone, the product description text may further describe its screen size, camera specifications, operating system version, etc. By obtaining these detailed information, we can fully understand the degree of match between the matching product and the target detection product in terms of brand, category, and key attributes, thereby providing basic data support for the subsequent factor matching score calculation.

[0091] Step S5320: preset a corresponding matching score based on the brand information, category information, product title text, and product description text; if the corresponding product in the matching product set matches the brand information, category information product title text, or product description text of the target detected product, obtain the corresponding matching score;

[0092] The corresponding matching scores are preset based on the brand information, category information, product title text and product description text. The matching score is set based on the importance of brand information, category information, product title text and product description text to the degree of product matching. For example, if the brand information of the product that matches the target detected product is completely consistent with the brand information of the target detected product, a higher brand matching score (for example, 3 points) is obtained, because brand information is one of the core identifiers of the product and has an important impact on the matching degree of the product. If the category information is consistent, a category matching score (for example, 2 points) is obtained, because the category information determines the basic classification of the product, and products of different categories may have essential differences even if the brand is the same. For the product title text and product description text, the matching score can be based on the key attributes (such as model, color, storage capacity, etc.), and if the match is successful, the corresponding text matching score (for example, 1 point) is obtained. For example, if the product title of the target detection product is "Apple iPhone 14Pro Max 512GB Space Black", and the product title of the matching product is "Apple iPhone 14Pro 256GB Silver", it is recognized that the brand information of the two is the same, but the model, storage capacity and color are different, so the brand information match is 3 points, and the other attribute matches are 0 points. In this way, the degree of match between the matching product and the target detection product in multiple dimensions such as brand, category, product title and product description can be quantified, providing a basis for the subsequent factor matching score calculation.

[0093] Step S5330: Add the matching scores of the brand information, category information, product title text, and product description text of each product to obtain the factor matching score of the corresponding product, so as to determine the corresponding product with the highest factor matching score.

[0094] After completing the calculation of the matching scores of each product in the matching product set and the target detection product in terms of brand information, category information, product title text and product description text, these matching scores of each product are added together to obtain the factor matching score of the product. The factor matching score reflects the comprehensive matching degree between the product and the target detection product in multiple dimensions. The higher the score, the higher the matching degree. For example, if the brand information of a matching product matches 3 points, the category information matches 2 points, the product title text matches 1 point, and the product description text matches 1 point, then the factor matching score of the product is 3+2+1+1=7 points. Calculate the factor matching score of each product in the matching product set one by one, and compare the scores of all products to determine the product with the highest score. The product with the highest score is considered to be the product that is most similar to the target detection product. In this way, the product that best matches the target detection product can be screened out from the matching product set, ensuring that subsequent analysis and judgment are based on the most relevant market data.

[0095] In this embodiment, through a multi-dimensional matching scoring mechanism, accurate screening of matching products is achieved, which significantly improves the accuracy and comprehensiveness of fraudulent product detection. This embodiment constructs a comprehensive information framework by acquiring key attributes such as brand information, category information, product title text, and product description text of each product in the matching product set and the target detection product, ensuring the multi-dimensionality and comprehensiveness of the matching score. In addition, corresponding matching scores are preset for brand information, category information, product title text, and product description text, and scores are dynamically assigned according to the matching situation. This dynamic scoring mechanism can effectively quantify the degree of matching between matching products and target detection products in multiple dimensions, avoiding the deviation that may be caused by single-dimensional scoring. This multi-dimensional and dynamic matching scoring mechanism not only improves the accuracy of detection, but also enhances the comprehensiveness of detection, providing the platform with a more effective means of detecting fraudulent products.

[0096] For further examples, see Figure 6 , determining a price threshold based on multiple commodity prices corresponding to the commodity with the highest score, comprising the following steps:

[0097] Step S5410: obtaining multiple commodity prices corresponding to the commodity with the highest score and the historical commodity prices of the commodity within a preset time, and calculating the average discount value of the commodity;

[0098] After determining the product with the highest score, it is necessary to further obtain multiple product prices of the product and its historical product prices within a preset time in order to calculate the average discount value of the product. Specifically, multiple product prices of the highest-scoring product at different time points or in different channels are extracted from the market price library. These product prices may include historical transaction prices, market research prices, industry standard prices, etc. At the same time, the historical product prices of the product within a preset time are also obtained. The preset time can be a period of time in the past (for example, the past 30 days, 90 days, or a year). The specific time range is set according to actual needs. By obtaining these price data, the price fluctuations of the product can be fully understood.

[0099] The average discount margin value of the product is calculated based on the acquired price data. The average discount margin value reflects the average discount level of the product within a preset time, and is achieved by calculating the difference between the historical price and the current price of the product. For example, assuming that the historical prices of the highest-scoring product in the past 30 days were 5,000 yuan, 5,200 yuan, and 5,100 yuan, respectively, and the current price is 4,800 yuan, calculate the discount margin between these historical prices and the current price (for example, 5,000 yuan - 4,800 yuan = 200 yuan, a discount margin of 4%), and then take the average of all discount margins to obtain the average discount margin value of the product. In this way, the price fluctuation of the product can be quantified, providing an important basis for the subsequent price threshold calculation.

[0100] Step S5420: determine the detection ratio of the corresponding commodity based on the average discount value of the commodity and a preset mapping relationship, and multiply the detection ratio by the lowest price among the prices corresponding to the commodity with the highest score to obtain a price threshold.

[0101] Based on the average discount value calculated in the previous step and the preset mapping relationship, determine the corresponding detection ratio. The mapping relationship is predefined, usually expressed in the form of a table or function, and is used to map the average discount value to the detection ratio. For example, assume that the corresponding relationship between the average discount value and the detection ratio is defined in the mapping relationship table: when the average discount value is 5%, the detection ratio is 0.9; when the average discount value is 10%, the detection ratio is 0.8; when the average discount value is 15%, the detection ratio is 0.7. According to the calculated average discount value, find the corresponding detection ratio in the mapping relationship table. For example, if the calculated average discount value is 10%, the corresponding detection ratio is 0.8. It should be noted that the average discount value can be expressed in the form of an interval. For example, when the average discount value is 5% to 10%, the detection ratio is 0.9.

[0102] The detection ratio is multiplied by the lowest price among the prices corresponding to the highest-scoring goods to obtain the price threshold. The lowest price refers to the lowest price recorded in the market price database for the highest-scoring goods, which reflects the lower limit of the reasonable price of the goods in the market. For example, assuming that the lowest price of the highest-scoring goods in the market price database is 5,000 yuan, and the detection ratio is 0.8, then the price threshold is 5,000 yuan × 0.8 = 4,000 yuan. In this way, a reasonable price threshold can be calculated based on the data in the market price database, combined with the average discount value and the mapping relationship.

[0103] In this embodiment, a reasonable price threshold can be calculated based on the data in the market price library, combined with historical price fluctuations and dynamic mapping relationships, to provide a more reliable basis for fraudulent product detection. This multi-level, dynamic price threshold calculation method not only improves the accuracy of detection, but also enhances the adaptability and robustness of detection, providing the platform with a more effective means of fraudulent product detection, thereby ensuring the transaction security and user experience of the platform.

[0104] For further examples, see Figure 7 , after determining a price threshold based on multiple commodity prices corresponding to the commodity with the highest score, and comparing the commodity price of the target detected commodity with the price threshold to determine whether the target detected commodity is a fraudulent commodity, comprising:

[0105] Step S6100: When the price of the target detected commodity is lower than the price threshold, a deviation ratio between the price of the target detected commodity and the price threshold is calculated;

[0106] If the price of the target detected commodity is lower than the price threshold, the deviation ratio between the commodity price and the price threshold is further calculated. The deviation ratio is used to measure the degree of deviation between the price of the target detected commodity and the reasonable market price, which is achieved by calculating the ratio of the commodity price of the target detected commodity to the price threshold. Specifically, the calculation formula of the deviation ratio is: deviation ratio = 1-(commodity price of the target detected commodity / price threshold) × 100%. For example, if the commodity price of the target detected commodity is 3,000 yuan and the price threshold is 4,000 yuan, the deviation ratio is 1-(3000 / 4000) × 100% = 25%. In this way, the price deviation degree of the target detected commodity can be quantified, providing data support for subsequent fraudulent commodity judgments.

[0107] Step S6200: If the deviation ratio is less than a preset ratio threshold, fraudulent product detection is performed on other products in the online store to which the target detected product belongs. If the prices of the other products are all higher than the corresponding price threshold, the target detected product is marked as an abnormal product to form an abnormal product set for further manual detection.

[0108] The deviation ratio calculated in the previous step is compared with the preset ratio threshold. The ratio threshold is pre-set and is used to determine whether the price deviation of the target detection product reaches the threshold that requires further processing. For example, assuming that the ratio threshold is 30%, if the deviation ratio of the target detection product is 25% (less than the ratio threshold), the other products of the online store to which the target detection product belongs are further detected for fraudulent products. Specifically, the relevant information of all other products in the store is obtained, and the price threshold of these products is calculated according to the same process as the target detection product, and the price of each product is compared with the corresponding price threshold. If the prices of all other products in the store are higher than the corresponding price threshold, it means that the prices of other products in the store are within a reasonable range, and the price of the target detection product is significantly lower than the reasonable price, and there may be abnormal detection. In this case, the target detection product is marked as an abnormal product and added to the abnormal product set for further manual detection and analysis. In this way, when abnormally low-priced products are found, the overall price of the store to which it belongs can be further verified to ensure that the detection results of abnormal products are more accurate and reliable, and to avoid misjudgment of the target detection product.

[0109] In this embodiment, the accuracy and reliability of fraudulent product detection are significantly improved through a multi-level detection and verification mechanism. The implementation of this embodiment avoids misjudgments caused by price fluctuations or special market conditions. This multi-level verification mechanism not only improves the accuracy of detection, but also reduces the possibility of misjudgment, ensuring that the detection results of abnormal products are more reliable. In addition, this embodiment achieves a comprehensive screening of abnormal products by combining automatic detection and manual review, which not only improves detection efficiency, but also ensures the accuracy of detection results, providing the platform with a more effective means of fraudulent product detection, thereby improving the transaction security and user experience of the platform.

[0110] See also Figure 8, a fraudulent product detection device provided to meet one of the purposes of the present application is a functional embodiment of the fraudulent product detection method of the present application. On the other hand, the device is a fraudulent product detection device provided to meet one of the purposes of the present application, including a target detection product determination module 5100, a similarity matching module 5200, a matching score calculation module 5300 and a fraudulent product judgment module 5400, wherein the target detection product determination module 5100 is used to determine the target detection product and related information based on the category information of the product to be detected, and the related information includes brand information, category information and multiple key attributes; the similarity matching module 5200 is used to convert the target detection product and the similarity matching module 5300 into a target detection product. The detected product is matched with the products in a preset market price library for similarity, and some products in the market price library that match the target detected product are roughly recalled to form a matching product set; a matching score calculation module 5300 is used to calculate the factor matching score of the products in the matching product set based on the brand information, category information and multiple key attributes of the corresponding products in the matching product set, and determine the product with the highest factor matching score; a fraudulent product judgment module 5400 is used to determine a price threshold based on multiple product prices corresponding to the product with the highest score, and compare the product price of the target detected product with the price threshold to determine whether the target detected product is a fraudulent product.

[0111] In a further embodiment, the target detection product determination module 5100 includes: a title text acquisition submodule, used to obtain the product title text of the product to be detected; a category information acquisition submodule, used to input the product title text into a pre-trained category recognition model to obtain the category information of the product to be detected; a category information matching submodule, used to match the category information of the product to be detected with multiple category information in a preset category rule library. If the match is successful, the product to be detected is determined as the target detection product.

[0112] In a further embodiment, the similarity matching module 5200 includes: a cosine similarity matching submodule, which is used to perform cosine similarity matching on the relevant information of the goods in the preset market price library and the relevant information of the target detection goods in the first stage, and screen out goods that are preliminarily matched with the target detection goods from the market price library based on a preset first similarity score threshold to form a preliminary matching goods set; a common subsequence matching submodule, which is used to calculate the longest common subsequence recall rate of the relevant information of the target detection goods and the relevant information of each product in the preliminary matching goods set in the second stage, and further screen out goods that are secondarily matched with the target detection goods from the preliminary matching goods set based on a preset recall rate threshold to form a final goods matching set.

[0113] In a further embodiment, the common subsequence matching submodule includes: a recall rate calculation submodule, which is used to calculate the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set; a recall rate comparison submodule, which is used to compare the longest common subsequence recall rate with a preset recall rate threshold, if the longest common subsequence recall rate is higher than the preset recall rate threshold, further determine whether the category information of the corresponding product is the same as the category information of the target detected product, if the same, the product is used to form a final product matching set; a product matching set formation submodule, which is used to further determine whether the cosine similarity of the corresponding product is higher than a preset second similarity threshold if the longest common subsequence recall rate is lower than the preset recall rate threshold, if it is higher than the second similarity threshold, the product is used to form a final product matching set.

[0114] In a further embodiment, the matching score calculation module 5300 includes: a data acquisition submodule, used to obtain the brand information, category information and multiple key attributes of the corresponding products in the matching product set and the target detection product, the multiple key attributes including product title text and product description text; a matching score calculation submodule, used to preset the corresponding matching score based on the brand information, category information, product title text and product description text, if the corresponding product in the matching product set matches the brand information, category information product title text or product description text of the target detection product, then the corresponding matching score is obtained; a matching score addition submodule, used to add the matching scores of the brand information, category information product title text and product description text of each product to obtain the factor matching score of the corresponding product, so as to determine the corresponding product with the highest factor matching score.

[0115] In a further embodiment, the fraudulent product judgment module 5400 includes: a discount value calculation sub-module, used to obtain multiple product prices corresponding to the product with the highest score and the historical product prices of the product within a preset time, and calculate the average discount value of the product; a price threshold calculation sub-module, used to determine the detection ratio of the corresponding product based on the average discount value of the product and a preset mapping relationship, and multiply the detection ratio by the lowest price among the prices corresponding to the product with the highest score to obtain a price threshold.

[0116] In a further embodiment, the fraudulent product judgment module 5400 includes: a deviation ratio calculation submodule, which is used to calculate the deviation ratio between the product price of the target detected product and the price threshold when the product price of the target detected product is lower than the price threshold; an abnormal product set formation submodule, which is used to perform fraudulent product detection on other products in the online store to which the target detected product belongs if the deviation ratio is less than a preset ratio threshold; if the product prices of the other products are all higher than the corresponding price threshold, the target detected product is marked as an abnormal product to form an abnormal product set for further manual detection.

[0117] In order to solve the above technical problems, the present application also provides a computer device. Fig. 9 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor may implement a fraudulent product detection method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor may execute the fraudulent product detection method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0118] In this embodiment, the processor is used to execute Figure 8 The memory stores the program code and various data required to execute the above modules or submodules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules / submodules in the fraudulent commodity detection device of this application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0119] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the fraudulent product detection method of any embodiment of the present application.

[0120] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0121] It will be understood by those skilled in the art that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, changed, combined, or deleted. Furthermore, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted. Furthermore, the steps, measures, and schemes in the various operations, methods, and processes in the prior art that are open source and in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0122] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A fraudulent commodity detection method, characterized in that: The steps include: Based on the category information of the product to be detected, determine the target product and its related information, wherein the related information includes brand information, category information, and multiple key attributes; Perform similarity matching between the target detected product and the products in the preset market price library, and recall some products in the market price library that match the target detected product to form a matching product set; Calculating factor matching scores of products in the matching product set based on brand information, category information, and multiple key attributes of corresponding products in the matching product set, and determining a product with the highest factor matching score; Based on the multiple commodity prices corresponding to the commodity with the highest score, a price threshold is determined, and the commodity price of the target detected commodity is compared with the price threshold to determine whether the target detected commodity is a fraudulent commodity.

2. The fraudulent commodity detection method according to claim 1, characterized in that: Based on the category information of the product to be detected, determine the target product to be detected and its related information, which includes brand information, category information and multiple key attributes, including: Get the product title text of the product to be tested; Input the product title text into a pre-trained category recognition model to obtain the category information of the product to be detected; The category information of the commodity to be detected is matched with multiple category information in a preset category rule library. If the match is successful, the commodity to be detected is determined as a target detection commodity.

3. The fraudulent commodity detection method according to claim 1, characterized in that: The target detected product is matched with the products in the preset market price library by similarity, and some products in the market price library that match the target detected product are roughly recalled to form a matching product set, including: In the first stage, the relevant information of the commodities in the preset market price library is preliminarily matched with the relevant information of the target detection commodity, and based on the preset first similarity score threshold, the commodities that preliminarily match the target detection commodity are screened out from the market price library to form a preliminarily matched commodity set; In the second stage, the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set is calculated, and based on a preset recall rate threshold, products that are secondarily matched with the target detected product are further screened out from the preliminary matching product set to form a final product matching set.

4. The fraudulent commodity detection method according to claim 3, characterized in that: In the second stage, the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set is calculated, and based on a preset recall rate threshold, products that are secondarily matched with the target detected product are further screened out from the preliminary matching product set to form a final product matching set, including: Calculating the longest common subsequence recall rate of the relevant information of the target detected product and the relevant information of each product in the preliminary matching product set; Comparing the longest common subsequence recall rate with a preset recall rate threshold, if the longest common subsequence recall rate is higher than the preset recall rate threshold, further determining whether the category information of the corresponding product is the same as the category information of the target detected product, if they are the same, using the product to form a final product matching set; If the longest common subsequence recall rate is lower than the preset recall rate threshold, it is further determined whether the cosine similarity of the corresponding product is higher than a preset second similarity threshold. If it is higher than the second similarity threshold, the product is used to form a final product matching set.

5. The fraudulent commodity detection method according to claim 1, characterized in that: Calculating factor matching scores of the products in the matching product set based on brand information, category information, and multiple key attributes of corresponding products in the matching product set, and determining a product with the highest factor matching score, including: Obtain brand information, category information, and multiple key attributes of the corresponding product in the matching product set and the target detection product, wherein the multiple key attributes include product title text and product description text; Presetting a corresponding matching score based on brand information, category information, product title text, and product description text, and obtaining a corresponding matching score if the corresponding product in the matching product set matches the brand information, category information, product title text, or product description text of the target detected product; The matching scores of the brand information, category information, product title text, and product description text of each product are added together to obtain the factor matching score of the corresponding product, so as to determine the corresponding product with the highest factor matching score.

6. The fraudulent commodity detection method according to any one of claims 1 to 5, characterized in that: Determining a price threshold based on multiple commodity prices corresponding to the commodity with the highest score includes: Obtain multiple commodity prices corresponding to the commodity with the highest score and the historical commodity prices of the commodity within a preset period of time, and calculate the average discount value of the commodity; The detection ratio of the corresponding commodity is determined based on the average discount value of the commodity and a preset mapping relationship, and the detection ratio is multiplied by the lowest price among the prices corresponding to the commodity with the highest score to obtain a price threshold.

7. The fraudulent commodity detection method according to any one of claims 1 to 5, characterized in that: After determining a price threshold based on multiple commodity prices corresponding to the commodity with the highest score, and comparing the commodity price of the target detected commodity with the price threshold to determine whether the target detected commodity is a fraudulent commodity, the method further includes: When the price of the target detected commodity is lower than the price threshold, calculating the deviation ratio between the price of the target detected commodity and the price threshold; If the deviation ratio is less than the preset ratio threshold, fraudulent product detection will be performed on other products in the online store to which the target detected product belongs. If the prices of the other products are all higher than the corresponding price threshold, the target detected product will be marked as an abnormal product to form an abnormal product set for further manual detection.

8. A fraudulent commodity detection device, characterized in that: include: A target detection commodity determination module is used to determine the target detection commodity and related information based on the category information of the commodity to be detected, wherein the related information includes brand information, category information and multiple key attributes; A similarity matching module is used to perform similarity matching between the target detected product and the products in the preset market price library, that is, to recall some products in the market price library that match the target detected product to form a matching product set; A matching score calculation module, used to calculate the factor matching scores of the products in the matching product set based on the brand information, category information and multiple key attributes of the corresponding products in the matching product set, and determine the product with the highest factor matching score; The fraudulent product judgment module is used to determine a price threshold based on multiple product prices corresponding to the product with the highest score, and compare the product price of the target detected product with the price threshold to determine whether the target detected product is a fraudulent product.

9. A computer device comprising a central processing unit and a memory, characterized in that: The central processing unit is used to call and run the computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: It stores a computer program implemented according to the method described in any one of claims 1 to 7 in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

Images