Commodity information matching method and device, equipment, medium and product

Abstract

This application relates to a method, apparatus, device, medium, and product for matching product information. The method includes: acquiring product information of a product entity to be detected, the product information containing multiple descriptive texts, and extracting its knowledge subgraph; retrieving multiple product entities from the knowledge graph whose knowledge subgraph matches the knowledge subgraph of the product entity to be detected, forming a candidate product set; acquiring a sample set corresponding to each product entity in the candidate product set; and performing semantic matching between the product information of the product entity to be detected and the sample set of product entities in the candidate product set, thereby determining the product entities that match the product entity to be detected. The technical solution of this application is applicable to product infringement detection scenarios. When it is necessary to determine whether a product entity to be detected infringes on certain product entities, by further refining the search based on completeness, it can not only avoid missed detections through the role of attribute overlap, but also accurately identify infringing products.

Description

Technical Field

[0001] This application relates to the field of e-commerce information technology, and in particular to a product information matching method and its corresponding apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the increasing maturity of internet technology, the e-commerce industry has also developed rapidly, with a large number of merchants choosing to sell their goods through e-commerce channels. The variety of goods is vast, and their quality varies greatly; many merchants sell infringing products. The sale of infringing goods seriously damages the legitimate interests of the original brand owners. Therefore, detecting infringing products from a massive amount of inventory is an effective means of protecting the legitimate interests of brand owners.

[0003] Although e-commerce platforms offer numerous methods for detecting counterfeit goods sold by merchants, various circumvention techniques employed by sellers still prevent complete detection of infringing products. A common existing method involves maintaining a brand terminology database and using it for fuzzy matching of product listings. However, this method cannot detect products where the brand name is absent from the text, leading to numerous false negatives. For example, some sellers intentionally remove the brand name from their products, leaving only the product description and model number to evade platform detection. Furthermore, many brand names are also common phrases that appear in non-branded product descriptions, causing false positives with traditional detection methods. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above-mentioned problems by providing a product information matching method and corresponding apparatus, computer equipment, computer-readable storage medium, computer program product, etc., and to meet the various purposes of this application, the following technical solutions are adopted:

[0005] In one aspect, a product information matching method is provided to suit one of the purposes of this application, comprising the following steps:

[0006] Obtain product information of the product entity to be detected. The product information contains multiple descriptive texts. Extract its knowledge subgraph. The knowledge subgraph contains the mapping relationship data between the attributes extracted from the descriptive texts and the attribute values.

[0007] A candidate product set is formed by retrieving multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected. The knowledge graph stores the knowledge subgraphs corresponding to multiple product entities in the product database.

[0008] Obtain a sample set corresponding to each product entity in the candidate product set. Each attribute in the knowledge subgraph corresponding to the sample set includes one or more descriptive texts. All descriptive texts in the sample set are organized in a uniform order.

[0009] The product information of the product entity to be detected is semantically matched one by one with the sample set of product entities in the candidate product set to determine the product entity that matches the product entity to be detected.

[0010] In some of the more detailed embodiments, obtaining the product information of the product entity to be detected includes the following steps:

[0011] In response to a product posting request from an online store, obtain the product information of the product entity to be detected corresponding to the request. The product information includes any one or more of the product title, product details, and description text corresponding to the product attribute data of the product entity.

[0012] The product information of the product entity to be detected is used to extract attributes, so as to extract the corresponding attribute values ​​from its description text according to different attributes, and form the mapping relationship data between attributes and attribute values.

[0013] Based on the preset knowledge graph structure, the mapping relationship data between the attributes and attribute values ​​of the product entity to be detected is constructed into a corresponding knowledge subgraph.

[0014] In some of the more detailed embodiments, before the step of obtaining the product information of the product entity to be detected, the following steps are included:

[0015] Construct a knowledge graph corresponding to the product database. The knowledge graph includes knowledge subgraphs corresponding to each product entity in the product database. The knowledge subgraphs contain mapping relationship data between attributes and attribute values ​​extracted from the description text of the product information of the corresponding product entity.

[0016] In some of the more detailed embodiments, a knowledge graph corresponding to the product database is constructed, including the following steps:

[0017] Create a knowledge graph to obtain product information for each product entity in the product database;

[0018] For each product entity, the product information is extracted to extract the corresponding attribute values ​​from its description text according to different attributes, thus forming the mapping relationship data between attributes and attribute values.

[0019] Based on the preset structure of the knowledge graph, the mapping relationship data between the attributes and attribute values ​​of each product entity is constructed into a knowledge subgraph of the corresponding product entity;

[0020] Obtain different description versions of product information for each product entity. Each description version contains descriptive text that matches the mapping relationship data between the various attributes and attribute values ​​of the product entity. Construct samples corresponding to each description version in a unified order, and combine all samples to form the sample set of the corresponding product entity, which is then stored in the knowledge subgraph of the corresponding product entity.

[0021] In a more detailed embodiment, a candidate product set is formed by retrieving multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected. This includes the following steps:

[0022] Obtain the mapping relationship data between attributes and attribute values ​​in the knowledge subgraph of the product entity to be detected, and use it as its attribute set;

[0023] The overlap degree of the attribute set of the product entity to be detected is calculated by matching the attribute set of each product entity in the knowledge graph with the attribute set of each product entity in the knowledge graph.

[0024] The set of product entities in the knowledge graph whose attribute overlap meets the preset conditions is selected as the candidate product set.

[0025] In a further embodiment, the product information of the product entity to be detected is semantically matched one by one with the sample set of product entities in the candidate product set to determine the product entities that match the product entity to be detected. This includes the following steps:

[0026] Calculate the first document similarity between the product information of the product entity to be detected and the sample set of product entities in the candidate product set. Use the corresponding attribute overlap of the product entity as the weight of the first document similarity to obtain a weighted similarity. Select product entities whose weighted similarity meets the preset conditions to form the first product set.

[0027] Calculate the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, and select product entities whose semantic similarity meets the preset conditions to form the second product set.

[0028] The product entities in the second product set are used as matching product entities to be detected, and are pushed to the terminal device that provides the product information of the product entities to be detected.

[0029] In a further embodiment, before calculating the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, the following steps are included:

[0030] The sample set of product entities in the knowledge graph is used as training samples to iteratively train the preset text feature extraction model until it converges.

[0031] The text feature extraction model is used to extract the sample set of each product entity in the first product set and the semantic vector of the product information of the product entity to be detected.

[0032] On the other hand, to suit one of the purposes of this application, a product information matching device is provided, comprising:

[0033] The information extraction module obtains the product information of the product entity to be detected. The product information contains multiple descriptive texts. The module extracts its knowledge subgraph, which contains the mapping relationship data between the attributes extracted from the descriptive texts and the attribute values.

[0034] The retrieval and matching module retrieves multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected, forming a candidate product set. The knowledge graph stores the knowledge subgraphs corresponding to multiple product entities in the product database.

[0035] The extraction and sorting module obtains the sample set corresponding to each product entity in the candidate product set. Each attribute in the knowledge subgraph corresponding to the sample set includes one or more descriptive texts. All descriptive texts in the sample set are organized in a uniform order.

[0036] The semantic matching module performs semantic matching on the product information of the product entity to be detected one by one with the sample set of product entities in the candidate product set, and determines the product entity that matches the product entity to be detected.

[0037] In a further embodiment, the information extraction module includes: an information extraction unit that responds to a product posting request from an online store and obtains product information of the product entity to be detected corresponding to the request; the product information includes any one or more of the product title, product details, and description text corresponding to the product attribute data of the product entity; an attribute extraction unit that extracts attributes from the product information of the product entity to be detected, so as to extract corresponding attribute values ​​from each of its description texts according to different attributes, and constructs mapping relationship data between attributes and attribute values; and a subgraph construction unit that constructs the mapping relationship data between the attributes and attribute values ​​of the product entity to be detected into a corresponding knowledge subgraph according to a preset knowledge graph structure.

[0038] In a further embodiment, the product information matching device further includes: a graph construction module, which constructs a knowledge graph corresponding to the product database. The knowledge graph includes knowledge subgraphs corresponding to each product entity in the product database. The knowledge subgraphs contain mapping relationship data between attributes and attribute values ​​extracted from the description text of the product information of the corresponding product entity.

[0039] In a further embodiment, the knowledge graph construction module includes: a knowledge graph creation unit, which creates a knowledge graph and obtains product information for each product entity in the product database; a mapping construction unit, which extracts attributes from the product information of each product entity to extract corresponding attribute values ​​from its description texts according to different attributes, thereby forming mapping relationship data between attributes and attribute values; a subgraph creation unit, which constructs the mapping relationship data between the attributes and attribute values ​​of each product entity into a knowledge subgraph of the corresponding product entity according to the preset structure of the knowledge graph; and a subgraph matching unit, which obtains different description versions of the product information of each product entity, each description version containing description text that matches the mapping relationship data between the attributes and attribute values ​​of the product entity, constructs samples corresponding to each description version in a unified order, and constructs a sample set of the corresponding product entity from all samples, storing them in the knowledge subgraph of the corresponding product entity.

[0040] In a further embodiment, the retrieval and matching module includes: a mapping and merging unit, which acquires the mapping relationship data between attributes and attribute values ​​in the knowledge subgraph of the product entity to be detected, as its attribute set; an overlap calculation unit, which performs overlap degree matching calculation between the attribute set of the product entity to be detected and the corresponding attribute sets of each product entity in the knowledge graph, and determines the attribute overlap degree between each product entity in the knowledge graph and the product entity to be detected; and a candidate filtering unit, which sets the product entities in the knowledge graph whose attribute overlap degree meets the preset conditions as a candidate product set.

[0041] In a further embodiment, the semantic matching module includes: a first filtering unit, which calculates a first document similarity between the product information of the product entity to be detected and a sample set of product entities in the candidate product set, uses the corresponding attribute overlap of the product entity as the weight of the first document similarity to obtain a weighted similarity, and filters out product entities whose weighted similarity meets a preset condition to form a first product set; a second filtering unit, which calculates a semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, and filters out product entities whose semantic similarity meets a preset condition to form a second product set; and a product push unit, which pushes the product entities in the second product set as matching product entities with the product entity to be detected to the terminal device that provides the product information of the product entity to be detected.

[0042] In a further embodiment, the semantic matching module further includes: a model training unit that uses a sample set of product entities in the knowledge graph as training samples to iteratively train a preset text feature extraction model until it converges; and a vector extraction unit that uses the text feature extraction model to extract semantic vectors of the sample set of each product entity in the first product set and the product information of the product entity to be detected.

[0043] In another aspect, a computer device provided for one of the purposes of this application includes a central processing unit and a memory, the central processing unit being used to invoke and run a computer program stored in the memory to perform the steps of the product information matching method described in this application.

[0044] In another aspect, a computer-readable storage medium is provided to suit another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the described product information matching method, which, when invoked by a computer, performs the steps included in the method.

[0045] In another aspect, a computer program product provided for another purpose of this application includes a computer program / instructions that, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0046] Compared with the prior art, this application has many advantages, including at least the following aspects:

[0047] First, based on the e-commerce detection and matching scenario, this application extracts the knowledge subgraph corresponding to the entity to be detected. It then calculates the attribute overlap between the mapping data of attributes and attribute values ​​in the knowledge subgraph of the entity to be detected and the mapping data of attributes and attribute values ​​in the knowledge subgraphs of various product entities in the knowledge graph. This allows for the selection of multiple knowledge subgraphs similar to the entity to be detected from the knowledge graph. These multiple knowledge subgraphs constitute a candidate product set. Next, the application recalls the descriptive text corresponding to the mapping data of attributes and attribute values ​​in the knowledge subgraphs of the candidate product set, and extracts the semantic vectors of the descriptive text and the semantic vectors of the product information of the entity to be detected. Finally, it performs semantic matching between the semantic vectors of the entity to be detected and the semantic vectors of the product entities in the candidate product set to select product entities that are the same as or similar to the entity to be detected. Based on this method, products that are the same as or similar to the entity to be detected can be efficiently selected from the product database.

[0048] Secondly, this application first utilizes the overlap between the attributes of the product entity to be detected and the product entities in the knowledge graph to achieve recall, thus achieving a comprehensive retrieval and obtaining a candidate product set. Then, it uses the product information of the product entity to be detected and the sample sets of each product entity in the candidate product set for semantic matching, achieving a fine ranking of each product entity in the candidate product set, thereby filtering out product entities that match the product entity to be detected. Since each product entity in the candidate product set pre-contains a corresponding sample set, and each attribute of the product entity includes one or more descriptive texts, these descriptive texts are different forms of expression for the same product entity, thus expanding the reference range of product information and providing richer reference information for semantic matching. Therefore, more accurate matching results can be obtained during semantic matching, ensuring precise retrieval.

[0049] Furthermore, the technical solution of this application is applicable to the scenario of commodity infringement detection. When it is necessary to determine whether the commodity entity to be detected infringes on certain historical commodity entities, by further refining the detection based on comprehensiveness, not only can the role played by attribute overlap be avoided, but also the positive role played by the reference information provided by the sample set can be used to overcome the problem of the lack of commodity information of the commodity entity to be detected, and thus accurately identify infringing commodities. Attached Figure Description

[0050] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0051] Figure 1 This is a flowchart illustrating a typical embodiment of the product information matching method of this application.

[0052] Figure 2 This is a flowchart illustrating the process of obtaining product information of the product entity to be detected in an embodiment of this application.

[0053] Figure 3 This is a flowchart illustrating the process of constructing a knowledge subgraph for the product entity to be detected, as described in an embodiment of this application.

[0054] Figure 4 This is a flowchart illustrating the process of constructing a knowledge graph corresponding to a product database in an embodiment of this application.

[0055] Figure 5 This is a flowchart illustrating the process of obtaining a candidate product set in an embodiment of this application.

[0056] Figure 6 This is a flowchart illustrating the process of determining a product entity that matches the product entity to be detected, as described in an embodiment of this application.

[0057] Figure 7 This is a flowchart illustrating the process of extracting semantic vectors in an embodiment of this application.

[0058] Figure 8 This is a schematic block diagram of the product information matching device of this application;

[0059] Figure 9 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0060] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0061] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated 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 say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0062] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0063] Those skilled in the art will understand that the terms "client," "terminal," and "terminal device" as used herein include both devices that receive wireless signals, devices that only possess wireless signal receiver capabilities without transmission capabilities, and devices with receiving and transmitting hardware, devices that have receiving and transmitting hardware capable of bidirectional communication over a bidirectional communication link. Such devices may include: cellular or other communication devices such as personal computers or tablets, having single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service) that can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant) that may include a radio frequency receiver, pager, internet / intranet access, web browser, notepad, calendar, and / or GPS (Global Positioning System) receiver; and conventional laptops and / or handheld computers or other devices that have and / or include radio frequency receivers. As used herein, "client," "terminal," and "terminal device" can be portable, transportable, installed in a means of transportation (air, sea, and / or land), or suitable and / or configured to operate locally and / or in a distributed manner, operating in any other location on Earth and / or in space. "Client," "terminal," and "terminal device" as used herein can also be a communication terminal, an internet access terminal, or a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or a smart TV, set-top box, etc.

[0064] The hardware referred to by the names "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 architecture, such as a central processing unit (including an arithmetic logic unit and a control unit), memory, input devices, and output devices. The computer program is stored in its memory, and the central processing unit loads the program stored in the secondary storage into the main memory to run it, execute the instructions in the program, and interact with the input and output devices to complete specific functions.

[0065] It should be noted that the concept of "server" used in this application can also be extended to the case of server clusters. Based on the network deployment principles understood by those skilled in the art, the servers should be logically divided. Physically, these servers can be independent of each other but accessible through interfaces, or they can be integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method in this application.

[0066] One or more of the technical features of this application, unless explicitly specified herein, can be deployed on a server and accessed by a client remotely calling the online service interface provided by the server, or can be directly deployed and run on a client to access the service.

[0067] Unless otherwise specified, the neural network models referenced or potentially referenced in this application may be deployed on a remote server and invoked remotely on the client, or deployed on a client with the capability to invoke directly. In some embodiments, when running on the client, the corresponding intelligence may be acquired through transfer learning in order to reduce the requirements on the client's hardware resources and avoid excessive consumption of the client's hardware resources.

[0068] Unless otherwise specified, all data involved in this application may be stored remotely on a server or on a local terminal device, as long as it is suitable for use by the technical solution of this application.

[0069] Those skilled in the art will understand that although the various methods in this application are described based on the same concept and thus present commonality among them, they can be performed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept; therefore, concepts expressed in the same way, as well as concepts that are appropriately changed for convenience but are expressed differently, should be understood equivalently.

[0070] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0071] The product information matching method of this application can be programmed into a computer program product and deployed on a client or server for execution. For example, in an exemplary application scenario of this application, it can be deployed on the server of an e-commerce platform. In this way, the method can be executed by human-computer interaction with the process of the computer program product through a graphical user interface by accessing the interface opened after the computer program product is run.

[0072] Please see Figure 1 The product information matching method of this application, in its typical embodiment, includes the following steps:

[0073] Step S1100: Obtain the product information of the product entity to be detected. The product information contains multiple descriptive texts. Extract its knowledge subgraph. The knowledge subgraph contains the mapping relationship data between the attributes extracted from the descriptive texts and the attribute values.

[0074] The product information refers to the product description information corresponding to the product posted by the user. Products posted on e-commerce platforms are defined using this product information, thereby defining a product entity at the data level to facilitate the retrieval of product information. Product information includes the corresponding descriptive text and product images. This product information typically includes multiple descriptive texts, including but not limited to title text, details text, and product attribute data.

[0075] The entity to be detected needs to be matched with identical or similar entities. In an exemplary e-commerce scenario, one or more merchant users have the right to sell a certain product, while other merchants do not. These other merchants, in order to gain sales profits, submit product information different from the original product to circumvent the e-commerce platform's infringement detection. To protect the interests of merchant users with the right to sell products, the e-commerce platform performs infringement detection on listed products to prevent merchant users without the right to sell a particular product from infringing on the interests of those with the right.

[0076] The mapping relationship between attributes and attribute values ​​involves segmenting the descriptive text of a product entity into words, classifying the segmented words according to attributes using an attribute extraction model, and then using these segmented words as attribute values ​​to obtain the mapping relationship data between attributes and attribute values. For example, for the product title "Brand X 2021 Summer UV Protection Ice Silk Loose Comfortable Windbreaker M15237", after segmentation, multiple segmented words can be obtained. These multiple segmented words are then classified according to attributes to form the mapping relationship data between attributes and attribute values. This mapping relationship data is represented as follows:

[0077] (“Brand”, “Brand X”), (“Suitable Season”, “Summer”), (“Function”, “UV Protection”), (“Fabric”, “Ice Silk”), (“Fit”, “Loose Fit”), (“Style”, “Windbreaker”), (“Model”, “M15237”)

[0078] Of course, the mapping data between attributes and attribute values ​​obtained by classifying the same descriptive text based on different mapping relationships may be slightly different. Those skilled in the art will naturally understand this, and the examples here should not limit the scope of the inventive spirit of this application.

[0079] After extracting attributes from the product information of the product entity to be detected and obtaining the mapping relationship data between attributes and attribute values, a knowledge subgraph corresponding to the product entity to be detected can be established according to the structure of the knowledge graph of this application, and the mapping relationship data can be stored in the knowledge subgraph.

[0080] Step S1200: Retrieve multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected, forming a candidate product set. The knowledge graph stores the knowledge subgraphs corresponding to multiple product entities in the product database.

[0081] This application provides a knowledge graph that represents the connections between various product entities, their attributes, and attribute values ​​in a product database in the form of a graph structure. Each product entity is organized according to this graph structure, thus generating a corresponding knowledge subgraph, which represents the correspondence data between the product entity, its attributes, and attribute values.

[0082] Once the server obtains the knowledge subgraph of the product entity to be detected, it retrieves knowledge subgraphs from the knowledge graph that meet the preset conditions for attribute overlap with the knowledge subgraph of the product entity to be detected, that is, it retrieves multiple product entities and constructs a candidate product set as candidate product entities.

[0083] Specifically, the server retrieves the mapping relationship data between multiple attributes and attribute values ​​contained in the knowledge subgraph of the product entity to be detected, and retrieves the mapping relationship data between multiple attributes and attribute values ​​contained in each of the knowledge subgraphs in the knowledge graph. It then calculates the number of overlaps between the mapping relationship data of the attributes and attribute values ​​in the knowledge subgraph of the product entity to be detected and the mapping relationship data of the attributes and attribute values ​​in the knowledge subgraphs of the knowledge graph, obtaining the attribute overlap degree. From the knowledge graph, it filters out knowledge subgraphs whose attribute overlap degree meets preset conditions, and the set of product entities corresponding to these knowledge subgraphs is used as the candidate product set. In other words, the server filters out multiple product entities that meet preset conditions from the product attribute library by calculating the attribute overlap degree, thereby reducing the amount of data to be processed in the next step and improving the data processing speed of the next step. Since the process of retrieving matching product entities for the product entity to be detected is determined by the overlap relationship between attributes, and attribute data is relatively accurate data when selling products, it can better ensure the effectiveness of product recall and ensure that as many candidate product entities as possible that are similar to the product entity to be detected can be obtained during the recall stage.

[0084] Step S1300: Obtain the sample set corresponding to each product entity in the candidate product set. Each attribute in the knowledge subgraph corresponding to the sample set includes one or more descriptive texts. All descriptive texts in the sample set are organized in a uniform order.

[0085] Obtain the knowledge subgraph corresponding to the product entity in the candidate product set, and retrieve the description text corresponding to the mapping relationship between the attribute and attribute value in the knowledge subgraph of the product entity from the product database.

[0086] Because the same product entity may have different versions of product information, the attribute values ​​extracted from the same attribute in different versions of product information may be different. In other words, the same attribute value may correspond to multiple attribute values.

[0087] The server's mapping relationship data between an attribute and its value in the knowledge subgraph of a product entity may recall multiple corresponding description texts from multiple versions of product information for the same product, and these multiple description texts belong to different versions of product information.

[0088] The server will retrieve one or more descriptive texts from the product database based on the mapping relationship between an attribute and its value in the knowledge subgraph of the candidate product set, and organize the retrieved one or more descriptive texts according to the attribute organization order corresponding to the knowledge subgraph where the mapping relationship between the attribute and its value is located.

[0089] Based on the mapping relationship between all attributes and attribute values ​​in a knowledge subgraph of the candidate product set, the server retrieves one or more descriptive texts corresponding to the mapping relationship data between each attribute from the product database. The descriptive texts corresponding to the mapping relationship between all attributes and attribute values ​​in a knowledge subgraph are organized in the same attribute organization order to form a knowledge subgraph or a sample set of product entities.

[0090] It is easy to understand that by providing a sample set for the same product entity, which contains different versions of descriptive text corresponding to the attributes of the product entity, the reference information of the product entity can be enriched.

[0091] In one embodiment, the mapping relationship data between each product entity and its sample set can be stored independently. In another embodiment, the sample set of each product entity can also be stored as one of its attributes in the knowledge subgraph of that product entity.

[0092] Step S1400: Semantically match the product information of the product entity to be detected with the sample set of product entities in the candidate product set, and determine the product entities that match the product entity to be detected.

[0093] After receiving the product information of the product entity to be detected or the sample set of product entities in the candidate product set, the server can call the pre-trained text feature extraction model to perform deep information extraction on the product information of the product entity to be detected or the sample set of product entities in the candidate product set, thereby obtaining the corresponding semantic vector.

[0094] The text feature extraction model described herein is pre-trained to a convergent state and is used to perform deep semantic representation learning on the input title text. Those skilled in the art will understand that as long as the text feature model is trained with sufficient samples to make it suitable for extracting deep semantic information from the product information of the product entity to be detected or from a sample set of product entities in the candidate product set, the text feature extraction model can be used in the technical solution of this application. The text feature extraction model is generally a neural network model, preferably a model based on BERT, as it is more suitable for processing sequential information such as text. Subsequent embodiments of this application will further reveal the preferred model selection method, which will not be discussed here.

[0095] After obtaining the semantic vectors of the product entity to be detected and the semantic vectors of product entities in the candidate product set, the server performs semantic matching between them. Specifically, semantic matching is performed by calculating the similarity between the semantic vectors of the product entity to be detected and the semantic vectors of product entities in the candidate product set, and this can be achieved in various ways. For example:

[0096] In one embodiment, after obtaining the semantic vector of the product entity to be detected and the semantic vectors of product entities in the candidate product set, a preset data distance algorithm can be used to calculate the data distance between the semantic vector of each product entity in the candidate product set and the semantic vector of the product entity to be detected. The data distance algorithm includes, but is not limited to, cosine similarity algorithm, Euclidean distance algorithm, Pearson correlation coefficient algorithm, Minkowski distance algorithm, Mahalanobis distance algorithm, Jaccard coefficient algorithm, etc. Those skilled in the art can choose any one of these data distance algorithms to implement, as long as it can calculate the distance between points in the data set.

[0097] After the data distance is determined, in an optional embodiment, it can be normalized to a numerical space such as [0,1] for ease of calculation. The larger the value, the closer the data distance is, which means that the probability of the detected product entity matching the corresponding product entity in the candidate product set is higher.

[0098] In another embodiment, the semantic vector index of the product entities in the candidate product set is constructed through the interface provided by the Faiss framework. Then, through the similarity calculation interface provided by the framework, the corresponding preset algorithm can be applied to obtain similarity data, thereby quickly calculating the similarity between the semantic vector of the product entity to be detected and the semantic vector of the product entities in the candidate product set.

[0099] The algorithm for calculating the similarity between the semantic vector of the product entity to be detected and the semantic vectors of product entities in the candidate product set can be flexibly set. After the similarity between the product entity to be detected and each product entity in the candidate product set is calculated based on the semantic vector, the product entities in the candidate product set that meet the preset matching conditions are filtered out, and the product entities filtered out from the candidate product set are matched with the product entity to be detected.

[0100] In this typical embodiment, the advantages of this application are demonstrated. On the server side, by structuring the product information of each product entity, a sample set corresponding to each product entity is pre-constructed, focusing on the descriptive text corresponding to the mapping relationship between each attribute and attribute value, thus enriching the reference information. Subsequently, when determining whether there is a product entity in the candidate product set that is the same as the product entity to be detected, the semantic vector obtained from the sample set is used for matching, which greatly improves the accuracy of matching and reduces the false negative rate.

[0101] Furthermore, the server obtains the semantic vector of the product entity to be detected and the semantic vectors of product entities in the candidate product set. It performs semantic matching based on the semantic vector of the product entity to be detected, calculates the similarity between the semantic vector of the product entity to be detected and the semantic vectors of product entities in the candidate product set, and determines the product entities that match the product entity to be detected from the candidate product set based on the similarity. This allows for the filtering of product entities that are the same as the product entity to be detected from the product database, improving the accuracy of retrieval and matching of infringing products on e-commerce platforms. Unlike traditional searches that rely solely on keywords, which cannot accurately retrieve and match products and lead to a large number of missed or false detections.

[0102] Compared with the prior art, this application has many advantages, including at least the following aspects:

[0103] First, based on the e-commerce detection and matching scenario, this application extracts the knowledge subgraph corresponding to the entity to be detected. It then calculates the attribute overlap between the mapping data of attributes and attribute values ​​in the knowledge subgraph of the entity to be detected and the mapping data of attributes and attribute values ​​in the knowledge subgraphs of various product entities in the knowledge graph. This allows for the selection of multiple knowledge subgraphs similar to the entity to be detected from the knowledge graph. These multiple knowledge subgraphs constitute a candidate product set. Next, the application recalls the descriptive text corresponding to the mapping data of attributes and attribute values ​​in the knowledge subgraphs of the candidate product set, and extracts the semantic vectors of the descriptive text and the semantic vectors of the product information of the entity to be detected. Finally, it performs semantic matching between the semantic vectors of the entity to be detected and the semantic vectors of the product entities in the candidate product set to select product entities that are the same as or similar to the entity to be detected. Based on this method, products that are the same as or similar to the entity to be detected can be efficiently selected from the product database.

[0104] Secondly, this application first utilizes the overlap between the attributes of the product entity to be detected and the product entities in the knowledge graph to achieve recall, thus achieving a comprehensive retrieval and obtaining a candidate product set. Then, it uses the product information of the product entity to be detected and the sample sets of each product entity in the candidate product set for semantic matching, achieving a fine ranking of each product entity in the candidate product set, thereby filtering out product entities that match the product entity to be detected. Since each product entity in the candidate product set pre-contains a corresponding sample set, and each attribute of the product entity includes one or more descriptive texts, these descriptive texts are different forms of expression for the same product entity, thus expanding the reference range of product information and providing richer reference information for semantic matching. Therefore, more accurate matching results can be obtained during semantic matching, ensuring precise retrieval.

[0105] Furthermore, the technical solution of this application is applicable to the scenario of commodity infringement detection. When it is necessary to determine whether the commodity entity to be detected infringes on certain historical commodity entities, by further refining the detection based on comprehensiveness, not only can the role played by attribute overlap be avoided, but also the positive role played by the reference information provided by the sample set can be used to overcome the problem of the lack of commodity information of the commodity entity to be detected, and thus accurately identify infringing commodities.

[0106] Please see Figure 2 In a further embodiment, step S1100, obtaining the product information of the product entity to be detected, includes the following steps:

[0107] Step S1110: Respond to the product publishing request from the online store and obtain the product information of the product entity to be detected corresponding to the request. The product information includes any one or more of the following: product title, product details, and description text corresponding to the product attribute data:

[0108] The server responds to the product publishing request from the online store, obtains the product information corresponding to the product publishing request, constructs the product information into a product entity at the data level, and uses this product entity as the product entity to be detected.

[0109] The server retrieves various descriptive texts from the product information of the product entity to be detected. For example, it retrieves the product title, "Brand X 2021 Summer Anti-UV Ice Silk Loose Comfortable Windbreaker M15237", from the product information.

[0110] Step S1120: Extract attributes from the product information of the product entity to be detected, so as to extract the corresponding attribute values ​​from each of its descriptive texts based on the attribute, and form the mapping relationship data between attributes and attribute values:

[0111] To obtain the mapping relationship data between attributes and attribute values ​​from the product information of product entities, each descriptive text needs to be segmented into words first. Tokenization can be implemented using an attribute extraction model based on a neural network model. This attribute extraction model includes a text feature extraction module and a conditional random field (CRF) module. The text feature extraction module is typically implemented based on the Transformer underlying network architecture, and can use basic network models suitable for processing sequence data, such as Lattice LSTM or BERT, to perform representation learning on the descriptive text to obtain text feature vectors. The CRF module, i.e., the CRF model, identifies attributes of the described text based on the text feature vectors to extract the attributes. Since the techniques used by these models for implementing named entity recognition are well known to those skilled in the art, they will not be elaborated upon. Of course, the attribute extraction model is pre-trained to convergence before being used in this application, enabling it to acquire the corresponding capabilities. This capability allows the model to perform representation learning based on the embedding vectors obtained from the descriptive text of this application to obtain text feature vectors corresponding to deep semantic information, and to perform attribute recognition based on these text feature vectors, obtaining each attribute according to the attribute recognition.

[0112] Based on the attribute extraction model, the server identifies multiple attributes and extracts corresponding attribute values ​​from multiple descriptive texts of the product information of the corresponding product entity. The acquired attribute values ​​are matched with attributes to determine which attribute corresponds to which attribute. Based on the mapping relationship between attributes and attribute values, the recognition result of the descriptive text is obtained. For example, the attribute extraction result for the product information "Brand X 2021 Summer UV Protection Ice Silk Loose Comfortable Windbreaker M15237" is: ("Brand", "Brand X"), ("Applicable Season", "Summer"), ("Function", "UV Protection"), ("Fabric", "Ice Silk"), ("Fit", "Loose"), ("Style", "Windbreaker"), ("Model", "M15237").

[0113] Step S1130: Based on the preset knowledge graph structure, construct the mapping relationship data between the attributes and attribute values ​​of the product entity to be detected into a corresponding knowledge subgraph:

[0114] After the server obtains the mapping relationship data between multiple attributes and attribute values ​​of the product entity to be detected, it sorts the mapping relationship data between these attributes and attribute values, and constructs a knowledge subgraph of the product entity to be detected using a preset knowledge graph data structure.

[0115] The above embodiments accurately extract the attribute values ​​corresponding to the product information through attribute extraction, which facilitates the accurate construction of its knowledge subgraph for the product entity to be detected. Subsequently, matching of the same or similar product entities is performed based on the knowledge subgraph. The attribute overlap is calculated based on the attribute values ​​of the knowledge subgraph, which can ensure the effective recall of as many candidate product entities as possible and achieve the purpose of retrieval completeness.

[0116] Please see Figure 3 In a more detailed embodiment, before the step of obtaining the product information of the product entity to be detected, the following steps are included:

[0117] Step S1000: Construct a knowledge graph corresponding to the product database. The knowledge graph includes knowledge subgraphs corresponding to each product entity in the product database. Each knowledge subgraph contains the mapping relationship between attributes and attribute values ​​extracted from the description text of the product information of the corresponding product entity.

[0118] The technical solution of this application uses the operating environment of an e-commerce platform as its application environment. This e-commerce platform can be an e-commerce service platform that offers independent website services, typically such as a cross-border e-commerce service platform. Because such platforms need to consider the network environment across different regions globally and the independence of each merchant, they configure each merchant's store as an independent website, thereby enabling the e-commerce platform to serve a large number of such independent websites.

[0119] Each of these independent websites possesses its own product database corresponding to the goods sold on the site, which includes product information for a large number of product entities. Therefore, it is easy to understand that by accessing the product database of each independent website, one can obtain a massive amount of product information for each corresponding independent website.

[0120] In this application, a knowledge graph is constructed based on the product database contained in the e-commerce platform or independent website where the product entity to be detected resides, corresponding to the product database of the application environment. The server constructs corresponding knowledge subgraphs for all or some product entities in the product database. Based on the graph architecture of the knowledge graph, multiple knowledge subgraphs in the knowledge graph share nodes with the same mapping relationship data between attributes and attribute values, thereby reducing the data volume of the knowledge graph, saving storage space, and facilitating data retrieval and association.

[0121] Please see Figure 4 In a more detailed embodiment, step S1000, constructing a knowledge graph corresponding to the product database, includes the following steps:

[0122] Step S1010: Create a knowledge graph and obtain product information for each product entity in the product database:

[0123] The server responds to the knowledge graph creation command and adapts to the product database contained in the e-commerce platform or independent site where the product entity to be detected is located, retrieving all or part of the product information of the product entity from the corresponding database.

[0124] Step S1020: Extract attributes from the product information of each product entity to extract corresponding attribute values ​​from its description text based on different attributes, thus forming mapping relationship data between attributes and attribute values.

[0125] Based on the attribute extraction model, attributes are extracted from the description text of each product entity in the corresponding database. The attribute extraction model extracts the corresponding attribute values ​​from the product information, and the attributes and attribute values ​​are matched to obtain the mapping relationship data between the corresponding product entity attributes and attribute values.

[0126] Step S1030: Based on the preset structure of the knowledge graph, construct the mapping relationship data between the attributes and attribute values ​​of each product entity into a corresponding knowledge subgraph for the product entity:

[0127] After obtaining the mapping relationship data between the attributes and attribute values ​​of the product entity, the server organizes the mapping relationship between the attributes and attribute values ​​of the product entity according to the graph structure of the knowledge graph to construct the knowledge subgraph of the product entity.

[0128] Step S1040: Obtain different description versions of the product information for each product entity. Each description version contains descriptive text that matches the mapping relationship data between the various attributes and attribute values ​​of the product entity. Construct samples corresponding to each version in a unified order, and combine all samples to form the sample set of the corresponding product entity, which is then stored in the knowledge subgraph of the corresponding product entity.

[0129] The technical solution of this application takes the operating environment of an e-commerce platform as its application environment. The product database contains massive amounts of product information. Different users may upload the same or different product information to the database for the same product. For example, multiple users may upload the same product with the same or different versions of product information. These multiple descriptions can use different words or phrases to represent the same information. For instance, for the same mobile phone, one description might be called "beautiful mobile phone," another "beautiful mobile phone," or yet another "beautiful handheld phone." If the product database stores these identical or different versions of data separately, it will repeatedly occupy the database's space resources, resulting in a waste of storage resources. The server performs overlap detection on the product information uploaded by users, aligning the identical or different versions of product information uploaded by different users for the same product. This allows the server to store multiple versions of the same product information in the same location, facilitating subsequent retrieval of this product information.

[0130] After obtaining product information for different versions of the same product entity, the server extracts the mapping relationship data between attributes and attribute values ​​from each version of the product information. Based on the attribute-value mapping relationship data extracted from each version of the product information, the server retrieves the corresponding description text from the corresponding version of the product information. Each version has multiple attribute-value mapping relationship data, thus each version can retrieve multiple description texts, which constitute a sample.

[0131] Because the same product entity has different versions of product information, for the same attribute, different versions of product information may have different attribute values. That is to say, the same attribute of the same product entity can be mapped to multiple attribute values, forming the mapping relationship data between attributes and attribute values.

[0132] After obtaining samples corresponding to product information for each version of the same product entity, the server organizes the descriptive versions contained in the samples according to attributes. Specifically, the descriptive texts recalled by the same attribute are grouped into one category, which facilitates the association and retrieval of descriptive texts with the same attribute in each sample. Thus, the samples are organized into a sample set, so that the server can call the data in the sample set and improve processing efficiency.

[0133] The same product entity may have different versions of product information. By obtaining different versions of product information for the same product entity, the server can create an accurate profile of that product entity. The server can then perform similarity matching between the product entity to be screened and the product entity, thereby improving the matching accuracy.

[0134] Please see Figure 5 In a further embodiment, step S1200, retrieving a set of candidate products from the knowledge graph that matches the knowledge subgraph of the product entity to be retrieved, includes the following steps:

[0135] Step S1210: Obtain the mapping relationship data between attributes and attribute values ​​in the knowledge subgraph of the product entity to be detected, as its attribute set:

[0136] The server retrieves the mapping relationship data between multiple attributes and attribute values ​​from the knowledge subgraph of the product entity to be detected, and stores the mapping relationship between the retrieved multiple attributes and attribute values ​​in the same attribute set.

[0137] Step S1220: Perform overlap matching calculations between the attribute set of the product entity to be detected and the corresponding attribute sets of each product entity in the knowledge graph to determine the attribute overlap between each product entity in the knowledge graph and the product entity to be detected.

[0138] The server retrieves the mapping relationships between multiple attributes and attribute values ​​from the knowledge subgraphs in the knowledge graph, and stores the retrieved mapping relationships in the corresponding attribute sets. Thus, the server can obtain the corresponding number of attribute sets based on multiple knowledge subgraphs of the knowledge graph or corresponding product entities.

[0139] The attribute overlap degree of the product entity to be detected is calculated by matching its attribute set with the attribute set of each product entity in the knowledge graph. This determines the attribute overlap degree between the product entity to be detected and the product entity in the knowledge graph. The example formula is as follows:

[0140]

[0141] Where, S new,i Set represents the degree of attribute overlap between the attribute set of the product entity to be detected and the attribute set of the product entity in the knowledge graph. new,i Set represents the number of instances where the mapping relationship data between attributes and attribute values ​​in the attribute set of the product entity to be detected overlaps with the mapping relationship data between attributes and attribute values ​​in the attribute set of the i-th product entity in the knowledge graph. i This represents the number of mapping relationship data between attributes and attribute values ​​in the attribute set of the i-th item in the knowledge graph.

[0142] For ease of calculation, the value range of attribute overlap is normalized to the numerical space of [0, 1], with limits of 0 and 1 respectively. The larger the value, the more similar the two are and the higher the attribute overlap. Conversely, the smaller the value, the greater the difference between the two and the lower the attribute overlap.

[0143] Step S1230: Select the set of product entities in the knowledge graph whose attribute overlap meets the preset conditions as the candidate product set.

[0144] For the product entity to be detected, the attribute overlap between it and each product entity corresponding to the knowledge graph has been calculated. Product entities with attribute overlap exceeding the preset conditions are selected from the product entities corresponding to the knowledge graph and sorted in descending order of attribute overlap. This constructs a candidate product set, and multiple product entities with high attribute overlap with the product entity to be detected are selected as the candidate product set.

[0145] The preset condition can be an attribute overlap threshold. In this embodiment, it is recommended that when the attribute overlap is greater than the attribute overlap threshold, the product entity be filtered out from the product database to construct a candidate product set. The attribute overlap threshold is selected and set within the range of 0.1 to 1. Those skilled in the art can flexibly adapt the specific value of the preset threshold according to the actual business situation.

[0146] By calculating the attribute overlap between the attribute set of the product entity to be detected and the attribute set of the corresponding product entity in the knowledge graph, multiple product entities with similar attribute overlap can be selected from the knowledge graph, while obviously dissimilar product entities are eliminated, reducing the amount of data processing in subsequent steps. At the same time, because attribute values ​​are relatively precise, recalling data based on attribute values ​​can achieve data comprehensiveness.

[0147] Please see Figure 6 Referring to the previous embodiment, step S1400, which involves semantically matching the product information of the product entity to be detected with the sample set of product entities in the candidate product set to determine the product entity that matches the product entity to be detected, includes the following steps:

[0148] Step S1410: Calculate the first document similarity between the product information of the product entity to be detected and the sample set of product entities in the candidate product set. Use the corresponding attribute overlap of the product entity as the weight of the first document similarity to obtain a weighted similarity. Select product entities whose weighted similarity meets the preset conditions to form the first product set.

[0149] To calculate the first document similarity between the product information of the product entity to be detected and the sample set of each product entity in the candidate product set, the document similarity is calculated based on the document similarity algorithm between the product information of the product entity to be detected and the sample set of each product entity in the candidate product set, thereby obtaining the first document similarity between the product entity to be detected and each product entity in the candidate product set.

[0150] The document similarity algorithm is an algorithm that can calculate the similarity between texts, such as BM25 and TP-IDF. In this embodiment, the BM25 algorithm is recommended. Those skilled in the art can make flexible modifications to the specific implementation according to the actual business scenario.

[0151] To obtain the first document similarity between the product information of the product entity to be detected and the sample set of each product entity in the candidate product set, and to use this as a measure of document similarity between the product information of the product entity to be detected and the sample set of each product entity in the candidate product set, an exemplary formula is as follows:

[0152] Srr new,j =S new,i .BM25(X new ,X J )

[0153] Among them, Srr new,j S represents the weighted similarity. new,i The attribute overlap calculated above is used as a weight in the formula, and BM25() represents the BM25 algorithm.

[0154] According to the formula, the sample set between the product information of the product entity to be detected and each product entity in the candidate product set is calculated based on the document similarity algorithm, and the attribute overlap calculated above is used as the weight to calculate the first document similarity. For the convenience of calculation, the value range of the first document similarity is normalized to a numerical space such as [0,1]. The larger the first document similarity value, the higher the document similarity between the two, and vice versa. The smaller the first document similarity value, the greater the difference between the two and the lower the document similarity.

[0155] For the product entity to be detected, the first document similarity between it and each product entity in the candidate product set has been calculated. Product entities in the candidate product set whose first document similarity meets the preset conditions are selected and sorted in descending order of first document similarity. Thus, the first product set is constructed, and multiple product entities with high first document similarity to the product entity to be detected are selected as the first product set.

[0156] The preset condition can be a first document similarity threshold. In this embodiment, it is recommended that when the first document similarity is greater than the first document similarity threshold, the product entity is selected from the candidate product set to construct a first product set. The first document similarity threshold is selected and set within the range of 01 to 1. Those skilled in the art can flexibly adapt the specific threshold value according to the actual business situation.

[0157] The first document similarity calculation is performed based on the product information of the product entity to be detected and the sample set of each product entity in the candidate product set. This allows for the selection of multiple product entities with similar first document similarity from the candidate product set, while obviously dissimilar product entities are removed, reducing the amount of data processing in subsequent steps.

[0158] Step 14200: Calculate the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, and select product entities whose semantic similarity meets the preset conditions to form the second product set.

[0159] A pre-trained, converged text feature extraction model is used to extract corresponding semantic vectors from the product information of the product entities to be detected, and to extract corresponding semantic vectors from the sample set of product entities in the first product set. The text feature extraction model can employ various neural network models suitable for extracting deep semantic information from text, such as AlBert, ERNIE, ELECTRA, and BERT. In a preferred embodiment, the BERT model is recommended as the text feature extraction model.

[0160] In one recommended embodiment, the BERT model is used as the text feature extraction model. The text feature extraction model extracts the corresponding semantic vectors from the product information of the product entity to be detected and the sample set of product entities in the first product set. An example formula is as follows:

[0161] V j =Bert(X) j )

[0162] Among them, V j Let represent the semantic vector of a product entity in the first product set, and j represent the j-th descriptive text in the sample set of the product entity in the first product set.

[0163] V new =Bert(X) new )

[0164] Among them, V new This represents the semantic vector of the product to be detected, and new represents the descriptive text of the product to be detected.

[0165] The deep features of the product information of the product entity to be detected are extracted by the BERT model to obtain the corresponding text vector. Similarly, the deep features of the sample set of product entities in the first product set are extracted by the BERT model to obtain the corresponding text vector, and the obtained semantic vector is stored in the product feature library.

[0166] After obtaining the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set is calculated.

[0167] The semantic vector of the product information of the product to be detected can be mapped one-to-one with the semantic vectors of the sample sets of each product entity in the first product set. The semantic similarity between the semantic vector and the sample sets of each product entity in the first product set can be calculated. This can be achieved based on any data distance algorithm such as cosine similarity, Euclidean distance, or Pearson coefficient. In a preferred embodiment, it is recommended to use a combination of cosine similarity algorithm to determine the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vectors of the sample sets of product entities in the first product set. An example formula is shown below as a semantic level measure of the semantic vectors of the product entity to be detected and the semantic vectors of the product entities in the first product set:

[0168] Sc new,j =Cosine(V new V j )

[0169] Among them, Sc new,j This indicates semantic similarity, and Cosine() represents the cosine similarity algorithm.

[0170] According to the formula, the semantic similarity between the product information of the product entity to be detected and the sample set of each product entity in the first product set is calculated based on the cosine similarity algorithm. For the convenience of calculation, the range of the semantic similarity value is normalized to a numerical space such as [0,1]. The larger the semantic similarity value, the higher the document similarity between the two, and vice versa. The smaller the semantic similarity value, the greater the difference between the two and the lower the document similarity.

[0171] For the product entity to be detected, the semantic similarity between it and each product entity in the first product set has been calculated. Product entities in the first product set whose semantic similarity meets the preset conditions are selected. The selected product entities are used to construct the second product set. Multiple product entities with high document visibility to the product entity to be detected are obtained as the second product set.

[0172] The preset condition can be a semantic similarity threshold. In this embodiment, it is recommended that when the semantic similarity is greater than the semantic similarity threshold, the product entity is filtered out from the first product set and collected in the second product set. The semantic similarity threshold is selected and set within the range of 0.1-1. Those skilled in the art can flexibly adapt the specific threshold value according to the actual business situation.

[0173] The second product set may include a product entity, which is the same product entity as the product entity to be detected. The product entity to be detected and the product entity in the second product set have different product information. That is to say, the product entity to be detected and the product entity in the second product set are essentially the same product entity, but their respective product information description texts are different.

[0174] The cosine similarity algorithm is used to calculate the semantic vector of the product entity to be detected and the semantic vectors of multiple product entities in the first product set. This allows the product entities that are the same as the product entity to be detected to be selected from the first product set, thereby improving the efficiency of e-commerce platforms in detecting the same product entities.

[0175] In one embodiment, the second product set may contain multiple product entities, which are the same as or have a high degree of similarity to the product entity to be detected. Alternatively, the multiple product entities in the second product set may be different versions of the same product entity.

[0176] Step 1430: The product entities in the second product set are selected as matching product entities to be detected and pushed to the terminal device that provides the product information of the product entities to be detected.

[0177] To transmit the similarity detection information of the product entities to be detected to the terminal device, the product information of the product entities in the second product set can be pushed to the display interface on the terminal device for display.

[0178] After receiving the product information of the product entities in the second product set, the terminal device can display it on the display interface, allowing merchants or backend users to obtain similarity detection information of the product entities to be detected, and determine whether the products uploaded by users infringe on the rights of merchants with sales rights.

[0179] Please see Figure 7 In a further embodiment, step S1420, which calculates the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, includes the following steps:

[0180] Step S1440: Use the sample set of product entities in the knowledge graph as training samples to iteratively train the preset text feature extraction model until it converges.

[0181] To extract the semantic vector of product information from a product entity, a text feature extraction model can be prepared. This model can be based on basic neural network models such as AlBert, ERNIE, ELECTRA, and BERT. It can be trained to convergence by those skilled in the art using sufficient training samples, so that it learns the ability to extract the deep semantic vector of the input product entity based on its product information.

[0182] The dataset used to provide the training samples can directly adopt the sample sets from the knowledge subgraphs of each product entity in the knowledge graph. One version of product information in this sample set, i.e., the descriptive text corresponding to each attribute of the product entity, can be used as a training sample, thereby obtaining a massive number of training samples that can be used to perform unsupervised training on the text feature extraction model.

[0183] Step S1450: Using the text feature extraction model, extract the sample set of each product entity in the first product set and the semantic vector of the product information of the product entity to be detected:

[0184] A text feature extraction model trained to convergence is used to extract corresponding semantic vectors from the sample sets of each product entity in the first product set, and to extract corresponding semantic vectors from the product information of the product entity to be detected. In one embodiment, the extracted semantic vectors can be associated and stored in the knowledge subgraph of the corresponding product entity for direct retrieval.

[0185] As can be seen from this embodiment, the sample set in the knowledge graph can not only provide training samples for the text feature extraction model, but also play the role of reference information in the data retrieval accuracy process, with significant economic benefits.

[0186] Please see Figure 8To meet one of the purposes of this application, a product information matching device is provided, which is a functional embodiment of the product information matching method of this application. The device includes an information extraction module 1100, a retrieval and matching module 1200, an extraction and sorting module 1300, and a semantic matching module 1400. Specifically: the information extraction module 1100 acquires product information of the product entity to be detected, the product information containing multiple descriptive texts, and extracts its knowledge subgraph, which contains mapping relationship data between attributes and attribute values ​​extracted from the descriptive texts; the retrieval and matching module 1200 retrieves the knowledge subgraph and its corresponding semantic matching value from the knowledge graph of the product entity to be detected. Multiple product entities that match the knowledge subgraph constitute a candidate product set. The knowledge graph stores the knowledge subgraphs corresponding to multiple product entities in the product database. The extraction and sorting module 1300 obtains a sample set corresponding to each product entity in the candidate product set. Each attribute in the knowledge subgraph of the sample set includes one or more descriptive texts, and the descriptive texts in all sample sets are organized in a uniform order. The semantic matching module 1400 performs semantic matching between the product information of the product entity to be detected and the sample set of product entities in the candidate product set, thereby determining the product entities that match the product entity to be detected.

[0187] In a further embodiment, the information extraction module 1100 includes: an information extraction unit that responds to a product release request from an online store and obtains product information of the product entity to be detected corresponding to the request, wherein the product information includes any one or more of the product title, product details, and description text corresponding to the product attribute data of the product entity; an attribute extraction unit that extracts attributes from the product information of the product entity to be detected, so as to extract corresponding attribute values ​​from each of its description texts according to different attributes, thereby forming mapping relationship data between attributes and attribute values; and a subgraph construction unit that constructs the mapping relationship data between the attributes and attribute values ​​of the product entity to be detected into a corresponding knowledge subgraph according to a preset knowledge graph structure.

[0188] In a further embodiment, the product information matching device further includes: a graph construction module, which constructs a knowledge graph corresponding to the product database. The knowledge graph includes knowledge subgraphs corresponding to each product entity in the product database. The knowledge subgraphs contain mapping relationship data between attributes and attribute values ​​extracted from the description text of the product information of the corresponding product entity.

[0189] In a further embodiment, the knowledge graph construction module includes: a knowledge graph creation unit, which creates a knowledge graph and obtains product information for each product entity in the product database; a mapping construction unit, which extracts attributes from the product information of each product entity to extract corresponding attribute values ​​from its description texts according to different attributes, thereby forming mapping relationship data between attributes and attribute values; a subgraph creation unit, which constructs the mapping relationship data between the attributes and attribute values ​​of each product entity into a knowledge subgraph of the corresponding product entity according to the preset structure of the knowledge graph; and a subgraph matching unit, which obtains different description versions of the product information of each product entity, each description version containing description text that matches the mapping relationship data between the attributes and attribute values ​​of the product entity, constructs samples corresponding to each description version in a unified order, and constructs a sample set of the corresponding product entity from all samples, storing them in the knowledge subgraph of the corresponding product entity.

[0190] In a further embodiment, the retrieval and matching module 1200 includes: a mapping and merging unit, which acquires the mapping relationship data between attributes and attribute values ​​in the knowledge subgraph of the product entity to be detected, as its attribute set; an overlap calculation unit, which performs overlap degree matching calculation on the attribute set of the product entity to be detected and the corresponding attribute sets of each product entity in the knowledge graph, to determine the attribute overlap degree between each product entity in the knowledge graph and the product entity to be detected; and a candidate filtering unit, which sets the product entities in the knowledge graph whose attribute overlap degree meets the preset conditions as a candidate product set.

[0191] In a further embodiment, the semantic matching module 1400 includes: a first filtering unit, which calculates a first document similarity between the product information of the product entity to be detected and the sample set of product entities in the candidate product set, uses the corresponding attribute overlap of the product entity as the weight of the first document similarity to obtain a weighted similarity, and filters out product entities whose weighted similarity meets a preset condition to form a first product set; a second filtering unit, which calculates a semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, and filters out product entities whose semantic similarity meets a preset condition to form a second product set; and a product push unit, which pushes the product entities in the second product set as matching product entities with the product entity to be detected to the terminal device that provides the product information of the product entity to be detected.

[0192] In a further embodiment, the semantic matching module 1400 further includes: a model training unit that uses a sample set of product entities in a knowledge graph as training samples to iteratively train a preset text feature extraction model until it converges; and a vector extraction unit that uses the text feature extraction model to extract semantic vectors of the sample sets of each product entity in the first product set and the product information of the product entity to be detected.

[0193] To address the aforementioned technical problems, embodiments of this application also provide computer equipment. For example... Figure 9 The diagram shows the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When executed by the processor, the computer-readable instructions enable the processor to implement a product search category identification method. The processor provides computing and control capabilities, supporting the operation of the entire computer device. The memory stores computer-readable instructions, which, when executed by the processor, enable the processor to execute the product information matching method of this application. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0194] In this embodiment, the processor is used to execute... Figure 8 The system contains the specific functions of each module and its sub-modules, and the memory stores the program code and various data required to execute these modules or sub-modules. A network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules / sub-modules in the product information matching device of this application, and the server can call the server's program code and data to execute the functions of all sub-modules.

[0195] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the product information matching method of any embodiment of this application.

[0196] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the method described in any embodiment of this application.

[0197] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0198] In summary, the technical solution of this application is applicable to the scenario of commodity infringement detection. When it is necessary to determine whether the commodity entity to be detected infringes on certain commodity entities, by further refining the detection based on completeness, not only can the role played by attribute overlap be avoided, but also the positive role played by the reference information provided by the sample set can be used to overcome the problem of the lack of commodity information of the commodity entity to be detected, and thus accurately identify infringing commodities.

[0199] Those skilled in the art will understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in this application can be alternated, modified, combined, or deleted. Furthermore, other steps, measures, and solutions in the various operations, methods, and processes discussed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted. Furthermore, steps, measures, and solutions in the prior art that are similar to those disclosed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted.

[0200] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A product information matching method, characterized in that, Includes the following steps: In response to a product publishing request from an online store, the system obtains product information of the product entity to be detected, and extracts its knowledge subgraph based on the product information. The product information is product description information corresponding to the product published to the e-commerce platform, used to define the corresponding product entity from the data level. It contains multiple descriptive texts, including title text, detail text, and any number of product attribute data. The knowledge subgraph contains mapping relationship data between attributes and attribute values ​​extracted from the descriptive text. A candidate product set is formed by retrieving multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected. The knowledge graph stores the knowledge subgraphs corresponding to multiple product entities in the product database. Obtain a sample set corresponding to each product entity in the candidate product set. Each attribute in the knowledge subgraph of the sample set includes multiple descriptive texts. All descriptive texts in the sample set are organized in a uniform order. The multiple descriptive texts in the same sample set describe the corresponding product entity of the sample set in different forms. The product information of the product entity to be detected is semantically matched one by one with the sample set of product entities in the candidate product set to determine the product entity that matches the product entity to be detected.

2. The product information matching method according to claim 1, characterized in that, Responding to product listing requests from online stores and obtaining product information for the entity to be tested includes the following steps: Responding to product release requests from online stores, obtain product information of the entity to be tested corresponding to the request; The product information of the product entity to be detected is used to extract attributes, so as to extract the corresponding attribute values ​​from its description text according to different attributes, and form the mapping relationship data between attributes and attribute values. Based on the preset knowledge graph structure, the mapping relationship data between the attributes and attribute values ​​of the product entity to be detected is constructed into a corresponding knowledge subgraph.

3. The product information matching method according to claim 1, characterized in that, Before responding to a product listing request from an online store and obtaining product information for the entity to be inspected, the following steps are included: Construct a knowledge graph corresponding to the product database. The knowledge graph includes knowledge subgraphs corresponding to each product entity in the product database. The knowledge subgraphs contain mapping relationship data between attributes and attribute values ​​extracted from the description text of the product information of the corresponding product entity.

4. The product information matching method according to claim 3, characterized in that, Constructing a knowledge graph corresponding to the product database includes the following steps: Create a knowledge graph to obtain product information for each product entity in the product database; For each product entity, the product information is extracted to extract the corresponding attribute values ​​from its description text according to different attributes, thus forming the mapping relationship data between attributes and attribute values. Based on the preset structure of the knowledge graph, the mapping relationship data between the attributes and attribute values ​​of each product entity is constructed into a knowledge subgraph of the corresponding product entity; Obtain different description versions of product information for each product entity. Each description version contains descriptive text that matches the mapping relationship data between the various attributes and attribute values ​​of the product entity. Construct samples corresponding to each description version in a unified order, and combine all samples to form the sample set of the corresponding product entity, which is then stored in the knowledge subgraph of the corresponding product entity.

5. The product information matching method according to claim 1, characterized in that, The process involves retrieving multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected, forming a candidate product set. This includes the following steps: Obtain the mapping relationship data between attributes and attribute values ​​in the knowledge subgraph of the product entity to be detected, and use it as its attribute set; The overlap degree of the attribute set of the product entity to be detected is calculated by matching the attribute set of each product entity in the knowledge graph with the attribute set of each product entity in the knowledge graph. The set of product entities in the knowledge graph whose attribute overlap meets the preset conditions is selected as the candidate product set.

6. The product information matching method according to claim 5, characterized in that, The process involves semantically matching the product information of the product entity to be detected one by one with the sample set of product entities in the candidate product set to determine the product entities that match the product entity to be detected. This includes the following steps: Calculate the first document similarity between the product information of the product entity to be detected and the sample set of product entities in the candidate product set. Use the corresponding attribute overlap of the product entity as the weight of the first document similarity to obtain a weighted similarity. Select product entities whose weighted similarity meets the preset conditions to form the first product set. Calculate the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, and select product entities whose semantic similarity meets the preset conditions to form the second product set. The product entities in the second product set are used as matching product entities to be detected, and are pushed to the terminal device that provides the product information of the product entities to be detected.

7. The product information matching method according to claim 6, characterized in that, Before calculating the semantic similarity between the semantic vector of the product information of the product entity to be detected and the semantic vector of the sample set of product entities in the first product set, the following steps are included: The sample set of product entities in the knowledge graph is used as training samples to iteratively train the pre-set text feature extraction model until it converges. The text feature extraction model is used to extract the sample set of each product entity in the first product set and the semantic vector of the product information of the product entity to be detected.

8. A product information matching device, characterized in that, include: The information extraction module is configured to respond to product publishing requests from online stores, obtain product information of the product entity to be detected, and extract its knowledge subgraph based on the product information. The product information is product description information corresponding to the product published to the e-commerce platform, used to define the corresponding product entity from the data level, and includes multiple descriptive texts, including title text, detail text and any number of product attribute data. The knowledge subgraph contains mapping relationship data between attributes and attribute values ​​extracted from the descriptive text. The retrieval and matching module is configured to retrieve multiple product entities from the knowledge graph that match the knowledge subgraph of the product entity to be detected, forming a candidate product set. The knowledge graph stores the knowledge subgraphs corresponding to multiple product entities in the product database. The extraction and sorting module is set to obtain the sample set corresponding to each product entity in the candidate product set. Each attribute in the knowledge subgraph corresponding to the sample set includes multiple descriptive texts. All descriptive texts in the sample set are organized in a uniform order. The multiple descriptive texts in the same sample set describe the corresponding product entity of the sample set in different forms. The semantic matching module is configured to perform semantic matching on the product information of the product entity to be detected one by one with the sample set of product entities in the candidate product set, and determine the product entity that matches the product entity to be detected.

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

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images