Commodity marking method and device, equipment and medium

Abstract

The invention relates to a commodity marking method and device, equipment and a medium in the technical field of e-commerce, and the method comprises the steps: determining prediction implicit representations corresponding to all layers in a label path corresponding to a hierarchical label system, wherein a target commodity title is mapped to the prediction implicit representations, and obtaining each instance demonstration classification set corresponding to the plurality of instance implicit representations matched with the instance demonstration classification sets. Constructing a plurality of corresponding classification instances corresponding to each layer, traversing each layer according to a father-child hierarchical relationship, obtaining a classification label of a father layer of the layer and a same-layer candidate label set formed by the classification labels of the same layer in a hierarchical label system, constructing a corresponding prompt text input large language model, and inputting the prompt text into the same-layer candidate label set according to the classification labels of the father layer and the same-layer candidate label set. And determining a target classification label to which the layer belongs from the same-layer candidate label set by referring to a plurality of classification instances according to a given instruction by the control model. And forming a target multi-layer label set by using the target classification labels of each layer output by the model. According to the method and the device, accurate hierarchical labels can be efficiently labeled for commodities.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a commodity marking method and corresponding devices, electronic equipment, and computer-readable storage media. Background Art

[0002] Hierarchical classification of products is an important task with wide applications and demands in the field of e-commerce, because the hierarchical classification results can accurately describe the products. Hierarchical product classification is a branch of multi-label product classification, which involves the systematic arrangement and classification of product data in the entire hierarchical label structure. The output labels are organized in a parent-child hierarchy, where higher-level labels contain broader concepts and sub-labels describe more specific sub-topic attributes.

[0003] In traditional technology, the Bert model trained by ordinary fine-tuning strategy is usually used to achieve hierarchical classification of goods. Since hierarchical product classification involves a deep hierarchical label structure and a wide range of label sets, the label set of contextual learning is very large, the classification accuracy is not high and it takes a long time.

[0004] In view of the shortcomings of traditional practices, the applicant has been engaged in research in related fields for a long time and has taken a different approach to solve the problems in the industry. Summary of the invention

[0005] The primary purpose of the present application is to solve at least one of the above problems and provide a commodity marking method and corresponding device, electronic equipment, and computer program product.

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

[0007] A product labeling method provided for one of the purposes of this application comprises the following steps:

[0008] Perform multi-level classification representation on the target product title to obtain the predicted implicit representation corresponding to each layer in the label path corresponding to the hierarchical label system in which the target product title is mapped;

[0009] From the classification instance library, multiple instance implicit representations matching the predicted implicit representation are determined, and instance demonstration classification sets corresponding to the respective instance implicit representations are obtained, including the corresponding instance product titles, instance multi-layer label sets and instance candidate label sets corresponding to each layer thereof, wherein the instance multi-layer label set includes the classification labels of each layer to which the instance product title belongs, and the instance candidate label set includes the classification labels of the parent layer in the instance multi-layer label set according to the classification labels of the corresponding layer, and each classification label of the sublayer of the classification label of the parent layer in the hierarchical label system is obtained;

[0010] Corresponding to each layer in the label path, constructing a corresponding plurality of classification instances, wherein the classification instance includes the instance product title in the corresponding instance demonstration classification set and the instance candidate label set of the layer, and the classification labels corresponding to the parent layer and the same layer of the layer in the instance multi-layer label set;

[0011] Traversing each layer in the label path to which the target product title belongs according to the parent-child hierarchical relationship, obtaining the classification label of the parent layer of the layer and the candidate label set of the same layer composed of the classification labels of the same layer in the hierarchical label system, embedding the target product title, the classification label of the parent layer, the candidate label set of the same layer, and the multiple classification instances corresponding to the layer into a preset prompt template, obtaining the corresponding prompt text input into the large language model, so as to control the model to determine the target classification label to which the layer belongs from the candidate label set of the same layer according to the given instructions and with reference to the multiple classification instances;

[0012] The target classification labels of each layer output by the large language model are obtained to form a target multi-layer label set of the target product title.

[0013] On the other hand, a product labeling device adapted to one of the purposes of the present application includes a label representation module, a classification set acquisition module, an instance construction module, a label classification module and a label set construction module, wherein the label representation module is used to perform multi-level classification representation on the target product title, and obtain the predicted implicit representation corresponding to each layer in the label path corresponding to the hierarchical label system mapped to the target product title; the classification set acquisition module is used to determine from the classification instance library a plurality of instance implicit representations matching the predicted implicit representation, and obtain the instance demonstration classification set corresponding to each instance implicit representation, including the corresponding instance product title, the instance multi-layer label set and the instance candidate label set corresponding to each layer, the instance multi-layer label set including the classification labels of each layer to which the instance product title belongs, and the instance candidate label set including the classification label of the parent layer in the instance multi-layer label set according to the classification label of the corresponding layer, and obtaining the classification labels of each sublayer of the parent layer classification label in the hierarchical label system; the instance construction module , used to construct corresponding multiple classification instances corresponding to each layer in the label path, wherein the classification instance includes the instance product title in the corresponding instance demonstration classification set and the instance candidate label set of the layer, as well as the classification labels corresponding to the parent layer and the same layer of the layer in the instance multi-layer label set; a label classification module, used to traverse each layer in the label path to which the target product title belongs according to the parent-child hierarchical relationship, obtain the classification label of the parent layer of the layer in the hierarchical label system, and the candidate label set of the same layer composed of each classification label of the same layer, embed the target product title, the classification label of the parent layer, the candidate label set of the same layer, and the multiple classification instances corresponding to the layer into a preset prompt template, obtain the corresponding prompt text input large language model, so as to control the model to determine the target classification label of the layer from the candidate label set of the same layer according to the given instructions and with reference to the multiple classification instances; a label set construction module, used to obtain the target classification labels of the layers output by the large language model, and constitute the target multi-layer label set of the target product title.

[0014] On the other hand, an electronic device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the commodity labeling method described in the present application.

[0015] On the other hand, a computer program product provided to meet another purpose of the present application includes a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any embodiment of the present application.

[0016] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0017] The present application firstly maps the target product title into a hierarchical label system through a multi-level classification representation to obtain the corresponding predicted implicit representation, thereby realizing in-depth semantic analysis and hierarchical prediction classification processing of the product title. Then, the instance demonstration classification set corresponding to multiple instance implicit representations matching the predicted implicit representation is determined from the classification instance library, and each layer in the label path is traversed by using the parent-child hierarchical relationship to construct the classification instance corresponding to the layer, and the candidate label set of the same layer and the classification label of the parent layer of the layer are determined, and the prompt text containing all the data related to this layer is constructed and input into the large language model, so that it can refer to the existing and verified classification instances, deeply and accurately understand the target product title to classify it layer by layer, thereby ensuring the accuracy of classification. Secondly, by determining the classification label of the product title layer by layer, refined classification is realized, without considering all the classification labels in the entire label system, avoiding global search in a large and complex label system, and erroneous interference of redundant classification labels on classification, thereby greatly reducing time consumption and ensuring the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0019] Figure 1 A schematic diagram of a typical network deployment architecture for implementing the technical solution of this application;

[0020] Figure 2 A schematic diagram of a typical embodiment of the commodity labeling method of the present application;

[0021] Figure 3 This is a principle block diagram of the commodity marking device of this application;

[0022] Figure 4 This is a schematic diagram of the structure of an electronic device used in this application.

[0023] Specific implementation method

[0024] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0025] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0026] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0027] like Figure 1 In the network architecture shown, the e-commerce platform 82 is deployed on the Internet to provide corresponding services to its users. Similarly, the devices 80 of the merchant users of the e-commerce platform 82 and the devices 81 of the consumer users are also connected to the Internet to use the services provided by the e-commerce platform.

[0028] The exemplary e-commerce platform 82 provides supply and demand matching of products and / or services to the general public with the help of Internet infrastructure. In the e-commerce platform 82, products and / or services are provided as commodity information. To simplify the description, the concepts of commodity, product, etc. are used in this application to refer to the products and / or services in the e-commerce platform 82, which may specifically be physical products, digital products, tickets, service subscriptions, other offline services, etc.

[0029] In reality, all entities can access the e-commerce platform 82 as users, use various online services provided by the e-commerce platform 82, and achieve the purpose of participating in the business activities achieved by the e-commerce platform 82. These entities can be natural persons, legal persons or social organizations, etc. Corresponding to the two types of entities, merchants and consumers in business activities, the e-commerce platform 82 has two types of users, merchant users and consumer users. All entities in the product distribution chain in business activities, including manufacturers, sellers, retailers, logistics providers, etc., can use online services in the e-commerce platform 82 as merchant users, while consumers in business activities, including real or potential consumers, can use online services in the e-commerce platform 82 as their corresponding consumer users. In actual business activities, the same entity can act as both a merchant user and a consumer user, and this should be understood flexibly.

[0030] The infrastructure used to deploy the e-commerce platform 82 mainly includes the backend architecture and frontend equipment. The backend architecture runs various online services through the service cluster, including middleware or frontend services for the platform, services for consumers, services for merchants, etc., to enrich and improve its service functions; the frontend equipment mainly covers the terminal devices used by users to access the e-commerce platform 82 as clients, including but not limited to various mobile terminals, personal computers, point-of-sale devices, etc. For example, merchant users can enter product information for their online stores through their terminal devices 80, or generate their product information using the interfaces opened by the e-commerce platform; consumer users can access the webpage of the online store implemented by the e-commerce platform 82 through their terminal devices 81, trigger the shopping process through the shopping buttons provided on the webpage, and call various online services provided by the e-commerce platform 82 during the shopping process, so as to achieve the purpose of shopping orders.

[0031] In some embodiments, the e-commerce platform 82 may be implemented by a processing facility including a processor and a memory, the processing facility storing a set of instructions that, when executed, cause the e-commerce platform 82 to perform the e-commerce and support functions involved in the present application. The processing facility may be part of a server, client, network infrastructure, mobile computing platform, cloud computing platform, fixed computing platform, or other computing platform, and provides electronic components, merchant equipment, payment gateways, application developers, marketing channels, transportation providers, customer equipment, point-of-sale equipment, etc. of the e-commerce platform 82.

[0032] The e-commerce platform 82 can be implemented as an online service such as cloud computing service, software as a service (SaaS), infrastructure as a service (IaaS), platform as a service (PaaS), desktop as a service (DaaS), hosted software as a service, mobile backend as a service (MBaaS), information technology management as a service (ITMaaS), etc. In some embodiments, the various functional components of the e-commerce platform 82 can be implemented to be suitable for operation on various platforms and operating systems. For example, for an online store, its administrator users enjoy the same or similar functions regardless of various embodiments such as iOS, Android, HomonyOS, or web pages.

[0033] The e-commerce platform 82 can realize the corresponding independent station for each merchant to run its corresponding online store, and provide the merchant with the corresponding business management engine instance for the merchant to establish, maintain and run one or more online stores in one or more independent stations. The business management engine instance can be used for content management, task automation and data management of one or more online stores, and various specific business processes of the online store can be configured through interfaces or built-in components to support the implementation of business activities. The independent station is the infrastructure of the e-commerce platform 82 with cross-border service functions. Merchants can maintain their online stores more centrally and autonomously based on the independent station. The independent station usually has a domain name and storage space dedicated to the merchant, and different independent stations are relatively independent. The e-commerce platform 82 can provide standardized or personalized technical support for a large number of independent stations, so that merchant users can customize their own business management engine instance and use this business management engine instance to maintain one or more online stores they own.

[0034] The online store can implement backend configuration and maintenance by having the merchant user log in to its business management engine instance as an administrator. With the support of various online services provided by the infrastructure of the e-commerce platform 82, the merchant user can configure various functions in its online store as an administrator, view various data, etc. For example, the merchant user can manage various aspects of its online store, such as viewing the online store's recent activities, updating the online store's product catalog, managing orders, recent visit activities, total order activities, etc.; the merchant user can also view more detailed information about the business and visitors to the merchant's online store by obtaining reports or metrics, such as showing a sales summary of the merchant's overall business, specific sales and participation data of active sales marketing channels, etc.

[0035] The e-commerce platform 82 may provide communication facilities and associated merchant interfaces for providing electronic communications and marketing, such as utilizing electronic message aggregation facilities to collect and analyze communication interactions between merchants, consumers, merchant devices, customer devices, point-of-sale devices, etc., aggregating and analyzing communications, such as for increasing the potential for providing product sales, etc. For example, a consumer may have a question related to a product, which may generate a dialogue between the consumer and the merchant (or an automated processor-based agent on behalf of the merchant), where the communication facility is responsible for interacting and providing the merchant with analysis on how to increase the probability of a sale.

[0036] See also Figure 2 A commodity marking method of the present application, in its typical embodiment, comprises the following steps:

[0037] Step S1100: Perform multi-level classification representation on the target product title to obtain the predicted implicit representation corresponding to each layer in the label path corresponding to the hierarchical label system in which the target product title is mapped;

[0038] Any product in the e-commerce platform can be used as a target product. The product title of the target product is used as the target product title. The target product is labeled according to the target product title to determine a target multi-layer label set composed of at least one target classification label to which the target product title belongs, where each classification label belongs to the target product.

[0039] The product title is a text describing the corresponding product, which can directly and clearly describe any one or more of the attributes, types, characteristics, etc. of the product in a concise expression, allowing buyers to quickly and accurately understand the product and attract buyers' attention. Usually, the product title is associated with the corresponding product and stored in a product database. The product database can be accessed through a corresponding data acquisition interface to obtain the product title, and the data acquisition interface can be flexibly implemented by technical personnel in this field.

[0040] It is not difficult to understand that in order to better manage the various commodities in the e-commerce platform, it is usually necessary to classify the commodities. Generally, a multi-level label system, i.e., a hierarchical label system, is defined, which is presented as a tree structure. Each commodity will be classified into the classification labels of each layer contained in a label path to which the commodity belongs. The hierarchical label system refines the classification of commodities step by step from the top layer to the bottom layer. There is only one classification label at the top layer of the system, usually represented by "root". It is a virtual label without actual classification meaning and is only used as the starting point of classification. Starting from the top layer classification label, each layer contains several classification labels. Each layer corresponds to the parent layer and the child layer relationship with the next layer. If the classification label of the layer has a classification label of a child layer that further refines its semantics, then the classification label of the layer is the classification label of the parent layer of the classification labels of these child layers. In addition, each child layer classification label has one and only one parent layer classification label, and the parent layer classification label has at least one classification label of the child layer.

[0041] A preset label classification representation model is adopted, and the target product title is taken as input. The semantics of the target product title is extracted through the hidden layer of the model, and multi-level classification mapping is performed according to the semantics. By mapping to the hierarchical label system, starting from the top-level classification label (the label is predicted by default), a classification label that best matches the semantics is predicted from the classification labels of each sub-layer of the predicted classification label layer by layer until the most specific classification label is reached. A label path is formed with the classification labels corresponding to each layer, and a predicted implicit representation containing the semantic representation of each classification label in the label path is output.

[0042] The label classification representation model is pre-trained to a convergence state, and has learned the ability to determine the predicted implicit representation corresponding to each layer in the label path corresponding to the mapping of the product title to the hierarchical label system. The label classification representation model can be flexibly trained by those skilled in the art according to the disclosure herein. The label classification representation model can be selected using the Bert model, or any other model such as Transformer Encoder, MPNet, BiLSTM, etc.

[0043] Step S1200: from the classification instance library, multiple instance implicit representations matching the predicted implicit representation are determined, and instance demonstration classification sets corresponding to the respective instance implicit representations are obtained, including the corresponding instance product title, the instance multi-layer label set and the instance candidate label set corresponding to each layer, the instance multi-layer label set including the classification labels of each layer to which the instance product title belongs, the instance candidate label set including the classification label of the parent layer in the instance multi-layer label set according to the classification label of the corresponding layer, and obtaining the classification labels of each sub-layer of the classification label of the parent layer in the hierarchical label system;

[0044] Using the cosine similarity algorithm, for each instance implicit representation in the classification instance library, the cosine similarity between each semantic representation in the predicted implicit representation and each semantic representation in the instance implicit representation is calculated, and each cosine similarity is added to obtain the similarity between the predicted implicit representation and the instance implicit representation. The similarity is sorted in descending order according to the similarity, and N instance implicit representations with a high ranking and a similarity exceeding a preset threshold are screened out. These instance implicit representations are matched with the predicted implicit representation. The preset threshold and N can be set as needed by those skilled in the art according to the disclosure herein.

[0045] The classification instance library includes multiple data pairs, each data pair includes an instance demonstration classification set and its corresponding instance implicit representation, the instance implicit representation includes the semantic representation of each layer of classification labels to which the instance product title in the instance classification set belongs, the first layer of classification labels in each layer of classification labels is root, and the classification labels of the remaining layers are classification labels of sublayers of the classification labels of the layer above it, and the semantic representation of each layer of classification labels can be obtained by using a text encoding model that has been pre-trained to convergence, using its learned ability to encode the semantic representation of the input text, and taking each classification label as input to obtain the corresponding semantic representation. The text encoding model can use an open source pre-trained Bert model, or other open source models such as the pre-trained GPT series, and the pre-training tasks of these models are aimed at acquiring the above capabilities.

[0046] In one embodiment, an empty classification instance library is constructed. Multiple product titles can be manually collected. For each product title, the product title is used as an example product title, and the example product titles are manually classified. In the hierarchical label system, starting from the top-level classification label (the label is determined by default), a classification label that best matches the semantics is determined from the classification labels of each sub-layer of the determined classification label layer by layer until the most specific classification label is reached, and the classification labels of each layer to which the example product title belongs are obtained to form an example multi-layer label set. In addition, a candidate label set corresponding to each layer in the example multi-layer label set is constructed, and each candidate label set contains the classification label of the corresponding layer in the example multi-layer label set. Specifically, for the classification label of each layer in the example multi-layer label set, the classification label of each sub-layer of the classification label of the parent layer in the hierarchical label system is obtained according to the classification label of the layer in the example multi-layer label set, and the candidate label set is formed. The example product title, the example multi-layer label set, and the instance candidate label set of each layer thereof constitute an example demonstration classification set, and the instance implicit representation corresponding to the instance multi-layer label constitutes a data pair, which is added to the classification instance library.

[0047] Step S1300: For each layer in the corresponding label path, a plurality of corresponding classification instances are constructed, wherein the classification instance includes the instance product title in the corresponding instance demonstration classification set and the instance candidate label set of the layer, and the classification labels corresponding to the parent layer and the same layer of the layer in the instance multi-layer label set;

[0048] It can be understood that the instance implicit representation corresponding to the instance multi-layer label set in the acquired instance demonstration classification set matches the predicted implicit representation, that is, each layer in the instance multi-layer label set corresponds to each layer in the label path. Therefore, for each layer in the label path, taking a single acquired instance demonstration classification set as an example, the instance product title in the instance demonstration classification set, the instance candidate label set at the same layer as this layer in the label path, and the classification label of the parent layer of this layer in the instance multi-layer label set constitute the demonstration classification basis text, and the classification label at the same layer as this layer in the label path in the instance multi-layer label set is used as the demonstration classification result text, and the demonstration classification basis text and the demonstration classification result text constitute a classification instance. Accordingly, a corresponding classification instance is constructed with each acquired instance demonstration classification set.

[0049] Step S1400: traverse each layer in the label path to which the target product title belongs according to the parent-child hierarchical relationship, obtain the classification label of the parent layer of the layer in the hierarchical label system, and the candidate label set of the same layer composed of the classification labels of the same layer, embed the target product title, the classification label of the parent layer, the candidate label set of the same layer, and the multiple classification instances corresponding to the layer into a preset prompt template, obtain the corresponding prompt text input large language model, so as to control the model to determine the target classification label to which the layer belongs from the candidate label set of the same layer according to the given instructions and with reference to the multiple classification instances;

[0050] Traverse each layer in the label path to which the target product title belongs layer by layer, wherein the corresponding layer of each traversal is the child layer of the corresponding layer of the previous traversal, and for each traversed layer, obtain the classification label of the parent layer of the hierarchical label system of the classification label of this layer, and the classification labels of the same layer constitute the candidate label set of the same layer, which is used to determine the target classification label of this layer, the classification label of the parent layer of the first layer of the traversal is empty, and the candidate label set of the same layer constituted by the classification labels of the first layer has only one classification label represented as root. The target product title, the classification label of the parent layer, and the candidate label set of the same layer are used as the basis text to be classified. Further, the basis text to be classified and the multiple classification instances corresponding to the traversed layer are embedded in the preset prompt template, and the corresponding prompt text is obtained and input into the large language model to obtain the output result of the large language model.

[0051] The large language model is suitable for text processing in the field of NLP. It is pre-trained with an extremely large corpus until convergence, acquires the ability to generate human language, and has a certain degree of accurate text semantic understanding and logical reasoning ability. The large language model includes OpenAI, Falcon, Chinchilla, PaLM, LLaMA 2, text-embedding-ada-002, etc. Those skilled in the art can choose one to implement as needed.

[0052] An exemplary prompt template is as follows:

[0053] “Please select the classification label that best matches the semantics of the target product title from the candidate label set of the same level in the given text to be classified. The following are some similar classification examples:

[0054] Classification example 1: {demonstration classification basis text, demonstration classification result text};

[0055] Classification example 2: {demonstration classification basis text, demonstration classification result text};

[0056] Classification example 3: {demonstration classification basis text, demonstration classification result text};

[0057] Text to be classified: {target product title, parent layer classification label, same layer candidate label set}. ”

[0058] Step S1500: Obtain the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title.

[0059] The output result of the large language model is obtained, that is, the classification labels corresponding to the layers are used as target classification labels, and the target classification labels arranged in a parent-child hierarchical relationship constitute a target multi-layer label set of the target product title.

[0060] It can be known from the typical embodiments of the present application that the technical solution of the present application has many advantages, including but not limited to the following aspects:

[0061] The present application firstly maps the target product title into a hierarchical label system through a multi-level classification representation to obtain the corresponding predicted implicit representation, thereby realizing in-depth semantic analysis and hierarchical prediction classification processing of the product title. Then, the instance demonstration classification set corresponding to multiple instance implicit representations matching the predicted implicit representation is determined from the classification instance library, and each layer in the label path is traversed by using the parent-child hierarchical relationship to construct the classification instance corresponding to the layer, and the candidate label set of the same layer and the classification label of the parent layer of the layer are determined, and the prompt text containing all the data related to this layer is constructed and input into the large language model, so that it can refer to the existing and verified classification instances, deeply and accurately understand the target product title to classify it layer by layer, thereby ensuring the accuracy of classification. Secondly, by determining the classification label of the product title layer by layer, refined classification is realized, without considering all the classification labels in the entire label system, avoiding global search in a large and complex label system, and erroneous interference of redundant classification labels on classification, thereby greatly reducing time consumption and ensuring the accuracy of classification.

[0062] In a further embodiment, after step S1500, obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included:

[0063] Step S1600: Acquire a user's historical behavior product set, wherein the historical behavior product set includes at least one data pair, wherein the data pair includes a corresponding historical behavior description text and a target multi-layer tag set of its associated products;

[0064] The user behavior includes any number of clicks, purchases, pre-purchase consultations, post-purchase consultations, adding to shopping carts, favorites, searches, positive comments, negative comments, sharing, following, returns, repeat purchases, etc.

[0065] The historical behavior description text is used to describe the corresponding user behavior of the user, such as clicks.

[0066] The commodities targeted by the user each time the user performs a user behavior, such as commodities clicked by the user, commodities positively commented by the user, etc., are taken as associated commodities associated with the corresponding user behavior.

[0067] It can be understood that user behavior can be captured by setting event listeners at the interaction points on the front end of the e-commerce platform, such as binding listening functions to click events on a product page or a purchase button. When the user performs these behaviors, the trigger function records the behavior description text and its associated products to form a data pair, which is then added to the user's historical behavior product set.

[0068] Step S1610: for each data pair in the historical behavior product set, the target multi-layer label set in the data pair is classified into the stage score set of the behavior stage to which the historical user behavior belongs;

[0069] It is understandable that the various user behaviors can be divided into stages in advance according to the depth of interaction and / or interest level between the user and the product, so as to define the behavioral stages to which the various user behaviors belong. In an exemplary embodiment, the user behaviors of searching, clicking, collecting, sharing, and consulting before purchase are grouped as belonging to the first behavioral stage, the user behaviors of adding to the shopping cart are grouped as belonging to the second behavioral stage, the user behaviors of purchasing, consulting after purchase, and positive evaluation are grouped as belonging to the third behavioral stage, returns and negative comments are grouped as belonging to the fourth behavioral stage, and the user behaviors of repeat purchases are grouped as belonging to the fifth behavioral stage. Further, an empty stage scoring set is created for each behavioral stage.

[0070] Taking a single data pair as an example, determine the behavior stage to which the user behavior described by the historical behavior description text in the data pair belongs, and add each classification label in the target multi-layer label set in the data pair to the stage scoring set of the behavior stage.

[0071] Step S1620: for different classification labels in each stage scoring set, determine the stage score corresponding to the classification label in the stage scoring set according to the repetition frequency of the classification label in the stage scoring set and the preset stage weight of the stage scoring set;

[0072] The preset stage weight corresponding to each stage score set can be preset to represent the preference degree of the user for the corresponding related products when the user performs the user behavior of the behavior stage. In an exemplary embodiment, the stage scores corresponding to the first behavior stage, the second behavior stage, the third behavior stage, the fourth behavior stage, and the fifth behavior stage are 0.5, 1.5, 4, -6, and 10.

[0073] Taking different classification labels in a single stage scoring set as an example, the number of times the classification label appears repeatedly in the stage scoring set is accumulated to obtain the repetition frequency. The repetition frequency is multiplied by the preset stage weight of the stage scoring set to obtain the stage score corresponding to the classification label in the stage scoring set.

[0074] Step S1630: for each different classification label, sum the stage scores corresponding to the classification label in each stage score set to which the classification label belongs, to obtain the user preference score of the classification label;

[0075] Taking a single classification label as an example, the stage scores corresponding to the classification label in each stage score set are added up to get the total score. Since the total score comprehensively considers the performance of the classification label in the corresponding behavioral stages, it can reflect the user's preference for the classification label. Therefore, the total score is used as the user preference score for the classification label.

[0076] Step S1640: Filter out the classification labels whose user preference scores meet the preset conditions and construct a user preference label portrait.

[0077] From the user preference scores of each different classification label, the classification labels whose user preference scores exceed a preset threshold are screened out, and these classification labels are used to form a user preference label portrait.

[0078] In this embodiment, by analyzing the user's historical behavior product set, based on the association between the user behavior and the multi-layer label set of the product, and dividing the user behavior into different stages and assigning different weights, the weighted performance of the classification labels of the comprehensive products in each relevant behavior stage is quantitatively expressed to express the user preference score of the classification label, which can more carefully capture the user's preference for the classification label of the product at different stages. Finally, by screening out the classification labels with user preference scores higher than the preset threshold, a user preference label portrait is constructed, which can comprehensively consider multiple aspects with different emphases to ensure the accuracy of the user preference label portrait.

[0079] In a further embodiment, step S1640, after filtering out the classification labels whose user preference scores meet the preset conditions and constructing the user preference label portrait, includes the following steps:

[0080] Step S1650: Obtain a user preference label portrait of each user, determine the user preference label portraits that contain the same first preset number of classification labels and are corresponding to active users, and form an active user group;

[0081] In one embodiment, in order to quickly retrieve user tag portraits containing the same first preset number of classification tags, the user preference tag portrait of each user is stored in advance as a document in ES (Elasticsearch), wherein the classification tag is used as a field of the document, and the unique identifier of the document is the unique identifier of the corresponding user. Accordingly, the automatic index construction function of ES automatically completes the construction of the corresponding inverted index, and the obtained inverted index can map each classification tag to the unique identifier of all documents containing the classification tag (i.e., the unique identifier of the corresponding user). Further, an aggregate query is constructed, wherein the distribution of all classification tags is first obtained by using terms aggregation, and then filter aggregation is used to filter out the user set corresponding to the classification tags containing the first preset number, and the user preference portraits of each user in the user set all contain the first preset number of classification tags. Further, the first behavior data of each user in the user set in the e-commerce platform is obtained, and for each user, the first behavior data of each user is multiplied by the preset behavior weight of the first behavior data, and the sum is obtained to obtain the active score of the user. According to the active score of each user in the user set, users whose active scores exceed the preset threshold are screened out, such users are confirmed as active users, and these active users are grouped into active user groups. The preset threshold value can be set as needed by those skilled in the art based on the disclosure herein.

[0082] Since each classification label contained in the user preference label portrait is different, the higher the first preset number is, the more detailed the representation of the user of the user preference label portrait containing the first preset number of classification labels is on the preference label, and the active user group obtained needs to reach a consensus on more common preference labels; on the contrary, the lower the first preset number is, the coarser the representation of the user of the user preference label portrait containing the first preset number of classification labels is on the preference label, and the active user group obtained can reach a consensus on a relatively small number of common preference labels. The specific numerical setting of the first preset number can be set by those skilled in the art as needed according to the disclosure herein.

[0083] The unique identifier of the user is used to uniquely refer to the user to distinguish the user from other users, such as an ID, and can be flexibly implemented by those skilled in the art.

[0084] The first behavior data includes any of the following: the number of recent logins, online time (unit: minutes), the number of interactions with the product before purchasing the product, the number of purchases, the number of interactions with the product after purchasing the product, etc. The interaction with the product before purchasing the product includes any operation behavior that can be performed on the product before the user purchases the product, such as like, favorite, forward, search, consult, add to shopping cart, etc. The interaction with the product after purchasing the product includes any operation behavior that can be performed on the product after the user purchases the product, such as comment, return. The recent period can be any one of the past three days, the past week, the past half month, the past month, the past three months, the past half year, etc., and can be set as needed by those skilled in the art. The preset behavior weights corresponding to the various first behavior data can be pre-set, such as the preset behavior weights of the number of logins, the online time (unit: minutes), the number of interactions with the product before purchasing the product, the number of purchases, and the number of interactions with the product after purchasing the product: 0.5, 0.01, 2, 5, 3.

[0085] Step S1651: when the number of users in the active user group exceeds a preset threshold, determine whether there is a parent-child hierarchical relationship between the classification tags, and determine a corresponding multi-level tag set according to the classification tags with a hierarchical relationship;

[0086] The preset threshold is used to measure whether the number of users in the active user group is large enough. When the number of users exceeds the preset threshold, it means that there are enough users in the active user group at this time. In this regard, it is determined whether there is a parent-child hierarchical relationship between the various classification labels preferred by these users, by traversing each classification label to see whether one of the classification labels is the classification label of the parent layer of the other classification label in the hierarchical label system. If so, it means that there is a parent-child hierarchical relationship between the two classification labels. These classification labels with parent-child hierarchical relationships are mapped to the hierarchical label system, and the classification labels corresponding to the corresponding label paths are determined. For each label path, a multi-layer label set is formed with the classification labels corresponding to the label path. The classification labels corresponding to the corresponding label path refer to that each layer in the label path corresponds to a single classification label, and there is a parent-child hierarchical relationship between the classification labels corresponding to every two layers. The preset threshold can be set as needed by those skilled in the art based on the disclosure herein.

[0087] Step S1652: Obtain exposure times of the products in the plurality of multi-level tag sets, and select products whose exposure times meet a preset condition as first marketing products;

[0088] Determine that the product title belongs to the product corresponding to the multi-layer tag set, obtain the exposure times of multiple products, and sort them in order from high to low according to the exposure times. The top N1% products can be used as first marketing products, and subsequently the first marketing information of these first marketing products is sent to the users of the active user group, which helps to further promote the purchase progress of these first marketing products and improve the conversion rate; or, the top N2%-N1% products can be used as first marketing products, and subsequently the first marketing information of these first marketing products is sent to the users of the active user group, which helps to further increase the decision-making opportunities of users whether to purchase these first marketing products; or, the top 1-N2%-N1% products can be used as first marketing products, and subsequently the first marketing information of these first marketing products is sent to the users of the active user group, which helps to make more users see these first marketing products.

[0089] The N2% and N1% are values ​​within 100%, respectively. Those skilled in the art can set them as needed according to the disclosure herein, for example, N2% and N1% are 10% and 70%, respectively.

[0090] Step S1653: Obtain the first marketing information of each of the first marketing products, and send it to users in the active user group.

[0091] For each of the first marketing products, marketing information of the first marketing product can be pre-constructed as the first marketing information. The first marketing information can be an advertising banner, email, private message to the e-commerce platform assistant, private message to the online store, etc., which is used to promote and / or introduce products to users, and / or marketing activities to introduce products to users, attract users' attention to the first marketing products and / or stimulate users' desire to purchase the first marketing products. The specific marketing content can be created by the operators of the e-commerce platform and / or the employees of the corresponding online stores.

[0092] In this embodiment, by using the user preference tag portrait of the user, active user groups with the same number of preference tags are screened out, which can more carefully understand the personalized needs of users and effectively improve the targeting and conversion rate of marketing activities. In this process, by analyzing the user's behavior data and setting behavior weights, the evaluation of user activity is further refined, ensuring that marketing information can accurately reach those active users who are most likely to make purchases.

[0093] In a further embodiment, step S1640, after filtering out the classification labels whose user preference scores meet the preset conditions and constructing the user preference label portrait, includes the following steps:

[0094] Step S1660: Obtain a user preference label portrait of each user, determine the user preference label portraits that contain the same second preset number of classification labels and are corresponding to high-value users, and form a high-value user group;

[0095] In one embodiment, in order to quickly retrieve user tag portraits containing the same second preset number of classification tags, the user preference tag portrait of each user is stored in advance as a document in ES (Elasticsearch), wherein the classification tag is used as a field of the document, and the unique identifier of the document is the unique identifier of the corresponding user. Accordingly, the automatic index construction function of ES automatically completes the construction of the corresponding inverted index, and the obtained inverted index can map each classification tag to the unique identifier of all documents containing the classification tag (i.e., the unique identifier of the corresponding user). Further, an aggregate query is constructed, wherein the distribution of all classification tags is first obtained by using terms aggregation, and then filter aggregation is used to filter out the user set corresponding to the classification tags containing the second preset number, and the user preference portraits of each user in the user set all contain the second preset number of classification tags. Further, the second behavior data of each user in the user set in the e-commerce platform is obtained, and for each user, the second behavior data of the user is multiplied by the preset behavior weight of the second behavior data, and the value score of the user is obtained by summing. According to the value score of each user in the user set, users whose value scores exceed the preset threshold are screened out, and such users are confirmed to be high-value users, and these high-value users are grouped into high-value user groups. The preset threshold value can be set as needed by those skilled in the art based on the disclosure herein.

[0096] Since each classification label contained in the user preference label portrait is different, the higher the second preset number is, the more detailed the representation of the user of the user preference label portrait containing the second preset number of classification labels is on the preference label, and the resulting high-value user group needs to reach a consensus on more common preference labels; on the contrary, the lower the second preset number is, the coarser the representation of the user of the user preference label portrait containing the second preset number of classification labels is on the preference label, and the resulting high-value user group only needs to reach a consensus on a relatively small number of common preference labels. The specific numerical setting of the second preset number can be set by a person skilled in the art as needed based on the disclosure herein.

[0097] The second behavior data includes any multiple items of the number of times the user adds items to a shopping cart, the number of purchases, the maximum amount of a single transaction, the average amount of consumption, etc., and the preset behavior weights corresponding to each item of the second behavior data can be pre-set, for example, the preset behavior weights of the number of times the user adds items to a shopping cart, the number of purchases, the maximum amount of a single transaction, and the average amount of consumption are 0.5, 2, 4, and 5 respectively.

[0098] Step S1661: When the number of users in the high-value user group exceeds a preset threshold, determine whether there is a parent-child hierarchical relationship between the classification tags, and determine a corresponding multi-level tag set based on the classification tags with a hierarchical relationship;

[0099] The preset threshold is used to measure whether the number of users in the high-value user group is large enough. When the number of users exceeds the preset threshold, it means that there are enough users in the high-value user group at this time. In this regard, it is determined whether there is a parent-child hierarchical relationship between the various classification labels preferred by these users, by traversing each classification label to see whether one of the classification labels is the classification label of the parent layer of the other classification label in the hierarchical label system. When it is, it means that there is a parent-child hierarchical relationship between the two classification labels. These classification labels with parent-child hierarchical relationships are mapped to the hierarchical label system, and the classification labels corresponding to the corresponding label paths are determined. For each label path, a multi-layer label set is formed with the classification labels corresponding to the label path. The classification labels corresponding to the corresponding label path refer to that each layer in the label path corresponds to a single classification label, and there is a parent-child hierarchical relationship between the classification labels corresponding to every two layers. The preset threshold can be set as needed by those skilled in the art based on the disclosure herein.

[0100] Step S1662: Acquire exposure effectiveness scores of the products in the plurality of multi-level tag sets, and select products whose exposure effectiveness scores meet preset conditions as second marketing products;

[0101] Determine that the product title belongs to the product corresponding to the multi-layer tag set, obtain exposure effectiveness scores of multiple products, and sort them in order from high to low according to the exposure effectiveness scores. The top N1% of the products can be used as second marketing products, and subsequently the second marketing information of these second marketing products is sent to the users of the high-value user group, which helps to further promote the repurchase or multiple purchase progress of these second marketing products and improve the conversion rate; or, the top N2%-N1% of the products can be used as second marketing products, and subsequently the second marketing information of these second marketing products is sent to the users of the high-value user group, which helps to further increase the decision-making opportunities of users whether to purchase these second marketing products; or, the top 1-N2%-N1% of the products can be used as second marketing products, and subsequently the second marketing information of these second marketing products is sent to the users of the high-value user group, which helps to make more users see these second marketing products.

[0102] The exposure effectiveness score is obtained in advance by obtaining the number of clicks, exposures, and purchases of the corresponding product after exposure, dividing the number of clicks after exposure by the number of exposures to obtain the click-through rate, dividing the number of purchases after clicks by the number of clicks after exposure to obtain the conversion rate, and then multiplying the click-through rate and the conversion rate by their respective weights and adding them together to obtain the exposure effectiveness score. The weights corresponding to the click-through rate and conversion rate can be set by those skilled in the art as needed, such as 0.3 and 0.7.

[0103] The N2% and N1% are values ​​within 100%, respectively. Those skilled in the art can set them as needed according to the disclosure herein, for example, N2% and N1% are 10% and 70%, respectively.

[0104] Step S1663: Obtain the second marketing information of each of the second marketing products, and send it to users in the high-value user group.

[0105] For each of the second marketing products, marketing information of the second marketing product can be pre-constructed as the second marketing information. The second marketing information can be an advertising banner, email, private message to the e-commerce platform assistant, private message to the online store, etc., which is used to promote and / or introduce products to users, and / or marketing activities to introduce products to users, attract users' attention to the second marketing products and / or stimulate users' desire to purchase the second marketing products. The specific marketing content can be created by the operators of the e-commerce platform and / or the employees of the corresponding online stores.

[0106] In this embodiment, by using the user preference tag portrait of the user, high-value user groups with the same number of preference tags are screened out, which can more carefully understand the personalized needs of users and effectively improve the targeting and conversion rate of marketing activities. In this process, by analyzing the user's behavior data and setting behavior weights, the evaluation of user value is further refined, ensuring that marketing information can accurately reach those high-value users who are most likely to make repeat purchases and / or multiple purchases.

[0107] In a further embodiment, after step S1500, obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included:

[0108] Step S1700: obtaining a search text provided by a user and a plurality of description text sets of products to be searched, wherein the description text sets include a target multi-layer tag set of the products to be searched;

[0109] E-commerce platforms usually provide users with a product search function, and users can use this function to provide search text describing the desired product to the server.

[0110] The server obtains the search text, takes each product in the e-commerce platform as a product to be searched, and obtains a description text set for each product to be searched, wherein the description text set includes texts describing the corresponding product to be searched, including a target multi-layer tag set and a search display text. The corresponding product to be searched is also preset with its respective SKU description texts, and these SKU description texts are placed in the description text set. The search display text is used for the text required to display the introduction of the product when the product to be searched is subsequently displayed to the user as a product search result, such as any one or more of the product price, product title, etc.

[0111] The single SKU description text usually describes one of the brand, model, configuration, grade, color, packaging capacity, unit, production date, shelf life, use, price, origin, etc. of the product.

[0112] Step S1710: determining whether the search text contains SKU description text; when the search text does not contain SKU description text, determining a first semantic matching score between the underlying nested semantic representation of each target multi-layer tag set and the search semantic representation of the search text;

[0113] A named entity recognition algorithm pre-trained to convergence can be used to identify the SKU description text in the search text. When the SKU description text cannot be identified, it means that the search text does not contain the SKU description text at this time. For each of the target multi-layer label sets, the classification labels of each layer of the target multi-layer label set are traversed according to the parent-child hierarchical relationship. According to the traversal order, for each classification label except the last traversed classification label, an attributive description text with an inclusion relationship with the classification label of its child layer is constructed. The attributive description texts corresponding to these classification labels and the last traversed classification are combined into a bottom nested text according to the order of each attributive description, and the text encoding model is used to determine the bottom nested semantic representation of the bottom nested text. The text encoding model is used to determine the search semantic representation of the search text, and then the cosine similarity algorithm is used to calculate the cosine similarity between each bottom nested semantic representation and the search semantic representation as the first semantic matching score.

[0114] To facilitate understanding of the exemplary examples, the classification labels of each layer in the target multi-layer label set are: clothing, tops, women's clothing, skirts, and one-piece dresses, and the corresponding attributive description texts are: in clothing, in tops, in women's clothing, in skirts, and the corresponding underlying nested texts are: in clothing, in tops, in women's clothing, in group clothing, and in dresses.

[0115] The named entity recognition model learns the ability to recognize SKU description text in input text, and the model can be flexibly implemented by those skilled in the art.

[0116] Step S1720: Filter out the underlying nested semantic representations whose first semantic matching scores meet preset conditions, obtain the search display text in the default product image and description text set corresponding to the product to be searched, and send them to the user;

[0117] The underlying nested semantic representations whose first semantic matching score exceeds a preset threshold are screened out, which means that the to-be-searched product corresponding to the underlying nested semantic representation matches the search text, and the default product image and search display text in the description text set of the to-be-searched product corresponding to the underlying nested semantic representation are obtained, and the to-be-searched products are used as product search results, and the product search results are arranged in descending order according to the first semantic matching score, and the default product image and search display text of each arranged product search result are filled in a grid layout. The preset threshold can be set as needed by those skilled in the art according to the disclosure herein.

[0118] Step S1730: when the search text contains SKU description text, determine a second semantic matching score between the underlying nested semantic representation of each target multi-layer tag set and the search semantic representation of the search text that does not contain the SKU description text, and determine whether each description text set contains the SKU description text;

[0119] When the named entity recognition model can identify the SKU description text in the search text, it means that the search text contains the SKU description text at this time. After separating the SKU description text from the search text, in order to avoid the interference of the SKU description text on the core semantics of the product description in the search text, the text encoding model is used to determine the search semantic representation of the search text that does not contain the SKU description text. The cosine similarity algorithm is used to calculate the cosine similarity between the search semantic representation and the underlying nested semantic representation of each target multi-layer tag set as the second semantic matching score.

[0120] Furthermore, it is queried whether the SKU description text exists in the description text set of each commodity to be searched.

[0121] Step S1740: filter out the underlying nested semantic representations whose second semantic matching scores meet the preset conditions, as well as the products to be searched corresponding to the SKU description text, obtain product images of the products to be searched that match the SKU description text, and the search display text in the description text set, and send them to the user.

[0122] Filter out the underlying nested semantic representations whose second semantic matching score exceeds the preset threshold and contains the SKU description text corresponding to the to-be-searched product, which means that the to-be-searched product matches the search text, obtain the product image and search display text in the description text set of the to-be-searched product corresponding to the underlying nested semantic representation that match the SKU description text, use these to-be-searched products as product search results, arrange the product search results in descending order according to the second semantic matching score, and fill the product images and search display text of the arranged product search results in a grid layout. The preset threshold can be set as needed by a person skilled in the art based on the disclosure herein.

[0123] It can be understood that each product is pre-set with multiple pictures of the product, such as the product pictures corresponding to the different SKU description texts of the product, and multiple product pictures in the product carousel, among which the product picture belonging to the main picture of the product is the default product picture.

[0124] In this embodiment, first, by using the target multi-layer tag set to construct a hierarchical and detailed semantic representation for each product, this provides depth and details for subsequent search matching. Then, when the user provides a search text, a description text set related to the product is collected, including the target multi-layer tag set and the search display text, and it is confirmed whether the search text contains the SKU description text. When the search text does not contain the SKU description text, the first semantic matching score between the underlying nested semantic representation of each target multi-layer tag set and the semantic representation of the search text is determined to filter the product; and when the search text contains the SKU description text, the second semantic matching score between the search text without the SKU description text and each underlying nested semantic representation is determined, and it is confirmed whether the product description text set contains the corresponding SKU description text. Finally, according to the corresponding matching score and the on-demand inclusion of the SKU description text, the most relevant products are selected and displayed to the user, including product images and search display texts that accurately match the user's search needs, so that the user can quickly and accurately find the required products, thereby bringing higher user satisfaction and potential economic benefits to the e-commerce platform. In addition, considering that the underlying nested semantic representation can accurately represent the semantics of the corresponding product, and is matched on demand with the search semantic representation of the search text that does not contain the SKU description text, the accuracy and reliability of the corresponding matching score are ensured.

[0125] In a further embodiment, after step S1500, obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included:

[0126] Step S1800: obtaining a target multi-layer tag set for a specified product and a target multi-layer tag set for a plurality of products to be recommended;

[0127] The designated product is designated by the user, and the user may designate the product by purchasing the product, clicking on the product, sharing the product, following the product, adding the product to a shopping cart, etc.

[0128] The target multi-layer tag set to which the product title of the designated product belongs is obtained, and each product in the e-commerce platform is used as a product to be recommended, and the target multi-layer tag set to which the product title of each product to be recommended belongs is obtained.

[0129] Step S1810: determining the degree of match between the target multi-layer tag set of each to-be-recommended product and the text encoding representation corresponding to the target multi-layer tag set of the specified product;

[0130] For each of the target multi-layer tag sets, the text encoding model is used to take each classification label in the target multi-layer tag set as input to obtain a text encoding representation corresponding to the semantic representation of each classification label.

[0131] The cosine similarity algorithm is used to calculate the cosine similarity between the target multi-layer label set of each recommended product and the text encoding representation corresponding to the target multi-layer label set of the specified product as the matching degree.

[0132] Step S1820: Filter out the recommended products whose matching degree meets the preset conditions as the recommended products of the designated products.

[0133] The recommended product with the highest matching degree exceeding the preset threshold is screened out, which means that the recommended product matches the designated product and is thus used as the recommended product for the designated product.

[0134] In this embodiment, by utilizing the hierarchical and detailed semantic representation constructed by the target multi-layer tag set for each product, the matching degree between the target multi-layer tag set of each recommended product and the text encoding representation corresponding to the target multi-layer tag set of the specified product is determined, and the recommended product with the better matching degree is selected as the recommended product for the specified product, which can ensure the accuracy and reliability of the matching degree and is very efficient.

[0135] In a further embodiment, after step S1500, obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included:

[0136] Step S1900: Obtain target multi-layer tag sets and exposure effectiveness scores of multiple products, and group products with the same target multi-layer tag set into the same product group;

[0137] The exposure effectiveness score is obtained in advance by obtaining the number of clicks, exposures, and purchases of the corresponding product after exposure, dividing the number of clicks after exposure by the number of exposures to obtain the click-through rate, dividing the number of purchases after clicks by the number of clicks after exposure to obtain the conversion rate, and then multiplying the click-through rate and the conversion rate by their respective weights and adding them together to obtain the exposure effectiveness score. The weights corresponding to the click-through rate and conversion rate can be set by those skilled in the art as needed, such as 0.3 and 0.7.

[0138] The number of commodities obtained can be determined by those skilled in the art as needed.

[0139] Obtain the target multi-layer label set and exposure effectiveness score of the product titles of multiple products, and confirm whether each target multi-layer label is the same. If they are the same, it means that the corresponding products are of the same type, so the same corresponding products are grouped into the same product group.

[0140] Step S1910: Sort the products in each product group according to the exposure effectiveness score, and select a plurality of top-ranked products from each product group to form corresponding high-conversion product sets;

[0141] Since the exposure effectiveness score can quantitatively represent the effectiveness that a product can achieve after exposure, the higher the exposure effectiveness score, the higher the effectiveness achieved; the lower the exposure effectiveness score, the lower the effectiveness achieved. Accordingly, for a single product group, the multiple products ranked high in the product group are products with higher effectiveness after exposure, so these products constitute a high conversion product set.

[0142] Step S1920: for each high-conversion product set, obtain the product titles of all products in the high-conversion product set, extract the corresponding high-conversion keyword set, and associate the high-conversion keyword set with the target multi-layer tag set corresponding to the high-conversion product set.

[0143] For a single high-conversion product set, the product titles of all products in the high-conversion product set are segmented to obtain multiple corresponding segmented texts, each of which includes each word in the corresponding product title. The TF-IDF algorithm is used to process the multiple segmented texts to determine the key score of the word among all the different word elements contained in these segmented texts. From these different word elements, these word elements are sorted in order from high to low according to the key score, and the top N word elements are selected as high-conversion keywords respectively. All high-conversion keywords constitute a high-conversion keyword set, and the high-conversion keyword set is associated with the target multi-layer label set corresponding to the high-conversion product set.

[0144] The key score is the product of the TF score and the IDF score of the word-gram. The TF score of the word-gram can be obtained by calculating the number of occurrences of the word-gram in the segmented text in which it is located and dividing it by the sum of the number of occurrences of each word-gram in the segmented text. The IDF score of the word-gram can be obtained by calculating the total number of corresponding word-gram texts of all segmented texts and dividing it by the total number of corresponding word-gram texts containing the word-gram, and then dividing the obtained quotient by the logarithm with the base 10.

[0145] In this embodiment, by utilizing the target multi-layer tag set to construct a hierarchical and detailed semantic representation for each product, combined with the exposure effectiveness score, the score comprehensively considers the click-through rate and conversion rate of the product, and can flexibly reflect the actual performance of different products in the market through weight adjustment. Products with the same characteristics and good performance can be classified into the same product group, and products with high exposure effectiveness scores can be screened out to form a high-conversion product set. Furthermore, the product titles in the high-conversion product set are segmented and processed by the TF-IDF algorithm to extract a high-conversion keyword set. These keywords not only reflect the core characteristics of the product, but also indicate the market's response to these characteristics. Associating these keywords with the target multi-layer tag set not only enhances the semantic richness of the product description, but also provides strong data support for the precision marketing, recommendation, and title optimization of the product.

[0146] See also Figure 3, a commodity labeling device provided to meet one of the purposes of the present application is a functional embodiment of the commodity labeling method of the present application. On the other hand, the device is a commodity labeling device provided to meet one of the purposes of the present application, including a label representation module 1100, a classification set acquisition module 1200, an instance construction module 1300, a label classification module 1400 and a label set construction module 1500, wherein the label representation module 1100 is used to perform multi-level classification representation on the target commodity title, and obtain the predicted implicit representation corresponding to each layer in the label path corresponding to the hierarchical label system mapped to the target commodity title; the classification set acquisition module 1200 is used to determine multiple instance implicit representations matching the predicted implicit representation from the classification instance library, and obtain the instance demonstration classification set corresponding to each instance implicit representation, including the corresponding instance commodity title, the instance multi-layer label set and the instance candidate label set corresponding to each layer, the instance multi-layer label set includes the classification labels of each layer to which the instance commodity title belongs, and the instance candidate label set includes the classification label of the parent layer in the instance multi-layer label set according to the classification label of the corresponding layer, and obtains the parent layer. The classification labels of the layer are the classification labels of the sublayers in the hierarchical label system; the instance construction module 1300 is used to construct corresponding multiple classification instances corresponding to each layer in the label path, and the classification instance includes the instance product title in the corresponding instance demonstration classification set and the instance candidate label set of the layer, as well as the classification labels corresponding to the parent layer and the same layer of the layer in the instance multi-layer label set; the label classification module 1400 is used to traverse each layer in the label path to which the target product title belongs according to the parent-child hierarchical relationship, obtain the classification label of the parent layer of the layer in the hierarchical label system, and the candidate label set of the same layer composed of each classification label of the same layer, embed the target product title, the classification label of the parent layer, the candidate label set of the same layer, and the multiple classification instances corresponding to the layer into a preset prompt template, obtain the corresponding prompt text input large language model, so as to control the model to determine the target classification label of the layer from the candidate label set of the same layer according to the given instructions and with reference to the multiple classification instances; the label set construction module 1500 is used to obtain the target classification labels of the layers output by the large language model to form the target multi-layer label set of the target product title.

[0147] In a further embodiment, after the tag set construction module 1500, it includes: a set acquisition submodule, which is used to obtain the user's historical behavior product set, and the historical behavior product set includes at least one data pair, and the data pair includes a target multi-layer tag set of a corresponding historical behavior description text and its associated product; a tag collection submodule, which is used to, for each data pair in the historical behavior product set, classify the target multi-layer tag set in the data pair into the stage score set of the behavior stage to which its historical user behavior belongs; a score determination submodule, which is used to, for different classification labels in each stage score set, determine the stage score corresponding to the classification label in the stage score set according to the repetition frequency of the classification label in the stage score set and the preset stage weight of the stage score set; a score acquisition submodule, which is used to, for each different classification label, sum the stage scores corresponding to the classification label in each stage score set to obtain the user preference score of the classification label; a portrait construction submodule, which is used to screen out classification labels whose user preference scores meet preset conditions and construct a user preference label portrait.

[0148] In a further embodiment, after the portrait construction submodule, it includes: a first user group construction submodule, which is used to obtain the user preference label portrait of each user, and determine the user preference label portraits that contain the same first preset number of classification labels and belong to each user corresponding to the active user, to constitute an active user group; a first set determination submodule, which is used to determine whether there is a parent-child hierarchical relationship between each of the classification labels when the number of users in the active user group exceeds a preset threshold, and determine the corresponding multi-level label set according to the classification labels with a hierarchical relationship; a first product screening submodule, which is used to obtain the exposure times of multiple products in the multi-level label sets, and screen out products whose exposure times meet preset conditions as first marketing products; a first marketing submodule, which is used to obtain the first marketing information of each of the first marketing products, and send it to users in the active user group.

[0149] In a further embodiment, after the portrait construction submodule, it includes: a second user group construction submodule, which is used to obtain the user preference label portrait of each user, determine the user preference label portrait containing the same second preset number of classification labels and the users corresponding to the high-value users, and constitute a high-value user group; a second set determination submodule, which is used to determine whether there is a parent-child hierarchical relationship between each of the classification labels when the number of users in the high-value user group exceeds a preset threshold, and determine the corresponding multi-level label set according to the classification labels with a hierarchical relationship; a second product screening submodule, which is used to obtain the exposure effectiveness scores of the products in multiple multi-level label sets, and screen out the products whose exposure effectiveness scores meet the preset conditions as the second marketing products; a second marketing submodule, which is used to obtain the second marketing information of each of the second marketing products, and send it to the users in the high-value user group.

[0150] In a further embodiment, after the tag set construction module 1500, it includes: a text acquisition submodule, which is used to acquire the search text provided by the user, and a plurality of description text sets of the products to be searched, wherein the description text sets include the target multi-layer tag set of the products to be searched; a first matching submodule, which is used to determine whether the search text contains SKU description text, and when the search text does not contain SKU description text, determine the first semantic matching score between the underlying nested semantic representation of each of the target multi-layer tag sets and the search semantic representation of the search text; a first result sending submodule, which is used to screen out the underlying nested semantic representation whose first semantic matching score meets the preset conditions, and obtain the default product image corresponding to the product to be searched. The search display text in the picture and description text set is sent to the user; a second matching submodule is used to determine the second semantic matching score between the underlying nested semantic representation of each of the target multi-layer tag sets and the search semantic representation of the search text without the SKU description text when the search text contains the SKU description text, and determine whether each description text set contains the SKU description text; a second result sending submodule is used to screen out the underlying nested semantic representation whose second semantic matching score meets the preset conditions and the product to be searched corresponding to the SKU description text, obtain the product picture of the product to be searched that matches the SKU description text, and the search display text in the description text set, and send them to the user.

[0151] In a further embodiment, after the tag set construction module 1500, it includes: a tag set acquisition submodule, which is used to acquire the target multi-layer tag set of the specified product and the target multi-layer tag sets of multiple products to be recommended; a matching degree determination submodule, which is used to determine the matching degree between the target multi-layer tag set of each product to be recommended and the text encoding representation corresponding to the target multi-layer tag set of the specified product; and a recommendation determination submodule, which is used to screen out the products to be recommended whose matching degrees meet the preset conditions as the recommended products of the specified product.

[0152] In a further embodiment, after the label set construction module 1500, it includes: a product grouping submodule, which is used to obtain target multi-layer label sets and exposure effectiveness scores for multiple products, and group the products of the same target multi-layer label set into the same product group; a conversion screening submodule, which is used to sort the products in each product group according to the exposure effectiveness score, and screen out multiple top-ranked products from each product group to form corresponding high-conversion product sets; a set association submodule, which is used to obtain the product titles of all products in each high-conversion product set, extract the corresponding high-conversion keyword set, and associate the high-conversion keyword set with the target multi-layer label set corresponding to the high-conversion product set.

[0153] In order to solve the above technical problems, the present application also provides an electronic device. Figure 4 As shown, a schematic diagram of the internal structure of an electronic device. The electronic device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the electronic device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement the product labeling method. The processor of the electronic device is used to provide computing and control capabilities to support the operation of the entire electronic device. The memory of the electronic device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the product labeling method described in this application. The network interface of the electronic device is used to connect and communicate with a terminal. Those skilled in the art will understand that Figure 4 The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0154] In this execution mode, the processor is used to execute Figure 3The memory stores the program code and various data required to execute the above modules or modules. The network interface is used to transmit data between user terminals or servers. The memory in this execution mode stores the program code and data required to execute all modules / modules in the commodity marking device of this application, and the server can call the program code and data of the server to execute the functions of all modules.

[0155] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the method of any embodiment of the present application.

[0156] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0157] To sum up, this application can efficiently label products with accurate hierarchical labels, and can apply the hierarchical labels of products and related data in the e-commerce platform in many aspects to plan the operating scenarios of the e-commerce platform, thereby improving the operating results of the e-commerce platform.

[0158] It will be understood by those skilled in the art that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, changed, combined, or deleted. Furthermore, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted. Furthermore, the steps, measures, and schemes in the various operations, methods, and processes in the prior art that are open source and in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0159] The above is only a partial implementation of the present application. It should be pointed out that ordinary technicians in this technical field can make several improvements and modifications without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A commodity labeling method, characterized in that: The steps include: Perform multi-level classification representation on the target product title to obtain the predicted implicit representation corresponding to each layer in the label path corresponding to the hierarchical label system in which the target product title is mapped; From the classification instance library, multiple instance implicit representations matching the predicted implicit representation are determined, and instance demonstration classification sets corresponding to the respective instance implicit representations are obtained, including the corresponding instance product titles, instance multi-layer label sets and instance candidate label sets corresponding to each layer thereof, wherein the instance multi-layer label set includes the classification labels of each layer to which the instance product title belongs, and the instance candidate label set includes the classification labels of the parent layer in the instance multi-layer label set according to the classification labels of the corresponding layer, and each classification label of the sublayer of the classification label of the parent layer in the hierarchical label system is obtained; Corresponding to each layer in the label path, constructing a corresponding plurality of classification instances, wherein the classification instance includes the instance product title in the corresponding instance demonstration classification set and the instance candidate label set of the layer, and the classification labels corresponding to the parent layer and the same layer of the layer in the instance multi-layer label set; Traversing each layer in the label path to which the target product title belongs according to the parent-child hierarchical relationship, obtaining the classification label of the parent layer of the layer and the candidate label set of the same layer composed of the classification labels of the same layer in the hierarchical label system, embedding the target product title, the classification label of the parent layer, the candidate label set of the same layer, and the multiple classification instances corresponding to the layer into a preset prompt template, obtaining the corresponding prompt text input into the large language model, so as to control the model to determine the target classification label to which the layer belongs from the candidate label set of the same layer according to the given instructions and with reference to the multiple classification instances; The target classification labels of each layer output by the large language model are obtained to form a target multi-layer label set of the target product title.

2. The commodity marking method according to claim 1, characterized in that: After obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included: Acquire a user's historical behavior commodity set, wherein the historical behavior commodity set includes at least one data pair, wherein the data pair includes a corresponding historical behavior description text and a target multi-layer label set of its associated commodity; For each data pair in the historical behavior product set, the target multi-layer label set in the data pair is classified into the stage score set of the behavior stage to which the historical user behavior belongs; For different classification labels in each stage scoring set, determine the stage score corresponding to the classification label in the stage scoring set according to the repetition frequency of the classification label in the stage scoring set and the preset stage weight of the stage scoring set; For each different classification label, sum the stage scores corresponding to the classification label in each stage score set to obtain the user preference score of the classification label; Filter out the classification labels whose user preference scores meet the preset conditions and construct a user preference label portrait.

3. The commodity marking method according to claim 2, characterized in that: After filtering out the classification labels whose user preference scores meet the preset conditions and constructing the user preference label portrait, the following steps are included: Obtain a user preference label portrait of each user, determine the user preference label portraits that contain the same first preset number of classification labels and are corresponding to active users, and form an active user group; When the number of users in the active user group exceeds a preset threshold, determining whether there is a parent-child hierarchical relationship between the classification tags, and determining a corresponding multi-level tag set according to the classification tags with a hierarchical relationship; Obtaining exposure times of multiple commodities in the multi-level tag set, and selecting commodities whose exposure times meet preset conditions as first marketing commodities; The first marketing information of each of the first marketing products is obtained and sent to users in the active user group.

4. The commodity marking method according to claim 2, characterized in that: After filtering out the classification labels whose user preference scores meet the preset conditions and constructing the user preference label portrait, the following steps are included: Obtaining a user preference label portrait of each user, determining the user preference label portraits that contain the same second preset number of classification labels and that are high-value users, to form a high-value user group; When the number of users in the high-value user group exceeds a preset threshold, it is determined whether there is a parent-child hierarchical relationship between the classification tags, and a corresponding multi-level tag set is determined according to the classification tags with a hierarchical relationship; Obtaining exposure effectiveness scores of the products in the plurality of multi-level tag sets, and selecting products whose exposure effectiveness scores meet preset conditions as second marketing products; The second marketing information of each of the second marketing products is obtained and sent to users in the high-value user group.

5. The commodity marking method according to claim 1, characterized in that: After obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included: Acquire a search text provided by a user and a plurality of description text sets of products to be searched, wherein the description text sets include a target multi-layer tag set of the products to be searched; Determine whether the search text contains SKU description text, and when the search text does not contain SKU description text, determine a first semantic matching score between the underlying nested semantic representation of each target multi-layer tag set and the search semantic representation of the search text; Filter out the underlying nested semantic representations whose first semantic matching scores meet preset conditions, obtain the search display text from the default product image and description text set corresponding to the product to be searched, and send them to the user; When the search text contains SKU description text, determining a second semantic matching score between the underlying nested semantic representation of each of the target multi-layer tag sets and the search semantic representation of the search text without the SKU description text, and determining whether each description text set contains the SKU description text; Filter out the underlying nested semantic representations whose second semantic matching scores meet the preset conditions, as well as the products to be searched corresponding to the SKU description text, obtain product images of the products to be searched that match the SKU description text, and the search display text in the description text set, and send them to the user.

6. The commodity marking method according to claim 1, characterized in that: After obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included: Obtain a target multi-layer label set for a specified product and a target multi-layer label set for multiple products to be recommended; Determine the degree of match between the target multi-layer label set of each to-be-recommended product and the text encoding representation corresponding to the target multi-layer label set of the specified product; Filter out the recommended products whose matching degree meets the preset conditions as the recommended products of the designated products.

7. The commodity marking method according to claim 1, characterized in that: After obtaining the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title, the following steps are included: Obtain target multi-layer label sets and exposure effectiveness scores for multiple products, and group products with the same target multi-layer label set into the same product group; Sort the products in each product group according to the exposure effectiveness score, and select a plurality of top-ranked products from each product group to form corresponding high-conversion product sets; For each high conversion product set, obtain the product titles of all products in the high conversion product set, extract the corresponding high conversion keyword set, and associate the high conversion keyword set with the target multi-layer tag set corresponding to the high conversion product set.

8. A commodity marking device, characterized in that: It includes: The label representation module is used to perform multi-level classification representation on the target product title, and obtain the predicted implicit representation corresponding to each layer in the label path corresponding to the hierarchical label system in which the target product title is mapped; A classification set acquisition module is used to determine multiple instance implicit representations that match the predicted implicit representation from the classification instance library, and obtain the instance demonstration classification set corresponding to each instance implicit representation, including the corresponding instance product title, the instance multi-layer label set and the instance candidate label set corresponding to each layer, the instance multi-layer label set includes the classification labels of each layer to which the instance product title belongs, and the instance candidate label set includes the classification label of the parent layer in the instance multi-layer label set according to the classification label of the corresponding layer, and obtains each classification label of the sublayer of the classification label of the parent layer in the hierarchical label system; An instance construction module is used to construct a corresponding plurality of classification instances corresponding to each layer in the label path, wherein the classification instance includes the instance commodity title in the corresponding instance demonstration classification set and the instance candidate label set of the layer, and the classification labels corresponding to the parent layer and the same layer of the layer in the instance multi-layer label set; The label classification module is used to traverse each layer in the label path to which the target product title belongs according to the parent-child hierarchical relationship, obtain the classification label of the parent layer of the layer in the hierarchical label system, and the candidate label set of the same layer composed of the classification labels of the same layer, embed the target product title, the classification label of the parent layer, the candidate label set of the same layer, and the multiple classification instances corresponding to the layer into a preset prompt template, obtain the corresponding prompt text input large language model, so as to control the model to determine the target classification label belonging to the layer from the candidate label set of the same layer according to the given instructions and with reference to the multiple classification instances; The label set construction module is used to obtain the target classification labels of each layer output by the large language model to form a target multi-layer label set of the target product title.

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

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

Images