Method for aggregating identical products and its device, equipment, medium, and product

Abstract

The present application relates to the field of e-commerce information technology, and discloses a method for aggregating identical products and its device, equipment, medium, and product. The method includes: obtaining a category label of a product object in a product database, wherein the category label is obtained by calling a pre-trained product classification model to classify the product information of the product object; obtaining a comprehensive feature vector corresponding to the product information of the product object, and calculating the similarity between product combinations consisting of two product objects under the same category signature, wherein the comprehensive feature vector is a deep semantic feature representation extracted by the product classification model; screening out the product combinations under each category signature whose similarity exceeds a preset threshold as identical product combinations; aggregating all identical product combinations to obtain a list of identical products corresponding to each category label. The present application can achieve accurate aggregation of identical products, and thereby construct a list of identical products to improve user experience.

Description

Technical Field

[0001] The present application relates to the field of e-commerce information technology, and in particular to a method for aggregating identical products and its corresponding apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] The phenomenon of listing the same products on the product interface displayed on the e-commerce platform often causes the same products to occupy a large amount of the limited display space of the graphical user interface of the user terminal device, which affects the shopping experience of buyers and users in browsing various styles, and also affects the success rate of product recommendations. One solution in the prior art is that the backend server of the e-commerce platform pre-determines the same products based on the similarity of the main pictures or product titles of each product object and aggregates the same products to respond to the event of obtaining the same aggregated products for the specified product triggered by the platform user, and push the same product collection corresponding to the product to the graphical user interface of the user terminal device. The same product collection is a collection of multiple same products of the same style as the product, and the display method on the graphical user interface is mostly a carousel display, thereby optimizing the problem of repeated appearance of the same products on the product interface display.

[0003] Generally speaking, most e-commerce platforms usually use the text information between product pairs as features to aggregate products of the same type, and judge whether they are the same type based on whether the features are similar, or judge whether the two products are the same type based on whether they have overlapping pictures. This usually leads to some problems. Specifically, the effect of the method based on image matching is relatively limited. The different shooting angles of the photos showing the two products may lead to the failure of identifying the same product. The two product titles and descriptions are almost the same, and it is obvious that the products should actually be the same type. However, since the final judgment is based only on the image matching dimension and lacks analysis of the title text matching dimension, the judgment of the same product is incorrect.

[0004] In order to enable the same products to be accurately identified and aggregated to reduce the space occupied by the same products in the graphical interface of product display, and to construct a list of the same products for user reference and improve user experience, the applicant has made corresponding explorations in this regard. Summary of the invention

[0005] The purpose of this application is to meet user needs and provide a method for aggregating similar products and its corresponding apparatus, computer equipment, non-volatile storage medium and computer program product.

[0006] In order to meet the various objectives of this application, this application adopts the following technical solutions:

[0007] A method for aggregating products of the same type provided for the purpose of this application includes the following steps:

[0008] Obtaining a category label of a commodity object in a commodity database, wherein the category label is obtained by calling a pre-trained commodity classification model to classify commodity information of the commodity object;

[0009] Obtaining a comprehensive feature vector corresponding to the product information of the product object, and calculating the similarity between product combinations consisting of two product objects signed by the same type of target, wherein the comprehensive feature vector is a deep semantic feature representation extracted by the product classification model;

[0010] Filter out the product combinations whose similarity in each category exceeds a preset threshold as the same product combination;

[0011] Aggregate all the same product combinations to obtain a list of the same products corresponding to each category tag.

[0012] In an extended embodiment, the commodity classification model performs the following steps to generate a category label for the commodity object, including the following steps:

[0013] Obtain product information of a product object in a product database, wherein the product information includes a title text and a product image;

[0014] Extracting corresponding text feature vectors and picture feature vectors from the title text and the product picture respectively;

[0015] Classify the commodity object according to the text feature vector and / or the image feature vector to obtain a category label corresponding to the commodity object;

[0016] The text feature vector and / or the image feature vector are stored as the comprehensive feature vector associated with the corresponding commodity object.

[0017] In a further embodiment, storing the text feature vector and / or the image feature vector as the comprehensive feature vector associated with the corresponding commodity object comprises the following steps:

[0018] Calling a preset text feature extraction model to extract text features from the product information to obtain its text feature vector;

[0019] Calling a preset image feature extraction model to extract image features of the product information and obtain its image feature vector;

[0020] After the image feature vector is normalized into a single high-dimensional vector with the same dimension as the text feature vector, it is concatenated with the text feature vector to obtain a comprehensive feature vector.

[0021] In a further embodiment, obtaining a comprehensive feature vector corresponding to the commodity information of a commodity object and calculating the similarity between commodity combinations consisting of two commodity objects signed by the same type of target includes the following steps:

[0022] Obtain the pre-stored comprehensive feature vector of the commodity object;

[0023] Calculate the cosine distance between two comprehensive feature vectors in the product combination composed of two product objects signed by the same type of target as the similarity between the two product objects;

[0024] The similarities between each pair of commodity objects in each commodity combination signed by each type of target are stored.

[0025] In a further embodiment, selecting the product combinations whose similarity in each category exceeds a preset threshold as the same product combinations includes the following steps:

[0026] determining a preset threshold value;

[0027] Obtain the similarity of the product combinations signed by the same category of targets, and compare it with the preset threshold;

[0028] When the corresponding similarity of the commodity combination exceeds the preset threshold, it is determined that the commodity combination is a combination of the same commodity.

[0029] In a further embodiment, determining the preset threshold value comprises the following steps:

[0030] The similarity of a batch of product combinations is randomly obtained to obtain the manual labeling information of whether they are the same or different items, so as to determine the corresponding two data sets;

[0031] Performing normal distribution statistics on the two data sets according to the similarity, and determining the similarity corresponding to the intersection of two corresponding normal distribution curves as a first threshold;

[0032] Acquire a predetermined second threshold value;

[0033] The first threshold and the second threshold are compared, and the highest one is used as the preset threshold.

[0034] In a further embodiment, all combinations of the same product are aggregated to obtain a list of the same product corresponding to each category tag, including the following steps:

[0035] For each category tag, a maximum greedy matching algorithm is applied to select multiple similar product families. Each similar product family is a set of multiple combinations of the same product corresponding to the same product object and includes multiple product objects. Different similar product families contain different product objects.

[0036] Determine the product with the largest inventory in the same similar product family as the representative product, and the rest as the represented products;

[0037] Convert each similar product family into a list of identical products.

[0038] A device for aggregating commodities of the same type proposed for the purpose of this application comprises:

[0039] Category label acquisition module: used to acquire category labels of commodity objects in the commodity database, wherein the category labels are obtained by calling a pre-trained commodity classification model to classify commodity information of the commodity objects;

[0040] Similarity calculation module: used to obtain the comprehensive feature vector corresponding to the commodity information of the commodity object, and calculate the similarity between the commodity combinations consisting of two commodity objects signed by the same category of targets. The comprehensive feature vector is a deep semantic feature representation extracted by the commodity classification model;

[0041] Same product screening module: used to screen out the product combinations whose similarity in each category exceeds a preset threshold as the same product combinations;

[0042] List generation module: used to aggregate all combinations of the same product and obtain a list of the same products corresponding to each category tag.

[0043] In a further embodiment, the category tag acquisition module includes:

[0044] A commodity information acquisition submodule, used to acquire commodity information of commodity objects in a commodity database, wherein the commodity information includes title text and commodity pictures;

[0045] A feature vector extraction submodule is used to extract the corresponding text feature vector and picture feature vector from the title text and the product picture respectively;

[0046] A category label generation submodule is used to classify the commodity object according to the text feature vector and / or the image feature vector to obtain a category label corresponding to the commodity object;

[0047] The comprehensive feature vector submodule is used to store the text feature vector and / or the image feature vector as the comprehensive feature vector associated with the corresponding commodity object.

[0048] In a further embodiment, the comprehensive feature vector module includes:

[0049] A text feature vector submodule is used to call a preset text feature extraction model to extract text features from the product information to obtain its text feature vector;

[0050] The image feature vector submodule is used to call a preset image feature extraction model to extract image features of the product information and obtain its image feature vector;

[0051] The comprehensive feature vector submodule is used to normalize the image feature vector into a single high-dimensional vector with the same dimension as the text feature vector, and then concatenate it with the text feature vector to obtain a comprehensive feature vector.

[0052] In a further embodiment, the similarity calculation module includes:

[0053] A comprehensive feature vector acquisition submodule is used to acquire a pre-stored comprehensive feature vector of a commodity object;

[0054] The cosine distance calculation submodule is used for similarity calculation. The cosine distance between two comprehensive feature vectors in the commodity combination composed of two commodity objects signed by the same type of target is used as the similarity between the two commodity objects.

[0055] The storage submodule is used to store the similarities between the two commodity objects in each commodity combination signed by each type of target.

[0056] In a further embodiment, the same item product screening module includes:

[0057] A threshold confirmation submodule, used to determine a preset threshold;

[0058] A threshold comparison submodule is used to obtain the similarity of the commodity combinations signed by the same type of target and compare it with the preset threshold;

[0059] The same item confirmation submodule is used to determine that the product combination is a same item combination when the corresponding similarity of the product combination exceeds the preset threshold.

[0060] In a further embodiment, the threshold confirmation submodule includes:

[0061] A data set confirmation unit is used to randomly obtain the similarity of a batch of product combinations and obtain the manual annotation information of whether they belong to the same or different items, so as to determine the corresponding two data sets;

[0062] A first threshold unit is used to perform normal distribution statistics on the two data sets according to the similarity, and determine the similarity corresponding to the intersection of two corresponding normal distribution curves as a first threshold;

[0063] A second threshold unit, used to obtain a predetermined second threshold;

[0064] The threshold comparison unit is used to compare the first threshold and the second threshold, and the highest one is used as the preset threshold.

[0065] In a further embodiment, the module for obtaining a list of identical products includes:

[0066] The similar product family submodule is used to select multiple similar product families for each category tag by applying the maximum greedy matching algorithm. Each similar product family is a set of multiple combinations of the same product corresponding to the same product object and includes multiple product objects. Different similar product families contain different product objects.

[0067] The representative product confirmation submodule is used to determine the product with the largest inventory in the same similar product family as the representative product, and the rest as the represented products;

[0068] The list generation submodule is used to convert each similar product family into a list of the same products.

[0069] 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 method for aggregating the same products described in the present application.

[0070] A computer-readable storage medium is provided to meet another purpose of the present application, which stores a computer program implemented according to the method for aggregating products of the same item in the form of computer-readable instructions. When the computer program is called and executed by a computer, the steps included in the method are executed.

[0071] A computer program product provided to meet another purpose of the present application includes a computer program / instructions, which, when executed by a processor, implement the steps of the method for aggregating the same products described in any embodiment of the present application.

[0072] Compared with the prior art, the advantages of this application are as follows:

[0073] This application combines the two dimensions of product images and title text, respectively extracts the semantic features of the product objects corresponding to the two dimensions to construct a comprehensive feature vector, calculates the similarity of the product combination composed of two product objects and compares it with a preset threshold to obtain a combination of the same product, sorts the product objects in the combination of the same product according to the maximum inventory of the product objects, and uses a greedy algorithm to select the product objects corresponding to the representative product and the represented product objects to construct a list of the same product according to the above sorting. The representative product has the largest inventory among the maximum inventory corresponding to the same product and ranks first in the list. Due to the combination of the deep semantic feature information of the two dimensions, the deep semantic feature information corresponding to the obtained product information is richer and the similarity calculated based on this is more accurate, and the products of the same product are accurately aggregated. Sorting according to the maximum inventory in the same list facilitates buyers who need to purchase a large number of products to make quick decisions and directly purchase representative products or the first few products on the list.

[0074] This application is based on the two dimensions of product images and title texts corresponding to the product objects, and has a high degree of comprehensive information. It can accurately aggregate products of the same style, and establish a list of products of the same style for buyers to refer to, thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] 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:

[0076] Figure 1 This is a flow chart of a typical embodiment of the method for aggregating products of the same item of the present application;

[0077] Figure 2 A schematic diagram of a process for generating a category label for a commodity object in an embodiment of the present application;

[0078] Figure 3 A schematic diagram of a process for constructing a comprehensive feature vector in an embodiment of the present application;

[0079] Figure 4 A schematic diagram of a process for calculating similarity based on a comprehensive feature vector in an embodiment of the present application;

[0080] Figure 5 A schematic diagram of a process for screening a combination of products of the same item in an embodiment of the present application;

[0081] Figure 6 A schematic diagram of a process for constructing a preset threshold in an embodiment of the present application;

[0082] Figure 7 A flowchart of the process of constructing a list of identical products in an embodiment of the present application;

[0083] Figure 8 It is a principle block diagram of a typical embodiment of the same commodity aggregation device of the present application;

[0084] Fig. 9 A schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[0085] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0086] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0087] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0088] It will be understood by those skilled in the art that the "client", "terminal" and "terminal device" used herein include both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capabilities, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers, tablet computers, which have single-line displays or multi-line displays or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service, personal communication system), which can combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant, personal digital assistant), which may include a radio frequency receiver, pager, Internet / intranet access, web browser, notepad, calendar and / or GPS (Global Positioning System, global positioning system) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with and / or including a radio frequency receiver. The "client", "terminal" and "terminal device" used herein may be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable for and / or configured to run locally, and / or in a distributed form, at any other location on the earth and / or in space. The "client", "terminal" and "terminal device" used herein may also be a communication terminal, an Internet terminal, a music / video playing terminal, for example, a PDA, a MID (Mobile Internet Device) and / or a mobile phone with a music / video playing function, or a smart TV, a set-top box and other devices.

[0089] The hardware referred to by the names such as "server", "client", and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit calls the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.

[0090] It should be pointed out that the concept of "server" referred to in this application can also be extended to the case of server clusters. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or integrated into a physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility, and should not use it to restrict the implementation of the network deployment method of this application.

[0091] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for access.

[0092] The neural network models cited or may be cited in this application, unless expressly specified, can be deployed on a remote server and remotely called on the client, or can be deployed on a client with sufficient device capabilities and directly called. Those skilled in the art should understand that as long as the device operating resources are qualified, they can be used as model training devices and model operating devices corresponding to the neural network model. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning, so as to reduce the requirements for the client hardware operating resources and avoid excessive occupation of the client hardware operating resources.

[0093] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as it is suitable for being called by the technical solution of this application.

[0094] Those skilled in the art should be aware that, although the various methods of the present application are described based on the same concept and thus present commonality to each other, unless otherwise specified, these methods can be independently executed. Similarly, for each embodiment disclosed in the present application, they are all proposed based on the same inventive concept, therefore, concepts with the same expression, and concepts that are appropriately changed for convenience despite different expressions, should be understood as equivalent.

[0095] Unless the mutually exclusive relationship between the embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct a new embodiment, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0096] A method for aggregating identical products of the present application can be programmed as a computer program product and deployed in a terminal device and / or server for execution. Thus, a client can access a user interface opened after the computer program product is executed in the form of a web program or application program to achieve human-computer interaction.

[0097] See also Figure 1 The present application discloses a method for aggregating products of the same type, in a typical embodiment, including steps S1100 to S1400, which are as follows:

[0098] Step S1100, obtaining a category label of a commodity object in a commodity database, wherein the category label is obtained by calling a pre-trained commodity classification model to classify commodity information of the commodity object;

[0099] One application scenario adapted by this application is that a merchant instance in an e-commerce platform hopes that the e-commerce platform will provide a list of identical products for its designated product object. Since the designated product object and the list of identical products belong to the same category tag, the e-commerce platform first needs to obtain the corresponding category tag in the product database for the product object designated by the merchant instance, so as to subsequently obtain the list of identical products under the same category tag as the designated product object. Specifically, the e-commerce platform can respond to the relevant request of the merchant instance and obtain the category tag of the product object designated by the merchant instance for the list of identical products. After confirming the category tag of the product object, the e-commerce platform can obtain the corresponding list of identical products under the same category tag based on the category tag of the product object.

[0100] The same product list is a product list constructed according to a certain order based on multiple product objects with relatively high similarity signed with the same type of target product object.

[0101] The product information is the image and text information corresponding to the product object stored in the product database, wherein the text information includes the title information of the product object for prominent display, the content information for introducing the product details, the attribute information for describing the product characteristics, etc., and the image information includes the main image of the product object for prominent display and the detail image for introducing the product details.

[0102] The category tag is the classification path of the category tree of the commodity classification framework of the e-commerce platform. The category tree is expanded layer by layer from multiple levels of categories. Each level includes multiple category tags. The parent category tag includes multiple child category tags, thus forming a "category tree". The category tree usually includes three or four levels, and generally does not exceed five levels. Each level in the category tree corresponds to multiple category tags. For a commodity object, each category tag in the multi-level classification structure constitutes its classification path, and each category tag in the classification path has a hierarchical affiliation.

[0103] Step S1200, obtaining a comprehensive feature vector corresponding to the product information of the product object, and calculating the similarity between product combinations consisting of two product objects signed by the same type of target, wherein the comprehensive feature vector is a deep semantic feature representation extracted by the product classification model;

[0104] The comprehensive feature vector is an image feature vector and a text feature vector generated by calling the corresponding title text and image information semantic features in the product information extracted by the product classification model. The comprehensive feature vector is formed by splicing the image feature vector and the text feature vector, and the comprehensive feature vector is matrixed and associated with the corresponding product objects and stored in the product database, which is used for subsequent steps to call the similarity calculation model to calculate the similarity of each pair of product objects.

[0105] The similarity is based on the category label of the specified commodity object, obtains the commodity object corresponding to the category label in the commodity database and obtains its corresponding comprehensive feature vector, calls the Faiss framework to build an index to represent the similarity between commodity objects, and based on the corresponding comprehensive feature vectors between the commodity objects, calculates the similarity between the commodity combination consisting of the obtained commodity object and the specified commodity object under the same category label, and constructs a similarity list based on the commodity combination and the similarity calculated thereof and stores it in the database for subsequent retrieval and acquisition of the corresponding commodity combination in the similarity list by using the similarity as the index value.

[0106] Step S1300, filtering out the product combinations whose similarity in each category exceeds a preset threshold as the same product combinations;

[0107] The preset threshold is a threshold obtained by comparing the similarity of each product combination in two mutually exclusive same-item data sets and non-same-item data sets with a manually set threshold. When the threshold obtained by statistics is larger than the manually set threshold, the preset threshold is determined to be the threshold obtained by statistics; otherwise, the preset threshold is determined to be the manually set threshold.

[0108] The similarity of the commodity combinations consisting of two commodity objects signed in each category is obtained, and it is determined whether the similarity of the commodity combinations exceeds a preset threshold, and the commodity combinations exceeding the preset threshold are screened out as the same commodity combinations.

[0109] Step S1400: aggregate all combinations of the same product to obtain a list of the same product corresponding to each category tag.

[0110] Get the same product combination under the same category tag, aggregate and build a list of the same products, and associate it with the current category tag for storage. According to the above logic, get the list of the same products corresponding to each category tag.

[0111] This embodiment integrates the feature information corresponding to the title text and the product image of the product object, realizes multimodal information fusion, and on this basis, performs similarity calculation processing according to the fused comprehensive feature information, thereby realizing the similarity calculation between the product objects in the e-commerce platform. Since its similarity calculation basis is not limited to the image information, but relies on the comprehensive feature information spliced ​​by the image feature information and the text feature information, the information volume is rich and the information granularity is finer, so the similarity value with high accuracy can be obtained more quickly, which is conducive to improving the efficiency of aggregation of the same items of products, finding similar products, and automatic push of similar products by the system in the e-commerce platform.

[0112] See also Figure 2 In an extended embodiment, the commodity classification model performs the following steps S1510 to S1540 for generating a category label for the commodity object:

[0113] Step S1510, obtaining product information of a product object in a product database, wherein the product information includes a title text and a product image;

[0114] The server of the e-commerce platform obtains the product information of the product object to generate the category label of the product object. The product information of the product object may be obtained from the product database of the merchant instance itself in one embodiment, or from the product database of the e-commerce platform itself in another embodiment. Regardless of the method, as long as the server used for this application can obtain the product information corresponding to the product object, the category label of the product object can be generated in the subsequent steps according to the product information.

[0115] The product information includes title text and product images. The title text mainly adopts one or more descriptive information representing the product attributes in the product object, and the product image mainly adopts the main image containing the product appearance of the product object. Generally speaking, the title text and product image can be obtained according to the preset quantity and content requirements. For example, the title text can be specified as the attribute information containing all attribute items in the title information of the product object, and the product image can only select the main product image mainly used to represent the product appearance. The specific implementation method can be flexibly handled by those skilled in the art on this basis.

[0116] Step S1520, extracting the corresponding text feature vector and picture feature vector from the title text and the product picture respectively;

[0117] In order to extract the corresponding feature information from the title text and product image in the product object, feature representation can be achieved with the help of a variety of pre-trained neural network models, so that the feature information can contain the corresponding summary text and deep semantic information of the product image. For example, a model such as Bert can be used to implement the word segmentation of the title text and the extraction of feature information therein, and a model such as Resnet can be used to implement the extraction of feature information in the product image. Not limited to the examples here, in theory, any network model suitable for extracting deep semantic information from the product image and title text can be used in this application to implement the corresponding feature extraction processing, and finally, a text feature vector corresponding to the title text and a picture feature vector corresponding to the product image are obtained.

[0118] Step S1530, classifying the commodity object according to the text feature vector and / or the image feature vector to obtain a category label corresponding to the commodity object;

[0119] In order to obtain the corresponding category label of the commodity object, the text feature vector and the image feature vector can be selected and discarded, and the corresponding preset classification neural network model can be called to obtain classification result information. The classification result information includes the similarity information of each level and each classification label in the category tree of the commodity object mapped to the e-commerce platform, and the similarity of the classification labels at each level is optimized according to their affiliation, and at least one classification path of the commodity object is determined to finally obtain the corresponding category label of the commodity object.

[0120] In one embodiment, the text description corresponding to the product SKU attribute in the title text of the product object is generally comprehensive and accurately expressed, so the text feature vector corresponding to the title text can be used alone. Accordingly, the preset text feature model is called to obtain the text feature vector. The text feature vector has deep semantic features that characterize the product attributes. Therefore, the text feature vector can be classified with the help of a multi-classifier such as Softmax to obtain the corresponding category label of the product object.

[0121] In another embodiment, the product image of the product object intuitively reflects the product appearance, so the image feature vector corresponding to the product image can be applied separately. Accordingly, the preset image feature model is called to obtain the image feature vector. The image feature vector has deep semantic features that represent the attributes of the product. Therefore, the image feature vector can be classified with the help of a classifier such as Softmax to obtain the corresponding category label of the product object.

[0122] In yet another embodiment, considering that relying solely on the title text or product image of the product object may not be able to fully represent the complete deep semantic information corresponding to the product object, the text feature model and the image feature model of the first two embodiments can be used to extract the text feature vector and the image feature vector corresponding to the title text and the product image of the product object respectively, and the two are simply concatenated into a comprehensive feature vector. Then, the comprehensive feature vector is classified by a multi-classifier such as Softmax to obtain the corresponding category label of the product object.

[0123] After confirming the category tag corresponding to the commodity object, the category tag is associated with the corresponding commodity object and stored in a commodity database, so as to obtain the corresponding commodity object according to the category tag.

[0124] Step S1540: storing the text feature vector and / or the image feature vector as the comprehensive feature vector associated with the corresponding commodity object.

[0125] According to the typical embodiments of the present application, it can be implemented flexibly to form a plurality of alternative embodiments, mainly in that the content of the obtained comprehensive feature vector is different due to the different selection of the text and image feature vectors. Specifically, in these alternative embodiments, the text feature vector and the image feature vector can be spliced ​​to form the comprehensive feature vector, or the text feature vector or the image feature vector can be used alone to form the comprehensive feature vector. That is to say, the present application does not strictly require the use of the image feature vector and the text feature vector to form the comprehensive feature vector in the present application, but the technical personnel in this field can flexibly choose the image feature vector and the text feature vector, and it should be possible to achieve the purpose of aggregating the same products of the present application. Of course, correspondingly, in accordance with the various alternative embodiments here, corresponding changes are also made in the stage of using the similarity calculation model to calculate the similarity of the comprehensive feature vector, that is, the corresponding model is called according to the corresponding comprehensive feature vector to perform the corresponding similarity calculation.

[0126] The comprehensive feature vector is associated with the corresponding commodity object and stored in the commodity database, so as to quickly retrieve the corresponding comprehensive feature vector through the commodity object for calculating the similarity between commodity objects and / or commodity classification.

[0127] See also Figure 3 In a further embodiment, step S1540, storing the text feature vector and / or the image feature vector as the comprehensive feature vector associated with the corresponding commodity object, comprises the following steps:

[0128] Step S1541, calling a preset text feature extraction model to extract text features from the product information to obtain its text feature vector;

[0129] In order to obtain the text feature vector of the product information, the title text information in the product information can be preprocessed, and the text information representing the product attributes in the title text information can be filtered according to the commonly used e-commerce product attribute text dictionary. Then, the Bert text model is called to extract text features from the text information and embed it to obtain the text feature vector corresponding to the title text information.

[0130] Step S1542, calling a preset image feature extraction model to extract image features of the product information and obtain its image feature vector;

[0131] In order to obtain the image feature vector in the product information, the product image in the product information can be preprocessed, or common image preprocessing operations can be used, such as using a Yolo network model to locate the product and remove image noise, etc., and then calling the Resnet image feature model to extract the semantic features that intuitively reflect the product appearance in the product image to obtain the image feature vector corresponding to the image in the product information.

[0132] Step S1543, normalizing the image feature vector into a single high-dimensional vector having the same dimension as the text feature vector, and concatenating the vector with the text feature vector to obtain a comprehensive feature vector.

[0133] The text feature vector extracted by the Bert model is usually a high-dimensional row vector, while the image feature vector is a matrix of multiple rows and columns. Therefore, in order to facilitate the feature splicing of the two, the image feature vector can be normalized into a row vector, the dimension of which is the same as that of the text feature vector. Then, the elements of each row and column of the image feature vector are filled into the normalized row vector in the order of row and column relationship to complete the normalized construction. At this point, the scale of the image feature vector and the text feature vector is unified, and then the two are simply spliced ​​to construct a comprehensive feature vector. The comprehensive feature vector obviously realizes the fusion of the feature information of the text feature vector and the image feature vector.

[0134] This embodiment further uses the text feature model and the image feature model to extract the deep semantic features corresponding to the title text and the product image for the product object. Finally, the deep semantic features of the text feature vector and the image feature vector are realized by a simple splicing method. It can be seen that the obtained comprehensive feature vector actually contains information about all aspects of the product object to a great extent, accurately and richly represents the product object, and provides a solid foundation for the accuracy of similarity calculation between product objects.

[0135] See also Figure 4In a further embodiment, step S1200, obtaining a comprehensive feature vector corresponding to the commodity information of a commodity object, and calculating the similarity between commodity combinations consisting of two commodity objects signed by the same type of target, comprises the following steps:

[0136] Step S1210, obtaining a pre-stored comprehensive feature vector of the commodity object;

[0137] Obtain a category tag pre-stored in a commodity database for associated commodity objects, and obtain a plurality of commodity objects whose category tags corresponding to the commodity objects in the commodity database are the category tags based on the associated storage relationship between the category tags and the commodity objects. Then, obtain a comprehensive feature vector corresponding to the commodity object based on the associated storage relationship between the comprehensive feature vector and the commodity objects. Finally, obtain a plurality of commodity objects based on the same category tag and their corresponding comprehensive feature vectors.

[0138] Step S1220, calculating the cosine distance between two comprehensive feature vectors in a commodity combination formed by two commodity objects signed by the same type of target as the similarity between the two commodity objects;

[0139] According to the comprehensive feature vectors corresponding to the commodity objects signed by the same type of targets, the indexes between the comprehensive feature vectors are constructed through the interface provided by the faiss framework. The cosine similarity algorithm is further applied to calculate the value of the index through the similarity calculation interface provided by it, so as to quickly calculate the cosine distance between the commodity combinations formed by each pair of commodity objects. Similarly, the cosine distance of the commodity combinations signed by each type of targets is calculated, and the obtained cosine distance is used as the similarity between the commodity combinations formed by each pair of commodity objects.

[0140] The cosine distance is calculated by normalizing the value obtained by dividing the dot product of the comprehensive feature vectors of the two commodity objects by the product of the module lengths corresponding to the comprehensive feature vectors. The final cosine distance value range is 0-1, and the limits are 0 and 1 respectively. When the cosine distance value is 1, the angle between the overlapping comprehensive vectors is 0, indicating that the similarity between the two commodity objects is 1. When the cosine distance value is 0, the angle between the separating comprehensive vectors is 180 degrees, indicating that the similarity between the two commodities is 0. The larger the cosine distance value, the higher the similarity between the two. Conversely, the smaller the value, the greater the difference between the two and the lower the similarity.

[0141] Step S1230, storing the similarities between the two commodity objects in each commodity combination signed by each type of target.

[0142] The similarity corresponding to the product combination composed of each pair of product objects is associated with the category tag to which they belong for storage, and similarity between the pair of product objects in each product combination under each category tag is stored in the same way, so as to facilitate obtaining the similarity of the product combination under the category tag by searching the category tag.

[0143] In this embodiment, the Faiss framework is used to realize the rapid calculation of the similarity of the commodity combination consisting of two commodity objects, and it can reduce the implementation cost of the technical solution of this application and improve the operation efficiency.

[0144] See also Figure 5 In a further embodiment, step S1300, screening out the product combinations whose similarity in each category target signing exceeds a preset threshold as the same product combination, comprises the following steps:

[0145] Step S1310, determining a preset threshold;

[0146] A preset threshold is obtained, where the preset threshold is a determination condition for the similarity of a commodity combination consisting of two commodity objects, and is used as a boundary in the same dimension for determining the similarity of each commodity object.

[0147] Step S1320, obtaining the similarity of the commodity combinations signed by the same category of targets, and comparing it with the preset threshold;

[0148] According to the associated storage relationship between the category tag and the product combination belonging to the same category tag, the product combination of the same category tag in the product database is obtained, and the product combination and its corresponding similarity are associated and stored, and finally the similarity corresponding to each product combination under the same category tag is obtained, and compared with the preset threshold. Similarly, the similarity corresponding to each product combination under the category tag is obtained and compared with the preset threshold.

[0149] Step S1330: When the corresponding similarity of the product combination exceeds the preset threshold, the product combination is determined to be a combination of the same product.

[0150] A comparison of the same dimension is performed based on the corresponding similarity of the product combinations signed by the same category of targets and the preset threshold. When the similarity of the former exceeds the latter, the product combinations signed by the same category of targets are determined to be the same product combinations. Similarly, the same product combinations signed by each category of targets are obtained accordingly.

[0151] In this embodiment, the commodity combinations consisting of two commodity objects are determined to be the same by judging their corresponding similarities and a preset threshold. This is more scientific and objective than simply sorting the similarities of the commodity combinations from high to low and selecting the top dozen to be the same.

[0152] See also Figure 6 In a further embodiment, step S1310, determining the preset threshold, comprises the following steps:

[0153] Step S1311, randomly obtaining the similarity of a batch of product combinations to obtain the manual labeling information of whether they belong to the same item or different items, thereby determining two corresponding data sets;

[0154] A batch of product combinations are randomly obtained, and the data sets of product combinations of the same item and product combinations of different items are obtained by manually marking whether they are the same item. The comprehensive feature vector corresponding to the product objects in the product combination is obtained, and the cosine distance between the product objects is quickly calculated as the similarity of the mutually constituted product combinations through the faiss module based on the comprehensive feature vector. Similarly, the similarity of the corresponding product combinations in the data sets of the same item combination and the data sets of the different item combination is calculated.

[0155] Step S1312, performing normal distribution statistics on the two data sets according to the similarity, and determining the similarity corresponding to the intersection of two corresponding normal distribution curves as a first threshold;

[0156] Data statistics are performed based on two dimensions: the similarity of product combinations in the data set of the same product combination and the data set of the non-same product combination and the number of product combinations corresponding to the similarity. The similarity of product combinations is used as the X-axis, and the number of product combinations corresponding to the similarity is used as the Y-axis. According to statistical knowledge, it can be obtained that the curve formed by the data in the two data sets conforms to the normal distribution. The similarity corresponding to the intersection of the curves is calculated based on this as the first threshold.

[0157] Step S1313, obtaining a predetermined second threshold;

[0158] In order to avoid the final preset threshold being too small, the second threshold is manually set to 0.8. The manually set second threshold is used as the lowest lower limit threshold of the preset threshold (the specific preset threshold needs to be implemented according to the following steps in this embodiment). The similarity corresponding to the product combination consisting of two product objects exceeds the second threshold, i.e., 0.8, to be determined as the same product combination. The second threshold can be flexibly adjusted by technicians in this field according to business needs. The second threshold set manually in advance is used to avoid the final preset threshold being too small to be used as the basis for judging whether the product objects in the product combination are the same.

[0159] Step S1314, comparing the first threshold and the second threshold, and taking the highest one as the preset threshold.

[0160] Compare the first threshold and the second threshold. When the first threshold is greater than the second threshold, the first threshold is confirmed to be the preset threshold. When the second threshold is greater than the first threshold, the second threshold is confirmed to be the preset threshold. In order to make the preset threshold more suitable for judging whether the commodity objects in the commodity combination are the same, when the second threshold is greater than the first threshold, the manually preset second threshold can be optionally abandoned, and other batches of commodity combinations can be replaced to obtain the first threshold multiple times in the same manner as above, until the first threshold is greater than the second threshold to obtain the first threshold as the preset threshold. The technicians in this technical field can make flexible adjustments accordingly.

[0161] In this embodiment, the implementation of the preset threshold is defined. The preset threshold is obtained by statistically analyzing the similarities corresponding to commodity combinations consisting of two commodity objects, and the second threshold is manually set to ensure the lower limit of the preset threshold. The maximum of the first threshold and the second threshold is selected as the preset threshold, which ensures that the preset threshold is not too small to affect the accuracy of the judgment result of whether the commodities are the same according to the preset threshold, and ensures that the obtained preset threshold is high enough and suitable for the application scenario of comparing the similarities corresponding to the preset threshold and most of the commodity combinations, so that the commodity combinations with high similarity are judged as the same commodity combinations.

[0162] See also Figure 7 In a further embodiment, step S1400, aggregating all combinations of the same product to obtain a list of the same product corresponding to each category tag, includes the following steps:

[0163] Step S1410: for each category tag, a maximum greedy matching algorithm is applied to select multiple similar product families, each similar product family is a set consisting of multiple combinations of the same product corresponding to the same product object and includes multiple product objects, and different similar product families contain different product objects;

[0164] Get the same product combination corresponding to each category tag to build a same product set, sort the same product set according to the product inventory, select a same product set and apply the maximum greedy matching algorithm to select the representative product and the represented product therein, remove the currently confirmed representative product and the represented product, repeat the above logic to select the representative product and the represented product for another same product set, until the unique representative product and the represented product in each same product set are obtained and the representative product and the represented product in each same product set are different, so as to build a similar product family.

[0165] Each similar product family is a collection of multiple identical product objects corresponding to the same product objects, each of which constitutes a product combination of the same type. The collection includes multiple product objects, namely, unique representative products and represented products. The product objects in different similar product families are different from each other.

[0166] Step S1420, determining the product with the largest inventory in the same similar product family as the representative product, and the rest as the represented products;

[0167] The maximum inventory quantity of the goods corresponding to the goods objects in each similar goods family is obtained, and the goods objects in the same similar goods family are sorted in descending order according to their corresponding maximum inventory quantities, and the goods with the largest inventory quantity are confirmed as representative goods, and the rest are confirmed as represented goods.

[0168] Step S1430, converting each similar product family into a list of products of the same item.

[0169] The similar product family is sorted by the largest inventory, and a corresponding list of the same product is constructed according to the sequence of the representative product and the represented product obtained by sorting.

[0170] Similarly, the same logic is used to process each similar product family to obtain a list of the same products corresponding to each category tag.

[0171] In one embodiment, to facilitate human-computer interaction, the same product list is constructed to highlight the representative product therein, and other represented products are expanded and displayed when the representative product is touched by the user, thereby obtaining an aggregate display effect on the interface. On this basis, the same product list is pushed to the graphical user interface of the user terminal device for user access.

[0172] In this embodiment, the same products are aggregated to construct a list of the same products, which simplifies the space occupied by the same products in the graphical interface, is particularly convenient for information output in the limited interface space of mobile devices, and is convenient for users to refer to the representative products and represented products in the list of the same products, and quickly understand the products with large inventory among the same products, thereby increasing the purchase rate of products with relatively low prices or price advantages and usually large purchase volumes, and improving user experience and increasing user stickiness.

[0173] Furthermore, by functionalizing each step in the method disclosed in the above embodiments, a device for aggregating the same commodity of the present application can be constructed. Figure 8In a typical embodiment, the device includes: a category tag acquisition module 1100: used to obtain the category tag of the commodity object in the commodity database, and the category tag is obtained by calling a pre-trained commodity classification model to classify the commodity information of the commodity object; a similarity calculation module 1200: used to obtain the comprehensive feature vector corresponding to the commodity information of the commodity object, and calculate the similarity between the commodity combinations consisting of two commodity objects under the same category, and the comprehensive feature vector is a deep semantic feature representation extracted by the commodity classification model; a same-item product screening module 1300: used to screen out the commodity combinations whose similarity exceeds a preset threshold under each category as the same-item product combinations; a list generation module 1400: used to aggregate all the same-item product combinations to obtain a same-item product list corresponding to each category tag.

[0174] In a further embodiment, the category tag acquisition module 1100 includes the following submodules:

[0175] Product information acquisition submodule: used to obtain product information of product objects in the product database, the product information including title text and product pictures; feature vector extraction submodule: used to extract the corresponding text feature vectors and picture feature vectors of the title text and product pictures respectively; category label generation submodule: used to classify the product objects according to the text feature vectors and / or picture feature vectors to obtain the corresponding category labels of the product objects; comprehensive feature vector submodule: used to store the text feature vectors and / or picture feature vectors as the comprehensive feature vector associated with the corresponding product object.

[0176] In a further embodiment, the comprehensive feature vector module includes:

[0177] The text feature vector submodule is used to call a preset text feature extraction model to perform text feature extraction on the product information to obtain its text feature vector; the picture feature vector submodule is used to call a preset picture feature extraction model to perform image feature extraction on the product information to obtain its picture feature vector; the comprehensive feature vector submodule is used to normalize the picture feature vector into a single high-dimensional vector with the same dimension as the text feature vector, and then splice it with the text feature vector to obtain a comprehensive feature vector.

[0178] In a further embodiment, the similarity calculation module 1200 includes the following submodules:

[0179] Comprehensive feature vector acquisition submodule: used to obtain the pre-stored comprehensive feature vectors of commodity objects; cosine distance calculation submodule: used to calculate the similarity and calculate the cosine distance between two comprehensive feature vectors in the commodity combination composed of two commodity objects signed by the same type of targets as the similarity between the two commodity objects; storage submodule: used to store the similarity between the two commodity objects in each commodity combination signed by each type of targets.

[0180] In a further embodiment, the same product screening module 1300 includes the following submodules:

[0181] Threshold confirmation submodule: used to determine the preset threshold; threshold comparison submodule: used to obtain the similarity corresponding to the product combination signed by the same type of target, and compare it with the preset threshold; same-item confirmation submodule: used to determine that the product combination is the same-item product combination when the corresponding similarity of the product combination exceeds the preset threshold.

[0182] In a further embodiment, the threshold confirmation submodule includes:

[0183] A data set confirmation unit is used to randomly obtain the similarity of a batch of commodity combinations and obtain the artificial annotation information of whether they belong to the same or different items, so as to determine the corresponding two data sets; a first threshold unit is used to perform normal distribution statistics on the two data sets according to the similarity, and determine the similarity corresponding to the intersection of two corresponding normal distribution curves as the first threshold; a second threshold unit is used to obtain a pre-given second threshold; a threshold comparison unit compares the first threshold and the second threshold, and the highest one is the preset threshold.

[0184] In a further embodiment, the module for obtaining a list of identical products includes:

[0185] The similar product family submodule is used to select multiple similar product families for each category tag by applying the maximum greedy matching algorithm. Each similar product family is a collection of multiple identical product combinations corresponding to the same product object and includes multiple product objects. Different similar product families contain different product objects. The representative product confirmation submodule is used to determine that the product with the largest inventory in the same similar product family is the representative product, and the rest are represented products. The list generation submodule is used to convert each similar product family into a list of identical products.

[0186] In order to solve the above technical problems, the embodiment of the present application also provides a computer device for running a computer program implemented according to the data updating method. Fig. 9 , Fig. 9 This is a basic structural block diagram of the computer device in this embodiment.

[0187] like Fig. 9As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected via a system bus. Among them, the non-volatile 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 data update method or a data scheduling 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 a data update method or a data scheduling method. 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.

[0188] In this embodiment, the processor is used to execute the specific functions of the category label acquisition module 1100, the similarity calculation module 1200, the same product screening module 1300, and the list generation module 1400 in the supply source recommendation device of the present invention, and the memory stores the program code and various data required to execute the above modules. 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 data update device or data scheduling device, and the server can call the program code and data of the server to execute the functions of all submodules.

[0189] The present application also provides a non-volatile storage medium, wherein the data updating method or data scheduling method is written into a computer program and stored in the storage medium in the form of computer-readable instructions. When the computer-readable instructions are executed by one or more processors, it means that the program is run in the computer, thereby enabling one or more processors to execute the steps of the data updating method or data scheduling method of any of the above-mentioned embodiments.

[0190] The present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the method for aggregating identical products described in any one of the embodiments of the present application.

[0191] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments 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 non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0192] In summary, this application can help merchants quickly and accurately aggregate products of the same price, reduce the graphical user interface space required for product display, and improve the user experience of users purchasing products on e-commerce platforms.

[0193] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0194] Those skilled in the art will appreciate that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

[0195] 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 method for aggregating products of the same type, characterized in that: The steps include: Obtaining a category label of a commodity object in a commodity database, wherein the category label is obtained by calling a pre-trained commodity classification model to classify commodity information of the commodity object; Obtaining a comprehensive feature vector corresponding to the product information of the product object, and calculating the similarity between product combinations consisting of two product objects signed by the same type of target, wherein the comprehensive feature vector is a deep semantic feature representation extracted by the product classification model; Screening out the product combinations whose similarity under each category target signing exceeds a preset threshold as the same product combination, including: determining the preset threshold; The similarity of a batch of product combinations is randomly obtained to obtain the manual labeling information of whether they are the same or different items, so as to determine the corresponding two data sets; Performing normal distribution statistics on the two data sets according to the similarity, and determining the similarity corresponding to the intersection of two corresponding normal distribution curves as a first threshold; Acquire a predetermined second threshold value; Comparing the first threshold value and the second threshold value, and taking the highest one as the preset threshold value; Obtain the similarity of the product combinations signed by the same category of targets, and compare it with the preset threshold; When the corresponding similarity of the product combination exceeds the preset threshold, the product combination is determined to be a combination of the same product; Aggregate all combinations of the same product to obtain a list of the same products corresponding to each category tag; The list of identical products is constructed so that a representative product is highlighted, and other represented products are expanded and displayed when the representative product is touched by the user, and the list of identical products is pushed to a graphical user interface of a user terminal device.

2. The method for aggregating products of the same item according to claim 1, characterized in that: The commodity classification model performs the following steps to generate a category label for the commodity object: Obtain product information of a product object in a product database, wherein the product information includes a title text and a product image; Extracting corresponding text feature vectors and picture feature vectors from the title text and the product picture respectively; Classify the commodity object according to the text feature vector and / or the image feature vector to obtain a category label corresponding to the commodity object; The text feature vector and / or the image feature vector are stored as the comprehensive feature vector associated with the corresponding commodity object.

3. The method for aggregating products of the same item according to claim 2, characterized in that: Storing the text feature vector and / or the image feature vector as the comprehensive feature vector associated with the corresponding commodity object comprises the following steps: Calling a preset text feature extraction model to extract text features from the product information to obtain its text feature vector; Calling a preset image feature extraction model to extract image features of the product information and obtain its image feature vector; After the image feature vector is normalized into a single high-dimensional vector with the same dimension as the text feature vector, it is concatenated with the text feature vector to obtain a comprehensive feature vector.

4. The method for aggregating products of the same item according to claim 1, characterized in that: Obtaining the comprehensive feature vector corresponding to the product information of the product object and calculating the similarity between the product combinations consisting of two product objects signed by the same type of target includes the following steps: Obtain the pre-stored comprehensive feature vector of the commodity object; Calculate the cosine distance between two comprehensive feature vectors in the product combination composed of two product objects signed by the same type of target as the similarity between the two product objects; The similarities between each pair of commodity objects in each commodity combination signed by each type of target are stored.

5. The method for aggregating products of the same type according to any one of claims 1 to 4, characterized in that: Aggregate all the same product combinations to obtain a list of the same products corresponding to each category tag, including the following steps: For each category tag, a maximum greedy matching algorithm is applied to select multiple similar product families. Each similar product family is a set of multiple combinations of the same product corresponding to the same product object and includes multiple product objects. Different similar product families contain different product objects. Determine the product with the largest inventory in the same similar product family as the representative product, and the rest as the represented products; Convert each similar product family into a list of identical products.

6. 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 5.

7. 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 5 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.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

Images