Information label generation method and apparatus, and computing device

Abstract

The invention discloses an information label generation method and device and computing equipment. The method comprises the following steps: performing entity identification on obtained product associated data to extract entity information of at least one product entity; the entity information comprises name information, brand information and category information; inputting the entity information into a category identification model, and obtaining a category label output by the category identification model; determining a target domain brand identification model matched with the class target label, and processing the entity information by using the target domain brand identification model and the general large model to generate a brand label; and determining a target domain name identification model matched with the class target label, and processing the entity information by using the target domain name identification model and the general large model to generate a name label. By adopting the scheme, the generation efficiency and the generation precision of the information label can be improved.

Description

Technical Field

[0001] This application relates to the field of big data analysis, specifically to an information tag generation method, apparatus, computing device, computer storage medium, and computer program product. Background Technology

[0002] Information tag mining is the process of extracting key features of information through analysis and processing. It is an important part of big data analysis and has a wide range of applications in business, industry and other scenarios.

[0003] Currently, a single model is typically used to generate information tags; however, the accuracy of information tags generated in this way is insufficient. Summary of the Invention

[0004] In view of the above problems, this application is made in order to provide an information tag generation method, apparatus, computing device, computer storage medium and computer program product that overcomes or at least partially solves the above problems.

[0005] According to a first aspect of this application, a method for generating information tags is provided, comprising:

[0006] Entity recognition is performed on the acquired product association data to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information, and category information;

[0007] Input the entity information into the category recognition model and obtain the category tag output by the category recognition model;

[0008] Determine the target domain brand recognition model that matches the target tag, and use the target domain brand recognition model and the general large model to process the entity information to generate brand tags;

[0009] A target domain name recognition model matching the class target tag is determined, and the entity information is processed using the target domain name recognition model and a general large model to generate name tags.

[0010] In one optional implementation, the step of inputting the entity information into the category recognition model and obtaining the category tag output by the category recognition model includes:

[0011] Multiple candidate categories are generated based on the entity information;

[0012] The candidate categories and the entity information are input into the category discrimination model to obtain the category tags output by the category discrimination model.

[0013] In an optional implementation, the method further includes:

[0014] A new category has been detected; update the category information.

[0015] The updated category information and old sample data are input into the map generation model to obtain the category map output by the map generation model;

[0016] Input the category map into the general large model to obtain new sample data output by the general large model;

[0017] The category discrimination model is updated using the new sample data.

[0018] In one optional implementation, the step of processing the entity information using the target domain brand recognition model and the general large model to generate brand tags includes:

[0019] Determine the target domain knowledge graph and / or target domain brand candidate model that match the aforementioned target tags;

[0020] Based on the target domain knowledge graph and / or the target domain brand candidate model, obtain candidate brands associated with the entity information;

[0021] Input the candidate brands and the entity information into the general large model to obtain the initial brand tags output by the general large model;

[0022] Retrieve the brand tags mapped from the initial brand tag from the standard brand library.

[0023] In one optional implementation, the step of processing the entity information using the target domain name recognition model and the general large model to generate name tags includes:

[0024] Determine the candidate model of the target domain name that matches the target tag of the class;

[0025] Based on the target domain name candidate model, obtain initial candidate names associated with the entity information;

[0026] The initial candidate names are supplemented with information based on the brand tags and / or the brand information to generate candidate names;

[0027] Input the candidate names and entity information into the general large model, and obtain the name tags output by the general large model.

[0028] In one optional implementation, the step of inputting the entity information into the category recognition model and obtaining the category tag output by the category recognition model includes: inputting the entity information into the category recognition model and obtaining the parent category tag output by the category recognition model;

[0029] The step of determining the target domain brand recognition model that matches the class target tag includes: determining the target domain brand recognition model that matches the parent class target tag;

[0030] The step of determining the target domain name recognition model that matches the class target tag includes: determining the target domain name recognition model that matches the parent class target tag;

[0031] After generating the name tag, the method further includes: generating a child class target tag based on the name tag and the parent class target tag.

[0032] In an optional implementation, after processing the entity information using the target domain brand recognition model and the general large model to generate brand tags, the method further includes: if the credibility of the brand tag is less than a preset credibility threshold, then regenerating the brand tag based on the parent class target tag and the name tag.

[0033] After processing the entity information using the target domain name recognition model and the general large model to generate name tags, the method further includes: if the credibility of the name tag is less than a preset credibility threshold, then re-execute the step of obtaining the category tag output by the category recognition model; if the credibility of the name tag is greater than or equal to the preset credibility threshold, then generate a child category tag based on the name tag and the parent category tag.

[0034] According to a second aspect of this application, an information tag generating apparatus is provided, comprising:

[0035] The entity extraction module is used to perform entity recognition on the acquired product association data in order to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information and category information;

[0036] The category recognition module is used to input the entity information into the category recognition model and obtain the category tags output by the category recognition model;

[0037] The brand recognition module is used to determine the target domain brand recognition model that matches the class target tag, and to process the entity information using the target domain brand recognition model and the general big model to generate brand tags;

[0038] The name recognition module is used to determine the target domain name recognition model that matches the class target tag, and to process the entity information using the target domain name recognition model and the general large model to generate name tags.

[0039] In one optional implementation, the category identification module is used to: generate multiple candidate categories based on the entity information;

[0040] The candidate categories and the entity information are input into the category discrimination model to obtain the category tags output by the category discrimination model.

[0041] In one optional implementation, the category identification module is used to: detect the addition of a new category and update the category information;

[0042] The updated category information and old sample data are input into the map generation model to obtain the category map output by the map generation model;

[0043] Input the category map into the general large model to obtain new sample data output by the general large model;

[0044] The category discrimination model is updated using the new sample data.

[0045] In one optional implementation, the brand recognition module is used to: determine a target domain knowledge graph and / or a target domain brand candidate model that matches the class target tag;

[0046] Based on the target domain knowledge graph and / or the target domain brand candidate model, obtain candidate brands associated with the entity information;

[0047] Input the candidate brands and the entity information into the general large model to obtain the initial brand tags output by the general large model;

[0048] Retrieve the brand tags mapped from the initial brand tag from the standard brand library.

[0049] In one alternative implementation, the name recognition module is used for:

[0050] Determine the candidate model of the target domain name that matches the target tag of the class;

[0051] Based on the target domain name candidate model, obtain initial candidate names associated with the entity information;

[0052] The initial candidate names are supplemented with information based on the brand tags and / or the brand information to generate candidate names;

[0053] Input the candidate names and entity information into the general large model, and obtain the name tags output by the general large model.

[0054] In one optional implementation, the category recognition module is used to: input the entity information into the category recognition model, obtain the parent category tag output by the category recognition model; and generate a child category tag based on the name tag and the parent category tag.

[0055] The brand recognition module is used to: determine the target domain brand recognition model that matches the parent class target tag;

[0056] The name recognition model is used to: determine the target domain name recognition model that matches the target tag of the parent class.

[0057] In one optional implementation, the brand recognition module is used to: if the credibility of the brand tag is less than a preset credibility threshold, then regenerate the brand tag based on the parent class target tag and the name tag;

[0058] The category identification module is used to: if the credibility of the name tag is less than a preset credibility threshold, then re-execute the step of obtaining the category tag output by the category identification model; if the credibility of the name tag is greater than or equal to the preset credibility threshold, then generate a child category tag based on the name tag and the parent category tag.

[0059] According to a third aspect of this application, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0060] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described information tag generation method.

[0061] According to a fourth aspect of this application, a computer storage medium is provided, wherein the storage medium stores at least one executable instruction, the executable instruction causing a processor to perform an operation corresponding to the above-described information tag generation method.

[0062] According to a fifth aspect of this application, a computer program product is provided, comprising at least one executable instruction that causes a processor to perform operations corresponding to the above-described information tag generation method.

[0063] This application embodiment first extracts entity information from product association data, processes the entity information using a category recognition model to obtain category tags, then determines the matching target domain brand recognition model and target domain name recognition model based on the category tags, processes the entity information using the target domain brand recognition model and a general model to obtain brand tags, and processes the entity information using the target domain name recognition model and a general model to obtain name tags, thereby improving the generation efficiency and accuracy of information tags.

[0064] The embodiments of this application first generate multiple candidate categories based on entity information, and then use a category discrimination model to obtain category tags, thereby significantly reducing the discrimination range of the category discrimination model and improving the efficiency of category tag generation.

[0065] In this embodiment of the application, when a new category is added, the latest category map is obtained by using the map generation model, and then the generation capability of the general large model is used to output new sample data. The category discrimination model is updated using the new sample data to ensure the discrimination accuracy of the subsequent category discrimination model.

[0066] This application embodiment utilizes pre-generated class target tags to determine the matching target domain knowledge graph and / or target domain brand candidate model, thereby obtaining candidate brands. This allows the general large model to obtain initial brand tags under the constraint of candidate brands, improving the accuracy and efficiency of brand tag generation. Furthermore, the initial brand tags output by the general large model are aligned with the standard brand library to obtain standardized brand tags, which facilitates subsequent tag management and analysis.

[0067] This application embodiment uses pre-generated class target tags to determine the matching target domain name candidate model, thereby obtaining the initial candidate name; then, it uses brand tags to supplement the initial candidate name with information to obtain the candidate name, so that the general large model can obtain name tags under the constraint of candidate names, thereby improving the accuracy and efficiency of name tag generation.

[0068] In this embodiment, the category recognition model is first used to generate parent category tags. After the name tags are determined, child category tags are generated based on the parent category tags and name tags, forming a complete category tag system and improving the integrity of information tags.

[0069] This application embodiment generates a new brand tag by using the generated name tag and the existing parent class target tag when the credibility of the current brand tag is low, thereby improving the accuracy of brand tag generation.

[0070] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0071] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0072] Figure 1 A schematic diagram of an operating environment provided for implementing at least one embodiment of this application is shown;

[0073] Figure 2A flowchart illustrating an information tag generation method provided in Embodiment 1 of this application is shown;

[0074] Figure 3 A flowchart illustrating a method for generating class labels according to Embodiment 2 of this application is shown;

[0075] Figure 4 A flowchart illustrating a brand label generation method provided in Embodiment 3 of this application is shown;

[0076] Figure 5 A flowchart illustrating a name tag generation method provided in Embodiment 4 of this application is shown;

[0077] Figure 6 A flowchart illustrating an information tag generation method provided in Embodiment 5 of this application is shown;

[0078] Figure 7 A flowchart illustrating a brand label generation method provided in Embodiment 5 of this application is shown;

[0079] Figure 8 A flowchart illustrating a name tag generation method provided in Embodiment 5 of this application is shown;

[0080] Figure 9 This illustration shows a schematic diagram of the structure of an information tag generation device according to Embodiment Six of this application;

[0081] Figure 10 A schematic diagram of the structure of a computing device provided in Embodiment 7 of this application is shown. Detailed Implementation

[0082] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0083] It should be noted that the product-related data and other user-related data involved in the embodiments of this application are all authorized or fully authorized information and data from all parties, and follow the principle of minimum necessity, obtaining only the information essential to providing services to users. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for selecting authorization or refusal.

[0084] Figure 1A schematic diagram of an operating environment provided for implementing various embodiments of this application is shown. This application can be applied to application environments including, but not limited to, client 2, server 4, and network 6.

[0085] in:

[0086] Server 4 can consist of one or more computing devices. These computing devices may include virtualized computing instances. Virtualized computing instances may include virtual machines, such as emulations of computer systems, operating systems, servers, etc. The computing devices can load virtual machines based on virtual images and / or other data that define specific software used for emulation (e.g., operating systems, dedicated applications, servers). As the demand for different types of processing services changes, different virtual machines can be loaded and / or terminated on one or more computing devices. A hypervisor can be implemented to manage the use of different virtual machines on the same computing device.

[0087] Server 4 can be configured to communicate with client 2, etc., via network 6. Network 6 includes various network devices such as routers, switches, multiplexers, hubs, modems, bridges, repeaters, firewalls, proxy devices, and / or similar devices. Network 6 may include physical links, such as coaxial cable links, twisted pair cable links, fiber optic links, or combinations thereof, or wireless links, such as cellular links, satellite links, Wi-Fi links, etc.

[0088] Server 4 can provide services such as storage, reading, downloading, writing, querying, and deleting, such as providing static resource download services to clients through multiple domain names.

[0089] Client 2 can be running Windows or Android. TM Electronic devices running operating systems such as iOS, including smartphones, tablets, laptops, virtual reality devices, gaming devices, set-top boxes, in-vehicle terminals, and smart TVs. Based on these operating systems, various applications, such as browsers, can run.

[0090] Example 1

[0091] Figure 2 The diagram shows a flowchart of an information tag generation method provided in Embodiment 1 of this application.

[0092] The information tag generation method provided in this application is mainly used for generating commercial information tags.

[0093] Specifically, such as Figure 2 As shown, the method includes the following steps:

[0094] Step S201: Perform entity recognition on the acquired product association data to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information and category information.

[0095] The process involves acquiring product-related data, which includes, but is not limited to, search data, posting data, browsing data, and / or operation data. Furthermore, this application does not limit the data format of the product-related data; for example, the product-related data can be text, video, and / or audio.

[0096] In one optional implementation, to improve the accuracy of information tag generation, the acquired product association data can be preprocessed after acquisition. This preprocessing may include conversion processing, that is, converting non-text data such as video and / or audio into text data for subsequent processing. This application embodiment does not limit the processing method for converting non-text data into text data. For example, audio content can be converted into text using a speech recognition algorithm (such as Automatic Speech Recognition, ASR); image content can be converted into text using an image text recognition algorithm (such as Optical Character Recognition); and / or, a multimodal large model can be used to recognize non-text data content and convert it into text. Furthermore, this preprocessing may also include: invalid data processing (such as meaningless character removal), duplicate data removal, etc.

[0097] Further, Named Entity Recognition (NER) is performed on the product-related data. This involves using appropriate entity recognition algorithms to extract the corresponding product entities from the product-related data. Each product entity corresponds to a triple containing entity information, including the product entity's name, brand, and category information. That is, each product entity's triple can be represented as (name, brand, category). If a certain type of information cannot be extracted, its value in the triple can be empty, and so on.

[0098] In one alternative implementation, entity recognition prompts can be generated, and the entity recognition prompts and product association data can be input into a general large language model to obtain entity information output by the general large language model.

[0099] Although step S201 can extract the entity information triplet of the product entity from the product association data, the initially extracted entity information cannot accurately and standardly reflect the entity's true name, brand, and category. Therefore, after extracting the entity information, subsequent steps S202-S204 are used to obtain accurate name tags, brand tags, and category tags.

[0100] Step S202: Input entity information into the category recognition model and obtain the category tag output by the category recognition model.

[0101] A category recognition model is pre-trained based on a machine learning algorithm. This model is used to process entity information to obtain matching category tags. The category recognition model can be a single model, such as a single discriminative model, or it can consist of multiple sub-models. This application does not limit the specific structure or training method of the category recognition model.

[0102] In the specific implementation process, for any entity, the entity information is provided to the category recognition model. After processing the entity information, the category recognition model obtains a category label that matches the entity information. Optionally, when performing category recognition, the category label can be obtained solely based on the category information in the entity information; or, category labels can be generated by combining name information and / or brand information with the category information. In this way, the name information and brand information can provide auxiliary information for category recognition, improving the accuracy of category label recognition.

[0103] Step S203: Determine the target domain brand recognition model that matches the class target tag, and use the target domain brand recognition model and the general large model to process entity information to generate brand tags.

[0104] Multiple domain ranges are predefined. For each domain range, brand training samples matching that domain range are obtained. The lightweight model is then trained using these matching brand training samples to obtain a domain brand recognition model matching that domain range. This results in multiple domain brand recognition models corresponding to different domain ranges. Each domain brand recognition model can accurately identify brand tags within its domain range.

[0105] In the specific implementation process, after obtaining the class tags of entity information, the class tags are matched with the domain scope of brand recognition models in various domains. The domain brand recognition model that matches the class tag is taken as the target domain brand recognition model. During the matching process, the target domain brand recognition model that matches the class tag can be determined using similarity algorithms or matching mini-models.

[0106] Furthermore, by combining a target domain brand recognition model with a general-purpose large model, entity information is processed comprehensively to obtain brand tags that match the entity information. Since the target domain brand recognition model has high accuracy in brand recognition within the corresponding domain and is a lightweight model with fast processing speed, while the general-purpose large model has high generalization ability, combining the target domain brand recognition model and the general-purpose large model to obtain brand tags can improve the accuracy and efficiency of brand tag generation. The general-purpose large model mentioned in this application refers to a large model with general processing capabilities, including but not limited to: general language models, multimodal large models, knowledge graph large models, retrieval enhancement models, etc.

[0107] In the brand tag recognition process, brand identification can be performed solely based on brand information within the entity information, or it can be performed by combining brand information with category and name information. The category and name information can then provide auxiliary information for brand tag generation, improving brand recognition accuracy. Optionally, the original entity information can be corrected using the already identified category tags to obtain corrected entity information. This involves replacing the category information in the original entity information with the already identified category tags. Subsequently, the corrected entity information is processed using a target domain brand recognition model and a general large model to generate brand tags, further improving brand recognition accuracy.

[0108] Step S204: Determine the target domain name recognition model that matches the class target tag, and use the target domain name recognition model and the general large model to process the entity information to generate name tags.

[0109] Multiple domain scopes are pre-defined. For each domain scope, name training samples matching that domain scope are obtained. These matching name training samples are used to train a lightweight model to obtain a domain name recognition model matching that domain scope. This results in multiple domain name recognition models corresponding to different domain scopes, each capable of accurately identifying name tags within its domain scope. In this embodiment, the name tags include Standard Product Units (SPUs), which are the smallest standardized unit defined for similar products.

[0110] In the specific implementation process, after obtaining the class tag of entity information, the class tag is matched with the domain scope of each domain name recognition model, and the domain name recognition model that matches the class tag is taken as the target domain name recognition model. During the matching process, the target domain name recognition model that matches the class tag can be determined using similarity algorithms or matching small models.

[0111] Furthermore, by combining a target domain brand recognition model with a general model, entity information is processed comprehensively to obtain name tags that match the entity information. Since the target domain name recognition model has high accuracy in name recognition within its corresponding domain and is a lightweight model with fast processing speed, while the general model has high generalization ability, combining the target domain name recognition model and the general model to obtain name tags can improve the generation accuracy and efficiency of name tags. The entity information used in the name tag recognition process includes not only name information but also category information and brand information. Therefore, category information and brand information can provide context for name tag generation, improving name recognition accuracy. Optionally, the original entity information can be corrected using the already identified category tags to obtain corrected entity information. This involves replacing the category information in the original entity information with the already identified category tags. Subsequently, the target domain name recognition model and the general model are used to process the corrected entity information to generate name tags, thereby further improving brand recognition accuracy.

[0112] In one optional implementation, steps S203 and S204 can be executed concurrently to improve the overall efficiency of information tag generation; or, steps S203 and S204 can be executed sequentially, for example, step S203 can be executed first and then step S204 can be executed. In this case, when executing step S204, the identified brand tags and / or category tags combined with entity information can be provided to the name tag generation process to further improve the accuracy of name tag generation.

[0113] In one alternative implementation, the above steps can be performed by the corresponding modules or by a multi-task model.

[0114] Therefore, the information tag generation method provided in this application first extracts entity information from product association data, processes the entity information using a category recognition model to obtain category tags, then determines the matching target domain brand recognition model and target domain name recognition model based on the category tags, processes the entity information using the target domain brand recognition model and a general model to obtain brand tags, and processes the entity information using the target domain name recognition model and a general model to obtain name tags, thereby improving the generation efficiency and accuracy of information tags.

[0115] Example 2

[0116] Figure 3 The diagram shows a flowchart of a class label generation method provided in Embodiment 2 of this application.

[0117] The class tag generation method provided in this application embodiment is a further optimization based on the information tag generation method in Embodiment 1.

[0118] Specifically, such as Figure 3 As shown, the method includes the following steps:

[0119] Step S301: Perform entity recognition on the acquired product association data to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information and category information.

[0120] Step S302: Generate multiple candidate categories based on entity information.

[0121] To improve category identification efficiency, this step first generates multiple candidate categories based on entity information. These candidate categories can be generated using one or more of the following methods:

[0122] Candidate category generation method 1: Generate candidate categories based on similarity with entity information. Specifically, calculate the similarity between entity information and different categories in the category library, and select the N categories with the highest similarity as candidate categories.

[0123] Candidate category generation method 2: A lightweight candidate category model is pre-trained, and entity information is input into the candidate category model to obtain multiple candidate categories.

[0124] Candidate category generation method three: Pre-construct a knowledge graph, such as a category knowledge graph and / or a multivariate knowledge graph, and obtain candidate categories that match the entity information based on this knowledge graph. For example, a category knowledge graph contains standard categories, category aliases, abbreviations, community terms, synonyms, etc., so standard categories and category aliases that match the category information in the entity information can be mined from the category knowledge graph. As another example, a multivariate knowledge graph is a knowledge graph that contains categories, brands, and / or names, so candidate categories can also be mined from this multivariate knowledge graph using brand information and / or name information in the entity information.

[0125] Step S303: Input the candidate categories and entity information into the category discrimination model and obtain the category tags output by the category discrimination model.

[0126] A pre-trained category discrimination model is used to determine category tags that match the current entity information from known categories. In actual implementation, candidate categories and entity information are input into the category discrimination model, which then selects the category tags from the candidate categories.

[0127] In one optional implementation, to improve the accuracy of the category discrimination model, when a new category is detected, the model can be updated promptly or at a preset time (e.g., when the system is idle). Specifically, upon detecting a new category, the category information is updated; the updated category information and old sample data are input into the category graph generation model to obtain the category graph output by the model; the category graph is then input into a general model to obtain new sample data output by the model; and the category discrimination model is updated using the new sample data. In this implementation, the category graph generation model fully and accurately mines the relationships between the updated categories based on the updated categories and old sample data, thereby generating the latest category graph. The general model combines the category graph and relevant prompts to generate new sample data. This eliminates the need for manual annotation and sample data generation, enabling automatic updates to the category discrimination model and improving its update efficiency.

[0128] In one alternative implementation, a standard category library can be used to construct category training samples, which are then used to train the category discrimination model. Since the training samples are generated based on standard categories, the category tags output by the trained category discrimination model can be aligned with the standard category library.

[0129] In an optional implementation, if the category training samples are not generated based on a standard category library, after obtaining the category tags output by the category discrimination model, the category tags are further mapped to the standard category library, and the standard categories in the standard category library that match the category tags are used as the final category tags. This implementation does not limit the specific mapping method; for example, mapping can be performed based on preset rule mapping, keyword matching, and / or similarity calculation.

[0130] Optionally, to further improve the accuracy of tag generation, this embodiment of the application can perform anomaly detection and / or anomaly correction after obtaining the category tag. That is, it determines whether the current category tag and product association data match a preset anomaly condition; if so, anomaly correction is performed. For example, if the co-occurrence frequency of a game character name and a game in the product association data reaches a preset threshold, but the final category tag is not a game category tag, then it is determined that an anomaly condition has been met, and anomaly correction is required. During anomaly correction, configured correction rules or a trained lightweight correction model can be used.

[0131] Therefore, the category tag generation method provided in this application first generates multiple candidate categories based on entity information, and then uses a category discrimination model to obtain category tags, thereby significantly reducing the discrimination range of the category discrimination model and improving the efficiency of category tag generation.

[0132] Example 3

[0133] Figure 4 A flowchart illustrating a brand label generation method provided in Embodiment 3 of this application is shown.

[0134] The brand label generation method provided in this application is a further optimization based on the information label generation method in Embodiment 1.

[0135] Specifically, such as Figure 4 As shown, the method includes the following steps:

[0136] Step S401: Perform entity recognition on the acquired product association data to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information and category information.

[0137] Step S402: Input entity information into the category recognition model and obtain the category tag output by the category recognition model.

[0138] Step S403: Determine the target domain knowledge graph and / or target domain brand candidate model that match the class target tag.

[0139] To avoid information interference between different domains and improve the recognition accuracy of brand tags, embodiments of this application can construct domain knowledge graphs for different domains, and these domain knowledge graphs can be independent of each other. Specifically, the domain knowledge graph can include a domain brand knowledge graph, which contains the relationships between different brand names (including standard brand names, brand nicknames, brand abbreviations, brand community terms, etc.) within the corresponding domain; and / or, the domain knowledge graph can include a domain multivariate knowledge graph, which includes not only brand nodes but also name nodes, category nodes, etc.

[0140] And / or, for different domain scopes, the lightweight model is trained in advance using the brand training samples corresponding to that domain scope to obtain the domain brand candidate model corresponding to that domain scope.

[0141] In actual implementation, the category tag is matched with the divided domain scope to determine the domain scope that matches the category tag. The domain knowledge graph corresponding to the matched domain scope is used as the target domain knowledge graph, and the domain brand candidate model corresponding to the matched domain scope is used as the target domain brand candidate model.

[0142] Step S404: Obtain candidate brands associated with entity information based on the target domain knowledge graph and / or the target domain brand candidate model.

[0143] This step uses a target domain knowledge graph and / or a target domain brand candidate model to perform initial brand screening, thereby narrowing the processing scope of the subsequent general model and improving the overall efficiency of brand tag generation.

[0144] Specifically, the following methods can be used to obtain candidate brands:

[0145] Implementation Method 1: Obtain candidate brands associated with entity information based on the target domain knowledge graph. Specifically, the brand information in the entity information is matched with the corresponding domain brand knowledge graph to obtain candidate brands associated with that brand information. For example, brand nicknames, community terms, etc., associated with the brand information can be obtained. And / or, the brand information and name information in the entity information are matched with the domain multivariate knowledge graph of the corresponding domain scope. This not only retrieves candidate brands that match the brand information but also candidate brands associated with the name information, ensuring that the subsequently generated brand tags have global optimality.

[0146] Implementation Method Two: Provide entity information to the target domain brand candidate model. The target domain brand candidate model can output multiple candidate brands with high relevance to the entity information through similarity calculation, model inference, and other methods. This improves the accuracy of candidate brand generation.

[0147] Implementation Method 3: Generate candidate brands by combining a target domain knowledge graph and a target domain brand candidate model. For example, initial candidate brands matching entity information can be obtained first through the target domain knowledge graph, and then the target domain brand candidate model can be used to further filter the initial candidate brands (e.g., by calculating the similarity between the initial candidate brands and standard brands in the standard brand library) to obtain the final candidate brands. This improves the accuracy of candidate brand generation.

[0148] Step S405: Input the candidate brands and entity information into the general large model and obtain the initial brand labels output by the general large model.

[0149] The generated candidate brand and entity information are provided to the general big model. The general big model selects the candidate brand that best matches the entity information from the candidate brands as the output result. The brand label output by the general big model is called the initial brand label.

[0150] Step S406: Obtain the brand tag mapped to the initial brand tag from the standard brand library.

[0151] The initial brand tags output by the general large model may not match the standard brands in the brand tag library. To align the brand tags with the standard brand library, this step further retrieves the brand tags mapped from the initial brand tags from the standard brand library. For example, similarity algorithms can be used to determine the brand tags mapped from the initial brand tags.

[0152] In one optional implementation, to further improve the accuracy of tag generation, this embodiment of the application can perform anomaly detection and / or anomaly correction after obtaining the brand tag. That is, it determines whether the current brand tag and product association data match preset anomaly conditions; if so, anomaly correction is performed. This embodiment of the application does not limit the specific correction method; for example, anomaly correction can be performed using configured correction rules or a trained lightweight correction model.

[0153] Therefore, the brand tag generation method provided in this application uses the pre-generated class target tags to determine the matching target domain knowledge graph and / or target domain brand candidate model, and then obtains candidate brands, so that the general large model obtains initial brand tags under the constraint of candidate brands, thereby improving the brand tag generation accuracy and brand tag generation efficiency; moreover, the initial brand tags output by the general large model are aligned with the standard brand library to obtain standardized brand tags, which facilitates subsequent tag management and analysis.

[0154] Example 4

[0155] Figure 5 A flowchart illustrating a name tag generation method provided in Embodiment 4 of this application is shown.

[0156] The name tag generation method provided in this application embodiment is a further optimization based on the information tag generation method in Embodiment 1.

[0157] Specifically, such as Figure 5 As shown, the method includes the following steps:

[0158] Step S501: Perform entity recognition on the acquired product association data to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information and category information.

[0159] Step S502: Input entity information into the category recognition model and obtain the category tag output by the category recognition model.

[0160] Step S503: Determine the target domain brand recognition model that matches the class target tag, and use the target domain brand recognition model and the general large model to process entity information to generate brand tags.

[0161] Step S504: Determine the target domain name candidate model that matches the class target tag, and obtain the initial candidate name associated with the entity information based on the target domain name candidate model.

[0162] To avoid information interference between different domains and improve the recognition accuracy of name tags, embodiments of this application can pre-train a lightweight model using name training samples corresponding to each domain to obtain a domain name candidate model for that domain.

[0163] In practice, target labels are matched with the defined domain scopes to determine the domain scope that matches the target label. The domain name candidate model corresponding to the matched domain scope is then used as the target domain name candidate model. Entity information is provided to the target domain name candidate model, which can output multiple initial candidate labels with high relevance to the entity information through similarity calculation, model inference, etc., thereby improving the accuracy of candidate label generation.

[0164] Step S505: Supplement the initial candidate names with information based on brand labels and / or brand information to generate candidate names.

[0165] Initial candidate names typically do not carry brand data. To improve the accuracy of the final name tag generation, this step supplements the initial candidate names with information such as the brand tags and / or brand information in the entity information, thereby ensuring that the candidate names carry brand data.

[0166] Step S506: Input the candidate names and entity information into the general large model, and obtain the name labels output by the general large model.

[0167] The generated candidate names and entity information are provided to the general large model. The general large model selects the candidate name that best matches the entity information from the candidate names as the output result, which includes a name label.

[0168] In one optional implementation, to further improve the accuracy of tag generation, this embodiment of the application can perform anomaly detection and / or anomaly correction after obtaining the name tag. That is, it determines whether the current name tag and product association data match preset anomaly conditions; if so, anomaly correction is performed. This embodiment of the application does not limit the specific correction method; for example, anomaly correction can be performed using configured correction rules or a trained lightweight correction model.

[0169] Therefore, the name tag generation method provided in this application uses the first generated class target tags to determine the matching target domain name candidate model, and then obtains the initial candidate name; then, it uses brand tags to supplement the initial candidate name to obtain the candidate name, so that the general large model can obtain name tags under the constraint of candidate names, thereby improving the name tag generation accuracy and name tag generation efficiency.

[0170] Example 5

[0171] Figure 6 A flowchart illustrating an information tag generation method provided in Embodiment 5 of this application is shown.

[0172] The name tag generation method provided in this application embodiment is a further optimization based on the information tag generation method in Embodiment 1.

[0173] Specifically, such as Figure 6 As shown, the method includes the following steps:

[0174] Step S601: Perform entity recognition on the acquired product association data to extract entity information of at least one product entity; input the entity information into the category recognition model to obtain the parent category tag output by the category recognition model.

[0175] The product-related data can include user search data and other product-related data. In this embodiment, a category recognition model is used to process entity information to obtain coarse-grained parent category tags, such as first-level category tags and / or second-level category tags.

[0176] Step S602: Generate brand labels based on parent class target labels and entity information.

[0177] The method for generating brand labels can be found by referring to... Figure 7 Steps to achieve:

[0178] S6021, retrieve the list of target brands that match the parent class target tag.

[0179] The brands belonging to this parent category target tag constitute a target brand list that matches the parent category target tag. The target brand list comes from the standard brand library and contains multiple brand names.

[0180] S6022, Determine whether there is a brand in the target brand list that matches the entity information; if yes, proceed to step S6023; if no, proceed to step S6024.

[0181] A keyword matching algorithm is used to find brands that match the entity information (specifically, brand information) from the target brand list. This embodiment does not limit the keyword matching algorithm; for example, the Aho-Corasick Automaton can be used for keyword matching.

[0182] S6023, Generate brand tags based on brands that match the entity information in the target brand list.

[0183] If a brand matching the entity information exists in the target brand list, it indicates a successful exact match, and the brand matching the entity information in the target brand list will be used as the brand tag.

[0184] S6024, determine the target domain brand recognition model that matches the parent class target tag, and use the target domain brand recognition model and the general large model to process entity information to generate brand tags.

[0185] If no brand matches the entity information in the target brand list, it means that the exact match failed. In this case, the target domain brand identification model that matches the parent class target tag is determined, and then the target domain brand identification model and the general model are combined to generate brand tags.

[0186] Step S603: Determine whether the credibility of the brand label is less than the preset credibility threshold; if yes, proceed to step S604; if no, proceed to step S605.

[0187] In actual implementation, a pre-trained credibility detection model is used. Relevant data generated during brand tag generation is provided to this model to obtain its output credibility score. For example, the generation method of the brand tag (e.g., whether it's generated via S6023 or S6024) and the generation result (e.g., generation anomalies, the model's confidence level for the tag during generation) can be provided to the credibility detection model, and its output credibility score can be obtained. For instance, if the brand tag is generated via S6023, the generated credibility score is higher than the preset credibility threshold (i.e., a successful match, high tag credibility); if the generation result contains anomalies such as inability to generate the tag or the presence of a preset anomaly, the generated credibility score is lower than the preset credibility threshold (i.e., an anomaly exists, extremely low tag credibility); if it's generated via S6024 and there are no generation anomalies, the credibility detection model obtains the final tag credibility score based on the confidence levels of each model for the tag during generation.

[0188] Step S604: Regenerate the brand label based on the parent class target label and the name label.

[0189] If the credibility of the current brand tag is less than the preset credibility threshold, a new brand tag will be generated based on the parent category target tag and the name tag. For example, the brand that belongs to the parent category target tag and the brand to which the name tag belongs will be used as the new brand tag.

[0190] Step S605: Adopt the brand label.

[0191] If the credibility of the current brand label is greater than or equal to the preset credibility threshold, then the brand label will be adopted as the final brand label.

[0192] Step S606: Generate name tags based on the parent class target tag and entity information.

[0193] The method for generating brand labels can be found by referring to... Figure 8 Steps to achieve:

[0194] S6061, Get the list of target names that match the parent class target tag.

[0195] The product names belonging to the parent class target tag constitute a target name list that matches the parent class target tag. This target name list comes from the standard name library and contains multiple brand names.

[0196] S6062, determine whether there is a name in the target name list that matches the entity information; if yes, proceed to step S6063; if no, proceed to step S6064.

[0197] A keyword matching algorithm is used to search for names that match the entity information (specifically, the name information) from the list of target names. This embodiment does not limit the keyword matching algorithm; for example, an Aho-Corasick automaton can be used for keyword matching.

[0198] S6063, Generate name tags based on the names in the target name list that match the entity information.

[0199] If a name matching the entity information exists in the target name list, it indicates a successful exact match, and the name matching the entity information in the target name list will be used as the brand tag.

[0200] S6064, determine the target domain name recognition model that matches the parent class target tag, and use the target domain name recognition model and the general large model to process entity information to generate name tags.

[0201] If no name matching the entity information is found in the target name list, it indicates that an exact match has failed. In this case, a target domain name recognition model matching the parent class target tag is determined, and then combined with the target domain name recognition model and the general model to generate name tags. The specific implementation process of this step can be found in other embodiments, and will not be repeated here.

[0202] Step S607: Determine whether the credibility of the name tag is less than the preset credibility threshold; if yes, proceed to step S601; if no, proceed to step S608.

[0203] The relevant data generated during the name tag generation process is provided to the credibility detection model to obtain the credibility output by the model. For example, the generation method of the name tag (e.g., whether it is generated by S6063 or S6064), the generation result (e.g., generation anomalies, the model's confidence level for the tag during the name tag generation process, etc.) can be provided to the credibility detection model, and the credibility output by the credibility detection model can be obtained. For example, if the name tag is generated by S6063, the generated credibility is higher than the preset credibility threshold (i.e., accurate matching, the current tag credibility is high); if the generation result contains generation anomaly information such as the inability to generate a tag or the existence of a preset anomaly, the generated credibility is lower than the preset credibility threshold (i.e., there is an anomaly, the current tag credibility is extremely low); if it is generated by S6064 and there is no generation anomaly information, the credibility detection model obtains the final tag credibility based on the combined confidence levels of each model for the tag during the tag generation process.

[0204] If the credibility of the name tag is less than the preset credibility threshold, then step S601 is executed again to obtain a new parent class target tag.

[0205] Step S608: Adopt the name tag.

[0206] If the credibility of the current name tag is greater than or equal to the preset credibility threshold, then the name tag will be adopted as the final name tag.

[0207] Step S609: Generate child class target tags based on name tags and parent class target tags.

[0208] Specifically, class tags that belong to the name tag and are subordinate to the parent class tag are taken as subset class tags, thereby improving the class tags.

[0209] Therefore, the information tag generation method provided in this application first uses a category recognition model to generate parent category tags, and then generates child category tags based on the parent category tags and name tags after the name tags are determined, thus forming a complete category tag system and improving the integrity of information tags.

[0210] In an optional implementation, after determining the class target tag, a target domain attribute extraction model matching the class target tag can be invoked to extract attribute tags from product association data and / or entity information. This implementation does not limit the specific attribute tag extraction method.

[0211] In an alternative implementation, appropriate extraction methods can also be used to extract user-related object tags (such as extracting tags of followed publishers), user-related hot activity tags, etc. from product-related data.

[0212] In one optional implementation, the various models involved in the embodiments of this application can be manually verified after one or more models output results, that is, the accuracy of the output results is manually verified, and then training samples and manual annotations are generated based on the manual verification results to achieve accurate model updates.

[0213] In one optional implementation, this embodiment provides a visualization interface that displays the analysis results of corresponding tags. For example, for any product-related data, various information tags extracted from that product-related data can be displayed; and / or, for any information tag, the trigger frequency percentage of that information tag can be displayed; and / or, for any information tag, the changing trend of that information tag over time can be analyzed, and so on. In summary, the embodiments of this application do not limit the subsequent analysis methods of the generated information tags.

[0214] Example 6

[0215] Figure 9 A schematic diagram of an information tag generation device according to Embodiment Six of this application is shown. Figure 9 As shown, the device 900 includes: an entity extraction module 910, a category recognition module 920, a brand recognition module 930, and a name recognition module 940.

[0216] The entity extraction module 910 is used to perform entity recognition on the acquired product association data in order to extract entity information of at least one product entity; wherein, the entity information includes name information, brand information and category information;

[0217] The category identification module 920 is used to input the entity information into the category identification model and obtain the category tag output by the category identification model;

[0218] The brand recognition module 930 is used to determine the target domain brand recognition model that matches the class target tag, and to process the entity information using the target domain brand recognition model and the general large model to generate brand tags;

[0219] The name recognition module 940 is used to determine the target domain name recognition model that matches the class target tag, and to process the entity information using the target domain name recognition model and the general large model to generate name tags.

[0220] In one optional implementation, the category identification module 920 is used to: generate multiple candidate categories based on the entity information;

[0221] The candidate categories and the entity information are input into the category discrimination model to obtain the category tags output by the category discrimination model.

[0222] In one alternative implementation, the category identification module 920 is used for:

[0223] A new category has been detected; update the category information.

[0224] The updated category information and old sample data are input into the map generation model to obtain the category map output by the map generation model;

[0225] Input the category map into the general large model to obtain new sample data output by the general large model;

[0226] The category discrimination model is updated using the new sample data.

[0227] In one alternative implementation, the brand recognition module 930 is used to: determine a target domain knowledge graph and / or a target domain brand candidate model that matches the class target tag;

[0228] Based on the target domain knowledge graph and / or the target domain brand candidate model, obtain candidate brands associated with the entity information;

[0229] Input the candidate brands and the entity information into the general large model to obtain the initial brand tags output by the general large model;

[0230] Retrieve the brand tags mapped from the initial brand tag from the standard brand library.

[0231] In one alternative implementation, the name recognition module 940 is used for:

[0232] Determine the candidate model of the target domain name that matches the target tag of the class;

[0233] Based on the target domain name candidate model, obtain initial candidate names associated with the entity information;

[0234] The initial candidate names are supplemented with information based on the brand tags and / or the brand information to generate candidate names;

[0235] Input the candidate names and entity information into the general large model, and obtain the name tags output by the general large model.

[0236] In one optional implementation, the category identification module 920 is configured to: input the entity information into the category identification model, obtain the parent category tag output by the category identification model; and generate a child category tag based on the name tag and the parent category tag.

[0237] The brand recognition module 930 is used to: determine the target domain brand recognition model that matches the parent class target tag;

[0238] Name recognition model 940 is used to: determine the target domain name recognition model that matches the target tag of the parent class.

[0239] In one optional implementation, the brand recognition module 930 is used to: if the credibility of the brand tag is less than a preset credibility threshold, then regenerate the brand tag based on the parent class target tag and the name tag;

[0240] The category identification module 920 is used to: if the credibility of the name tag is less than a preset credibility threshold, then re-execute the step of obtaining the category tag output by the category identification model; if the credibility of the name tag is greater than or equal to the preset credibility threshold, then generate a child category tag based on the name tag and the parent category tag.

[0241] Therefore, the information tag generation device provided in this application first extracts entity information from product association data, processes the entity information using a category recognition model to obtain category tags, then determines the matching target domain brand recognition model and target domain name recognition model based on the category tags, processes the entity information using the target domain brand recognition model and a general model to obtain brand tags, and processes the entity information using the target domain name recognition model and a general model to obtain name tags, thereby improving the generation efficiency and accuracy of information tags.

[0242] Example 7

[0243] Figure 10 A schematic diagram of a computing device according to Embodiment Seven of this application is shown. The specific embodiments of this application do not limit the specific implementation of the computing device.

[0244] like Figure 10 As shown, the computing device may include: a processor 1002, a communications interface 1004, a memory 1006, and a communications bus 1008.

[0245] The processor 1002, communication interface 1004, and memory 1006 communicate with each other via communication bus 1008. Communication interface 1004 is used to communicate with other network elements such as clients or other servers. The processor 1002 executes program 1010, specifically performing the relevant steps described in the embodiment of the information tag generation method for computing devices.

[0246] Specifically, program 1010 may include program code that includes computer operation instructions.

[0247] The processor 1002 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0248] Memory 1006 is used to store program 1010. Memory 1006 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device. Program 1010 can specifically be used to cause processor 1002 to perform the operations described in the method embodiments above.

[0249] Example 8

[0250] Embodiment 8 of this application provides a non-volatile computer storage medium storing at least one executable instruction or computer program that enables a processor to perform the operation corresponding to the information tag generation method in any of the above method embodiments.

[0251] Example 9

[0252] Embodiment 9 of this application provides a computer program product, which includes at least one executable instruction or computer program that enables a processor to perform the operation corresponding to the information tag generation method in any of the above method embodiments.

[0253] In summary, based on the computing device, computer storage medium, and computer program product provided in this embodiment, entity information is first extracted from product association data. The entity information is then processed using a category recognition model to obtain category tags. Next, a matching target domain brand recognition model and target domain name recognition model are determined based on the category tags. The entity information is then processed using the target domain brand recognition model and a general model to obtain brand tags, and the entity information is processed using the target domain name recognition model and a general model to obtain name tags, thereby improving the efficiency and accuracy of information tag generation.

[0254] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of this application are not directed to any particular programming language. It should be understood that the content of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of this application.

[0255] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0256] Similarly, it should be understood that, in order to simplify this application and aid in understanding one or more of the various inventive aspects, features of the embodiments of this application are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of this application. However, this method of disclosure should not be construed as reflecting an intention that the claimed application requires more features than expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0257] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0258] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0259] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of this application. This application can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0260] It should be noted that the above embodiments are illustrative of this application and not restrictive, and those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. An information label generating method characterized by comprising: The method comprises: performing entity recognition on the obtained product association data to extract entity information of at least one product entity; wherein the entity information comprises name information, brand information and category information; inputting the entity information into a category recognition model to obtain a category label output by the category recognition model; determining a target domain brand recognition model matched with the category label, and processing the entity information by using the target domain brand recognition model and a general large model to generate a brand label; determining a target domain name recognition model matched with the category label, and processing the entity information by using the target domain name recognition model and the general large model to generate a name label.

2. The method of claim 1, wherein, The method further comprises: detecting an increase of a new category, and updating category information; inputting the updated category information and old sample data into a graph generation model to obtain a category graph output by the graph generation model; 3. The method of claim 2, wherein, inputting the category graph into the general large model to obtain new sample data output by the general large model; updating the category discrimination model by using the new sample data. The processing of the entity information by using the target domain brand recognition model and the general large model to generate a brand label comprises: determining a target domain knowledge graph and / or a target domain brand candidate model matched with the category label; obtaining a candidate brand associated with the entity information based on the target domain knowledge graph and / or the target domain brand candidate model; 4. The method according to any one of claims 1-3, characterized in that, inputting the candidate brand and the entity information into the general large model to obtain an initial brand label output by the general large model; obtaining a brand label mapped from the initial brand label from a standard brand library. The processing of the entity information by using the target domain name recognition model and the general large model to generate a name label comprises: determining a target domain name candidate model matched with the category label; obtaining an initial candidate name associated with the entity information based on the target domain name candidate model; 5. The method according to any one of claims 1-4, characterized in that, supplementing information of the initial candidate name according to the brand label and / or the brand information to generate a candidate name; inputting the candidate name and the entity information into the general large model to obtain a name label output by the general large model. The inputting of the entity information into the category recognition model to obtain a category label output by the category recognition model comprises: inputting the entity information into the category recognition model to obtain a parent category label output by the category recognition model; The determination of a target domain brand recognition model matched with the category label comprises: determining a target domain brand recognition model matched with the parent category label; The determination of a target domain name recognition model matched with the category label comprises: determining a target domain name recognition model matched with the parent category label.

6. The method according to any one of claims 1-5, characterized in that, ​ ​ ​ After the name tag is generated, the method further includes generating a sub-category tag according to the name tag and the parent-category tag.

7. The method of claim 6, wherein, After the brand tag is generated by processing the entity information using the target domain brand identification model and the general large model, if the credibility of the brand tag is less than a preset credibility threshold, the method further includes regenerating a brand tag according to the parent-category tag and the name tag. After the name tag is generated by processing the entity information using the target domain name identification model and the general large model, if the credibility of the name tag is less than a preset credibility threshold, the method further includes re-executing the step of obtaining the category tag output by the category identification model; and if the credibility of the name tag is greater than or equal to the preset credibility threshold, generating a sub-category tag according to the name tag and the parent-category tag.

8. An information label generating apparatus characterized by comprising: The method comprises: An entity extraction module is configured to perform entity recognition on the obtained product-related data to extract entity information of at least one product entity, wherein the entity information includes name information, brand information, and category information. A category identification module is configured to input the entity information into a category identification model to obtain a category tag output by the category identification model. A brand identification module is configured to determine a target domain brand identification model matched with the category tag, and process the entity information using the target domain brand identification model and a general large model to generate a brand tag. A name identification module is configured to determine a target domain name identification model matched with the category tag, and process the entity information using the target domain name identification model and a general large model to generate a name tag.

9. A computing device, comprising: The processor, the memory, the communication interface, and the communication bus are in communication with each other. The memory is configured to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the information tag generation method in any one of claims 1-8. The storage medium stores at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the information tag generation method in any one of claims 1-8.

10. A computer storage medium, characterized in that The storage medium stores at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the information tag generation method in any one of claims 1-8.

11. A computer program product, characterised in that, ​

Images