Method, device, computer equipment and medium for identifying identical products

Abstract

The present disclosure provides a method, apparatus, computer device, and medium for identifying identical products. The identical product identification method includes obtaining a pair of similar products, wherein the pair includes two products to be identified and the two products have matching appearances; and determining whether the two products to be identified are identical based on matching attributes and sales specifications of the two products to be identified. Identical products identified by the present disclosure can be priced similarly.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technology, and more specifically, to a method, apparatus, computer equipment, and medium for identifying identical products. Background Art

[0002] With the rapid development of e-commerce, more and more e-commerce platforms have emerged. Existing e-commerce platforms include Amazon, Alibaba.com, AliExpress (AE), and International Core Business Unit (ICBU).

[0003] Against this backdrop, connecting product data across multiple e-commerce platforms and forming a connected product database is crucial for e-commerce companies with multiple platforms to gain insight into market conditions, and identifying identical products is a key step. Furthermore, merchants have a strong need to measure their competitiveness. If e-commerce platforms provide functionality to monitor the shipment and pricing of identical products both on and off the platform, merchants will gain insights into the platform's and their own competitiveness, enabling them to optimize their operational strategies. Identifying identical products at comparable prices is equally crucial.

[0004] However, in existing e-commerce scenarios, identifying identical products often involves recalling products with identical appearances. Products with identical appearances are not necessarily priced comparable. Therefore, building a price monitoring system requires more rigorous and thorough technology for identifying identical products. Summary of the Invention

[0005] In view of this, the present disclosure aims to improve the method for identifying identical products so that recalled identical products can be compared in price.

[0006] According to a first aspect of the present disclosure, a method for identifying identical products is provided, comprising:

[0007] Obtaining a similar product pair, wherein the similar product pair includes two products to be identified and the two products to be identified have matching appearances;

[0008] According to the matching of the attributes and the matching of the selling specifications of the two commodities to be identified, it is determined whether the two commodities to be identified are the same commodity.

[0009] Optionally, the attribute matching of the two commodities to be identified is determined by the following steps:

[0010] Performing standardization conversion on the attribute information of the two commodities to be identified respectively to obtain two standardized attribute information sets, wherein the standardized attribute information sets correspond to the commodities to be identified in a one-to-one manner;

[0011] Performing an intersection operation on the two standardized attribute information sets to obtain an attribute information intersection;

[0012] Determining whether the attributes in the intersection of the attribute information include a target attribute, where the target attribute is an attribute that determines the price of the commodity to be identified;

[0013] Wherein, when the attributes in the intersection of the attribute information include the target attribute, the attributes of the two commodities to be identified match.

[0014] Optionally, the matching of the selling specifications of the two commodities to be identified is determined by the following steps:

[0015] Standardizing the sales specification information of the two commodities to be identified respectively to obtain two standardized sales specification information, wherein the standardized sales specification information corresponds to the commodities to be identified one by one;

[0016] Comparing the two standardized sales specification information;

[0017] The standardized sales specification information is the sales quantity under the standard unit. When the sales quantities in the two pieces of standardized sales specification information are consistent, the sales specifications of the two commodities to be identified match.

[0018] Optionally, obtain similar product pairs, including:

[0019] Scoring the similarity between the target product and the product to be matched based on the product feature information to obtain a similarity score, wherein the product feature information includes image feature information of the product;

[0020] Determine whether the target product and the product to be matched form a similar product pair based on the similarity score.

[0021] Optionally, the similarity between the target product and the product to be matched is scored based on the product feature information to obtain a similarity score, including:

[0022] After extracting the first feature vector and the second feature vector, comparing the similarity between the target product and the product to be matched based on the first feature vector and the second feature vector;

[0023] and, determining the similarity score based on the obtained comparison result;

[0024] The first feature vector is a feature vector representing the commodity feature information of the target commodity, and the second feature vector is a feature vector representing the commodity feature information of the commodity to be identified;

[0025] The first feature vector and the second feature vector are each obtained by concatenating multiple feature vectors extracted by multiple different neural networks, or are each obtained by concatenating multiple feature vectors extracted by multiple different methods.

[0026] Optionally, there are multiple products to be matched, and the similarity between the target product and the products to be matched is scored based on the product feature information to obtain a similarity score, including:

[0027] After extracting the plurality of second feature vectors, performing feature clustering on the plurality of second feature vectors to form a plurality of clusters;

[0028] Searching for a class with the smallest distance between the class center and the first eigenvector from the multiple classes to obtain a target class, wherein the second eigenvector in the target class is a target eigenvector;

[0029] comparing the similarity between the target product and the to-be-matched product corresponding to the target feature vector based on the first feature vector and the target feature vector;

[0030] and, determining a similarity score between the target product and the product to be matched corresponding to the target feature vector based on the obtained comparison result;

[0031] The first feature vector is a feature vector representing the commodity feature information of the target commodity, and the second feature vector is a feature vector representing the commodity feature information of the commodity to be identified.

[0032] Optionally, the product feature information further includes at least one of the following: text feature information of the product, and user behavior feature information on the product.

[0033] Optionally, when it is determined according to the similarity score that the target product and the product to be matched form a similar product pair, the target product and the product to be matched are two products to be identified, and the identification method further includes:

[0034] Determining whether the similarity score meets a first threshold condition to obtain a first determination result;

[0035] Among them, when the first judgment result is that the similarity score does not meet the first threshold condition, it is not necessary to determine whether the two commodities to be identified are the same commodity based on the attribute matching and sales specification matching of the two commodities to be identified.

[0036] Optionally, when the first judgment result is that the similarity score meets the first threshold condition, the identification method further includes:

[0037] Determining whether the target product and the first-level category of the to-be-matched product match to obtain a second determination result;

[0038] Among them, when the second judgment result is that the primary category of the target product and the product to be matched does not match, there is no need to determine whether the two products to be identified are the same product based on the attribute matching and sales specification matching of the two products to be identified.

[0039] Optionally, when the second judgment result is that the target product and the to-be-matched product have a matching primary category, the identification method further includes:

[0040] Determining whether the similarity score meets a second threshold condition;

[0041] and, if the similarity score satisfies the second threshold condition, determining whether the two commodities to be identified are the same commodity based on a matching condition of attributes and a matching condition of sales specifications of the two commodities to be identified;

[0042] The first threshold condition is a general condition for commodities, the second threshold condition is a condition specific to the target commodity, and the second threshold condition is stricter than the first threshold condition.

[0043] Optionally, determining whether the two commodities to be identified are the same commodity according to a matching condition of attributes and a matching condition of selling specifications of the two commodities to be identified includes:

[0044] After determining that the attributes and sales specifications of the two commodities to be identified match, determine whether the leaf categories of the two commodities to be identified match, and, if the leaf categories of the two commodities to be identified match, determine that the two commodities to be identified are the same commodity.

[0045] According to a second aspect of the present disclosure, there is provided a platform server, comprising:

[0046] a memory for storing computer executable code;

[0047] A processor is configured to execute the computer executable code to implement any one of the methods described in the first aspect.

[0048] According to a third aspect of the present disclosure, a computer-readable medium is provided, comprising a computer-executable code, wherein when the computer-executable code is executed by a processor, any one of the methods described in the first aspect is implemented.

[0049] In the disclosed embodiment, whether two commodities to be identified are the same commodity is determined based on the matching of their attributes and selling specifications. Therefore, the final determination of the same commodity is the result obtained after comprehensive consideration of the commodity appearance, attributes and selling specifications. This makes the matching of commodities of the same type more rigorous, so that the determined same commodities have the same value and comparable prices. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The above and other objects, features and advantages of the present disclosure will become more apparent through description of the embodiments of the present disclosure with reference to the following drawings, in which:

[0051] Figures 1A-1B An interface state diagram is shown when the method for identifying identical products according to an embodiment of the present disclosure is applied;

[0052] Figure 2 A flowchart of identifying identical products according to an embodiment of the present disclosure is shown;

[0053] Figure 3 A flowchart of obtaining similar product pairs according to one embodiment of the present disclosure is shown;

[0054] Figure 4 A schematic diagram illustrating an execution method for identifying identical products according to an embodiment of the present disclosure is shown;

[0055] Figure 5 Shows examples of product categories on different e-commerce platforms;

[0056] Figure 6 A schematic diagram of attribute matching execution according to an embodiment of the present disclosure is shown;

[0057] Figure 7 A schematic diagram of executing sales specification matching according to an embodiment of the present disclosure is shown;

[0058] Figure 8 A schematic diagram of the structure of a device for identifying identical products according to an embodiment of the present disclosure is shown;

[0059] Figure 9 A structural diagram of a computer device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0060] The present disclosure is described below based on examples, but the present disclosure is not limited to these examples. Certain specific details are described in detail in the detailed description of the present disclosure below. Those skilled in the art will appreciate that the present disclosure is fully understood without these details. To avoid obscuring the essence of the present disclosure, well-known methods, processes, and procedures have not been described in detail. The accompanying drawings are not necessarily drawn to scale.

[0061] In the e-commerce field, the identification of similar products with comparable prices can be used in a variety of scenarios. For example, it can identify similar products to a target product (i.e., the product to be identified as similar) on an e-commerce platform, thereby helping the e-commerce company that owns the e-commerce platform gain insight into market conditions or helping merchants on the platform gain insight into the platform and its own competitiveness.

[0062] When identifying the same product as a target product on an e-commerce platform, a notification message such as Figure 1A The page shown is displayed on the display interface of the terminal device to prompt the user to enter the same product to be identified. Then, the user copies the web link of the product to be identified on the e-commerce platform to Figure 1A In the page shown, Figure 1A "https: / / detail.tmall.com / item.htm?spm=a230r.1.14.22.74ec239fFZBKCy&id=651346556119&ns=1&abbucket=9&sku_properties=5919063:6536025". Next, you can push the following information to the user's terminal device: Figure 1B The page shown, thus Figure 1B The page shown shows the same product to the user. Figure 1B The letter "X" sequence in the middle represents a web link of the same product on the e-commerce platform, and the letter "Y" sequence represents a web link of another same product on the e-commerce platform. Of course, there may be more than two identical products to be identified. In this case, the user can drag Figure 1B Use the slider to view more of the same item.

[0063] In the prior art, the identification of identical products mostly refers to recalling products with the same appearance. Products with the same appearance are not necessarily comparable in price. In view of this, the embodiments of the present disclosure provide a method for identifying identical products. The main idea is to combine the appearance, attributes and selling specifications of the products to determine the identical products of the target product, thereby making the matching of identical products more rigorous, and ultimately the identical products determined for the target product have the same value and comparable price as the target product. The identification method can be executed by a platform server, which provides an interface to the user's terminal device for the user to provide the identical products to be identified and push the identification results to the user.

[0064] Figure 2 FIG2 is a flow chart of a method for identifying identical products according to an embodiment of the present disclosure. Figure 2 , the identification method of the same product includes:

[0065] Step S110, obtaining a similar product pair, wherein the similar product pair includes two products to be identified and the appearance of the two products to be identified matches;

[0066] Step S120 , determining whether the two commodities to be identified are the same commodity based on the matching of the attributes and the matching of the selling specifications of the two commodities to be identified.

[0067] The method for identifying the same product provided by the embodiment of the present disclosure determines whether the target product and the product to be matched are the same product for the target product. It should be understood that in practice, it is necessary to search for the same product of the target product from the massive amount of products to be matched on the same platform and / or different platforms, which will consume a lot of time. In order to save time, the method for identifying the same product provided by the embodiment of the present disclosure is generally divided into two processes: coarse screening and fine screening, wherein the coarse screening process is mainly performed by the above-mentioned step S110, and the fine screening process is mainly performed by the above-mentioned step S120. The purpose of the coarse screening process is to obtain products that may be the same as the target product, that is, to obtain similar product pairs including the target product and products that may be the same as the target product, thereby narrowing the data range that needs to be screened in the fine screening process and reducing resource waste. The purpose of the fine screening process is to eliminate similar product pairs that are not the same product among many similar product pairs, so that the final determined same product has the same value and comparable price.

[0068] The coarse screening process and the fine screening process are described in detail below.

[0069] (1) Coarse screening process

[0070] (1) Similarity retrieval steps

[0071] Figure 3 The following is a flowchart for obtaining similar product pairs. Figure 3 Step S110, obtaining similar product pairs, including:

[0072] Step S111: scoring the similarity between the target product and the product to be matched based on the product feature information to obtain a similarity score;

[0073] Step S112: determining whether the target product and the product to be matched form a similar product pair based on the similarity score.

[0074] The product feature information includes the image feature information of the product, thereby ensuring that the appearance of the similar product pairs matches.

[0075] In step S111, the target product can be Figure 1A or Figure 1BThe information shown is provided by the user (such as a staff member or merchant of an e-commerce company); the products to be matched can be determined by a preset search range of the same products. For example, the search range of the same products of the target product is preset to the same e-commerce platform, and the products to be matched are the massive products on the same e-commerce platform; for another example, the preset search range of the same products of the target product is an e-commerce platform other than the e-commerce platform where the target product is located, and the products to be matched are the massive products on different e-commerce platforms.

[0076] In an optional embodiment, step S111 scores the similarity between the target product and the product to be matched based on the product feature information to obtain a similarity score, including: after extracting the first feature vector and the second feature vector, comparing the similarity between the target product and the product to be matched based on the first feature vector and the second feature vector; and then determining the similarity score based on the comparison result. Among them, the first feature vector is a feature vector that represents the product feature information of the target product, which can be extracted from the picture of the target product; the second feature vector is a feature vector that represents the product feature information of the product to be identified, which can be extracted from the picture of the product to be identified. In the embodiment of the present disclosure, the picture of the target product comes from the e-commerce platform where the target product is located; the picture of the product to be identified comes from the e-commerce platform where the product to be identified is located.

[0077] Specifically, when the first eigenvector represents the image feature information of the target product and the second eigenvector represents the image feature information of the product to be matched, the first eigenvector and the second eigenvector can be extracted using one of the methods such as convolutional neural network, SIFT algorithm, HOG algorithm or gray level co-occurrence matrix.

[0078] Convolutional Neural Networks (CNN) is a type of neural network (NNs). Its structure mainly includes input layer, convolution layer, ReLU (Rectified Linear Unit) layer, pooling layer and fully connected layer. Among them, the convolution layer is the core layer for building a convolutional neural network. Unlike conventional neural networks, the neurons in each layer of a convolutional neural network are arranged in three dimensions. Through training, the interconnected relationships between a large number of nodes within the convolutional neural network are adjusted, so that the convolutional neural network has the function of processing information and has relatively impressive performance. The function of processing information, for example, is to extract feature vectors representing image feature information from images.

[0079] After training, convolutional neural networks of different structures can all have the function of extracting feature vectors with considerable performance, but the extracted feature vectors will vary depending on the feature information they represent. In view of this, when using a convolutional neural network to extract the above-mentioned first and second feature vectors, the same trained convolutional neural network can be used for extraction. In this case, the convolutional neural network used directly outputs the first and second feature vectors. In addition, multiple trained convolutional neural networks with different structures can also be used for extraction. In this case, each convolutional neural network used outputs a first sub-feature vector and a second sub-feature vector. The multiple first sub-feature vectors are concatenated to form the above-mentioned first feature vector, and the multiple second sub-feature vectors are concatenated to form the above-mentioned second feature vector. In the above two cases, the latter case uses multiple trained convolutional neural networks with different structures, so the first feature vector obtained can more comprehensively represent the product feature information of the target product, and the second feature vector obtained can also more comprehensively represent the product feature information of the product to be matched. It should be understood that the multiple convolutional neural networks used to extract the multiple second sub-feature vectors in the latter case are the multiple convolutional neural networks used to extract the multiple first sub-feature vectors.

[0080] The method of splicing multiple first sub-feature vectors can be to arrange the elements of the first sub-feature vector output by each convolutional neural network from front to back according to the arrangement sequence of the convolutional neural network. For example, multiple convolutional neural networks with different structures that have completed training include a first convolutional neural network and a second convolutional neural network, wherein the first sub-feature vector output by the first convolutional neural network is (1,2,3), and the second sub-feature vector output by the second convolutional neural network is (2,5,6,7), then the first eigenvector is (1,2,3,2,5,6,7). The method of splicing multiple second sub-feature vectors is the same as the method of splicing multiple first sub-feature vectors mentioned above, which will not be repeated here.

[0081] The SIFT algorithm is an algorithm for extracting SIFT (Scale Invariant Feature Transform) features. SIFT features are based on the characteristics of local appearance points of interest on an object. They are independent of image size and rotation, and therefore highly tolerant to changes in lighting, noise, and micro-perspective. Due to these properties, SIFT features are highly significant and relatively easy to extract. The HOG algorithm is an algorithm for extracting HOG (histogram of oriented gradients) features. HOG features are constructed by calculating and statistically analyzing the directional gradient histograms of local image regions. Their advantage is that they remain highly invariant to geometric and optical deformations. In other words, HOG features are highly robust to environmental changes. Therefore, for the same product photographed in different environments on an e-commerce platform, the SIFT and HOG features remain invariant. Therefore, determining the similarity between the target product and the matching product based on SIFT and HOG features can effectively mitigate the adverse effects of different shooting environments. The ease of SIFT feature extraction makes the SIFT algorithm well-suited for feature extraction of the large number of products on e-commerce platforms.

[0082] The gray-level co-occurrence matrix is ​​a common method for describing texture by studying the spatial correlation of grayscale, thereby extracting the texture features of an image. Because texture is formed by the recurrence of grayscale distribution in spatial locations, a certain grayscale relationship exists between two pixels separated by a certain distance in image space, which is the spatial correlation of grayscale described above.

[0083] Of course, when the first feature vector represents the image feature information of the target product and the second feature vector represents the image feature information of the product to be matched, the first feature vector and the second feature vector can also be extracted in parallel using multiple methods such as convolutional neural networks, SIFT algorithms, HOG algorithms, or gray-level co-occurrence matrices. That is, multiple methods extract feature vectors from the image of the target product in parallel, one method extracts a feature vector from the image of the target product, and multiple methods concatenate multiple feature vectors extracted from the image of the target product to obtain the first feature vector; similarly, multiple methods extract feature vectors from the image of the product to be identified in parallel, thereby concatenating multiple feature vectors extracted from the image of the product to be identified to obtain the second feature vector. It should be understood that the multiple methods used to extract feature vectors from the image of the target product described above are the same as the multiple methods used to extract feature vectors from the image of the product to be identified. The method of concatenating feature vectors here is similar to the method of concatenating multiple first sub-feature vectors described above, and therefore will not be repeated. In the method for extracting the first feature vector and the second feature vector, since the first feature vector and the second feature vector are each formed by splicing multiple feature vectors extracted by multiple methods, the first feature vector can represent the product feature information of the target product in multiple aspects, and the second feature vector can also represent the product feature information of the product to be matched in multiple aspects.

[0084] After extracting the first and second eigenvectors, the similarity between the target product and the product to be matched can be measured by calculating the cosine distance or Euclidean distance between the first and second eigenvectors. The calculated cosine distance or Euclidean distance is the comparison result obtained by comparing the similarity between the target product and the product to be matched.

[0085] Cosine distance, also known as cosine similarity, uses the cosine value of the angle between the first eigenvector and the second eigenvector in the eigenvector space to measure the difference between the target product and the product to be matched. The smaller the angle between the first eigenvector and the second eigenvector, the more similar the target product and the product to be matched are.

[0086] Unlike cosine distance, Euclidean distance refers to the distance between the endpoints of each feature vector in the feature vector space and is directly related to the position coordinates of the endpoints of each feature vector. The closer the endpoints of the first feature vector are to the endpoints of the second feature vector, the more similar the target product and the product to be matched are.

[0087] It should be noted that for multi-dimensional feature vectors, cosine distance primarily distinguishes differences in direction and is insensitive to the numerical values ​​of each dimension. Euclidean distance, on the other hand, can reflect the differences in numerical features across dimensions and is therefore more commonly used for analyses that require differences based on the magnitude of the dimensions. Therefore, in practice, the decision to use cosine distance or Euclidean distance can be made based on the type of feature vector being analyzed for similarity.

[0088] In practice, the identification of identical products is also performed across two different e-commerce platforms. This involves recalling identical products from one e-commerce platform and those from the other. In this scenario, the feature vectors of each product on one e-commerce platform can be extracted, followed by the feature vectors of each product on the other. The cosine distance or Euclidean distance between the feature vectors of a product on one platform and the feature vectors of a product on the other platform is then calculated to compare the similarity between the two products, ultimately determining the similarity between any product on one platform and any product on the other.

[0089] After calculating the cosine distance or Euclidean distance between the first eigenvector and the second eigenvector, a similarity score is determined. The more similar the target product and the to-be-matched product are, the smaller the similarity score. In other embodiments, the more similar the target product and the to-be-matched product are, the larger the similarity score.

[0090] In another optional embodiment, step S111 scores the similarity between the target product and the product to be matched based on the product feature information to obtain a similarity score, including: based on the images of the target product and the product to be matched on the e-commerce platform, comparing the similarity between the target product and the product to be matched by one of the mean hash algorithm, the perceptual hash algorithm and the difference hash algorithm. Since the mean hash algorithm, the perceptual hash algorithm and the difference hash algorithm are existing algorithms for identifying image similarity based on pixel values, the comparison of the similarity between the target product and the product to be matched by each of these three algorithms will not be described in detail here. After obtaining the comparison result, the similarity score can be determined based on the comparison result obtained as described in the previous embodiment.

[0091] In another optional embodiment, there are multiple products to be matched, and step S111 scores the similarity between the target product and the product to be matched based on the product feature information to obtain a similarity score, including: after extracting multiple second feature vectors, performing feature clustering on the multiple second feature vectors to form multiple classes; searching the class with the smallest distance between the class center and the first feature vector from the multiple classes to obtain a target class, and the second feature vector in the target class is the target feature vector; comparing the similarity between the target product and the product to be matched corresponding to the target feature vector based on the first feature vector and the target feature vector; and determining the similarity score between the target product and the product to be matched corresponding to the target feature vector based on the comparison result obtained; wherein the first feature vector is a feature vector representing the product feature information of the target product, and the second feature vector is a feature vector representing the product feature information of the product to be identified.

[0092] Specifically, the above-mentioned feature clustering method adopts, for example, the K-means clustering algorithm or the density-based spatial clustering of applications with noise (DBSCAN) algorithm; the distance between the cluster center and the first eigenvector can be the Euclidean distance as described above; based on the first eigenvector and the target eigenvector, the similarity between the target product and the product to be matched corresponding to the target eigenvector can be compared using the cosine distance or Euclidean distance as described in the above embodiment. The similarity score between the target product and the product to be matched corresponding to the target eigenvector is determined based on the comparison result, or the similarity score can be determined based on the representational meaning of the similarity score as described above.

[0093] It should be noted that this embodiment only determines the similarity score between the target product and the to-be-matched product corresponding to the target feature vector. For classes other than the target class among multiple classes, the to-be-matched products corresponding to each second feature vector are considered dissimilar to the target product, that is, they are considered unable to form a similar product pair with the target product.

[0094] This embodiment is based on clustering multiple second feature vectors to form multiple classes. Since similarity scores with the target product are determined only for the target product corresponding to each target feature in the target class, the number of matching products for which similarity scores need to be calculated is effectively reduced. The determination of target classes significantly narrows the search scope for similar product pairs, thereby reducing the overall time spent identifying identical products to the target product from among multiple matching products.

[0095] Furthermore, the product feature information may also include at least one of the following: text feature information of the product, and feature information of user behavior toward the product.

[0096] Specifically, product text feature information can be extracted from at least one of the following fields related to the product on the e-commerce platform: product attribute list, product title, product details, and the product category-property-value knowledge system (Category-Property-Value). The text feature information extracted from these fields is the text information describing the product characteristics. Existing text processing methods can be used to extract text feature information from these fields.

[0097] In the case where the product feature information includes text feature information of the product, step S111 scores the similarity between the target product and the product to be matched based on the text feature information, thereby obtaining a text similarity score.

[0098] The behavioral characteristics of users towards products can be extracted from their historical behaviors towards products. For example, a historical behavior of a user on an e-commerce platform is "clicking product a and then clicking product b." The behavioral characteristics of the user towards product a extracted from this historical behavior is "after clicking product a, the user clicked a similar product to product a, namely product b." Artificial Neural Networks (ANNs) can be used to extract behavioral characteristics. Artificial Neural Networks, also referred to as Neural Networks (NNs) or Connection Models, are models that mimic the behavioral characteristics of animal neural networks and perform distributed parallel information processing. After being trained with a large amount of historical user behavior data, they can achieve the purpose of extracting behavioral characteristics of users towards products from their historical behaviors towards products.

[0099] In the case where the product feature information includes the product's behavioral feature information, step S111 will score the similarity between the target product and the product to be matched based on the behavioral feature information, thereby obtaining a behavioral similarity score. It should be noted that there are generally two behavioral similarity scores. The first one (also called the first behavioral similarity score) indicates that the target product and the product to be matched form a similar product pair based on the behavioral feature information; the second one (also called the second behavioral similarity score) indicates that the target product and the product to be matched do not form a similar product pair based on the behavioral feature information. As long as one of the behavioral feature information of the target product and the behavioral feature information of the product to be matched determines that the target product and the product to be matched form a similar product pair, the above-mentioned first behavioral similarity score is obtained; otherwise, the above-mentioned second similarity score is obtained.

[0100] The similarity between the target product and the product to be matched is scored based on the product's image feature information to obtain an image similarity score. If at least one of a text similarity score and a behavioral similarity score is also obtained, then in step S112, it is necessary to determine whether the target product and the product to be matched form a similar product pair based on the multiple similarity scores obtained. For example, the determination of whether the target product and the product to be matched form a similar product pair is made based on the sum of the multiple similarity scores or the weighted average of the multiple similarity scores.

[0101] It should be noted that, as long as the similarity score does not indicate that the target product and the to-be-matched product are completely dissimilar, step S112 determines that the target product and the to-be-matched product form a similar product pair. After the target product and the to-be-matched product form a similar product pair, the target product and the to-be-matched product are the two to-be-identified products in step S110.

[0102] The above-mentioned step S110 performed through step S111 and step S112 is a similarity search performed based on product feature information, which aims to preliminarily screen out similar product pairs from a large number of products. This is the initial step of the coarse screening process.

[0103] (2) Coarse-grained threshold filtering step

[0104] Figure 4 FIG. 1 is a schematic diagram showing the execution of the method for identifying identical products provided by an embodiment of the present disclosure. Figure 4 When the similarity search determines that the target product and the product to be matched form a similar product pair, the method for identifying the same product further includes a coarse-grained threshold filtering step, specifically as follows: determining whether the similarity score meets a first threshold condition to obtain a first judgment result; wherein, when the first judgment result is that the similarity score does not meet the first threshold condition, there is no need to perform the above-mentioned step S120, that is, there is no need to determine whether the two products to be identified are the same product based on the attribute matching and the selling specification matching of the two products to be identified.

[0105] The value range of the aforementioned similarity score can be normalized to a range of 0 to 1. For example, a similarity score of 0 indicates that the target product and the product to be matched are completely identical, while a similarity score of 1 indicates that the target product and the product to be matched are completely different, corresponding to the aforementioned setting of "the more similar the target product and the product to be matched, the smaller the similarity score." Based on this, whether the first threshold condition is met, that is, whether the similarity score is less than the score threshold in the first threshold condition, if not, the first judgment result is that the similarity score does not meet the first threshold condition; otherwise, the first judgment result is that the similarity score meets the first threshold condition.

[0106] In this step, if the first judgment result is that the similarity score does not meet the first threshold condition, there is no need to perform the above step S120, that is, the similar product pairs formed by the target product and the product to be matched have been filtered out of the range of the same product through the coarse-grained threshold filtering step.

[0107] (3) Coarse-grained category mapping filtering step

[0108] After the coarse-grained threshold filtering step is executed, if the first judgment result is that the similarity score meets the first threshold condition, then the similar product pair formed by the target product and the product to be matched is not filtered out of the range of the same product. Figure 4 In this case, the method for identifying identical products further includes a step of coarse-grained category mapping and filtering, specifically as follows: determining whether the first-level categories of the target product and the product to be matched match to obtain a second judgment result. Wherein, if the second judgment result is that the first-level categories of the target product and the product to be matched do not match, there is no need to perform the above-mentioned step S120, that is, there is no need to determine whether the two products to be identified are identical products based on the matching of the attributes and the matching of the selling specifications of the two products to be identified.

[0109] Specifically, a product category refers to a collection of products that share one or more common attributes. It represents a group of products with these attributes. Product categories are organized into multiple levels, depending on the granularity of the attributes. The coarsest granularity is called a first-level category, and the finest granularity is called a leaf category. For example, in the category "Clothing - Women's Clothing - Dresses," Clothing is a first-level category, Women's Clothing is a second-level category, and Dresses is a leaf category.

[0110] It should be noted that the same e-commerce platform adopts a unified category naming rule, and different e-commerce platforms are likely to adopt different category naming rules. Therefore, the same product may have different product categories on different e-commerce platforms. Figure 5 Tables 1 and 2 in the table show the product categories of four products, namely humidifier parts, vacuum cleaner parts, washing machine parts, and electric heaters, on AE and ICBU, respectively. The first-level category names of each product are the same. Of course, the first-level category names of other products on AE and ICBU may also be different. In addition, Figure 5 The leaf category names of each product are the same. This is because the leaf category names of each e-commerce platform are all product names. Since the product names are unified, the leaf category names of the same product on different e-commerce platforms are the same.

[0111] It should be emphasized that if the target product and the product to be matched come from different e-commerce platforms, then when judging whether the first-level categories of the target product and the product to be matched match, it is necessary to first complete the category mapping according to the category mapping relationship. Category mapping refers to the process of finding the category mapping relationship between the category systems of different e-commerce platforms and completing the category conversion. In the embodiment of the present disclosure, it can be that the first-level category of the product to be matched is converted to the e-commerce platform where the target product is located according to the category mapping relationship, and then it is judged whether the converted first-level category is the same as the first-level category of the target product; it can also be that the first-level category of the target product is converted to the e-commerce platform where the product to be matched is located according to the category mapping relationship, and then it is judged whether the converted first-level category is the same as the first-level category of the product to be matched.

[0112] The category mapping relationship used in the embodiment of the present disclosure can be generated based on a product other than the target product and the product to be matched. Figure 5 For the product of CEC Heater, it can be determined that a category mapping relationship is formed between the first-level category name "household appliances" on AE and the first-level category name "household appliances" on ICBU.

[0113] In this step, if the second judgment result is that the first-level categories of the target product and the product to be matched do not match, there is no need to execute the above step S120. That is, the similar product pairs that are not filtered out by the coarse-grained threshold filtering step are filtered out of the range of the same product through the coarse-grained category mapping filtering step.

[0114] This concludes the coarse screening process. If the similar product pair formed by the target product and the product to be matched has not been filtered out, the similar product pair enters the fine screening process.

[0115] (2) Fine screening process

[0116] (1) Steps of fine-grained threshold filtering

[0117] The second judgment result is that the target product and the to-be-matched product have a matching primary category, and the similar product pairs formed by the target product and the to-be-matched product have not been filtered out. In this case, refer to Figure 4 The method for identifying identical products further includes a step of fine-grained threshold filtering, specifically as follows: determining whether the similarity score satisfies a second threshold condition; and, if the similarity score satisfies the second threshold condition, executing the above-mentioned step S120, i.e., determining whether the two products to be identified are identical products based on the matching of the attributes and the matching of the selling specifications of the two products to be identified.

[0118] It should be noted that the first threshold condition is a universal condition for all products, meaning that the score threshold used in the first threshold condition applies to all products; while the second threshold condition is a condition specific to the target product, meaning that the score threshold used in the second threshold condition applies only to the target product. The second threshold condition can be understood with reference to the first threshold condition, except that the second threshold condition is stricter than the first threshold condition. Under the setting of "the more similar the target product and the product to be matched, the lower the similarity score," the score threshold used in the second threshold condition is lower than that used in the first threshold condition, thus enabling fine-grained threshold filtering of similar product pairs using the second threshold condition.

[0119] Different industries have different similarity requirements for the same product. Therefore, the score threshold used in the second threshold condition is set to different values ​​depending on the industry to which the product belongs. For example, under the setting "The more similar the target product and the matching product, the lower the similarity score," the second threshold condition would require a score threshold of 0.25 for the general industry, 0.13 for the jewelry industry, and 0.15 for the bicycle industry. This is because jewelry within the jewelry industry can have different prices depending on the details, and bicycles within the bicycle industry can have different prices depending on the details of their components.

[0120] In this step, if the similarity score does not meet the second threshold condition, step S120 is not performed. The fine-grained threshold filtering step is used to filter out similar product pairs from the same product range. This step is fine-grained based on industry knowledge. Based on the similarity distribution of each industry, the second threshold condition (primarily determining the score threshold used in the second threshold condition) that meets the expected precision and recall rate is determined. This eliminates similar product pairs in each industry that are clearly not the same product.

[0121] (2) Attribute matching step and sales specification matching step

[0122] Combine Figure 4 If the similarity score meets the second threshold condition, step S120 is executed, that is, the attribute matching step and the sales specification matching step are first executed.

[0123] During the execution of the attribute matching step:

[0124] The two products to be identified may come from two different e-commerce platforms. There are differences in the expression of product attributes on the two different e-commerce platforms, including differences in the expression of attribute names and attribute values. In this case, the attribute matching of the two products to be identified is determined by the following steps: the attribute information of the two products to be identified is standardized, including the standardized conversion of attribute names and attribute values, to obtain two standardized attribute information sets, and then the two standardized attribute information sets are matched. Figure 6, two standardized attribute information sets, namely the target product attribute set and the product attribute set to be matched; then, perform an intersection operation on the two standardized attribute information sets. If the attribute value of a certain attribute in the two standardized attribute information sets is the same, then after the intersection operation, the attribute belongs to the attribute in the obtained attribute information intersection; determine whether the attributes in the attribute information intersection include the target attribute.

[0125] It should be noted that the target attribute may be one or more. Only when the attributes in the attribute information intersection include all target attributes, the attributes of the two to-be-identified products are determined to match. Otherwise, the attributes of the two to-be-identified products are determined to not match.

[0126] Combine Figure 6 The attribute information of the product to be identified is extracted based on the product attribute list on the e-commerce platform and the following fields related to the product to be identified: the product title, product details, product category-attribute-attribute value knowledge system, wherein the product attribute list provides the standard attribute expression of the e-commerce platform. Therefore, according to the standard attribute expression provided by the product attribute list, the product attribute information can be extracted from the product title, product details, product category-attribute-attribute value knowledge system in a targeted manner, and, according to the standard attribute expression provided by the product attribute list, the attribute mapping information required for the standardized conversion can be determined, thereby using the determined attribute mapping information to quickly perform standardized conversion on the extracted product attribute information.

[0127] In practice, only one of the two products to be identified may have the standard attribute expression of the e-commerce platform where it is located consistent with the attribute expression standard used in the above-mentioned standardized conversion. In this way, the attribute information of the other product to be identified will no longer use the standard attribute expression of the e-commerce platform where it is located after the standardized conversion.

[0128] The target attribute is an attribute that determines the price of the product to be identified, and is usually provided by industry knowledge. If the target attributes of the two products to be identified do not match in this step, then in step S120 it is determined that the two products to be identified are not the same product.

[0129] For example, the target product is a 10m waterproof electronic watch sold on AE, and the matching product is a 30m waterproof electronic watch sold on Alibaba.com with the same appearance and other characteristics. Since waterproof performance is a target attribute in the watch industry, and the attributes in the intersection of the attribute information of the two electronic watches do not include the target attribute, the attributes of the two electronic watches do not match, and therefore the two electronic watches are not the same product.

[0130] During the sales specification matching step:

[0131] The matching of the selling specifications of the two commodities to be identified is determined by the following steps: the selling specification information of the two commodities to be identified is respectively standardized to obtain two standardized selling specification information, and the standardized selling specification information corresponds one-to-one to the commodities to be identified; and the two standardized selling specification information are compared.

[0132] Combine Figure 7 , it is possible to specifically extract the sales specification information from the product title, product details and other fields based on the standard attribute expression provided by the product attribute list. The sales specification information includes the sales quantity and the sales unit. In view of the different sales units on different e-commerce platforms, the unit conversion table is called to perform standardized conversions on the sales specification information of the target product and the sales specification information of the product to be matched, thereby obtaining the sales quantity under the standardized unit of the target product (i.e. the standardized sales specification information of the target product) and the sales quantity under the standardized unit of the product to be matched (i.e. the standardized sales specification information of the product to be matched). After comparison, if the sales quantity under the standardized unit of the target product is consistent with the sales quantity under the standardized unit of the product to be matched, then the sales specifications of the two products to be identified match; otherwise, the sales specifications of the two products to be identified do not match.

[0133] For example, the target product is 5000g of Northeastern rice sold on AE, and the product to be matched is 10kg of Northeastern rice sold on Alibaba.com. The standard unit of weight in the unit conversion table is kg. Therefore, the standardized selling specification information of the target product is 5kg, and the standardized selling specification information of the product to be matched is 10kg. Since "5" and "10" do not match, the selling specifications of the two products to be identified do not match.

[0134] If it is determined in this step that the selling specifications of the two commodities to be identified do not match, then in step S120 it is determined that the two commodities to be identified are not the same commodity.

[0135] Combine Figure 4 , there is no order difference between the attribute matching step and the selling specification matching step, that is, the attribute matching step can be executed first, or the selling specification matching step can be executed first. It should be understood that since the attributes of the two commodities to be identified do not match, it is determined that the two commodities to be identified are not the same commodity. Therefore, after executing the attribute matching step first, if it is determined that the attributes of the two commodities to be identified do not match, the selling specification matching step will no longer be executed, otherwise it is necessary to execute the selling specification matching step. Similarly, since the selling specifications of the two commodities to be identified do not match, it is determined that the two commodities to be identified are not the same commodity. Therefore, after executing the selling specification matching step first, if it is determined that the selling specifications of the two commodities to be identified do not match, the attribute matching step will no longer be executed, otherwise it is necessary to execute the attribute matching step.

[0136] The above-mentioned steps of determining whether the attributes of two products to be identified match, as well as the steps of determining whether the selling specifications of two products to be identified match, solve the problem of differences in product representation across platforms, and are therefore of great significance for connecting product data between different platforms and even international e-commerce platforms and forming a connected product database.

[0137] (3) Fine-grained category verification steps

[0138] Reference Figure 4 If step S120 determines that the attributes of the two commodities to be identified match and the selling specifications match after the attribute matching step and the selling specification matching step are executed, then step S120 also includes a fine-grained category verification step, specifically as follows: judging whether the leaf categories of the two commodities to be identified match, and, when the leaf categories of the two commodities to be identified match, determining that the two commodities to be identified are the same commodity. Since the leaf category names of the same commodity are the same even on different e-commerce platforms, this step can be a direct comparison of whether the leaf categories of the two commodities to be identified are consistent. If they are consistent, it is determined that the leaf categories of the two commodities to be identified match, thereby completing the process of identifying the same commodity. Since commodities with the same leaf category are not necessarily comparable in price, but the probability of commodities with different leaf categories being comparable in price is very small, the fine-grained category verification step is placed at the end of the fine screening process, which is of great significance to ensuring the correctness of the final determined same commodity.

[0139] Reference Figure 4 After determining that the target product and the product to be matched are the same product, data rules can be applied to the same product formed by the target product and the product to be matched according to the needs of subsequent processes, so that they can be stored in the database for easy use in subsequent processes.

[0140] The embodiment of the present disclosure combines the appearance, attributes and selling specifications of the goods to determine whether two goods are the same, which makes the matching of the same goods more rigorous; the matching method of the appearance, attributes and selling specifications of the goods can solve the problem of expression differences between platforms, and can therefore be applied to recalling the same goods from different e-commerce platforms; moreover, the identification of the same goods adopts a process from coarse to fine and front and back verification, wherein the coarse screening determines the range of the same goods and the fine screening accurately locates the same goods. This process makes the recall of the same goods faster and less resource-consuming. At the same time, the threshold filtering from coarse granularity to fine granularity and the category verification from coarse granularity to fine granularity are used in the two front and back screenings, thereby ensuring that the scale of the goods to be matched recalled in each step will not be too large, and ensuring that the same goods can be accurately searched from the goods to be matched that are not too large in scale. The method for identifying the same goods provided by the embodiment of the present disclosure provides technical support for connecting the product data between multiple e-commerce platforms and forming a connected product database, and has good market prospects in the e-commerce field.

[0141] According to one embodiment of the present disclosure, Figure 8 As shown, a device for identifying the same product is also provided, including:

[0142] A coarse screening unit 110 is configured to obtain similar product pairs, where the similar product pairs include two products to be identified and the appearance of the two products to be identified match;

[0143] The fine screening unit 120 is used to determine whether two commodities to be identified are the same commodity based on the matching of their attributes and selling specifications.

[0144] Optionally, the coarse screening unit 110 is used to: score the similarity between the target product and the product to be matched based on product feature information to obtain a similarity score, wherein the product feature information includes image feature information of the product; and determine whether the target product and the product to be matched form a similar product pair based on the similarity score.

[0145] Optionally, when it is determined based on the similarity score that the target product and the product to be matched form a similar product pair, the target product and the product to be matched are two products to be identified, and the coarse screening unit 110 is further used to: determine whether the similarity score meets a first threshold condition to obtain a first judgment result; wherein, when the first judgment result is that the similarity score does not meet the first threshold condition, the fine screening unit does not need to determine whether the two products to be identified are the same product based on the attribute matching and sales specification matching of the two products to be identified.

[0146] Optionally, when the first judgment result is that the similarity score meets the first threshold condition, the coarse screening unit 110 is also used to: determine whether the first-level category of the target product and the product to be matched matches to obtain a second judgment result; wherein, when the second judgment result is that the first-level category of the target product and the product to be matched does not match, the fine screening unit does not need to determine whether the two products to be identified are the same product based on the attribute matching and sales specification matching of the two products to be identified.

[0147] Optionally, when the second judgment result is that the first-level category of the target product and the product to be matched matches, the fine screening unit 120 is also used to: judge whether the similarity score meets the second threshold condition; and, when the similarity score meets the second threshold condition, determine whether the two products to be identified are the same product based on the attribute matching and sales specification matching of the two products to be identified; wherein the first threshold condition is a condition that is general to products, the second threshold condition is a condition that is specific to the target product, and the second threshold condition is stricter than the first threshold condition.

[0148] Optionally, the fine screening unit 120 determines whether the two commodities to be identified are the same commodity based on the attribute matching and selling specification matching of the two commodities to be identified, and performs the following steps: after determining that the attributes of the two commodities to be identified match and the selling specification matching of the two commodities to be identified, judge whether the leaf categories of the two commodities to be identified match, and, if the leaf categories of the two commodities to be identified match, determine that the two commodities to be identified are the same commodity.

[0149] Since the implementation details of the above device have been described in the detailed introduction of the method embodiment above, they are not repeated here to save space.

[0150] According to an embodiment of the present disclosure, the method for identifying the same product can be: Figure 9 The platform server 800 is implemented. Figure 9 800 according to an embodiment of the present disclosure. Figure 9 The platform server 800 shown is merely an example and should not limit the functionality and scope of use of the embodiments of the present disclosure.

[0151] like Figure 9 As shown, the platform server 800 is implemented as a general-purpose computing device. Components of the platform server 800 may include, but are not limited to, the at least one processing unit 810, the at least one storage unit 820, and a bus 830 connecting various system components (including the storage unit 820 and the processing unit 810).

[0152] The storage unit stores program codes, which can be executed by the processing unit 810, so that the processing unit 810 performs the steps of various exemplary embodiments of the present disclosure described in the description of the exemplary method above. For example, the processing unit 810 can perform the following steps: Figure 2 The steps shown in .

[0153] The storage unit 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 8201 and / or a cache memory unit 8202 , and may further include a read-only memory unit (ROM) 8203 .

[0154] The storage unit 820 may also include a program / utility 8204 having a set (at least one) of program modules 8205, such program modules 8205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0155] Bus 830 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0156] The platform server 800 can also communicate with one or more external devices 700 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more terminal devices that enable users to interact with the platform server 800, and / or any device that enables the computer device 800 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication can occur via an input / output (I / O) interface 850. Furthermore, the computer device 800 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 860. As shown, the network adapter 860 communicates with other modules of the platform server 800 via a bus 830. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the platform server 800, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0157] It will be understood by those skilled in the art that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling related hardware through instructions, and the instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. To this end, the present disclosure also provides a computer-readable medium on which a computer executable code is stored, and when the computer executable code is executed by the processor, the various processes of the various embodiments of the above identification method are implemented. Among them, computer-readable media, such as U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., can store computer executable code.

[0158] Since the code stored in the readable medium can execute the steps of any of the identification methods provided in the embodiments of the present disclosure, the beneficial effects that can be achieved by any of the identification methods provided in the embodiments of the present disclosure can be achieved. For details, please refer to the previous embodiments and will not be repeated here. The specific implementation of each of the above operations can be referred to the previous embodiments and will not be repeated here.

[0159] It should be understood that the above description is merely a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, many variations of the embodiments of this specification are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure shall be included within the scope of protection of the present disclosure.

[0160] It should be understood that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.

[0161] It should be understood that the foregoing description of this specification is based on specific embodiments. Other embodiments are within the scope of the claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0162] It should be understood that an element described herein in the singular or shown in the drawings as only one does not limit the number of the element to one. In addition, modules or elements described or shown herein as separate may be combined into a single module or element, and modules or elements described or shown herein as single may be split into multiple modules or elements.

[0163] It should also be understood that the terms and expressions used herein are for descriptive purposes only, and the one or more embodiments of this specification should not be limited to these terms and expressions. The use of these terms and expressions does not mean to exclude any equivalent features of the illustrations and descriptions (or portions thereof), and it should be recognized that various modifications that may exist should also be included in the scope of the claims. Other modifications, variations, and substitutions may also exist. Accordingly, the claims should be deemed to cover all such equivalents.

Claims

1. A method for identifying identical products, comprising: Obtaining a similar product pair, the similar product pair including two products to be identified and having matching appearances, scoring a similarity between a target product and the products to be matched based on product feature information to obtain a similarity score, wherein the product feature information includes image feature information of the products, and determining whether the target product and the products to be matched form a similar product pair based on the similarity score; When it is determined according to the similarity score that the target product and the product to be matched form a similar product pair, the target product and the product to be matched are two products to be identified, and whether the similarity score meets a first threshold condition is determined to obtain a first judgment result; If the similarity score satisfies the first threshold condition, determining whether the target product and the first-level category of the to-be-matched product match to obtain a second determination result; If the second judgment result is that the target product and the to-be-matched product have a matching primary category, determining whether the similarity score meets a second threshold condition; When the similarity score meets the second threshold condition, determine whether the two commodities to be identified are the same commodity based on the attribute matching and sales specification matching of the two commodities to be identified. After determining that the attributes of the two commodities to be identified match and the sales specification matching, judge whether the leaf categories of the two commodities to be identified match, and, if the leaf categories of the two commodities to be identified match, determine that the two commodities to be identified are the same commodity.

2. The identification method according to claim 1, wherein: The two products to be identified come from two different e-commerce platforms, and the product attributes of the two different e-commerce platforms are expressed differently. The attribute matching of the two products to be identified is determined by the following steps: Performing standardization conversion on the attribute information of the two commodities to be identified respectively to obtain two standardized attribute information sets, wherein the standardized attribute information sets correspond to the commodities to be identified in a one-to-one manner; Performing an intersection operation on the two standardized attribute information sets to obtain an attribute information intersection; Determining whether the attributes in the intersection of the attribute information include a target attribute, where the target attribute is an attribute that determines the price of the commodity to be identified; Wherein, when the attributes in the intersection of the attribute information include the target attribute, the attributes of the two commodities to be identified match.

3. The identification method according to claim 1, wherein: The matching of the selling specifications of the two commodities to be identified is determined by the following steps: Standardizing the sales specification information of the two commodities to be identified respectively to obtain two standardized sales specification information, wherein the standardized sales specification information corresponds to the commodities to be identified one by one; Comparing the two standardized sales specification information; The standardized sales specification information is the sales quantity under the standard unit. When the sales quantities in the two pieces of standardized sales specification information are consistent, the sales specifications of the two commodities to be identified match.

4. The identification method according to claim 3, wherein: Based on the product feature information, the similarity between the target product and the product to be matched is scored to obtain a similarity score, including: After extracting the first feature vector and the second feature vector, comparing the similarity between the target product and the product to be matched based on the first feature vector and the second feature vector; and, determining the similarity score based on the obtained comparison result; The first feature vector is a feature vector representing the commodity feature information of the target commodity, and the second feature vector is a feature vector representing the commodity feature information of the commodity to be identified; The first feature vector and the second feature vector are each obtained by concatenating multiple feature vectors extracted by multiple different neural networks, or are each obtained by concatenating multiple feature vectors extracted by multiple different methods.

5. The identification method according to claim 3, wherein: There are multiple products to be matched, and similarity scores are scored between the target product and the products to be matched based on product feature information to obtain similarity scores, including: After extracting the plurality of second feature vectors, performing feature clustering on the plurality of second feature vectors to form a plurality of clusters; Searching for a class with the smallest distance between the class center and the first eigenvector from the multiple classes to obtain a target class, wherein the second eigenvector in the target class is a target eigenvector; comparing the similarity between the target product and the to-be-matched product corresponding to the target feature vector based on the first feature vector and the target feature vector; and, determining a similarity score between the target product and the product to be matched corresponding to the target feature vector based on the obtained comparison result; The first feature vector is a feature vector representing the commodity feature information of the target commodity, and the second feature vector is a feature vector representing the commodity feature information of the commodity to be identified.

6. The identification method according to claim 3, wherein: The commodity feature information also includes at least one of the following: text feature information of the commodity, and user behavior feature information about the commodity.

7. The identification method according to claim 3, wherein: If the first judgment result is that the similarity score does not meet the first threshold condition, it is not necessary to determine whether the two commodities to be identified are the same commodity based on the matching of the attributes and the matching of the selling specifications of the two commodities to be identified.

8. The identification method according to claim 7, wherein: When the second judgment result is that the primary categories of the target product and the product to be matched do not match, it is not necessary to determine whether the two products to be identified are the same product based on the matching of the attributes and the matching of the selling specifications of the two products to be identified.

9. The identification method according to claim 8, wherein: The first threshold condition is a condition common to all commodities, the second threshold condition is a condition specific to the target commodity, and the second threshold condition is stricter than the first threshold condition.

10. A platform server, comprising: a memory for storing computer executable code; A processor, configured to execute the computer executable code to implement the identification method according to any one of claims 1 to 9.

11. A computer-readable medium comprising computer-executable codes, wherein when the computer-executable codes are executed by a processor, the identification method according to any one of claims 1 to 9 is implemented.

Images